Volume 1439, Issue 1 p. 30-70
ORIGINAL ARTICLE
Free Access

New York City Panel on Climate Change 2019 Report Chapter 2: New Methods for Assessing Extreme Temperatures, Heavy Downpours, and Drought

Jorge E. González

Jorge E. González

Mechanical Engineering Department, NOAA-CREST Center, The City College of New York, New York, New York

Search for more papers by this author
Luis Ortiz

Luis Ortiz

Mechanical Engineering Department, NOAA-CREST Center, The City College of New York, New York, New York

Search for more papers by this author
Brianne K. Smith

Brianne K. Smith

Department of Earth and Environmental Science, CUNY-Brooklyn College, New York, New York

Search for more papers by this author
Naresh Devineni

Naresh Devineni

Department of Civil Engineering, NOAA-CREST Center, The City College of New York, New York, New York

Search for more papers by this author
Brian Colle

Brian Colle

School of Marine and Atmospheric Sciences, Stony Brook University, Stony Brook, New York

Search for more papers by this author
James F. Booth

James F. Booth

Department of Earth and Atmospheric Sciences, The City College of New York, New York, New York

Search for more papers by this author
Arun Ravindranath

Arun Ravindranath

Department of Civil Engineering, NOAA-CREST Center, The City College of New York, New York, New York

Search for more papers by this author
Lea Rivera

Lea Rivera

Department of Civil Engineering, NOAA-CREST Center, The City College of New York, New York, New York

Search for more papers by this author
Radley Horton

Radley Horton

Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York

Search for more papers by this author
Katie Towey

Katie Towey

Department of Earth and Atmospheric Sciences, The City College of New York, New York, New York

Search for more papers by this author
Yochanan Kushnir

Yochanan Kushnir

Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York

Search for more papers by this author
Danielle Manley

Danielle Manley

Center for Climate Systems Research, Columbia University, New York, New York

Search for more papers by this author
Daniel Bader

Daniel Bader

Center for Climate Systems Research, Columbia University, New York, New York

Search for more papers by this author
Cynthia Rosenzweig

Cynthia Rosenzweig

NASA Goddard Institute for Space Studies, New York, New York

Search for more papers by this author
First published: 15 March 2019
Citations: 20

[Correction added on June 12, 2019, after first online publication: In the author list, the eighth author's name was changed from “Luis Rivera” to “Lea Rivera.”]

Contents

  • 2.1
  • Introduction
  • 2.2
  • NPCC3 approach
  • 2.3
  • Extreme temperatures and humidity
  • 2.4
  • Heavy downpours and inland flooding
  • 2.5
  • Droughts
  • 2.6
  • Conclusions and recommendations
  • 2.1 Introduction

    This New York City Panel on Climate Change (NPCC3) chapter builds on the projections developed by the second New York City Panel on Climate Change (NPCC2) (Horton et al., 2015). It confirms NPCC2 projections as those of record for the City of New York, presents new methodology related to climate extremes, and describes new methods for developing the next generation of climate projections for the New York metropolitan region. These may be used by the City of New York as it continues to develop flexible adaptation pathways to cope with climate change. The main topics of the climate science chapter are:
    • (1) Comparison of observed temperature and precipitation trends to NPCC2 2015 projections.
    • (2) New methodology for analysis of historical and future projections of heatwaves, humidity, and cold snaps.
    • (3) Improved characterization of observed heavy downpours.
    • (4) Characterization of observed drought using paleoclimate data.
    • (5) Suggested methods for next-generation climate risk information.

    The focus of NPCC3 is on high-risk events involving extreme temperatures, extreme precipitation, and drought. Current trends are presented using historical climate records of high temperature, cold snaps, humidity, and extreme precipitation for the New York metropolitan region. The geographical span of the New York metropolitan region considered here includes, in addition to New York City, adjacent sections of New Jersey such as Newark, Jersey City and Elizabeth, as well as other nearby locations in New York such as Yonkers and Long Island. Historical records of droughts in the Delaware watershed region are also examined. Each climate extreme is analyzed for detection of current trends, and future projections are updated for high-temperature extremes as a test of new methods that could be utilized by NPCC4.

    These represent finer temporal and spatial resolutions that may be of practical use to key stakeholders in New York City for planning purposes and/or emergency responses. They include local projections of extreme heat and demonstrate the role of the heterogeneous landscape of the city in each process (e.g., how the urban heat island (UHI) affects city neighborhoods differently). Each section of the chapter presents definitions, baselines, methods, and projections, along with uncertainties and recommendations for future work.

    2.2 NPCC3 approach

    As in NPCC2, NPCC3 makes use of definitions, measurements, baselines, and scenarios to represent how the probabilities of climate events may change in the future. Here, the focus is on extreme events. For most climate hazards, the definitions of extremes are consistent with the NPCC2, specifically for extreme heat, cold spells, and precipitation.

    NPCC3 confirms the temperature and precipitation projections of NPCC2 as those of record for use in planning. Based on emerging science, NPCC3 introduces a new methodology for analyzing heat and precipitation extremes that could be used for developing future projections of record in NPCC4.

    In NPCC2, temperature analyses included projections of average temperature changes and changes in heat waves and hot days. NPCC3 explores new methodologies for downscaling heat extremes and introduces new metrics to analyze historical and projected humidity. For precipitation, NPCC2 developed quantitative projections for average rainfall and daily maximum rainfall events, and NPCC3 introduces a methodology for quantifying projections for sub-daily heavy downpour rain events.

    In addition, NPCC3 examined how current observations of temperature and precipitation changes compare to projected changes from NPCC2 into the 2020s time slice, which encompassed the time period from 2010 to 2039. Figure 2.1 shows the results of this analysis and demonstrates that observations from 2010 to 2017—the period for which both observed data and NPCC2 projections are available to compare—have been largely consistent with projected changes in average conditions for both temperature (Fig. 2.1a) and precipitation (Fig. 2.1b). However, these comparisons should be viewed with caution because of the role that natural variation plays in the short term.

    Details are in the caption following the image
    Observations at Central Park (1900–2017) compared to the 2020s (2010–2039) time slice of NPCC2 projected changes for (a) average annual temperature and (b) average annual precipitation. Colored lines represent the 10th, 25th, 75th, and 90th percentiles of model projections across RCPs 4.5 and 8.5 for 35 GCMs. Shading shows the central range of projections between the 25th and 75th percentiles. Vertical dotted lines represent the range of the 2020s time slice from 2010 to 2039. Observed data are from the United States Historical Climatology Network (USHCN), and climate projections are from the Coupled Model Intercomparison Project Phase 5 (CMIP5). Note: These comparisons should be viewed with caution because of the role that natural variation plays in the short term.

    As NPCC3 shifts from a focus on average conditions to extremes, the baselines in some cases vary according to the relevance of the period for the extreme event researched and the period for which data are available. To the extent possible, consistency with NPCC2 is maintained. For example, the baseline for heat waves is 1971–2000, which is the same as NPCC2. However, NPCC3 uses summer months for extreme heat events (June, July, and August) for three reference weather stations, while NPCC2 used the whole year with one reference weather station.

    NPCC3 uses bias-corrected statistical downscaling and develops future projections for extreme heat based on summer seasons only and includes high-resolution dynamical downscaling at 1 km for selected time slices. Summer humidity is included in the projections as a new heat-related variable. The section of extreme temperatures closes with a short view of cold spells and winter extremes.

    The section on urban flooding makes use of shorter, more detailed records of satellite and radar data to demonstrate the spatial distribution of these extreme events at sub-hourly time resolution.

    For droughts, a much longer precipitation record based on tree rings is used to capture decadal variations in the New York City watershed region, and reconstructions of inflows to reservoirs are used to understand how frequently extreme droughts have occurred in the past.

    To create the new extreme event projections, bias-corrected statistical downscaling is used (see Section 2.3). In the Appendix, we provide an example of dynamic downscaling, a method that can capture the role of the urban built environment in magnifying heat events and mitigating flooding events.

    Model outputs from the fifth phase of the Coupled Model Intercomparison Project (CMIP5; Taylor et al., 2012) are used for projections of extreme heat. Methods for calculating future projections are consistent with NPCC2 but are updated to account for climate model biases in simulating the distribution of temperature (National Climate Assessment; Walsh et al., 2014). Results are provided in 30-year intervals centered on the 2020s, 2050s, and 2080s as defined by NPCC2. The ensemble of CMIP5 results includes two representative concentration pathways (RCPs) (see Box 2.1).

    2.3. Extreme temperature and humidity

    Summer (defined as the months of June, July, and August) temperatures are expected to increase in New York City throughout the 21st century (Horton et al., 2015), leading to more frequent and intense extreme heat events known as heat waves. Here, we follow the definition of heat waves according to the National Weather Service (NWS), that is, an interval of 3 (or more) consecutive days with temperatures of at least 90 °F (32.22 °C).

    Heat waves affect a wide range of human activities. These effects include increasing energy demand (Schaeffer et al., 2012; Sailor, 2001; Santamouris, 2014) and mortality (Knowlton et al., 2007; Luber and McGeehin, 2008; Anderson and Bell, 2010; Rosenthal et al., 2014). Moreover, higher temperatures associated with urbanization, a phenomenon called the Urban Heat Island (UHI) (Oke, 1982), exacerbate the impacts of extreme heat events (Li and Bou-Zeid, 2013; Ramamurthy and Bou-Zeid, 2016; Ramamurthy et al., 2017; Ortiz et al., 2018).

    New York City, being the most populated urban area in the United States with over 8 million people (U.S. Census Bureau, 2018), has a large human and economic incentive to understand and mitigate the negative impacts of these events now and in the future.

    Extreme heat projections have primarily been developed on global (Meehl and Tebaldi, 2004) or continental scales (Gao et al., 2012), with less work focusing on local urban projections that require accounting for finer-scale processes and feedbacks that may affect the occurrence and characteristics of high-temperature events. An example of these processes is the soil moisture-heat wave feedback, wherein dry soil conditions may amplify heat waves by reducing available moisture for evaporative cooling (Seneviratne et al., 2006; Lorenz et al., 2010; Fischer et al., 2007). Cities may amplify these feedbacks by reducing exposed soil area, greatly reducing the capacity for water retention near the land surface (Li and Bou-Zeid, 2013; Ramamurthy and Bou-Zeid, 2016; Ramamurthy et al., 2017).

    Other relevant city-scale processes include waste heat from buildings and transportation (Taha, 1997; Ichinose et al., 1999; Offerle et al., 2005), lower surface reflectivity of built surfaces (Taha et al., 1988; Morini et al., 2016; Ramamurthy et al., 2015) and increased heat storage in buildings and built structures (Oke et al., 1981; Arnfield and Grimmond, 1998).

    Humidity content of the atmosphere can play an adverse role in how humans react to high heat conditions (Davis et al., 2016; Hass et al., 2016). As air becomes more saturated with water vapor, the human body becomes less able to shed excess heat through evaporative cooling of perspiration. This can lead to exacerbation of high-temperature impacts such as fatigue and heat exhaustion.

    This section presents extreme heat and specific humidity projections for New York City using new methods, accounting where possible for urban effects via statistical processing of global climate model (GCM) simulation data. This statistical processing, or downscaling, is necessary because global models have, in general, very coarse spatial resolution (>100 km2) and are thus not able to resolve coastlines, topography, and land cover.

    The downscaling technique used by NPCC3 is histogram matching. It aims to adjust the model representations of observed climate by correcting their mean and variance to match a representative set of observations in the target domain (see Appendix 2.B). This differs from the bias adjustment procedure of NPCC2 that combined GCM results with station records to downscale the projections to the New York metropolitan region using the “delta method” (Horton et al., 2015), where mean monthly projected changes are applied to daily observations.

    In NPCC3, as in NPCC2, the climate projections are based on multiple climate models, driven by two RCPs—RCP4.5 (referred to as medium emissions) and RCP8.5 (referred to as high emissions) (see Box 2.1). The aim of this approach is to capture the uncertainties emerging from the range of model results as well as those related to the impacts of future industrial activity, energy use, and technology on greenhouse gases (GHGs), aerosol emissions, and land use change. For consistency, NPCC3 uses the same baseline period as the NPCC2 (1971–2000). Definitions and methods are detailed in Table 2.1.

    3 Box 2.1. Definitions and terms

    Climate change

    Climate change refers to a significant change in the state of the climate that can be identified from changes in the average state or the variability of weather and that persists for an extended time period, typically decades to centuries or longer. Climate change can refer to the effects of (1) persistent anthropogenic or human-caused changes in the composition of the atmosphere and/or land use, or (2) natural processes such as volcanic eruptions and Earth's orbital variations (IPCC, 2013).

    Global climate models (GCMs)

    A GCM is a mathematical representation of the behavior of the Earth's climate system over time that can be used to estimate its sensitivity to atmospheric concentrations of greenhouse gases (GHGs), aerosols, and land use change. Each model simulates physical exchanges among the ocean, atmosphere, land, and ice.

    Representative concentration pathways (RCPs)

    RCPs are sets of trajectories of concentrations of GHGs, aerosols, and land-use changes developed for climate models as a basis for long-term and near-term climate-modeling experiments (Moss et al., 2010). RCPs describe different climate futures based on different amounts of climate forcings. These data are used as inputs to GCMs to project the effects of these drivers on future climate. The NPCC uses sets of GCM simulations driven by two RCPs, known as 4.5 and 8.5. The set of GCM simulations driven by RCP 4.5 is defined here as a medium-emissions scenario, and that by RCP 8.5 as a high-emissions scenario.

    Climate change risk information

    On the basis of the selection of the RCPs and GCM simulations, local climate change information is developed for key climate variables—temperature, precipitation, and associated extreme events. These results and projections reflect a range of potential outcomes for the New York metropolitan region.

    Climate hazard

    A climate hazard is a weather or climate state such as a heat wave, flood, high wind, heavy rain, ice, snow, or drought that can cause harm and damage to people, property, infrastructure, land, and ecosystems. Climate hazards can be expressed in quantified measures, such as flood height in feet, wind speed in miles per hour, and inches of rain, ice, or snowfall that are reached or exceeded in a given period of time.

    Uncertainty

    Uncertainty denotes a state of incomplete knowledge that results from lack of information, natural variability in the measured phenomenon, instrumental and modeling errors, and/or from disagreement about what is known or knowable (IPCC, 2013).

    Table 2.1. Methods for heat extremes used in NPCC2 and NPCC3
    Methods NPCC2 (2015) NPCC3 (2018)
    Definition of heat wave
    • Three or more consecutive days at or above 90 °F
    • Three or more consecutive days at or above 90 °F
    Metrics
    • Number of heat waves per year
    • Duration of heat wave (number of days average heat wave lasts)
    • Total days at or above 90 °F per year
    • Total days at or above 100 °F per year
    • Number of heat waves per year
    • Duration of heat wave (number of days average heat wave lasts)
    • Average maximum temperature during heat wave (heat wave intensity)
    • Total days above 90 °F per year
    • Total days above 100 °F per year
    • Annual mean specific humidity
    Baseline years 1971–2000 1971–2000
    Baseline reference locations
    • Central Park
    • Central Park (extreme heat)
    • LaGuardia Airport (extreme heat, humidity)
    • JFK Airport (extreme heat)
    Baseline values
    • 2 heat waves per year
    • 4 days in duration
    • 18 days at or above 90 °F
    • 0.4 days at or above 100 °F
    • 1.1 heat waves per year
    • 4 days in duration
    • 94.2 °F maximum temperature
    • 9.6 days above 90 °F
    • 0.27 days above 100 °F
    • 0.0123 kgvapor/kgair mean specific humidity
    Future time slices
    • 2020s (2010–2039)
    • 2050s (2040–2069)
    • 2080s (2070–2099)
    • 2020s (2010–2039)
    • 2050s (2040–2069)
    • 2080s (2070–2099)
    Methodology
    • 2 RCPs (4.5 and 8.5)
    • 35 GCMs
    • 10th, 25th, 75th, and 90th percentilesc across both RCPs and 35 GCM outputs
    • Delta method used for GCM bias correction
    • 2 RCPs (4.5 and 8.5)
    • 26a GCMs
    • 10th, 25th, 75th, and 90th percentiles across both RCPs and 26 GCM outputs
    • GCM mean and variance bias correction to temperature station recordsb.
    • a NPCC3 uses a smaller GCM ensemble than the 35-model ensemble used in NPCC2 because not all GCMs provide specific humidity as a standard output.
    • b Bias correction is performed on GCM data in order to improve how representative a single grid box is to local New York City conditions.
    • c The Xth percentile is defined as the value that X percent of the outcomes are the same or lower than.

    2.3.1 Observed trends in summer heat waves

    Historical trends of daily maximum summer temperature in New York were analyzed using Central Park weather station, John F. Kennedy (JFK), and LaGuardia Airports during June, July, and August (Fig. 2.2). Central Park has the longest historical record, dating back to 1900, where the average annual daily maximum summer temperature has been rising at an average of 0.2 °F per decade from 1900 to 2013. JFK and LaGuardia weather stations go back to 1970, where average annual daily maximum summer temperatures have been increasing at a rate of 0.5 °F per decade and 0.7 °F per decade, respectively.

    Details are in the caption following the image
    Annual average daily maximum summer temperatures. (June, July, August) in Central Park from 1900 to 2013, LaGuardia (LGA) airport from 1970 to 2013, and John F. Kennedy (JFK) airport from 1970 to 2013. Solid lines represent linear trends for each station. Station records were obtained from the U.S. Historical Climatology Network (USHCN) Version 2.5 (Menne et al., 2013). *Central Park trend is significant at 0.01 level, while LGA and JFK trends are positive but not significant, possibly due to shorter record length.

    The distance between these weather stations provide insights into processes that affect temperatures near the surface, such as sea breezes1 and the UHI. Sea breeze effects appear in stations located close to Long Island's southern shore (e.g., JFK), with lower daily maximum temperatures compared to their in-land counterparts (Fig. 2.2).

    Sea breeze impacts on temperatures show that geospatial heterogeneity of the urban landscape plays a role in near-surface temperatures, and therefore impact occurrences of extreme heat. The weather station located at JFK, which experiences afternoon sea breezes, has a mean summer maximum temperature of 80.6 °F, whereas the other stations have a mean value of 82.7 °F, which is 2.1 °F higher. This is consistent with climatological studies (e.g., Gedzelman et al., 2003) of the UHI in the region, which have found that afternoon summer sea breezes may shift the center of the urban heat island west and north, toward New Jersey and The Bronx.

    2.3.2 New methods for projected changes in heat waves

    This section describes new methods and results for NPCC3 projected changes in heat waves. Projections are presented in two formats:
    • Graphical time series: Results shown as a time series are further broken out into the two RCP scenarios used in NPCC3 (see Fig. 2.3). These include a medium-emissions scenario (RCP4.5) (Thomson et al., 2011) and a high-emissions scenario (RCP8.5) (Riahi et al., 2011).
    • Time slices table: Results are summarized for the 10th, 25th, 75th, and 90th percentile of climate model outcomes across both RCP 4.5 and 8.5 scenarios in Table 2.2, averaged by 30-year time slices centered around the 2020s (2010–2039), 2050s (2040–2069), and 2080s (2070–2099) decades.
    Details are in the caption following the image
    Results from new bias-corrected projections for future assessments of (a) mean daily maximum temperature, (b) event frequency, (c) mean event duration, and (d) mean event intensity compared to the 1971–2000 base period. Solid lines represent the multimodel mean of a 26 global climate model ensemble, while shaded bands show 95% confidence intervals. Black lines indicate observations from three GHCN stations—Central Park, JFK, and LaGuardia—between 1971 and 2000.
    Table 2.2. Results from new projection methods for future assessments of heat wave across 52-member ensemble (26 models, RCPs 4.5 and 8.5) for New York City
    10th percentile 25th percentile 75th percentile 90th percentile
    Heat waves per year (average number 2020s 1 2 4 5
    of events per year) 2050s 2 3 5 6
    Baselinea 1.1 2080s 2 3 5 7
    Mean heat wave duration (average 2020s 3 4 6 8
    heat wave length in days) 2050s 4 5 9 13
    Baseline 4 2080s 4 6 15 27
    Mean heat wave intensity (average 2020s 91.8 92.5 94.5 95.7
    maximum temperature during 2050s 92.6 93.5 95.4 96.5
    heat wave event in °F) 2080s 93.2 94.2 97.1 99.1
    Baseline 94.2 °F
    Days above 90 °F (average number 2020s 6 11 25 34
    of days per year) 2050s 15 24 46 56
    Baseline 9.6 2080s 24 35 63 75
    Days above 100 °F (average number 2020s 0 0 0 2
    of days per year) 2050s 0 0 4 8
    Baseline 0.27 2080s 0 1 13 27
    • a Baseline refers to 1971–2000 average characteristics for Central Park, LaGuardia, and JFK.Note: NPCC3 confirms the temperature projections of NPCC2 as those of record that should be used for planning.

    2.3.2.1 Heat wave frequency, duration, and intensity

    Heat wave characteristics considered here are their frequency (events/year), mean event duration (number of days/event), and intensity (average maximum temperature/heat wave). NPCC2 had previously analyzed frequency and mean event duration; heat wave intensity is a new metric in NPCC3. While a new methodology is tested here that is different from NPCC2, NPCC3 confirms the use of NPCC2 projections as the projections of record for New York City to plan for extreme heat. The new methodologies presented in NPCC3 could be used in developing new projections of record in NPCC4.

    Using a composite observed temperature record derived by averaging the daily maximum temperature over the three New York City stations, results from 26 GCMs were bias corrected in order to project distributions of heat waves for the NPCC3 time slices following the methods of Piani et al. (2010) and Hawkins et al. (2013). (See Appendix 2.C for detailed methods.) The mean and standard deviation of a given variable were used to adjust the model distribution against the target observed distribution. For each GCM, the closest land grid point was selected, as was done in NPCC2, and the distribution of maximum daily temperature at this point was bias corrected against the city's composite maximum temperatures. This method is referred to as a “single-point” bias correction.

    The previous NPCC2 approach may have resulted in a bias toward slightly cooler projected extreme temperatures compared to those projected using the NPCC3 bias-correction methods, particularly toward the warmer periods in the 2080s time slice. These changes may be due to the correction to the variance that, at least partially, addresses the fact that GCM grid boxes near coasts may include water.

    NPCC3 analysis of the bias-corrected single-point projections shows overall increase across all heat wave metrics throughout the 21st century (Fig. 2.3). To highlight the sensitivity to emission scenarios, we present the response to medium-emission and high-emission scenarios separately. In Table 2.2, the projections are based on the distribution of multimodel results showing the 10th, 25th, 75th, and 90th percentile outcomes across both RCP scenarios, as was done in NPCC2.

    Mean daily maximum temperature (Fig. 2.3a) shows a nearly linear trend in the high-emissions scenario (RCP8.5), whereas the rate of change in the medium-emissions scenario (RCP4.5) slows after 2040.

    The number of heat waves per year (Fig. 2.3b) shows far less deviation between the two emissions scenarios. Both scenarios increase at a pace of about one additional yearly event every 20 years until 2060, where growth slows down considerably. This may be due to consecutive events coalescing into very long heat waves, which becomes more likely as heat waves increase in length and frequency. It is also an artifact of the definition of heat wave used, which establishes an unchanging temperature threshold through the entire century. As mean temperatures increase, meeting the 90 °F on consecutive days becomes more likely. Uncertainty in projections as described by confidence intervals increase over time, with a spread of 1 event in the first half of the century that grows to a spread of about two events by end of century.

    Mean event duration projections (Fig. 2.3c) are similar across the scenarios in the first half of the century, growing by around 2 days per 20-year period. However, the high-emissions scenario projections show accelerated growth in the latter half of the century, as well as more spread in the model ensemble, with an uncertainty band spanning about 10 days, compared to about 2 days in the first half. This accelerated increase in event duration may explain the stabilization of event frequency projections in Figure 2.3(b), as events may aggregate into longer heat waves.

    Mean intensity, defined as the mean of event maximum temperatures, shows large interannual variation (Fig. 2.3d), with projected values that increase from about 93 °F early in the century, to 95–98 °F by the end of the century. Confidence interval bands increase slightly throughout by end of century, reaching an ensemble spread of about 1 °F.

    2.3.2.2 New methods for warm day analyses

    Additional key metrics of extreme heat explored are number of days above 90 and 100 °F in the summer season (Table 2.2). Projected days above 90 °F are expected to become more likely as summer temperatures increase. By the 2080s, projections show 24 (10th percentile) to 75 (90th percentile) days above 90 °F compared to the 1971–2000 baseline (10 days).2

    2.3.3 Methods for assessing trends in humidity

    The humidity content of the atmosphere can play an adverse role in how humans react to high heat conditions (Davis et al., 2016; Hass et al., 2016). We present projections of daily mean specific humidity based on a 26 multimodel ensemble, across medium- and high-emissions scenarios, as in Section 2.3.2. For each model, the land grid point closest to New York City is used. Due to a lack of specific humidity records from all weather stations, GCM humidity was bias corrected based on LaGuardia Airport only. In addition, the 1971–2000 baseline for specific humidity is based only on the LaGuardia Airport weather station. Humidity is a new metric being considered by the NPCC3.

    Results show an increase between the 2020s and 2080s time slices of around 9% at each period's 10th percentile, while changes in the 90th percentile represent a 16% increase (Table 2.3). The uncertainty in these projections as characterized by the model ensemble 95% confidence bands (Fig. 2.4) is relatively large. Increases in specific humidity combined with increasing temperatures might lead to higher heat index (see Box 2.2), which has major consequences for human health and is a driver of peak energy demand for space cooling, as air conditioning systems remove sensible (temperature-related) and latent (moisture-related) heat from buildings. To assess combined air moisture and temperature impacts, concurrent hourly values must be used, rather than daily outputs from the model ensemble.

    3.3 Box 2.2. Key humidity definitions

    Specific humidity: A measure of the amount of water in the atmosphere; the mass fraction of water vapor per unit mass of moist air.

    Absolute humidity: Mass of air per unit volume of moist air.

    Relative humidity: The ratio of water vapor pressure to the saturation vapor pressure. It measures how saturated with water vapor the atmosphere is. As air becomes more saturated with water vapor, it becomes more difficult for the human body to shed excess heat through evaporative cooling of perspiration.

    Heat index: A measure of the combined effects of temperature and relative humidity. It is defined by the National Weather Service.

    See Appendix 2.C of this chapter for an expanded discussion of how climate change is projected to impact the heat index. NPCC3 recommends further testing of this methodology for the development of new projections of record in NPCC4.

    Table 2.3. Specific humidity projections across the 52-member ensemble (26 models, RCPs 4.5 and 8.5) for New York City
    Baseline 10th percentile 25th percentile 75th percentile 90th percentile
    Mean specific humidity 2020s 0.0123 0.0115 0.0119 0.0127 0.0131
    (kgvapor/kgair) 2050s 0.0123 0.0122 0.0126 0.0138 0.0146
    2080s 0.0123 0.0125 0.0130 0.0149 0.0164
    • Note: Baseline period for humidity is 1971–2000 at the LaGuardia Airport weather station.
    Details are in the caption following the image
    Specific humidity projections from the bias-corrected 26-member global climate model ensemble and across the medium (RCP4.5) and high (RCP8.5) emissions scenarios compared to the 1971–2000 baseline period at the LaGuardia Airport weather station. Shaded bands represent 95% confidence intervals across the ensemble. Specific humidity refers to the amount of water vapor in the atmosphere and combined with increasing temperatures can lead to a higher heat index.

    2.3.4 Cold snaps

    NPCC3 confirms the analysis of NPCC2 days with minimum temperatures below 32 °F as the projections of record for New York City planning. NPCC3 further examines historical extreme cold events, using two measures of extreme cold (Boyle, 1986; de Vries et al., 2012; Chen et al., 2013; Peterson et al., 2008; Efthymiadis et al., 2011):
    • A day below freezing occurs whenever minimum temperature is equal to or less than 32 °F
    • A cold day occurs whenever its minimum temperature is equal to or less than the 10th percentile of daily minimum temperature of a given year.

    Other definitions vary, including the use of standard deviations (Vavrus et al., 2006). Cold spell changes have been reported on regional scales (e.g., Europe, de Vries et al., 2012; China, Zhang et al., 2017; Northeast United States, Thibeault and Seth, 2014) and for global scales (Vavrus et al., 2006; Konrad, 1996) using GCM ensembles and long-term climate records. In most cases, cold days have shown decreases, and notably in Northern latitudes, it has been found that accelerated decreases of cold spells outpace increases in summer maxima (Thibeault and Seth, 2014).

    We used data from Central Park to establish a benchmark for cold spells. The 10th percentile threshold for cold days at this station was computed from the entire 1900–2017 record, with a value of 24.08 °F. In general, cold days per year decreased by 1.46 days every decade between 1900 and 2017, while days below freezing temperatures decreased at a rate of 1.85 days per decade (Fig. 2.5). This results in recent years having, on average, about 22 fewer days below freezing and 17 fewer cold days than in 1900. The rate of change of these trends is slightly lower than those reported for the entire Northeast by Thibeault and Seth (2014).

    Details are in the caption following the image
    Observed annual number of cold days at Central Park Station (1900–2017). Solid straight lines represent the linear trend in days below freezing temperatures (top) and cold days (bottom). A day below freezing occurs whenever minimum temperature is equal to or less than 32 °F. A cold day occurs whenever its minimum temperature is equal to or less than the 10th percentile of daily minimum temperature of a given year. Station records obtained from the U.S. Historical Climatology Network (USHCN). Both trends are significant at the 0.001 level.

    For the case of New York City, the attribution of these rapid decreases of cold spells may be a combined effect of global warming and urbanization. Urbanization leads to the UHI effect, which tends to have a larger effect in the winter.

    2.3.5 Polar vortex and extreme cold events

    The impact of global warming on climate implies an overall decrease in the number of cold extremes, while the number of warm extremes increases (Horton et al., 2015). However, recent persistent winter events of record cold weather in the Northeast United States and in other Northern Hemisphere regions raise concern of a possible connection to climate change.

    Both the science community (Screen et al., 2015) as well as the public (Lyons et al., 2018) have been engaged in research and discussion about cold air outbreaks associated with the Polar Vortex. An aspect of these discussions is the connection between the gradual disappearance of Arctic sea ice due to the polar amplification of global warming, the increase in atmospheric “blocking” events, and the slowing down and deepening of the wavy circulation in the midlatitudes (Screen and Simmonds, 2010; Overland et al., 2015). With the increase in amplitude and slowdown of atmospheric waves, cold air can flow down from the Arctic deep into the midlatitudes, and vice versa, warm air flows north. This creates protracted deviations from normal conditions in either place.

    In early January of 2014, a large cold air mass moved from Canada into the northern Great Plains states and made its way slowly to the Northeast. The unusual cold weather in the eastern half of the United States did not abate until April. At the same time, other areas in the Northern Hemisphere experienced record warm winter weather.

    Shorter events similar to this have happened since, as was the case during winter months in 2017–2018 and 2019. These events were connected to stratospheric warming, where the low-pressure vortex that is usually centered on the North Pole moves equatorward. This change in circulation is communicated down to the troposphere and results in anomalous weather situations during the winter season (Kretschmer et al., 2018; Screen et al., 2018).

    There has been much debate whether such events are linked to the gradual melting of sea ice in the Arctic, and it appears that the answer is that there is a link (Overland et al., 2015; Screen et al., 2018). This was shown in climate models (Zhang et al., 2018) and is consistent with the observation that polar vortex events are on the rise (Kretschmer et al., 2018).

    There is, however, no evidence that cold air outbreaks in the United States have increased as a result of this or other phenomena (Screen et al., 2015). The increase in polar vortex events was found to influence surface weather in Siberia, where a significant cooling of the average winter weather has been detected, in contrast with the observed warming elsewhere around the globe (Kretschmer et al., 2018; Zhang et al., 2018).

    2.3.6 Summary and future research directions for extreme temperature and humidity

    New methodologies for projections of heat wave characteristics for the New York metropolitan region were tested in NPCC3 using bias-corrected climate model projections. For the early part of the century (2020s), these results are consistent with those of NPCC2. In the later part of the century (2050s and 2080s), the NPCC3 results display the potential for more intense heat events with longer durations.

    Results show large changes across all heat wave metrics throughout the 21st century. The high-emissions scenario (RCP8.5) projects, in many cases, several times larger effects than the medium-emissions scenario. The uncertainty of the projections increases through time.

    The new NPCC3 methods include humidity, which is projected to increase by more than 30% from baseline values. These increases in atmospheric humidity with extreme temperatures are likely to have large societal implications reflected in public health and energy demands.

    NPCC3 confirms the NPCC2 projections for heat waves, hot days, and cold days as those of record for New York City in planning for the impacts of climate change and recommends the incorporation of the new methodologies into revised projections of record in NPCC4.

    Future work in projecting extreme heat and humidity for the NPCC should be directed to incorporating the spatial distribution of these extreme heat events to account for coastal influence and UHI effects (e.g., sea breeze effects). This may require using regional climate models (RCMs) to dynamically downscale projections to finer spatial scales within the New York metropolitan region. Carrying this out for an ensemble of GCMs and RCMs will require large computational efforts. New methods may be needed to account for uncertainties in dynamic downscaling. See Appendix 2.C for an example of the possible approach, utilizing one GCM and one RCM for two time slices, as a potential guide for new research directions in NPCC4.

    2.4 Heavy downpours and urban flooding

    NPCC2 projected quantitative changes in daily extreme rainfall amounts for 1 inch, 2 inches, and 4 inches (Table 2.4). NPCC2 also included a qualitative projection in relation to extreme rainfall, stating that heavy downpours in the New York metropolitan region are very likely to increase by the 2080s (Horton et al., 2015).

    Table 2.4. NPCC2 projected changes in heavy rainfall days (Horton et al., 2015)a
    Baseline Low estimate Middle range High estimate
    Heavy rainfall days (1971–2000) (10th percentile) (25th to 75th percentile) (90th percentile)
    2020s
    Number of days rainfall ≥1 inch 13 13 14–15 16
    Number of days rainfall ≥2 inches 3 3 3–4 5
    Number of days rainfall ≥4 inches 0.3 0.2 0.3–0.4 0.5
    2050s
    Number of days rainfall ≥1 inch 13 13 14–16 17
    Number of days rainfall ≥2 inches 3 3 4–4 5
    Number of days rainfall ≥4 inches 0.3 0.3 0.3–0.4 0.5
    2080s
    Number of days rainfall ≥1 inch 13 14 15–17 18
    Number of days rainfall ≥2 inches 3 3 4–5 5
    Number of days rainfall ≥4 inches 0.3 0.2 0.3–0.5 0.7
    • a Projections are based on 35 GCMs and two RCPs. Baseline data are for the 1971–2000 base period. Projections show the low estimate (10th percentile), middle range (25th–75th percentile), and high estimate (90th percentile) 30-year mean values from model-based outcomes. Note: NPCC3 confirms the use of these NPCC2 projections as those of record for city policy and planning purposes.

    NPCC3 does not provide new projections for heavy rainfall and confirms the NPCC2 projections as those of record for city planning and adaptation. It provides new analyses of the dynamics of heavy rainfall events in the New York metropolitan region recommended for use in developing new projections of record in NPCC4.

    NPCC3 focuses on observed annual rainfall (see Section 2.2) and observed heavy rainfall days in recent years compared to the NPCC2 2020s time slice projections. NPCC3 also analyzes the types of storm systems associated with heavy rainfall events, and the regional drivers of historical flash flooding events. This section also conducts a trend analysis of sub-daily heavy precipitation events at the 1-, 3-, 6-, and 24-h duration. Finally, this section explores ways to illustrate the spatial variation of urban flooding events. It is recommended that this work serve as a foundation for new projections of record for heavy rainfall that are to be developed in NPCC4.

    NPCC3 analyses of heavy downpours build on NPCC2 projections for daily extreme rainfall by more closely examining the past and present rainfall across New York City and across timescales. Additionally, NPCC3 includes observations of urban flooding (definition in Table 2.5), in New York City and surrounding areas. NPCC3 refocuses discussion from daily extreme rainfall to sub-daily “heavy downpours,” defined as rarely occurring rainfall at less than daily timescales that can produce urban flooding. NPCC3 lays the groundwork for a new set of future projections in NPCC4 using these metrics.

    Table 2.5. NPCC3 rainfall and urban flooding definitions
    Term Definition
    Daily extreme rainfall Rainfall depths at the high end of the rainfall probability distribution; defined as the number of days per year that exceed 1 inch, 2 inches, and 4 inches of rainfall averaged across New York City
    Heavy downpours Rainfall at 1-, 3-, 6-, and 24-h durations that may cause urban flooding; for statistical analyses, annual maximum values are used heavy rainfall intensities may differ across New York City rain gauges.
    Days of known flooding Urban flooding identified by experts (emergency managers and National Weather Service) through either New York City Hazard Mitigation Reports or the National Centers for Environmental Information (NCEI) Storm Events Database.
    Urban flooding Surface flooding of an urban (generally over 20% impervious) area. Urban flooding is caused by rain falling faster than local conveyance systems (sewers or streams) can transmit it. When available, streamflow data from small urban streams in the New York City metropolitan region are used as a proxy for New York City urban flooding data sets.

    Extreme rainfall is defined as a rainfall amount that is a rare event, that is, one that approaches the end of the probability distribution of all events. In NPCC2, daily extreme rainfall in the current climate was represented by the number of occurrences of rainfall above 1 inch, 2 inches, or 4 inches per day at the Central Park weather station in New York City.

    Extreme rainfall measured at Central Park has significant year-to-year variation such that no statistically significant trends in extreme rainfall can be identified (Horton et al., 2015). (A statistically significant trend indicates that this trend in extreme rainfall would be unlikely to occur by chance). NPCC2 did note that the heaviest 1% of daily rainfalls have increased by approximately 70% between 1958 and 2011 in the Northeast (Horton et al., 2015). NPCC2 used the observed measurements as a baseline (Horton et al., 2015) for projections of extreme rainfall (Table 2.6; Horton et al., 2015).

    Table 2.6. NPCC2 daily extreme rainfall analyses (from Horton et al., 2015)
    Extreme precipitation methods NPCC2 (2015)
    Definitions
    • Individual days per year with rainfall at or above 1 inch
    • Individual days per year with rainfall at or above 2 inches
    • Individual days per year with rainfall at or above 4 inches
    Metric
    • Number of days per year with rainfall reaching at or above daily rainfall total
    Baseline years 1971–2000
    Baseline values
    • 13 days (1 inch)
    • 3 days (2 inches)
    • 0.3 days (4 inches)
    Future time slices 30-year time slices for
    • 2020s (2010–2039)
    • 2050s (2040–2069)
    • 2080s (2070–2099)
    Methodology
    • 2 RCPs (4.5 and 8.5)
    • 35 GCMs
    • 10th, 25th, 75th, and 90th percentiles across both RCPs and 35 GCM outputs

    2.4.1 Extreme rainfall and heavy downpours

    This section focuses on extreme rainfall by describing the approaches to heavy downpours in NPCC2 and NPCC3, studying regional drivers of daily and sub-daily heavy rainfall, providing a revised historical analysis of heavy rainfall across New York City, and summarizing new research projecting future changes in heavy downpours in the region.

    2.4.1.1 Extreme daily rainfall and links to tropical and extratropical cyclones

    NPCC2 results included projections for extreme rainfall in the 2020s (2010–2039), 2050s (2040-2069), and 2080s (2070–2099) compared to the baseline (1971–2000) at Central Park. Data for the first 8 years of the 2020s time slice can now be compared to projections—as has been done above with climate averages in Section 2.2. However, several caveats should be kept in mind. First, the fewer years of data, the more likely that year-to-year variations will outweigh any longer-term climatic trends in data. Second, if there is a substantial climatic trend in rainfall extremes, the first 8 years of a 30-year time slice may look substantially different than the last 8 years of the time slice.

    Table 2.7 shows the 2020s (defined as 2010–2039) extreme daily rainfall projections from NPCC2 compared to actual rainfall data from 2011 to 2017 and the baseline values from 1970 to 2003. The first observed 8 years of the 2020s time slice show that observed heavy daily rainfall totals have fallen within the low- to middle-range estimate of the projected amounts.

    Table 2.7. Comparison of NPCC2 daily extreme rainfall projections for the 2020s time slice (2010–2039) to observed values at Central Park (2011–2017) and baseline values (1971–2000)
    Heavy rainfall days Baseline values (1971–2000) NPCC2 2020s low estimate (10th percentile) NPCC2 2020s middle range (25th–75th percentile) NPCC2 2020s high estimate (90th percentile) Observed values (2011–2017)
    Number of days ≥1 inch 13 13 14–15 16 14.1
    Number of days ≥2 inches 3 3 3–4 5 2.7
    Number of days ≥4 inches 0.3 0.2 0.3–0.4 0.5 0.4
    • Note: These comparisons should be viewed with caution because of the role that natural variation plays in the short term.

    NPCC3 provides supplemental information by conducting a benchmark analysis on the seasonal characteristics governing daily precipitation extremes in the recent historical record, as well as an overview of recent studies of precipitation extremes that have occurred since NPCC2.

    The primary large weather systems that affect New York City are cyclones. Cyclones refer to low-pressure regions where air converges and causes uplift. Cyclones can include extratropical cyclones, caused by mid-latitude weather fronts (e.g., Nor'Easters) and tropical cyclones, which originate in the tropical oceans (e.g., hurricanes). The tracks for both types of cyclones are used to associate extreme precipitation with storms.

    The precipitation data used in this section are daily weather station data from JFK, LaGuardia, and Newark International Airports available from the Integrated Surface Database (Smith et al., 2011). The extremes are defined as 24-h precipitation for the top 1% of days for the record, per station, which translates to 138 days per site. The baseline period used, 1979–2016, was selected based on the availability of satellite-era gridded sea-level pressure (SLP) reanalysis data (Dee et al., 2011). SLP fields are used to track extratropical cyclone centers via the numerical algorithm of Bauer et al. (2016). Cyclone association is calculated by checking if a cyclone center is within 1000 km of New York City on the date of the precipitation event. However, all cyclones that ended up being associated with a precipitation extreme passed within 500 km of New York City. Additional details of the analysis method are in Towey et al. (2018).

    Daily (24-h) precipitation extremes during the 1979–2016 period occurred most often in August, but more than one event occurred in each month of the year. For events in winter months, the precipitation more likely fell as snow, but the snow water equivalent is used for this analysis.

    Figure 2.6 summarizes the annual cycle for extreme 24-h precipitation events. Extratropical cyclones cause the largest number of extreme 24-h precipitation events in each month of the year. Tropical cyclone and non-cyclone events tend to occur in summer and early fall. Other events may simply be related to small-scale storms. These types of storms are likely to drop all their precipitation in a short time period and be associated with shorter-term heavy rainfall. They are the focus of the remainder of this section. For a more detailed analysis of extreme rainfall and cyclones, see Appendix 2.D.

    Details are in the caption following the image
    Total number of extreme 24-h precipitation events at three New York metropolitan region airports (JFK, LaGuardia, and Newark) per month from 1979 to 2016 correlated with storm type. If more than one airport measured a precipitation extreme on the same date, then that event is only counted once in this figure. ETC refers to extra-tropical cyclones and TC to tropical cyclones; TC + ETC are hybrid storms that display characteristics of both types. Cyclones are identified using Modeling, Analysis, and Prediction (MAP) Climatology of Midlatitude Storm Area (MCMS; Bauer et al., 2016). Tropical cyclones are distinguished by the presence of HURicane DATabase tracks. “Other” refers to high precipitation events differing from tropical and extratropical cyclones that may be related to small-scale storms.

    2.4.1.2 Regional outlook on heavy rainfall at sub-daily scales

    Building upon the daily rainfall extremes analysis in NPCC2, NPCC3 also reviews methods to examine heavy downpours that often drive urban flooding, with analysis of sub-daily events occurring at 1- to 6-h timescales (Smith et al., 2013). Rainfall that drives urban and flash flooding in the Northeast is typically temporally and spatially concentrated and is most often caused by thunderstorms (Fig. 2.7; Smith and Smith, 2015).

    Regarding projections, NPCC2 included a qualitative projection that downpours are “very likely” to increase by the 2080s. NPCC3 does not update this prediction but does establish heavy downpours as an additional quantitative variable to be included in projections in NPCC4.

    Details are in the caption following the image
    Proportion of recent flash flood events (streamflow >92 ft3/s/mi2 or 1 m3/s/km2) on small (<5.8 mi2 or 15 km2) USGS gauged streams in the Northeastern United States caused by (A) thunderstorms and (B) tropical cyclones. (Adapted from Smith and Smith, 2015).  Thunderstorms are distinguished by the presence of lightning flashes, while tropical cyclones are distinguished by the presence of HURicane DATabase tracks. Data are for the entire period of record for each stream gauge, generally from the mid-1980s to 2015.

    Since NPCC2 there have been a few new studies using GCM projections of precipitation extremes in the Northeast United States. The results from these studies are consistent with the NPCC2 Report in that they project an increase in precipitation, both in terms of the mean and extremes for the region (Ning et al., 2015). These precipitation changes are expected to occur in both winter and summer seasons (Fan et al., 2014).

    However, the uncertainty in these precipitation projections is much larger than the uncertainty in the modeled temperature projections. One reason for the uncertainty is the presence of modes of natural variability, such as the North Atlantic Oscillation,3 that affect precipitation in the New York metropolitan region. These can have strong interannual impacts on the location and the types of cyclones that generate at least half of the strong precipitation events for the region (see, e.g., Hall and Booth, 2017). Additionally, the issue of characterizing and projecting Northeast U.S. precipitation extremes and their relationship to natural climate variability has been found to be more complicated than understanding the cyclone tracks (Ning and Bradley, 2014).

    Recent studies have projected future rainfall intensity–duration–frequency (IDF) curves for New York State. These project future rainfall extremes for durations longer than 1 h using a set of downscaled GCMs and RCMs (Castellano and DeGaetano, 2015; DeGaetano and Castellano, 2017; Castellano and DeGaetano, 2017). Future IDF curves were developed by using change factors calculated as change in rainfall between past observations and downscaled climate model projections of future rainfall. These projections, which can be found online at http://ny-idf-projections.nrcc.cornell.edu/, are more certain for rainfall durations of more than 24-h than for those under 24 hours. These are peer-reviewed, local, sub-daily rainfall projections available for New York City, used in development of the New York City Climate Resiliency Design Guidelines (http://www.nyc.gov/resiliency).

    2.4.1.3 Heavy downpours: past trends and baselines

    For NPCC3, historical trends in heavy downpours were analyzed using hourly data from multiple New York City area NOAA rain gauges at Central Park, LaGuardia Airport, JFK Airport, and Newark Airport. This analysis allows for an investigation of trends in short-duration heavy downpours. Heavy downpours, defined as the annual maximum hourly, 3-hourly, 6-hourly, and daily rainfall depths were analyzed for change points. Change points indicate that the median rainfall depth has significantly changed in that year, while trends indicate a gradual and continuous shift in medians throughout the time-period (see Fig. 2.8 for examples). Methods for determining trends in extreme rainfall included the nonparametric Pettitt Test (Pettitt, 1979), nonparametric Mann-Kendall Test (Mann, 1945; Kendall, 1975), and Sen's Slope (Sen, 1968).

    Details are in the caption following the image
    Annual maximum 3-h rainfall (in) at Central Park (a) and JFK (b) rain gauges. Central Park displays a change point, a change in median rainfall depth, in 1967 (vertical line), and JFK displays an increasing trend.

    Historic trends in heavy downpours are difficult to establish. The natural annual variation in rainfall maxima is generally more significant than trends over time, so that few statistically significant change points or trends can be found in the rainfall record (see Appendix 2.D for specific results).

    Statistical results indicate that change points can be detected at the Central Park rain gauge in the mid-1960s at the 3- to 24-h timescale for annual maximum rainfall. Change points can also be detected at the Newark Airport rain gauge in 1971 for 1- and 24-h annual maximum rainfall. These change points may represent a meteorological regime shift associated with the wetter years after the mid-1960s drought, or they may represent changes in recording. After accounting for change points, few rainfall records exhibit statistically significant trends in annual maxima. The only significant trend is for the 3-h annual maximum rainfall depth at the JFK Airport rain gauge (Fig. 2.8). Across rain gauges and timescales, it appears that there may have been an upward shift in extreme rainfall in the late 1960s to early 1970s, but that there has been no consistent trend in heavy downpours across the city.

    In order to define current baseline and spatial variation for heavy downpours in New York City, a high-resolution rainfall dataset for 1 km2 and 15-min intervals was developed using warm season rainfall from 2001 to 2015 for the Fort Dix, NJ (KDIX) SR-88D (Weather Surveillance Radar, 1988 Doppler) radar in Mount Holly, New Jersey operated by the NWS (see Appendix 2.D for details of methods).

    Baseline data for heavy downpours in New York City indicate a spatial variation in rainfall depth (Fig. 2.9). Days of known flooding (from Hazard Mitigation Reports and National Centers of Environmental Information (NCEI)) vary in rainfall depth between 0.9 and 1.25 inches across New York City. Rainfall on flooding days is at a maximum (1.25 inches) over the geographic center of the city (North Brooklyn and Northwest Queens), while some areas of high rainfall extend to the northeast. This region of high rainfall rates can be related to the urban impact on convective rainfall with higher rainfall within the city center. Figure 2.10 shows average maximum rainfall rates on days of known flooding at sub-daily timescales.

    Details are in the caption following the image
    Daily average rainfall for days of known flooding (2001–2015). Rainfall data from Staten Island are missing due to a blocked radar band. Image adapted from Smith and Rodriguez (2017).
    Details are in the caption following the image
    Boxplot for NYC-averaged max rainfall rates over 30 min, 1-, 3-, and 24-h for the 86 days of known flooding (2001–2015). Staten Island is excluded due to a blocked radar band. Whiskers represent the 10th and 90th percentiles, the box represents the 25th and 75th percentiles, and the line represents the median. Image from Smith and Rodriguez (2017).

    2.4.1.4 Heavy downpour projections

    Predicting and understanding heavy downpours in New York City is difficult, and the science is not yet available to accurately project future heavy rainfall at 1- to 6-h timescales. Short-duration heavy downpours are likely more sensitive to atmospheric conditions than longer-duration extreme rainfall, and results from longer-duration studies are not directly applicable to shorter-duration rainfall (Westra et al., 2014). Additionally, heavy downpours can vary spatially across the city, and therefore rainfall results from Central Park may not be applicable across all five boroughs. Finally, when using rainfall projections for design purposes such as in the New York City Climate Resiliency Design Guidelines (NYC Mayor's Office of Recovery and Resiliency, 2018), it is important to recognize that smaller areas can experience higher rainfall rates than larger areas. This is often solved using Area Reduction Factors, which scale rainfall intensity by the areal extent of coverage (e.g. Wright et al., 2013).

    Urban impacts. Heavy downpours in New York City and other cities are affected by several physical processes including urban modification of rainfall and interactions with the land–sea boundary. Patterns of urban modification of rainfall have been found in Chicago (Changnon, 1968), Cleveland (Huff and Changnon, 1973), St. Louis (Changnon, 1979), San Antonio and Dallas (Shepherd et al., 2002), Houston (Burian and Shepherd, 2005), Indianapolis (Niyogi et al., 2011), Atlanta (Wright et al., 2012; McLeod et al., 2017), Baltimore (Smith et al., 2012), and Charlotte (Wright et al., 2013).

    Urban areas can change rainfall patterns through UHI effects, urban-induced roughness (i.e., buildings interrupting air flow), and aerosols caused by pollution (Shepherd, 2005; Shepherd, 2013). These effects influence the path and development of thunderstorms, resulting in different rainfall patterns depending on the atmospheric setting. Generally, a weak UHI increases rainfall over city centers, while strong UHIs increase rainfall around the urban fringe, particularly downwind of urban areas (Bornstein and LeRoy, 1990; Shepherd, 2013). Short-duration heavy rainfall that produces flooding in urban areas is typically driven by warm-season thunderstorms with the most extreme rain rates occurring in the evening (Ntelekos et al., 2007). These storms are also the most influenced by urbanization (Smith et al., 2013).

    A handful of studies have attempted to understand the spatial patterns of observed extreme rainfall in New York City as it is affected by urbanization. Bornstein and LeRoy (1990) investigated the impacts of the city on paths of thunderstorms. They found that the UHI can cause convection, or lifting of air, while the city roughness (buildings and structures) can induce divergence, or the separation of air; together these result in rainfall minima within the city and rainfall maxima surrounding and downwind of the city, especially on days with a strong UHI (Bornstein and LeRoy, 1990).

    Yeung et al., 2011, used high-resolution radar rainfall fields from the Fort Dix, NJ radar and Weather Research and Forecasting modeled storm events to investigate the role of urban areas on convective storm tracks in the greater New York City region. The results showed an increased number of days exceeding 1 inch of rainfall over New York City (on average, 9 days per summer season).

    Recently, 1-h, 4 km2 multisensor Stage IV rainfall data (Lin and Mitchell, 2005) were used to classify rainfall in New York City down to 1-h events (Hamidi et al., 2017). The results showed that rainfall extremes have substantially higher rainfall rates at a 1-h scale in summer, and that summer extreme rainfall is more localized and associated with frontal systems than is winter extreme rainfall. Furthermore, Queens is most likely, and Staten Island is least likely, to experience high-intensity large areal extent 1-h summertime precipitation extremes (Hamidi et al., 2017).

    Land–sea boundary effects. The land–sea boundary also plays an important role in storm development and spatial patterns of extreme rainfall. The sea breeze can cause air to converge at low levels, thus creating uplift for thunderstorms (Weckwerth, 2000) and producing strong convection (Wilson and Megenhardt, 1997). In regions with both urbanization and land–sea boundaries to the east, the land–sea boundary tends to increase convergence, provide a source of moisture for thunderstorms, and increase rainfall intensity (Ryu et al., 2016). It is difficult, however, to disentangle the effects of sea breezes and the urban influence, as urbanization can affect the location of sea-breeze fronts (Carter et al., 2012). Furthermore, New York City's location with its multiple water bodies creates an array of different sea breeze fronts across the city (Colle et al., 2003; Novak and Colle, 2006). In the case of New York City, where the sea breeze generally is located to the southeast of the urban center, cooler air tends to stabilize the lower layers of incoming thunderstorms, generally causing them to weaken as they cross the city center (D. Rind, personal communication).

    2.4.1.5 Effects of climate change on heavy downpours

    Climate change is likely to influence the complex dynamics of urban and sea breeze–modified heavy downpours. Heavy downpours are closely tied to the amount of available moisture in the air, which is in turn influenced by air temperature. As the climate warms, the Clausius–Clapeyron relationship indicates that a warmer atmosphere can have higher ratios of water vapor to air at saturation; this is likely to increase rates of heavy downpours with climate change (Trenberth et al., 2003).

    However, increases of rainfall intensity with temperature have been observed at much higher ratios than predicted by the Clausius–Clapeyron equation, especially for sub-daily extreme rainfall (Westra et al., 2014). It has also been projected that there will be more convective storms over the Northeast United States during the later 21st century (Li and Colle, 2016), which will additionally increase heavy downpours and flooding.

    Interactions between these complex mechanisms are difficult to predict, but some paths forward have been proposed (Arnbjerg-Nielsen et al., 2013). These include a determination of the storm types that drive extreme rainfall in New York City (Figs. 2.6 and 2.7). This will help to clarify a path forward and should be considered in future work of the NPCC.

    2.4.2 Urban flooding

    Increases in extreme rainfall are expected to increase urban flooding because an increase in water volume should increase flood peaks (Ashley et al., 2005; Melillo et al., 2014). At national and regional scales, however, the ability to detect this trend is difficult. Several studies have examined streamflow records for a connection between high-flow events and climate change at national and regional scales. While some studies have found significant trends in high-flow streamflow (Groisman et al., 2001a; Groisman et al., 2001b; Juckem et al., 2008; Sagarika et al., 2014), others have not (Douglas et al., 2000; McCabe and Wolock, 2002; Small et al., 2006; Villarini et al., 2009; Hirsch and Ryberg, 2011).

    These studies typically analyze watersheds that are undisturbed, but a few regional studies have attempted to discern changes in the urban flood record with climate change (Yang et al., 2013 in Milwaukee; Rouge and Cai, 2014 in Chicago). Yang notes that for changes in flood response it is difficult to disentangle signals of “large-scale climate change, regional climate change induced by urbanization, and contrasting runoff generation mechanisms associated with land surface properties.”

    Urban flooding was not covered in NPCC2. Here, NPCC3 establishes current baselines and past trends in urban flooding for New York City and the surrounding area. These can be used in the next generation of projections for urban flooding that will be derived from quantitative projections of heavy downpours to be developed by NPCC4.

    2.4.2.1 Urban flooding past trends and baselines

    In order to investigate trends, streamflow data for flash flooding in small watersheds near New York City was used as a proxy (see Appendix 2.E). Annual peak streamflows in several small (less than 15 km2) watersheds in the U.S. Census-designated New York City urban region were analyzed for change points and statistically significant trends, similarly to the heavy downpour rainfall data.

    Trends in urban flooding are difficult to establish via this flash flooding proxy (for full results see Appendix 2.D). One New York City–region stream record, for the Mahwah River near Suffern, NY, has a statistically significant change point in 1967, which is similar to the change point in rainfall extremes (see Section 2.4.1.3). A different stream, Jumping Brook near Neptune City, NJ, has a statistically significant negative trend in annual peaks. Changes in annual peaks vary across the 14 streams with both increasing and decreasing change points and trends.

    These changes are likely to be less representative of climate change (which would show more consistent patterns across the streams) and are more representative of a direct human impact on flash flooding. Nearly all the streams are highly managed through regulations, upstream diversions, and channel modifications. These urbanization effects, including the effects of impervious surfaces (Leopold, 1968), are likely to have larger impacts on the frequency and intensity of flash floods than does climate change (e.g., Yang et al., 2013; Rouge and Cai, 2014).

    Futhermore, the combination of urbanization and climate change requires flooding and stormwater management to be assessed in a nonstationary framework—that is a framework in which historic flood and runoff occurrence is not strictly relied upon to predict the probability of future flooding events (Milly et al., 2008). This appears to be particularly evident with compound flooding (flooding caused by the combination of heavy downpours and storm surge) occurrence. Research has shown the number of compound flooding events in New York City to be increasing as weather patterns shift and sea levels rise, to cause larger precipitation amounts and more storm surge (Wahl et al., 2015).

    Urban flooding baselines and variation across the city were analyzed using 311 flood report data (https://data.cityofnewyork.us/Social-Services/311-Service-Requests-from-2010-to-Present/erm2-nwe9). Typically, urban flooding is detected as a quick rise in stream depth or flow over a certain threshold (as used in Appendix 2.D). However, within the limits of New York City, there are no small stream channels to use for flood analyses. A major source of flood data is the 311 database, which records citizen phone calls to report street or highway flooding.

    These 311 flood calls are biased due to two major issues. First, population density differences across the city make floods much more likely to be seen and reported in dense areas of the city. Second, different communities may report flooding at different rates due to perceptions of the likelihood of a response to their reports. These differences in reporting can also be observed in other types of 311 calls; for example, noise complaints, broken street lights, and other reports are also geographically biased.

    This underlying bias of all types of 311 reports was used to correct the bias in flooding 311 reports. The number of 311 flood calls within a 1 km2 area around a point was divided by the number of all 311 calls within the same area (Fig. 2.11). This allows for a distinction of how often floods are reported as compared to other 311 issues.

    Details are in the caption following the image
    Flood observations based on 311 calls for the period 2004–2015. Left panel: Flood observations based on 311 calls, normalized by all 311 observations. Units are in flood observations per all observations in 1 km2; Right panel: New York City sewer type. Image from Smith and Rodriguez (2017).

    Additionally, the New York City 2019 Hazard Mitigation Plan with its annual updates and flood reports from the National Climatic Data Center (https://www.ncdc.noaa.gov/stormevents/) were used to define days of known flooding. These reports are generated by experts (NWS staff and emergency managers), which can make them more accurate. However, standards in reporting have changed over time, and event reporting has increased in more recent years.

    Baseline data based on 311 calls for urban flooding indicate substantial spatial variation across New York City from 2004 to 2015 (Fig. 2.11). Flooding appears to occur most often in areas near the coast and areas without combined sewers: Staten Island, Jamaica Bay, and eastern Queens. The flooding pattern has been analyzed in comparison to rainfall and other potential factors, including elevation, impervious surfaces, and population density, to determine the drivers of flooding in New York City (Smith and Rodriguez, 2017). Results indicate that high groundwater tables influence flooding along the coast, while intense 1-h to 1-day rainfalls cause flooding farther inland. Flooding in Staten Island is primarily caused by wintertime extratropical cyclones (Smith and Rodriguez, 2017).

    Results from the 311 data indicate that differences in flooding across the city are likely related to rainfall patterns, proximity to the coast, impervious coverage, and differing sewer coverage. The New York City 2019 Hazard Mitigation Plan additionally includes irregular topography, soil infiltration rate, and soil storage capacity as factors that influence flooding location. Figure 2.11 indicates the similar patterns between flood occurrence and sewer type, including: combined sewers, which collect both sewage and stormwater into one system; separate sewers, which have separate systems for sewage and stormwater; parks, which do not require sewers; and other, which includes any other means of stormwater conveyance, including direct drainage into local waterways.

    2.4.3. Future research on heavy downpours and urban flooding

    Improved projections of future heavy downpours and urban flooding in New York City will require substantially more research. Potential areas of future research include:

    Heavy downpours
    • Analysis of natural climate variability. A more complete characterization is needed of present-day variability in storms and flooding. Because the New York metropolitan region experiences such large shifts in temperatures due to the annual cycle as well as large year-to-year variability per season, it is often difficult to determine the strength of the signal of climate change relative to the noise of natural climate variability. Quantifying this relationship for different precipitation metrics would help decision-makers prioritize hazard-specific responses to the projected changes.
    • Precipitation downscaling. Results from GCMs can be dynamically downscaled, in which the outputs from the GCMs are used to force higher-resolution RCMs centered around the area of interest. The RCMs should include urban features related to the New York metropolitan region, such as its large proportion of impervious surface, tall buildings, and location near the sea. The proposed future methods for extreme heat events could be used as a basis for projecting heavy downpours as well. However, research in this area should evaluate how well RCMs can be used in conjunction with GCMs for projections of heavy downpours and urban flooding.
    • Given that daily precipitation maxima are associated primarily with extra-tropical cyclones, research could assess whether there has been any convergence in global climate model projections on future storm track changes relevant to New York City.
    Urban flooding
    • Urban flood modeling. While projecting future heavy downpours is a task that requires substantially more research, modeling urban flooding in New York City may be developed in the near future. Several open-source academic models exist or are in development (e.g., Downer and Ogden, 2004; Goodrich et al., 2010; Sanders et al., 2008), and commercial urban flood models (MIKE SHE, InfoWorks ICM) are available as well. Utilization of such models will allow for understanding more clearly the relationships between rainfall intensity, duration, and frequency, and their effects on urban flooding in the region. These models could be used to assess current and future flood risks as heavy downpour projections become available.4
    • Increased urban flood observations. It is difficult to determine urban flood risk in New York City and to validate urban flood models due to lack of data. Urban flooding is typically measured by depth of streamflow in small catchment streams, but this is difficult in New York City because its surface streams have been buried. Recent advances in environmental sensing using microcontrollers may indicate a path forward for urban flood data collection. In recent years, there have also been substantial efforts to identify flood risk outside the typical streamflow methods through citizen science reporting (Cheung et al., 2016; Poser and Dransch, 2010).

    2.5 Droughts

    NPCC1 reported the potential future changes in droughts for the city using the 12-month average Palmer Drought Severity Index (PDSI) (NPCC, 2010). It was projected that the frequency of drought will approximately double by the 2050s and will be five times greater by the 2080s. This NPCC3 report focuses on drought indices developed for the city's major reservoir system using paleoclimate data.

    The drought of record in the New York metropolitan region is the one that occurred in the early to mid-1960s (Namias, 1966). It stands as a warning of the potential vulnerability of New York City to severe water shortages. Many of the operating rules governing water management for the region depend largely on performance testing using the 1960s drought as the standard (Kolesar and Serio, 2011, Devineni et al., 2013, Ravindranath et al., 2016).

    Since reliable observed streamflow data in the region often date back only to the 1950s, this section addresses questions as to the longer-term drought risk including the characterization of drought duration, severity and return period through paleoclimate data analyses.

    Hydrologic reconstructions of streamflow from tree-rings spanning the past several centuries can provide a more complete picture of the range of variability at the decadal or longer time scales. Other paleoclimate studies using pollen assemblages suggest drought conditions from ∼800 to 1300 AD as well (Pederson et al., 2005).

    These paleoclimate studies can place the short instrumental record into a more long-term perspective. Previous work (Devineni et al., 2013, Woodhouse et al., 2006, Nowak et al., 2012; Stockton and Jacoby, 1976) have demonstrated the utility of paleo climate streamflow reconstructions in providing a more objective evaluation of operating rules for reservoir systems. Consequently, for NPCC3, we developed reconstructions of the Pepacton, Cannonsville, and Neversink (PCN) reservoir inflows (Fig. 2.12) using tree-ring chronologies in the upper Delaware River basin.5 We used these extended reservoir inflow records to develop long-term drought profiles on duration, severity and return periods under different water demand thresholds. Table 2.8 provides key definitions for terms used to discuss drought throughout this section.

    Details are in the caption following the image
    New York City's Water supply system. The Cannonsville, Pepacton, and Neversink reservoirs of the Delaware Watershed are analyzed by NPCC3 using long-term drought records from tree-ring data. Source: NYCDEP.
    Table 2.8. NPCC3 drought definitions
    Term Definition
    Reservoir inflow Streamflow (amount of water) coming into reservoirs.
    Reconstruction Estimate of streamflow for past period using trees proven to be good estimators of observed streamflow during the period of gauged record. This is typically developed using statistical models that capture the relationship between tree growth index and the observed streamflow record during the overlapping period. This statistical model is applied to the prior period.
    Cumulative deficit Accumulated water deficit over an n-year period. Deficit for each year is defined as the difference between water demand (reservoir releases) and water supply (reservoir inflows).

    2.5.1 Methods of analysis

    This section briefly presents the methodology employed for reconstructing reservoir inflow and for deriving drought indicators. Data description and technical details of the model structure are provided in Appendix 2.E. Full details of the methods can be found in Devineni et al. (2013).

    2.5.1.1 Reservoir inflow reconstructions

    We developed the PCN reservoir inflow reconstructions using a statistical regression model. Instrumental data (i.e., inflows for the three reservoirs during the observation period since 1928) were provided by the New York City Department of Environmental Protection. Tree-ring width measurements that represent paleoclimate data for the Delaware watershed date back to 1754. These are available from the Tree Ring Laboratory at the Lamont-Doherty Earth Observatory (LDEO).

    Given data from the three reservoirs and eight local tree-ring chronologies as predictor variables, the statistical model provides regression equations for each reservoir that are used to reconstruct the streamflow. The period over which the reconstruction was done is 1754–1927. The resulting outputs are simulations of annual average daily streamflow from 1754 to 2000 for the three reservoirs.

    2.5.1.2 Drought indicators

    We constructed a drought index to characterize the regional drought with explicit consideration of water demand. We developed the drought index on instrumental streamflow data first to gain an understanding of the observed drought risk since 1928. Then, we applied it to the reservoir inflows reconstructed from the tree-ring data.

    2.5.2 Results

    This section presents the results of the streamflow reconstructions and drought analyses for the instrumental period and the paleo-reconstructed period.

    2.5.2.1 Combined inflows from tree-ring data and incidence of observed drought

    The general trends of combined reservoir inflow from tree-ring data from 1754 to 2000 are shown in Figure 2.13. While the 1960s drought is the most severe in the extended record, the tree-ring analysis show that there were regimes with less severe but longer drought durations (e.g., 1830–1860, 1790–1810).

    Details are in the caption following the image
    Reconstruction of combined annual average daily inflow from eight tree-ring chronologies in the Pepacton, Cannonsville, and Neversink reservoirs, which supply approximately 50% of the New York City water supply (DRBC 2018). Since tree growth is dependent on climate and since each tree-ring represents a season of growth, tree-ring measurements provide information on hydrological indicators over a tree's life span that can be used to understand variations in climate.

    By examining this historical record, we found that there are at least eight incidences of historical drought lasting 5 consecutive years or longer occurring in the region since 1750 (Table 2.9). Six of these occurred in the paleo record period, and two were observed in the instrumental period. This indicates there is a potential for persistent drought in the New York metropolitan region in the future.

    Table 2.9. Incidence of historical drought of at least 5 consecutive years in the New York metropolitan region in the paleo record (1754–1927) and the instrumental record (1928–1999)
    Drought duration Years
    Paleo record
    10 years 1764–1773
    11 years 1791–1801
    5 years 1803–1807
    9 years 1852–1860
    6 years 1883–1888
    5 years 1909–1913
    Instrumental record
    5 years 1929–1933
    7 years 1961–1967

    2.5.3 Summary and future work

    Long-term drought risk for the New York City water supply system is developed based on tree-ring reconstructions for PCN reservoir inflows. The streamflow reconstructions reveal droughts with a longer duration than the 7-year major drought seen in the instrumental period (1961–1967). If the variability of streamflow as seen from the long paleoclimate tree-ring record (246 years) were to continue into the future, increases in regional water demand due to population increase and climate change could affect the duration of droughts. This is important from a drought risk and planning perspective.

    Hydrologic reconstructions provide a more complete picture of how streamflows have varied in the New York metropolitan region water supply area. However, longer-term water planning decisions should also be informed by climate scenarios, such as the New York City Panel on Climate Change (Moody and Brown, 2012; Steinschneider and Brown, 2012; NAS, 2018; Rosenzweig and Solecki, 2018).

    Given current understanding of seasonal to interannual climate variability, and of climate change, NPCC4 could develop an approach for regularly updating the drought estimates using climate observations and models tuned to prediction at different timescales. Consequently, future work should involve drought risk characterization and modeling that embraces paleo-reconstructions, climate model hindcasts utilizing such metrics as PDSI, observed trends, and near-term and longer-term projections in a rigorous way to understand climate risk and formulate management and adaptation strategies at decision-relevant scales.

    2.6 Conclusions and recommendations

    NPCC3 confirms the use of NPCC2 projections as those of record for decision-making in the City of New York. It analyzed how recent climate trends compare to the projections for the region. Further, it has begun to develop and test new methods for observations and projections to be used in resilience planning. Using expanded observations, bias correction, and RCMs, these methods can provide quantitative analyses for heat extremes, heavy downpours, and droughts. They are available for developing the next full set of NPCC projections.

    Based on these and other methods, the next generation of global and RCM outputs will be used in upcoming NPCC assessments to create a new unified set of projections for decision making in the New York metropolitan region. The methods tested by NPCC3 utilizing GCM and RCM ensembles and scenarios will enable the updated identification of climate change “hotspots” of vulnerability at finer spatial scales within the city and across the region.

    However, GCMs may not yet be able to simulate the forcings required for RCMs to model some finer-scale extreme events such as convective thunderstorms.

    Key findings

    Observations and projections

    • Observed annual temperature and precipitation trends between 2010 and 2017 fell largely within the NPCC2 projected range for the 2020s time period. [These comparisons should be viewed with caution because of the role that natural variation plays in the short term.]
    • Observations of increasing heavy rainfall between 2011 and 2017 fell largely within the NPCC2 projected range 2020s time period. [These comparisons should be viewed with caution because of the role that natural variation plays in the short term.]
    • NPCC3 confirms the use of the NPCC2 2015 projections for decision-making by the city and region.

    Extreme heat

    • Increasing decadal trends in annual daily average maximum summer temperatures in June, July, and August varied across the city. Central Park has experienced an increasing trend of 0.2 °F per decade from 1900 to 2013. Since 1970, JFK average annual daily maximum summer temperatures have been rising at a rate of 0.5 °F per decade, and LaGuardia at 0.7 °F per decade.
    • New projection methods for extreme heat events were developed and tested for the New York metropolitan region. The test includes bias correction, a method that adjusts the mean and variance of GCM results to match a representative set of observations from the region, and high-resolution regional climate modeling.

    Heavy downpours

    • New studies support NPCC2 projected increases in precipitation, in terms of the mean and extremes for the region. These precipitation changes are expected to occur in both the winter and summer seasons. However, uncertainty in these precipitation projections is larger than the uncertainty in temperature projections.
    • A change point in sub-daily heavy rainfall events can be detected at the Central Park rain gauge in the mid-1960s for the annual maxima of 3-hourly rainfall; the only significant trend found was for 3-h annual maximum rainfall depth at the JFK rain gauge.
    • Extratropical cyclones cause the largest number of extreme 24-h precipitation events in New York City in every month out of the year.
    • Rainfall that drives urban and flash flooding in the Northeast is typically temporally and spatially concentrated and is most often caused by thunderstorms.
    • Days of known flooding vary spatially across New York City in rainfall depth between 0.9 and 1.25 inches. Rainfall on flooding days is at a maximum (1.25 inches) over the geographic center of the city (North Brooklyn and Northwest Queens).
    • Urban flooding appears to occur most often in areas near the coast and areas without combined sewers: Staten Island, Jamaica Bay, and eastern Queens. Results from 311 call data indicate that differences in flooding across the city are likely related to rainfall patterns, proximity to the coast, impervious surfaces, and differing sewer coverage.
    • The groundwork for future projections was established by refocusing the discussion from daily to sub-daily rainfall extremes.

    Droughts

    • While there has not been a major drought since the 1960s in the New York metropolitan region, analysis based on tree-rings from about the last 250 years shows that 10-year or longer droughts have occurred. Thus, the possibility of future droughts should be considered in planning.

    Recommendations for research

    • Future NPCC research can improve the utility of quantitative heat wave projections by working with the New York City Department of Health and Mental Hygiene and the National Weather Service (NWS). Together these groups can investigate how best to evaluate need for future revisions of heat advisory criteria that consider changing combined effects of temperature and humidity (i.e., heat index).
    • Relevant research areas include examination of thresholds of heat and humidity effects on human health, and strategies to design interventions that will be effective in New York City's hotter climate.
    • Research is needed to determine benchmarks for sub-hourly extreme precipitation and associated flooding events using satellite data and rain gauges at Central Park, LaGuardia, JFK, and Newark. Using these benchmarks, researchers should aim to improve sub-hourly extreme precipitation projections that consider urban meteorological effects and identify neighborhoods likely to be flooded.
    • Improved characterization is needed of likely large-scale conditions that may lead to extreme drought based on further tree-ring analysis in the region.
  • 1 Sea breezes form due to temperature differences between the air over land and ocean and are a feature of New York City coastal areas (Haurwitz, 1947; Childs and Raman, 2005).
  • 2 Although warm days can occur outside of the traditional summer months, here we only consider June, July, and August, as extreme heat days are more likely to occur during this period.
  • 3 The North Atlantic Oscillation, or NAO, is a fluctuation in sea-level atmospheric pressure between the Icelandic Low and Azores High. It influences climate patterns in the Northern Hemisphere.
  • 4 See the following website for information about the Town & Gown: Citywide Stormwater Resiliency study: https://www1.nyc.gov/assets/ddc/downloads/town-and-gown/active-rfps/CitywideStormwaterResiliencyStudyT+GRFP.pdf
  • 5 While this NPCC report uses the Delaware River Basin as a drought proxy for the New York metropolitan region, the city measures and monitors water supply availability on a whole-system basis including the Delaware, Catskill and the Croton systems.
  • 6 Current computational power limits high-resolution RCM simulations to relatively time slices compared to those presented at the GCM level. However, the authors feel that the added spatial granularity provide significant value.
  • Appendix 2.A: NPCC3 Global Climate Models

    Table 2.A.1. Global climate models used in NPCC3 ensemble for extreme heat and humidity
    Center Model
    Commonwealth Scientific and Industrial Research Organization—Bureau of Meteorology (Australia) ACCESS1-0
    ACCESS1-3
    Canadian Centre for Climate Modeling and Analysis (Canada) CanESM2
    National Center for Atmospheric Research (USA) CCSM4
    Centro Euro-Mediterraneo per i Cambiamenti Climatici (Italy) CMCC-CM
    CMCC-CMS
    Centre National de Recherches Méteorologiques/Centre Eurorpéen de Rechercheet Formation Avencée en Calcul Scientifique (France) CNRM-CM5
    Commonwealth Scientific and Industrial Research Organization/Queensland Climate Change Centre of Excellence (Australia) CSIRO-Mk3-6-0
    NOAA Geophysical Fluid Dynamics Laboratory (USA) GFDL-ESM2G
    GFDL-ESM2M
    NASA Goddard Institute for Space Studies (USA) GISS-CM3
    GISS-E2-H
    GISS-E2-R
    Met Office Hadley Centre (UK) HadGEM2-AO
    HadGEM2-CC
    HadGEM2-ES
    Institut Pierre-Simon Laplace (France) IPSL-CM5A-LR
    IPSL-CM5A-MR
    IPSL-CM5B-LR
    Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute/National Institute for Environmental Studies/Japan Agency for Marine-Earth Science and Technology (Japan) MIROC-ESM
    MIROC-ESM-CHEM
    MIROC5
    Max Planck Institute for Technology (Germany) MPI-ESM-LR
    MPI-ESM-MR
    Meteorological Research Institute (Japan) MRI-CGCM3
    Institute for Numerical Mathematics (Russia) INM-CM4

    Appendix 2.B: NPCC3 Bias Correction Methods for Heat Waves

    The bias correction technique corrects for both differences in model mean and standard deviation using a linear model, or:
    urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0001

    TBC refers to the bias-corrected temperature record. In the equation, T refers to the temperature records, and σ refers to the standard deviation of temperatures. Subscripts Obs and GCM refer to observations and climate model data, respectively, while REF and RAW refer to the reference (2006–2015) and entire projection periods (2006–2099), respectively. The over bar (–) marker denotes use of the average for the specified dataset and time period. The calculation of the bias correction is performed by weighting the difference between the observed reference and total period data by the ratio of the observed to climate model standard deviations, and adding the “weighted difference” to the observations to produce a time series.

    The mean values of the four urban stations are used as observations for the training period of 2006–2015 for each model of the ensemble. The correction is then carried on for the three 30-year periods of interest (2020s, 2050s, and 2080s). The bias-corrected distributions are presented in Figure 2.B.1 for the complete ensemble of GCM daily maximum temperatures, which shows the bias-corrected distributions are much closer to the observations.

    Details are in the caption following the image
    Sample of bias-corrected GCM distributions for maximum temperatures for the training period of 2006–2015. KDE refers to a Kernel Density Estimate, a representation of the probability of occurrence of a given value in the data set.

    Appendix 2.C: Potential New Methods for NPCC4 Extreme Heat Projections

    The output from GCMs can be dynamically downscaled, in which GCM outputs are used to drive high-resolution RCMs. This approach has led to development of regional, or limited area models (Dickinson et al., 1989; Giorgi et al., 1993; Skamarock et al., 2008). This is a potential method for the next generation of NPCC climate change projections for New York City for use in adaptation planning and implementation. It can be a useful approach because GCMs used for quantifying future changes currently do not have adequate resolution to realistically simulate many extreme weather events, such as extreme heat, tropical storms, rapidly deepening nor'easters, severe convective storms, and heavy rainfall. Since these models are often run at 100–300 km grid spacing, much of the uncertainty originates from not properly resolving atmospheric dynamics for this weather. Further, GCM physics have large uncertainties at coarse resolution, since they do not resolve finer-scale processes such as the UHI and sea breezes. See Tables 2.C.1 and 2.C.2 for details.

    Table 2.C.1. Summary of approach used for regional climate model simulations
    Simulation approach
    Regional Climate Model Weather Research and Forecasting (WRF) model version 3.8
    Initial and boundary conditions Community Earth System Model version 1 (CESM1)
    Baseline years 2006–2010
    Simulation period June 1st to August 31st
    Scenarios RCP4.5 (medium emissions)
    RCP8.5 (high emissions)
    Table 2.C.2. Physics options used in WRF simulations
    Parameterization Reference
    Convection Kain-Fritsch (Kain, 2004)
    Microphysics WSM6 (Hong and Lim, 2006, p. 6)
    Boundary Layer Mellow-Yamada-Janjic (Nakanishi and Niino, 2006)
    Land Surface Noah land surface model (Tewari et al., 2004)
    Urban Physics BEP (Martilli et al., 2002)
    BEM (Salamanca et al., 2010)
    Cooling Tower (Gutiérrez et al., 2015b)
    Urban Drag Coefficient (Gutiérrez et al., 2015a)

    Dynamical downscaling uses output from GCMs as initial and boundary conditions for high-resolution model run centered on the region of interest. In order to resolve clouds and urban-specific processes (e.g., anthropogenic heat and radiation blocking), this will require resolutions around 4-km grid spacing or less; however, most current downscaling simulations use ∼20-km grid spacing. To address uncertainties in dynamically downscaled simulations, multi-simulation ensembles are employing varying boundary conditions, physics parameterizations, and grid spacing should be employed. Examples of this ensemble approach include the North American Regional Climate Change Assessment Program (NARCCAP; Mearns et al., 2009) for the contiguous United States at 50 km resolution.

    Another approach is using pseudo-global warming (PGW; Kimura and Kitoh, 2007). In the PGW approach, the ensemble mean monthly temperature changes from the GCMs are added to the historical reanalysis data, which in turn is used for initial conditions (ICs)/boundary conditions (BCs) for the future high-resolution regional domain runs. This approach is cheaper, since separate runs are not needed for each GCM, but this approach does not include any large-scale flow changes in the future from the GCM, since only temperature perturbations are added.

    2.C.1 High-resolution dynamical downscaling

    To test the approach for the New York metropolitan region, GCM projections are downscaled using an urbanized version of the Weather Research and Forecast Model (WRF; Skamarock et al., 2008) developed and maintained by the National Center for Atmospheric Research (NCAR). Model physics are based on Gutiérrez et al. (2015a; Gutiérrez et al. (2015b)) and are summarized in Table 2.C.2. In contrast to GCMs, dynamical downscaling is performed by embedding, or nesting, models of higher spatial complexity within each other. In this case, three domains (one parent, two nested) are used, with horizontal grid spacing of 9, 3, and 1 km (Fig. 2.C.1). The high-resolution domain covers the New York metropolitan region with results presented for New York City.

    Details are in the caption following the image
    Urban canopy parameters for NYC derived from PLUTO. These parameters are used to calculate interactions between the atmosphere and buildings.

    Urban parameterizations require use of urban canopy parameters, such as urban land use, building plant area fraction (Fig. 2.C.1, left), and building heights (Fig. 2.C.1, right) to represent energy and momentum exchanges between the atmosphere and built environment (Fig. 2.C.1). These parameters have been derived from the Property Land Use Tax-lot Output (PLUTO) (NYC Open Data. Primary Land Use Tax Lot Output (PLUTO). https://data.cityofnewyork.us/City-Government/Primary-Land-Use-Tax-Lot-Output-PLUTO-/xuk2-nczf/data.) made publicly available since 2013.

    Model initial and boundary conditions are taken from a bias-corrected Community Earth System Model (CESM) data set provided by NCAR (Bruyère et al., 2015), which corrects biases in the intra-annual variation for all meteorological variables using ERA-Interim Reanalysis. The bias correction technique follows the work from Holland et al. (2010) as applied by Bruyère et al. (2014). The correction method separates the GCM and reanalysis signal into a seasonally varying term and a perturbation term (containing the model's climate signal). The seasonal mean is the corrected mean using the reanalysis's historical seasonally varying mean, while keeping the model's climate perturbation. This method was found by Bruyère et al. (2014) to produce more realistic patterns of wind shear and tropical cyclone generation for the historic period. The projection ensemble is summarized in Table 2.C.1. Correcting all model variables was shown to decrease CESM1 cold temperature biases when used as input to a regional model. Finally, three time periods are selected6: historical (2006–2010), mid-century (2045–2049), and end of century (2095–2099).

    2.C.2 RCM results

    Spatial variation of heat wave changes is shown for two time slices (2045–2049 and 2095–2099; Fig. 2.C.2) for a typical year (i.e., median values). In general, sea breeze fronts, which typically develop in the afternoon due to land-ocean air temperature differences, play a crucial role in determining projected changes by moderating high temperatures near the coast.

    Details are in the caption following the image
    Median projections of event frequency for New York City.

    The RCM simulations show that the number of heat waves per year is projected to stabilize after 2045 in the medium-emissions scenario. This stabilization occurs because the land surface warms more than the ocean and this differential warming causes sea breezes to moderate the number of heat waves that take place. This is similar to current conditions in which sea breeze circulations prevent parts of Brooklyn and Queens from experiencing as many and as severe heat waves as parts of Manhattan and the Bronx. Projections for Manhattan and The Bronx, however, show increases in heat waves by two to four events per year for both time periods.

    In the high-emissions scenario (RCP8.5), the sea breeze is weakened due to increased ocean air temperatures, leading to Brooklyn and Queens experiencing higher event frequencies of between five and seven additional events per year. This is even greater than projected heatwave increases in Manhattan and The Bronx of four to six events. Larger increases near the coast may also be due to historically lower temperatures due to the local sea breeze.

    The heat index (Rothfusz, 1990), combining temperature and relative humidity, is often used as a metric of how heat affects humans. Projections show that the heat index is expected to increase with time across projections, with end of century changes ranging between 6 and 8 °F in medium-emissions scenario to 12–16 °F in high-emissions scenario (Fig. 2.C.3). Changes in heat index are, in general, slightly larger over Manhattan and the northern part of Brooklyn in all scenarios and time slices except in end-of-century high-emissions scenario, where a similar pattern as that observed in event frequency projections emerges, with heat index increasing at a faster rate toward the coast.

    Details are in the caption following the image
    Median 2 pm afternoon heat index for 2045–2049 and 2095–2099 periods compared to the 2006–2010 baseline.
    Although multimodel high-resolution ensembles were not used in this study due to computational cost, changes in internal model variability across time slices and emissions scenarios are explored. Sources of uncertainty in urban modeling may include:
    • Representation of urban environment: Although this study uses relatively high-resolution urban canopy parameters, building-atmosphere interactions are heavily parameterized, depending on values averaged over grid points.
    • Limited domain size: Due to computational limitations, high-resolution urban climate models run on a relatively small domain. Local conditions, in particular heat waves, are impacted by large scale synoptic processes that may occur thousands of miles away from New York City. Any uncertainties in the input model will be carried over in boundary and initial conditions used in these simulations.

    Appendix 2.D: Methods of Extreme Rainfall Analyses

    Table 2.D.1 summarizes the results of the extratropical and tropical cyclone associated analysis. If neither type of cyclone was associated with the precipitation event, the event is labeled noncyclone. In addition to identifying the total number of cyclone-associated events, we calculate the percentage of cyclones that cause extremes. To do this, we divide the number of cyclones associated with a precipitation extreme by the total number of cyclones that pass within 500 km of New York City. For extratropical cyclones, 7.5% of the storms caused a precipitation extreme. For tropical cyclones, the number is much higher, at 30%. This probabilistic calculation cannot be made for noncyclones, because the storm-type for those events is not known. At least some of those events are most likely associated with are quasi-linear convective systems (Lombardo and Colle, 2012), which are sometimes grouped with frontal systems (Kunkel et al., 2011). The dominant cause of extreme daily rainfall events for all airport stations out of these storm types is extratropical cyclones.

    Table 2.D.1. Association of daily precipitation extremes at New York metropolitan region airports from 1979 to 2016
    Storm type JFK Airport station LaGuardia Airport station Newark Airport station
    Extratropical cyclones 92, 4, 7.4% 92, 3, 7.3% 96, 5, 7.9%
    Tropical cyclone 19, 4, 30.3% 17, 3, 27% 13, 5, 24%
    Noncyclone 23 26 24
    • Note: When three values are given, the first is the number of isolated cyclones, the second is the number of extratropical plus tropical cyclones that occurred, the final number is the percentage of cyclone that pass within 500 km of New York City that caused extreme precipitation at the given station.

    Radar was processed with the Hydro-NEXRAD algorithms (Seo et al., 2011) and corrected with a daily multiplicative bias (as in Smith et al., 2012) using rain gauges from the National Oceanic and Atmospheric Administration Meteorological Assimilation Data Ingest system (NOAA MADIS, https://madis.ncep.noaa.gov).

    Methods for determining trends in extreme rainfall and proxy-stream flash flooding included the nonparametric Pettitt Test (Pettitt, 1979), nonparametric Mann-Kendall Test (Mann, 1945, Kendall, 1975), and Sen's Slope (Sen, 1968). Complete results from these analyses are shown in Tables 2.D.2 and 2.D.3.

    Table 2.D.2. Statistical analyses of rainfall data in New York City region
    Parameter CP LGA EWR JFK
    Hourly record Time period 1948–2013 1948–2013 1948–2013 Hourly record
    One hour Change point 1971 (+35%)** One hour
    Trend
    Three hours Change point 1967 (+42%)** Three hours
    Trend 0.2 mm/year**
    Six hours Change point 1966 (+34%)** Six hours
    Trend
    Daily Change point 1965 (+17%)** 1971 (+30%)** Daily
    Trend
    Daily record Time period 1869–2017 1940–2017 1893–2017 Daily record
    • Note: Rainfall gauges located at Central Park, LaGuardia Airport, Newark Airport, and JFK Airport. Statistics are for 1-, 3-, 6-h, and daily annual rainfall maxima. Change points are shown in year and (change in averages), while trends are shown in Sen's Slope.
    • **A value is significant at a 5% level.
    • *A value is significant at a 10% level.
    Table 2.D.3. Statistical analyses of streamflow data in the U.S. census New York City urban region
    USGS gauge ID Location Time frame Number of years Change point Trend (cfs/year) Notes
    01374654 Carmel, NY 1996–2012 17 Regulated flow
    01374930 Baldwin Place, NY 1996–2016 21 2011 (–38%)* Occasional regulation
    01381400 Morristown, NJ 1996–2015 20 Diversion upstream
    01387450 Suffern, NY 1959–1998 40
    2001–2015 15 1967 (+154%)** Well withdrawals upstream
    01392210 Passaic, NJ 1977–1999 21
    01399670 Whitehouse Station, NJ 1978–2015 37 14.09* Occasional regulations and upstream releases
    01401650 Belle Mead, NJ 1991–2015 25 Some irrigation regulation in summer
    01403150 Martinsville, NJ 1980–2015 35 3.84*
    01403400 Seeley Mills, NJ 1967–2015 49 Temporarily moved 1969–1979
    01403535 Watchung, NJ 1980–2015 36
    01403540 Watchung, NJ 1973–2015 43 Occasional regulation, channel modified in 1991 and 1997
    01407290 Marlboro, NJ 1980–2015 35 1999
    (–25%)*
    01407705 Neptune City, NJ 1967–2014 48 Diversion upstream, a portion is regulated
    01407760 Neptune City, NJ 1967–2016 50 –7.47** Upstream diversion water supply and golf courses
    • Note: Statistics are for annual peak (maximum) in instantaneous streamflow. Change points are shown in year and (change in averages), while trends are shown in Sen's Slope.
    • **A value is significant at a 5% level.
    • *A value is significant at a 10% level.

    Appendix 2.E: Methods of Tree Ring Analysis and Drought Analysis

    The PCN reservoir inflows were developed using a Bayesian regression model. Given data from three streamflow gages and eight local tree-ring chronologies (that date back to 1754) as predictor variables, the Bayesian model provides regression equations for each reservoir that are used to hindcast the streamflow. Annual average daily streamflow (June–May) was assumed to follow a lognormal distribution. The Bayesian regression models used to produce this partially pooled reconstruction explain around 60% of the streamflow variance and validate best against withheld data. The posterior probability distributions of the reconstructed combined reservoir inflow from the Bayesian regression model during the period 1754–2000 are shown in Figure 2.13. The record period common to all selected trees determined the time span of the reconstructions. The reconstructions of the combined reservoir inflow are presented as time series composed of the median of the posterior distribution for each year, as the reconstructions for each year are estimates of the posterior distribution of the annual average daily inflow for those years. The record of observed PCN combined inflow data is shown using the 11-year low-pass filtered values (blue color line during the instrumental period (1928–2000)). Similar low-pass filtered values are also shown for the median inflows (red color line) during the reconstruction period to visualize the general trend in the data.

    2.E.1. Drought index (methods)

    We developed the drought index to capture the effect of drought over multiple years. The index is based on the sequent peak algorithm (Loucks et al., 1981). It quantifies the water reservoir drawdown for meeting the demand. The steps for the computation are as follows:
    urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0002
    urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0003
    where Deficitt refers to the accumulated annual deficit, Dt refers to the annual water demand, St refers to the annual water supply and n is the total number of years under considerations. The maximum accumulated deficit estimated over the n-year period is defined as the Severity of the drought. It measures the potential impact of multiyear droughts (Etienne et al., 2016).

    2.E.2. Drought profile based on the reconstructed reservoir inflow data

    The demand-specific drought index is applied to the simulations of the reconstructed PCN combined inflows with a demand threshold of 950 million gallons per day (MGD) of annual average daily flow to develop the long-term drought risk profile. Figure 2.E.1 presents the joint probability distribution of the drought duration and severity as seen from the paleo records. The worst drought event in the instrumental period (the 1960s drought of 6 years and a cumulative deficit of 1000 MGD) is shown as a red circle in the figure. It is evident from the paleo streamflow data that the drought of the record, the 1960s drought, is still an extreme event relative to a long-term drought risk profile. The probability of exceedance of the 6-year drought duration is urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0004, an approximate average return period of 16 years if drought length is of concern. The probability of exceedance of the 1000 MGD cumulative deficit (drought severity) is urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0005, an approximately average return period of 33 years if drought severity is of concern. However, if combined variables of duration and severity are of interest, the probability of joint exceedance urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0006urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0007, an approximate average return period of 166 years. Hence, while a drought of a 6-year length occurs more frequently than the drought of a 1000 MGD severity, the recurrence of the joint drought as worse as the 1960s is anomalous.

    Details are in the caption following the image
    The joint drought profile for a demand of 950 MGD annual average daily outflow. The contour plot shows the joint probability distribution of drought duration and severity. The drought of the record (1960s drought of 6 years and 1000 MGD cumulative deficit) is shown as a red circle on the contour plot.

    2.E.3. Drought profile based on the reconstructed reservoir inflow data and changing demand

    It is important to note that the drought stress is always relative to the demand of the region. The above analysis is shown for a demand of 950 MGD of annual average daily flow as a benchmark water demand. We have chosen this threshold given this is the average PCN combined reservoir release (including diversions to New York City, conservation, and directed releases) for the last 5 years (USGS, 2018). To investigate the effect of water demand on drought stress, we have applied the drought index for four different thresholds, 950 MGD, 1000 MGD, 1050 MGD and 1100 MGD. Any average demand greater than 1100 MGD will exceed the average combined reservoir inflow.

    The joint probability distributions of drought duration and drought severity (long-term drought profiles) for various water demand levels is shown in Figure 2.E.2. We observe from these distributions that the drought duration is changing at a rate faster than the drought severity with increasing demand. As the water demand of the region increases, from a long-term planning perspective, the critical metric to focus on will be the length of drought. Drought stress is experienced in terms of its persistence. This can also be seen from Figure 2.E.3, which shows the individual distributions for each of these thresholds along with the drought of the record from the instrumental period.

    Details are in the caption following the image
    The joint drought profile for varying demands.
    Details are in the caption following the image
    The distributions of drought duration and severity for varying demands.

    The streamflow reconstructions reveal droughts with a longer duration than the duration of the drought seen in the instrumental period (1960s drought). Joint distributions of duration and severity are developed for various demand levels to get a better perspective of the long-term drought profile. Based on a demand level that matches the average reservoir releases for the last 5 years, the worst drought of the record in the instrumental period is 6-year drought with a 1000 MGD cumulative deficit. This event has a joint return period of 166 years when contextualized with the long-term drought profile. However, the drought stress is very sensitive to regional water demand. A marginal increase in the demand from the 950 MGD level will lead to droughts that are longer and more severe, and their joint occurrence becomes more frequent. A comparison of duration versus severity metrics indicates that the rate of change with respect to demand levels is much faster for the drought duration.

    2.E.4. Observed droughts

    For the period of 1928–2000, annual average daily inflows and cumulative reservoir deficit was calculated based on a total demand of 950 MGD of annual average daily flow (Fig. 2.E.4). Note that 950 MGD is approximately the average reservoirs’ release for the recent 5 years.

    Details are in the caption following the image
    Annual average daily inflows and cumulative deficit (drawdown) of the combined Pepacton, Cannonsville, and Neversink (PCN) reservoir during the instrumental period (1928–2000). The blue line shows the observed PCN reservoir combined inflow. The red line (inverted) indicates the cumulative deficit.

    In the decade of the 1960s, the reservoirs had extensive drawdown, making it the worst drought of the instrumental period. The observed duration of the drought is 6 years, from 1961 to 1967. The severity of the drought, measured as the cumulative deficit, is approximately 1000 MGD. The recovery period of this drought is 5 years. While there are other periods with small to moderate droughts, there is no other period in the instrumental record that has a drought as severe as the 1960s drought.

    Table 2.E.1 summarizes the individual and joint probability of exceedances and return periods of the drought duration and severity. Evidently, they are very sensitive to the demand. While the droughts stress for a demand level consistent with the water releases for the past 5 years is moderate, the drought stress is more likely and reoccurs more frequently for a marginal increase in the demand levels.

    Table 2.E.1. Summary of the probability of exceedances and the return periods of the droughts for four different demand levels
    Demand
    950 MGD 1000 MGD 1050 MGD 1100 MGD
    urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0008 urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0009 urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0010 urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0011
    urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0012 urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0013 urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0014 urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0015
    Exceedance probability urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0016 0.03 0.04 0.45 0.95
    urn:x-wiley:00778923:media:nyas14007:nyas14007-math-0017 0.006 0.028 0.40 0.94
    Return period Severity 33 years 25 years 2.2 years 1 years
    Joint 166 years 36 years 2.5 years 1 year