Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
83 CMIP climate projections have been published and periodically updated over the past several years. Many of the projections are available on an annual or even daily basis for many years into the future. The U.S. DOT has developed the CMIP Climate Data Process- ing Tool, which can process climate model outputs from CMIP3 and CMIP5 into relevant statistics for transportation planners. (The tool and user guide can be downloaded at https:// www.fhwa.dot.gov/environment/?sustainability/resilience/adaptation_framework/modules/ ?index.cfm?moduleid=4#tools.) Specifically, the tool is designed to analyze data that can be down- loaded from the U.S. Bureau of Reclamationâs downscaled CMIP3 and CMIP5 Climate and Hydrology Projections website (http://gdo-dcp.ucllnl.org/downscaled_cmip_projections). Readers who are considering doing their own in-house data gathering and analysis could download this tool and access the website as directed in the user guide in order to access rele- vant climate data for their given airport location. The user guide provides step-by-step instruc- tions for downloading, processing, and interpreting available climate projections. Both CMIP3 and CMIP5 can provide daily projections of a variety of climate measures, including minimum and maximum surface air temperatures, precipitation rates (mm/day), and humidity, and both provide projections that have been downscaled to a spatial resolution of 1/8 degree (which in the continental United States translates to a rectangular cell grid approximately 7.5 miles on each side). While the more recent CMIP5 projections have not been tested as thoroughly as CMIP3, CMIP5 is widely considered to be the best available science presently, and it has two key advan- tages. First, it provides continuous daily projections from the present out to 2099, compared to CMIP3, which has projections only for the years 2046â2065 and 2081â2099. Second, some of the CMIP5 data have been further downscaled to a finer resolution in order to provide better risk exposure estimates in a localized area. LOCA is a statistical downscaling technique that uses history to add improved fine-scale detail to global climate models. LOCA has been applied to 32 global climate models under Scenarios RCP4.5 and RCP8.5 from the CMIP5 archive; it currently provides estimates of maximum daily temperature and daily precipitation at a spatial resolution of 1/16th degree (a rectangular area of less than 4 miles on each side), covering North America from central Mexico through Southern Canada.19 Aside from the finer resolution, the LOCA technique is thought to result in better estimates of extreme climate days, particularly estimates of precipitation. The LOCA website has links to download sites containing the latest LOCA data. It is important to note that the ACROS high-temperature projections described in Chapter 3 (suggested for use in an initial screening analysis) were prepared before LOCA-based data became available, so the projections may differ. Thus, one must be cautious if trying to compare the ACROS projections of high temperatures to the projections described here. A P P E N D I X D Accessing Available Climate Projections
84 Climate Resilience and BenefitâCost Analysis: A Handbook for Airports Data Collection Strategies for High Temperatures One of the primary objectives of this handbook is to present a methodology for evaluating the increasing occurrence of high temperatures that may force airlines to impose weight restric- tions on takeoffs (or in extreme cases, cancel flights entirely). It is currently possible to obtain LOCA-based high-temperature projections at specific locations across the United States out to the year 2099. Exactly how much data needs to be collected will vary. For purposes of analyzing the impact of high temperatures on takeoffs, the projection of daily maximums is important because each day that temperatures exceed some threshold value may necessitate a weight restriction. In addition, there are three other factors that will affect the actual data collection: â¢ Selection of nearby geographical grid points relevant for the airport: Climate science best practices suggest use of at least four adjacent LOCA-based grid points near the airport. â¢ Relevant time horizon: It may be sufficient to focus on a time horizon that goes to the end of the expected life of an airportâs runway(s); however, the analyst may also wish to consider longer time frames to get a more general view of the potential long-term effects of increasing temperatures. â¢ Number of climate models: Unless specific information dictates otherwise, it is suggested that projections be obtained from all 32 available climate models. As noted in Chapter 4, each model makes individual point predictions, and it is the variation in the predictions across the different models for a given scenario and future date that reveals the uncertainty in those projections. As an example, if data projections for 2020 through 2099 were obtained from all 32 models for four grid points, the total number of high-temperature observations would be 365 days Ã 80 years Ã 4 grid points Ã 32 models = 3,737,600. While this is a large number, it could be reduced significantly by grouping the actual maximum temperatures into a small number of categories that represent the count of the number of times in a given year that the high temperature is at the indi- cated level. Thus, the data could be organized to look something like what is shown in Exhibit D-1. MODEL YEAR GRID_ID H100 H102 H104 H106 H108 H110 H112 H114 H116 H118 H120 H122 H124 H126 H128 ACCESS1-0 2020 1 15 16 13 22 29 15 18 1 0 0 0 0 0 0 0 ACCESS1-0 2020 2 18 13 13 23 27 22 13 4 0 0 0 0 0 0 0 ACCESS1-0 2020 3 16 16 12 22 25 19 19 1 0 0 0 0 0 0 0 ACCESS1-0 2020 4 16 14 16 21 28 18 15 2 0 0 0 0 0 0 0 ACCESS1-0 2021 1 24 21 15 21 19 11 7 3 2 0 0 0 0 0 0 ACCESS1-0 2021 2 30 18 18 18 20 12 9 2 2 0 0 0 0 0 0 ACCESS1-0 2021 3 25 16 15 22 21 11 8 3 2 0 0 0 0 0 0 ACCESS1-0 2021 4 32 20 16 20 18 11 8 1 2 0 0 0 0 0 0 â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ ACCESS1-0 2090 1 7 13 13 7 18 25 28 31 15 5 1 1 1 0 0 ACCESS1-0 2090 2 8 9 14 11 14 24 32 32 13 6 2 2 0 0 0 ACCESS1-0 2090 3 8 11 10 11 17 22 30 32 15 6 1 1 1 0 0 ACCESS1-0 2090 4 11 8 14 8 20 26 32 32 8 6 1 2 0 0 0 ACCESS1-3 2020 1 19 33 32 18 7 10 6 0 0 0 0 0 0 0 0 ACCESS1-3 2020 2 22 32 39 13 10 9 6 1 0 0 0 0 0 0 0 ACCESS1-3 2020 3 20 34 34 15 10 11 5 0 0 0 0 0 0 0 0 ACCESS1-3 2020 4 21 37 37 12 9 11 3 0 0 0 0 0 0 0 0 ACCESS1-3 2021 1 9 22 27 26 19 14 5 2 0 0 0 0 0 0 0 ACCESS1-3 2021 2 14 16 33 22 19 18 5 2 0 0 0 0 0 0 0 ACCESS1-3 2021 3 9 17 29 27 18 14 8 2 0 0 0 0 0 0 0 ACCESS1-3 2021 4 15 20 28 25 18 17 3 2 0 0 0 0 0 0 0 â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ ACCESS1-3 2090 1 12 15 20 20 37 33 19 15 5 3 1 0 0 0 0 ACCESS1-3 2090 2 12 14 16 23 35 32 30 13 5 1 2 0 0 0 0 ACCESS1-3 2090 3 11 15 18 24 34 35 22 12 6 3 1 0 0 0 0 ACCESS1-3 2090 4 13 16 16 25 34 40 18 12 5 1 2 0 0 0 0 â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ â¦ Exhibit D-1. High-temperature data.
Accessing Available Climate Projections 85 Each row represents a unique model/year/grid-point combination, and the âHâ columns represent counts of annual days at or above the indicated (Fahrenheit) temperature but below the next columnâs temperature. For example, the row for the model named ACCESS1-0/Year 2020/ Grid_ID 4 projects that there will be 16 days with daily high temperatures between 100Â°Fâ102Â°F, 14 days between 102Â°Fâ104Â°F, and so forth. Summarizing the data in these 2-degree increments between 100 and 128 drastically cuts down on the number of total data points but still retains a full range of high temperatures that should be relevant for purposes of estimating weight restric- tions at airports across the entire United States.20 This specific data organization is used directly in the Excel-based high-temperature template that was developed as part of this project and is described in detail in Appendix E. One could use such a data set as part of a Monte Carlo simulation analysis (discussed in Appendix C) that would reflect the uncertainty of the incidence of high temperatures projected across the different climate models. Check for Potential Bias Correction It is important to recognize that a given modelâs historical accuracy for a specific loca- tion may be systematically off (i.e., biased) even though the model does well overall. In this case, climate science best practices suggest that one should test to see whether some sort of bias correction is needed before sampling from the available model projections. For daily maximum temperature, for example, while one would not expect the daily projections from a GCM model to match actual daily temperatures, one would want to check for any system- atic differences in the range or distribution of such temperatures over a representative time periodâsay, 10 to 20 years. So one could gather actual historical data of daily maximum temperatures for the loca- tion of interest over, say, 20 years, plus corresponding projected temperatures from a given model. One approach that has been used in the climate science field is to order both sets of data from low to high, place them into 20 5-percentile bins, and then compute the mean of each bin. The absolute difference between the model mean and the observed historical mean in each corresponding bin (measured in degrees) could then be used as a bias correction factor for the model; this would account for any systematic variations in both the distribution and range (spread) of temperatures. In practice, however, one would also have to account for the likelihood that the overall temperature range itself could rise over the very long term. One way to implement the correction factors in such a situation would be to separate the relevant future years into successive 20-year cohorts (to match the length of the historical test period), compute 5-percentile bins for each, and then apply the relevant bias correction factor to the projections in each bin. Sampling and Weighting Strategies Depending on the nature of the available data projections, there are different sampling strategies one could use. For the present case where there are multiple models providing different projec- tions of future high temperatures, one could randomly select a single model for each separate simulation and use its projections for every year out to the end of the analysis period. This approach essentially assumes that the number of unique possible future outcomes is limited to the number of different models available. If one is drawing from a large number of models, then this may well be a reasonable strategy. However, if the collection of models includes some that are considered outliers (i.e., very different from other models), then it also means that the range of results may be quite sensitive to these outlier models.
86 Climate Resilience and BenefitâCost Analysis: A Handbook for Airports To dampen the influence of such models yet not exclude them completely, another possibility would be to randomly select (for each simulation) a model for each year of the analysis period. This will tend to reduce the influence of outlier models because it is unlikely they would be randomly selected year after year within a given simulation draw; at the same time, this will also likely increase the year-to-year variability within a simulation since the draws are coming from different models. A third possibility would be to combine the projections across models to compute a multi- model mean and fit a statistical distribution from which one could take simulation draws. One issue here is that the mean is likely changing over time (as temperatures increase), so one could in principle fit a separate distribution for each year. Short of that, one could aggregate over, say, each 10-year period and fit a single distribution for each decade to cut down on the computa- tional burden (see, for example, Coffel et al. 2017). Whichever sampling strategy is chosen, one might also want to consider implementing a weighting strategy so that better models are given more weight (and therefore a higher prob- ability of being sampled) than lesser ones. Sanderson et al. (2016) provide a weighting assess- ment for the 32 climate models mentioned previously; a modified version of these weights is used in the numerical examples provided in Appendix F, as well as in the Excel-based template for high temperatures. Data for Sea Level Rise CMIP5 projections also are available for sea level rise. But the nature of the projections is very different than the daily high temperatures from multiple models discussed previously. Specifi- cally, further analysis is required to translate projections into the likelihood of flooding risks at specific locations. One can combine estimates of the likelihood of extreme water events based on historical data with estimates of future sea level rise in order to obtain projections of future extreme water events. Fortunately, analyses published by NOAA are directly relevant and can be used for these purposes. First, NOAA has undertaken an extensive analysis of historical extreme water levels (EWLs) in the United States at 112 long-term stations of the National Water Level Observa- tion Network operated by the Center for Operational Oceanographic Products and Services (CO-OPS) (Zervas 2013). The data are analyzed to quantify probabilities of exceedance and the return periods (average length of time between exceedances of a given water level). NOAA uses a statistical model to characterize the distribution of EWL values, resulting in the estimation of an âexceedance probability curveâ as a function of the return period (Zervas 2013). For example, the exceedance probability curve for the station at Kings Point/Willets Point in New York is shown in Exhibit D-2. This is the station closest to LaGuardia Airport, which is about 6 miles to the southwest. Reading off the graph at, say, the 10-year return period shows a water level of about 1.5 meters. This means that, based on the historical data, this location would expect an extreme water event of at least 1.5 meters approximately every 10 years; the 10-year return period translates into a 10% probability on an annual basis. It is important to note that the water level is relative to the mean higher high water (MHHW) vertical datum established by CO-OPS, which is the average height of the diurnal high tide recorded at the station each day.21 There are three parameters used to define the curve for each station; these parameters are contained in Appendix I, Table A of the Zervas report. Following Gilleland and Katz (2016), the specific formula22 for finding the extreme water level is given as a function of these parameters
Accessing Available Climate Projections 87 plus the desired annual probability of occurrence. For example, using the parameters for Kings Point/Willets Point and setting p = 0.10 (implying a return period of 10 years) gives a result of 1.539, which is exactly consistent with what is shown graphically in Exhibit D-2. Note that the exceedance curve provides exactly the type of information needed to conduct a Monte Carlo simulation of annual water levels. For example, if one were considering a time period of, say, 50 years, for each year a random number could be drawn between 0 and 1 repre- senting the annual probability of occurrence, and then the implied extreme water level could be computed from the curve. In accordance with the curve, in many years there would be relatively low water levels (around 1 meter), while in other years there might be higher levels (up to about 3 meters at the extreme). As described, the estimated exceedance curve for each location is based entirely on historical data. But sea levels generally are expected to rise in the future, which presumably would affect these local flood events. To address this, NOAA has also modeled projected changes in local sea level rise (Sweet et al. 2017). An accompanying data file shows projections for almost 2,000 different coastal locations worldwide.23 There are six different scenarios considered for future sea level change, identified by the overall projected global mean sea level (GMSL) rise by 2100: â¢ Low (GMSL = 0.3 meters), â¢ Intermediate-low (0.5 meters), â¢ Intermediate (1.0 meters), â¢ Intermediate-high (1.5 meters), â¢ High (2.0 meters), and â¢ Extreme (2.5 meters). Local projections are given at 10-year intervals out to 2100 for each scenario. For example, the projected RSL rise for Willets Point in New York is shown in Exhibit D-3.24 How likely is each of these scenarios? Recalling the earlier discussion of RCP global climate scenarios, these GMSL scenarios have been estimated to have the exceedance probabilities under three of the four available RCP scenarios (as shown in Exhibit D-4).25 Thus one can select an RCP scenario and match the listed probabilities to the local RSL projections shown in Exhibit D-3. For example, under RCP8.5, the intermediate GMSL (or higher) scenario is estimated to occur 17% of the time. Increases in local sea level will shift extreme water levels upward by the same amount, assuming no change in local tidal magnitudes in the future. Under this assumption, if the intermediate Source: Zervas 2013. Exhibit D-2. Exceedance probability curve for Kings Point/Willets Point, NY.
88 Climate Resilience and BenefitâCost Analysis: A Handbook for Airports GMSL rise scenario were to occur, then in 2050, the local relative sea level rise for Willets Point would be projected as 41 cm (from Exhibit D-3), and one could then add this increment directly to whatever extreme water level projection were shown from the curve in Exhibit D-2.26 As suggested earlier, the nature of the uncertainty associated with these projected flood events is different from the uncertainty related to the high temperature projections described previously. The latterâs uncertainty comes directly from the variance in 32 different climate modelsâ projections of daily high temperatures. In contrast, the uncertainty in the sea level rise projections comes from probabilistic estimates of extreme water events as reflected in exceedance probability curves combined with six different projections of the likelihood of localized sea level rise. The specific methodology outlined previously for estimating future localized flood risks was implemented in the Excel-based extreme water template developed as part of this project. As described in Appendix C, one can make random draws as part of a Monte Carlo simulation that will reflect the uncertainty of the incidence of future extreme water events implied by the exceed- ance probability curves combined with projected future sea level rise. Finally, it is important to note that, while the ACROS SLR projections described in Chapter 3 (suggested for use in an initial screening analysis) assume RCP8.5, they are in fact based on older projections than the more up-to-date 2017 NOAA estimates cited previously. Thus, one must be cautious if trying to compare the ACROS projections of SLR to the projections described here. GMSL Scenario RSL in 2020 (cm) RSL in 2030 (cm) RSL in 2040 (cm) RSL in 2050 (cm) RSL in 2060 (cm) RSL in 2070 (cm) RSL in 2080 (cm) RSL in 2090 (cm) RSL in 2100 (cm) Low 5 9 14 19 24 29 32 36 38 Intermediate-low 6 12 18 24 31 37 42 47 51 Intermediate 9 19 29 41 54 69 85 102 118 Intermediate-high 12 26 40 57 77 100 126 153 182 High 16 32 52 77 108 139 173 217 262 Extreme 14 35 61 90 129 169 215 270 326 Source: Sweet et al. 2017. Exhibit D-3. Projected RSL rise for Willets Point, NY. Source: Sweet et al. 2017, Table 4. Exhibit D-4. GMSL scenario probabilities.