• Mitigation
• Adaptation
• Sustainability
• Energy

• Ongoing & Current Research
• Resources & Publications
• Policy & Guidance
• Webinars
• Workshops & Peer Exchanges
• Newsletter
• Contacts

FHWA → Environment → Climate Change → Adaptation → Ongoing And Current Research → Gulf Coast Study → Phase 2/Task 2 → Mobile Variability

Climate Variability and Change in Mobile, Alabama

Appendix C: Additional Information on the Temperature, Precipitation, and Streamflow Analyses

C.1. Data Available for Each Station and Variable

Historical data from five National Oceanic and Atmospheric Administration (NOAA) Global Historical Climatology Network (GHCN) stations in the Mobile region were analyzed to investigate existing climatic trends and baseline conditions. Two of the stations (Coden and Mobile Airport) are located in Mobile County and three of the stations are located in neighboring Baldwin County. Each station records daily minimum and maximum temperature and total daily precipitation. The Baldwin County stations have the longest record of temperature measurements dating back to 1915, while the Mobile County stations' temperature observations began in 1948. Precipitation measurements in Baldwin County began between 1912 and 1918, while precipitation measurements for the two Mobile County stations began in 1948 and 1956, respectively. The data availability for each station is plotted in a timeline in Figure 87.

Figure 87 : Data Available for Each Station and Variable

C.2. Climate Modeling Overview

Projections of environmental variables can be developed using a number of different tools, ranging from a simple relationship trend analysis to sophisticated climate models. Deciding what tool to use depends on factors such as the nature of the variable, the costs, and the use of the variable in the climate impact assessment.

Climate models can be used when a study requires projections of temperature, precipitation, wind, or relative humidity, as well as other environmental variables. Projections of sea level rise and specific localized storm events are not included in this discussion. In this study, these two sets of projections were produced through a qualitative review of the literature and scenario-based modeling, as discussed in Section 2.7 and Section 2.8.

This appendix first provides some important background information on climate models, including a discussion of the uncertainty inherent in climate models, and then describes the approach used in this study to identify optimum modeling parameters (e.g., future time horizons of interest) to inform transportation vulnerability assessments.

C.2.1. Introduction to Climate Models

Climate models simulate how climate may change in the future (see textbox titled, "What is a Climate Model?"). Climate model projections indicate how the climate may respond to future variations in greenhouse gas emissions. Climate models can project temperature, precipitation, specific humidity, winds, and a number of other physical parameters. Climate models are not intended to provide a projection of future weather (e.g., the temperature and precipitation on May 12, 2050), but are intended to provide an average of twenty to thirty years of simulated weather. Hence, climate projections are generally averaged over a twenty to thirty year period.

There are a number of different climate models developed and currently maintained by research groups around the world. Using the same inputs and initialization schemes, different climate models may project different degrees of warming and changes in precipitation. Climate sensitivity is a metric climate scientists can use to compare and evaluate models. Climate sensitivity describes projected global temperature change in response to a doubling of carbon dioxide emission.

The IPCC Fourth Assessment Report (AR4) (2007a) determined the climate sensitivity projected by 23 well-known climate models from 14 modeling groups. The climate sensitivity of these models ranges from 3.8°F (2.1°C) to 7.9°F (4.4°C), with a climate model ensemble average of approximately 5.9°F (3.3°C). Some climate models (e.g., the Hadley Center for Climate Predictions and Research HadCM3 model) simulate a warmer future than others (e.g., the National Center of Atmospheric Research (NCAR) Parallel Climate Model (PCM)).

Some climate models may "perform" better than others at representing observations for particular regions of the globe. However, the standard best practice is to average across climate models that represent the span of climate sensitivity.¹ This is why an ensemble average, which averages the projections across a large number of climate models, is generally preferred. These climate projections can be accessed from the publically available WCRP CMIP3 database.

C.2.2. Introduction to Emission Scenarios²

An emission scenario is a plausible representation of future greenhouse gas (GHG) emissions based on a set of assumptions about driving forces (such as demographic, socioeconomic, and technological change), and their key relationships. Emission scenarios describe how greenhouse gas emissions may change over time. In 2000, the IPCC published multiple emission scenarios accounting for variations in demographics, socioeconomics, and technology that are used as inputs into climate models. Emission scenarios vary from low to high levels of greenhouse gases emissions over time. Scenarios that have similar demographic, socioeconomic, and technological-change storylines are grouped into scenario families. The four scenario families that comprise the IPCC Special Report: Emissions Scenarios (SRES) set are A1, A2, B1 and B2. These scenarios do not include additional climate policies that differ from current practices (i.e., no new mitigation efforts are considered).

Figure 88 illustrates the differences in carbon dioxide emissions across several IPCC SRES emission scenarios. The differences between scenarios become increasingly noticeable from the near-term projections (e.g., 2020) to the end-of-century projections.

Figure 88 : Simulated Carbon Dioxide Emissions from 1990 to 2100 by Emission Scenario (IPCC 2007a)

Climate models are "run" under an emission scenario. As illustrated by Figure 89, the climate model ensemble run under the high (A1FI) emission scenario³ projects the greatest amount of global surface warming, at approximately 7.2°F (4.0°C) by the end of the century. Meanwhile, the low (B1) emission scenario⁴ projects the lowest amount of warming at about (3.2°F (1.8°C). The moderately-high (A2) emission scenario⁵ is considered a moderately-high emission scenario projecting a global surface warming of 6.1°F (3.4°C) by the end of the century.

Figure 89 . Multimodal Simulated Change in Global Surface Temperatures, as a Function of Emission Scenario, Relative to 1980 to 1999

Source: IPCC 207a. The bars on the right of the figure provide the climate model ensemble mean and the shading provides likely ranges across the models

The scientific community has not assigned probabilities to the emission scenarios suggesting which is more or less likely to occur; hence, each emission scenario should be considered with equal probability.

The IPCC Fifth Assessment Report (AR5), scheduled for release in 2013-2014, will present climate projections run with a set of new integrated socioeconomic, emissions, and climate scenarios. Past scenario development was conducted in a mainly sequential form, with socioeconomic and emission scenarios developed first and climate change projections developed based on those scenarios. The new and integrated scenarios will allow the modeling of climate system responses to human activities in parallel with emission scenario development. As a result, the AR5 will include scenarios that explore approaches to climate change mitigation in addition to the traditional "no climate policy" scenarios used in previous assessments.

C.2.3. Dealing with Uncertainty in Climate Projections

There is considerable confidence in the capability of climate models to simulate temperature projections,⁶ particularly at the continental scale, but less confidence in climate models ability to project precipitation.⁷ This difference in confidence should be qualitatively considered when incorporating risk and vulnerability assessment results into future planning.

There are three main sources of uncertainty in climate model simulations:⁸

1. Natural variability (the unpredictable nature of the climate system)
2. Model uncertainty (the ability to accurately model the Earth's many complex processes)
3. Scenario uncertainty (the ability to project future societal choices such as energy use)

The relative contribution of each uncertainty component to the climate model simulation's overall uncertainty varies with time. Hawkins and Sutton (2009) investigated how these relative contributions change over time when considering the global decadal mean surface air temperature (the approximate values provided here are for qualitative discussion purposes only).⁹ Most notably, scenario uncertainty is relatively minimal in the near-term but is the greatest contribution to total uncertainty by end-of-century. The model uncertainty represents a large portion of the total uncertainty throughout the time period, and is a dominant contributor by near-term and mid-century. Meanwhile, natural variability is a significant contributor to total uncertainty in the near-term, but becomes much less significant by end-of-century.

• Near-term: A large portion (21%) of near-term uncertainty is from natural variability. Most (70%) uncertainty is from model uncertainty. Scenario uncertainty contributes minimally (9%) to total uncertainty.
• Mid-century: Most uncertainty by mid-century is from model and scenario uncertainty, which contribute 54% and 39% of total uncertainty, respectively. Natural variability is a much smaller contributor to total uncertainty, contributing only 7%.
• End-of-century: Most uncertainty at end-of-century (70%) is a result of scenario uncertainty. Model uncertainty contributes 27% of total uncertainty and natural variability contributes less than 3%.

These uncertainties also change relative to each other for projections on different spatial scales. Natural variability becomes a greater source of uncertainty at finer scales.¹⁰ This is one reason why incorporating downscaled projections expands the potential uncertainty in climate projections.

As climate science progresses, the degree of uncertainty will likely be reduced-particularly for regional-scale projections. Hawkins and Sutton (2009) suggest that the uncertainty associated with regional projections in the near-term, dominated by model uncertainty and natural variability, could be significantly reduced through scientific progress.

The uncertainty around each of these components should be considered when conducting vulnerability assessments, making decisions, and implementing risk-averse policies. Various techniques can be used to address uncertainty, including probabilistic approaches to quantify uncertainty, modeling various emission scenarios to produce a wide range of future possibilities, comparing present-day model results with observations, and engaging expert judgment to express uncertainty based on level of agreement and amount of evidence.¹¹

In this study, a number of uncertainties are qualitatively addressed:

• Scenario uncertainty is intrinsically incorporated in the study design by providing projections driven by three different emissions scenarios: 'low' (B1), 'moderately-high' (A2), and 'high' (A1FI). Using these emission scenarios is a way to bound the projections, though it is quite possible that actual emissions in the 21st century will be above this range.
• Model uncertainty is captured, to some extent, by comparing projections across a number of climate models.

In addition to the common uncertainties inherent in climate modeling, this study incorporates an additional layer of uncertainty by using statistically downscaled temperature and precipitation projections. Downscaling climate model projections allows scientists to incorporate local conditions, such as the effect of local topography or prevailing sea breezes, by tailoring larger-scale climate model results to a finer-scale analysis. However, using downscaled data introduces an additional degree of model uncertainty and natural variability into the projections that is not quantified here. Downscaling further assumes that the relationship between today's observed data and modeled data remains stationary over time.

C2.4. Approach to Identify Optimum Climate Modeling Parameters

Two steps were taken to determine the best climate model projections to use in this study:

1. Optimum climate modeling parameters were determined
2. Available climate projection data sets were evaluated against these parameters.

This process determined what time periods of projections would be useful, the number of climate models to use, which emission scenarios to use, and whether and how to downscale the data. These determinations are summarized below.

Time Periods

Climate projections are generally provided for twenty to thirty year periods. The shorter, twenty year periods are more likely to be affected by natural variability, particularly in the near-term. However, the shorter time period reduces the impact from climate simulations that continue to evolve with time (i.e., climate model simulations used in this study will not project a stationary future but one that is constantly evolving with time). The longer, thirty year periods are preferred for extreme event analyses as it provides a longer data record.

Because they are preferred for extreme event analyses, thirty year periods were selected for this study. This study provides climate projections in three time periods: near-term (2010-2039), mid-century (2040-2069) and end-of-century (2070-2099).

A reference baseline period of 1980 to 2009 was chosen. This period replicates the weather and climate that transportation planners currently plan for. However, it may underestimate the magnitude of projected change, which might have been larger if an earlier baseline period was chosen. In addition, this choice in time period may not reflect the design values (such as 24-hour precipitation return periods) that were used to build current infrastructure, as most infrastructure was built decades ago.

Number of Climate Models

The scientific climate community recommends averaging across as many climate models as possible when developing projections for impact/risk assessments.¹² The models selected had to provide continuous daily output, robust results, and ideally capture the breadth of climate model sensitivity.¹³ As a result, ten climate models were selected to generate an ensemble average for the A2 and B1 emission scenarios and four climate models were selected for the A1FI emission scenario ensemble average.

Emission Scenarios

To capture a range of possible futures, three emission scenarios were used for a scenario-based analysis: a low emission scenario, B1; a moderately-high emission scenario, A2; and a high emission scenario, A1FI.

If resources do not allow multiple scenarios to be used, a high or moderately-high emission scenario (such as A1FI or A2) might be used so that the impact assessment captures the highest degree of risk to temperature change, or a more moderate emission scenario (such as A1B) might be used if planners prefer a more conservative set of estimates. As noted in the findings of this report, changes in baseline and extreme precipitation events may not change in proportion to the rate of increasing greenhouse gas emissions. As a result, it is useful to understand how available climate model simulations for the study area change as a function of emission scenario, before committing to one or more scenarios to inform the impact assessment.

C.2.5. Downscaling Techniques

Over the past two decades, climate models have provided results at increasingly finer spatial resolution. As the spatial resolution increases, the details in the topography of the land, such as mountainous regions and coastlines, become more obvious. However, higher spatial resolutions require increased computational resources and continued evaluation of the physics represented in the model. For example, at finer spatial scales, some of the terms that are not relevant at large scales may need to be reconsidered for inclusion.

For many transportation impact assessments, even these finer-scaled climate models produce projections at spatial resolutions that are too coarse to be informative. For example, the current spatial resolution of climate models (i.e., the size of the model grid cell) ranges from a surface area of about 180 miles by 180 miles (288 kilometers by 288 kilometers, or about 2.8 degrees by 2.8 degrees, varying by latitude) down to about 60 miles by 60 miles (96 kilometers by 96 kilometers, or about 1 degree by 1 degree, varying by latitude). This spatial resolution may be acceptable for some assessments and not for others.

To determine the optimum spatial resolution for a given study, the temporal resolution required for the assessment should also be considered. The types of questions poised in this study to determine downscaling technique included:

• Will daily, monthly, or annual averages need to be provided for the variable?
• Does the study area tend to be affected by coastal sea breezes?
• Is Mobile's weather significantly affected by changes in small-scale topography?
• These questions were prompted by the current state-of-the-literature describing when downscaling techniques were preferred for specific locations. A number of methods have been developed to further reduce or "downscale" the spatial resolution of climate projections from climate models. These methods can be divided into two distinct techniques: statistical downscaling and dynamic downscaling.

Statistical downscaling

Statistical downscaling determines a statistical relationship between locally observed data and large-scale modeled data over a historical time period. This relationship is then applied to climate model data for future time periods. A number of locations may be used to determine the best algorithm for statistical downscaling. This technique assumes the relationship between the larger-scale modeled variable and the local observations will not change over time.

The performance of statistical downscaling is largely constrained by the accuracy of the climate model to simulate regional temperature, humidity, and circulation patterns.¹⁴ In addition, accuracy is affected by whether the observation set used to train the algorithms captures the range of local weather conditions.¹⁵

Statistical downscaling is relatively affordable, which allows this method to be applied at a number of observation locations, for various emission scenarios, and for a number of climate models. Impact assessments informed by statistically downscaled data can include a number of climate models run with a multitude of emission scenarios.

Dynamic downscaling

Dynamic downscaling uses a global climate model to drive a regional climate model (RCM). Regional climate models are fine resolution models that incorporate enhanced algorithms and topography and can be nested within the climate model. Though not always done, this allows the regional climate model output to be incorporated into the climate model simulation.

Dynamic downscaling is appropriate when the global climate model does not adequately represent a region's climate.¹⁶ Because the technique is very computationally intensive and regional climate models require a significant development effort, dynamic downscaling is quite costly. Generally, studies of impact assessments informed by dynamic downscaling rely on one climate model and a few emission scenarios.

Selecting a technique

Deciding whether and how to downscale climate projections depends on a number of factors, such as whether the data will be used to inform additional modeling (such as hydrologic modeling), the scale of the variables that are needed to inform the assessment, the availability of data from public sites, and the study's budget and timeline.

This study uses downscaled data, because the data will inform transportation asset-specific (i.e., fine spatial scale) assessments at a coastal location (affected by factors such as coastal sea breezes). Statistical downscaling was selected because of its ability to simulate future changes in temperature and precipitation for a continuous time period and its capacity to affordably produce a number of climate projections by climate model and emission scenario. The Asynchronous Regional Regression Model (ARRM) method of statistical downscaling, in particular, was applied because it is capable of downscaling at daily timescales.¹⁷

C.3. Detailed Temperature and Precipitation Projections Methodology

This Appendix describes the methodology that was used to develop projections of future temperature and precipitation in the Mobile region. Projections of temperature and precipitation were produced using up to ten climate models, run under three emission scenarios, for three future time periods. To account for local influences, the large-scale climate data was statistically downscaled using the Asynchronous Regional Regression Model (ARRM) method to the locations of the five GHCN observation stations.

Under this method, historical statistical relationships between weather and climate were derived by comparing the local temperature and precipitation observations to the modeled climate data.¹⁸ After deriving and testing them, the relationships were used to translate future modeled climate projections of temperature and precipitation to the individual station level. This technique assumes the relationship between weather and climate does not change over time.

The statistical downscaling methodology was cross-validated and bias-corrected for each of the five station locations. The results of this cross-validation process showed:

• The downscaling method had a very low bias of +/- 0.2°F (0.1°C) modeling minimum and maximum temperature for a majority of the days from 1980 to 2009. However, for the few days that were observed to have extreme temperature values (i.e., either very cold or very hot), the biases were as high as +/- 1.0°F (0.6°C); biases were slightly higher for modeling the minimum temperature than for modeling the maximum temperature.¹⁹
• The downscaling method was not as accurate at simulating precipitation, due to the higher spatial and temporal variability associated with precipitation. Nevertheless, for the days from 1980 to 2009 with low observed amounts of precipitation, the modeled precipitation amounts tended to be 5 to 15% lower than the observational data. For the days with the highest observed amounts of precipitation (i.e., for the 99th percentile of precipitation), the models tended to overestimate observational data by 20 to 30%.²⁰

The remainder of this appendix is presented in three sections, corresponding to the steps taken to identify and present relevant projections:

• Identification of relevant climate projections
• Identification of statistically significant climate projections
• Determination of reporting format

C.3.1. Identification of Relevant Climate Projections

The climate "wish list" described in Appendix B.2 provided a starting point for the type of climate projections useful for impact assessments. The "wish list" does not define the thresholds or probability of occurrence best suited for assessing impacts on local transportation in Mobile, Alabama. Relevant thresholds and probabilities were defined based on three strategies:

• Discussions among local transportation engineers and planners about how local transportation infrastructure, maintenance, and operations in Mobile are impacted by today's weather hazards
• An extensive literature review, summarized in the Assessing Sensitivity report
• Meetings of transportation infrastructure engineers, federal transportation planners, and climate scientists

Table 28 provides a list of the temperature and precipitation weather hazards and climatic averages that were deemed useful for this study. These variables can be directly estimated using the daily downscaled precipitation and temperature data.

The modeled results for all variables were provided for each emission scenario, for each of the five station locations, and for the baseline and projected time frames (i.e., 1980-2009, 2010-2039, 2040-2069, 2070-2099). The results were then averaged across the statistically downscaled climate models.²¹

Table 28 : Temperature and Precipitation Variables Developed for this Study

An asterisk denotes a variable or percentile that does not provide robust quantitative results (per communication with Dr. Katharine Hayhoe) and its use should be limited to qualitatively informing the impact assessment.

Variable	TransportationMode	Methodology	Related Figures and Tables
Temperature
Annual, seasonal, and monthly average minimum, maximum, and mean temperature for each 30-year time period(9 components)	Airports (runway design)	For each 30-year period, the daily minimum, maximum, and mean temperature corresponding to each month, season, or year were averaged for each station location, climate model, and emission scenario. Then, the 30-year average was determined for each station location, climate model, and emission scenario. Averages and standard deviations were then calculated across climate models for each station location and emission scenario. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region.	Main Body – Figures 10, 11, 12, 13, 14, 15, 16, 17Appendix C – Tables 30, 31, 32Appendix E – Tables 56, 57, 58, 59, 60, 61, 62, 63, 64
Mean, 50th and 95th*percentile of high daily maximum temperature and the warmest day of the year for each 30-year time period(4 components)	Rail (AREMA rail design, buildings)	For each 30-year period, the daily maximum temperature for each year was identified. This resulted in a total of 30 data points in each time period for each climate model, station location, and emission scenario. The mean, 50th, and 95th percentile levels were estimated from this set of 30 data points using a quantile distribution and then averaged across climate models for each station location and emission scenario. The warmest day in summer for the 30-year period was estimated in the same way. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region.	Main Body – Figures 22, 23Appendix C – Table 33Appendix E – Tables 65, 66, 67, 68
Seasonal and annual number of days and maximum consecutive days of maximum temperatures at or above 95°F (35°C), 100°F (38°C), 105°F (41°C), and 110°F (43°C) during each 30-year time period(16 components)	Civil, Geotech, Pavement	For each 30-year period, the number of days where the maximum temperature was at or above 95°F, 100°F, 105°F, and 110°F was counted for each year. This resulted in 30 data points in each time period (one for each year), for each climate model, station location, and emission scenario. The 30 data points were averaged to estimate the annual number of days at or above each high temperature threshold for each climate model, station location, and emission scenario. The mean and standard deviation was then determined across the climate models for each station location and emission scenario. The process was repeated to obtain seasonal projections. The maximum consecutive days of high temperature for each threshold was likewise calculated. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region.	Main Body – Figures 18, 19, 20, 21Appendix C – Tables 33, 34Appendix E – Tables 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84
Mean; 5th*, 25th, 50th, 75th, and 95th* percentile; and minimum value for the average minimum air temperature over four consecutive days in winter and the average maximum temperature over four consecutive days in summer for each 30-year time period(14 components)	Bridge, Rail	For each winter in the 30-year period, the average of the minimum air temperature for any four consecutive days for each year was estimated for each climate model projection, emission scenario, and location. The 5th, 25th, 50th, 75th, and 95th percentile; mean; and coldest period across the 30 data points was estimated using a quantile distribution for each climate model, emission scenario, and location. For each summer in the 30-year period, the average of the maximum air temperature for any four consecutive days was estimated for each year, ultimately providing the 5th, 25th, 50th, 75th, 95th percentile; mean; and hottest period across the 30 data points for each climate model, emission scenario, and location. The average across climate models for each location was determined, and then averaged across station locations to provide an average for the Mobile region.	Main Body – Figures 24, 27Appendix C – Tables 35, 36Appendix E – Tables 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98
The mean, 1st*, 5th *, 10th, and 50th percentile of the coldest day of the year during each 30-yr time period(5 components)	Multi (pavement design)	Using the daily minimum temperatures, the coldest minimum temperature for each year was identified for each climate model, emission scenario, and station location. Across the 30 data points for each time period, the mean, 1st, 5th, 10th, and 50th percentile was calculated using the quantile distribution for each climate model, emission scenario, and station location. The average across climate models for each location was determined, and then averaged across station locations to provide an average for the Mobile region.	Main Body – Figures 25, 26Appendix C – Table 36Appendix E – Tables 99, 100, 101, 102, 103
Maximum 7-day average air temperature per year with the % probability of occurrence during each 30-yr period (mean, 50th, 90th, 95th*, 99th*percentile) for each 30-yr time period(5 components)	Multi (pavement design - asphalt)	Using the daily maximum temperature, the maximum 7-day average temperature for each year was determined. This produced a total of 30 data points in each time period, for each climate model, emission scenario, and station location. Across the 30 data points, the mean, 50th, 90th, 95th, and 99th percentile was estimated using the quantile distribution for each climate model, emission scenario, and station location. The average across climate models for each location was determined, and then averaged across station locations to provide an average for the Mobile region.	Appendix C – Table 35Appendix E – Tables 104, 105, 106, 107, 108
Precipitation
Annual, seasonal, and monthly total precipitation for each 30-year time period(3 components)	Multi (pavement design)	Daily precipitation corresponding to each month, season, or year was summed for each year, station location, climate model, and emission scenario. Then the 30-year average of each sum was determined. Averages and standard deviations were calculated across climate models for each station location and emission scenario. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region.	Main Body – Figures 32, 33Appendix C – Table 37 , Table 38Appendix E – Tables 109, 110, 111
Precipitation for 24-hour period with a 0.2%*, 1%*, 2%, 5%, 10%, 20%, and 50% probability of occurrence(7 components)	Multi (drainage, liquid storage)	The day with the maximum total daily precipitation for each year was found for each emission scenario, climate model, and station location. This produced a total of 30 data points for each time period. Across the 30 data points, the daily precipitation representing each probability of occurrence was estimated by fitting the 30 data points to the Gumbel extreme value distribution for each emission scenario, climate model, and station location. Averages and standard deviations were calculated across climate models for each station location and emission scenario. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region.	Main Body – Figures 36, 37Appendix C – Table 40Appendix E – Tables 112, 113, 114, 115, 116, 117, 118
Occurrence of precipitation for 24-hour period based on today's 0.2%*, 1%*, 2%, 5%, 10%, 20%, and 50% occurrence probabilities(7 components)	Multi (drainage)	For the 1980 to 2009 time period, the value of the occurrence probabilities using the maximum total daily precipitation was identified using the results of the variable above for each climate model, emission scenario, and station location. For each of the future time periods, the day with the maximum total daily precipitation for each year was found for each emission scenario, climate model, and station location. This produced a total of 30 data points Across these 30 data points, the occurrence probabilities were determined by applying a Gumbel extreme value distribution. These fitted distributions provided the new probabilities associated with the historical value of each baseline occurrence probabilities . Averages and standard deviations were calculated across climate models for each station location and emission scenario. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region.	Appendix C – Table 41Appendix E – Tables 119, 120, 121, 122, 123, 124, 125
Exceedance probability of precipitation across four consecutive days: 0.2%, 1%, 2%, 5%, 10%, 20%, 50%;Exceedance probability of precipitation across two consecutive days: 0.2%, 1%, 2%, 5%, 10%, 20%, 50%(14 components)	Pipeline	For each time period, a sum of daily precipitation was calculated for every four consecutive days. This produced a total of 10,950 data points. The data was ranked from high to low, and the exceedance probabilities of 0.2%, 1%, 2%, 5%, 10%, 20%, and 50% were then determined for each climate model, emission scenario, and station location. Averages and standard deviations were calculated across climate models for each station location and emission scenario. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region. This was repeated for the two-day exceedance probabilities.	Main Body – Figures 38, 39Appendix C – Table 42Appendix E – Tables 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139
Largest three-day total precipitation each season(1 component)	Multi	The maximum three-day total precipitation for each season was identified for each year. This produced 30 data points for each of the four seasons. The 30 data points were averaged to produce the average maximum three-day total for each season. For purposes of discussion, the results were averaged across station locations to produce an average for the Mobile region	Main Body – Figures 34, 35Appendix C – Table 39Appendix E – Table 140

The occurrence probabilities of the 24-hour precipitation variable were estimated by fitting the Gumbel Extreme Value (GEV) distribution to the maximum daily precipitation projected for each year in the 30 year period for each climate model, emission scenario, and station location.²² As the extreme events are not based on 100 years or more of data, there is a large amount of uncertainty associated with these extreme event projections.²³ Precipitation events at a finer temporal scale (e.g., 3-hour events) are not investigated in this study as the downscaling approach applied is only appropriate to the daily time scale.

The two-day, three-day, and four-day precipitation events were provided by Dr. Katharine Hayhoe. The two-day and four-day precipitation events were developed by applying a quantile distribution²⁴ to the running sum for each 30 year period (e.g., this method does not fit the data to a theoretical distribution but, in essence, "bins" the modeled data into percentiles). The seasonal three-day events are simply the climate model ensemble average of the heaviest three-day event for each season.

C.3.2. Identification of Statistically Significant Climate Projections

To focus the study on climate projections that represent a robust projected change from baseline conditions, a statistical test (apaired t-test) was used to identify significant (p<0.05) changes, i.e., climate projections that are statistically different from simulations of today's climate. The statistical test compared the projected climate model mean for each climate variable, emission scenario, station location, and future time period to the corresponding climate model baseline of 1980 to 2009. Other statistical tests are available which require additional analyses. The paired t-test was chosen for the following reasons:

• It is simple to apply (i.e., replicable) and explain.
• It is accepted by the statistical community and reasonable.
• It requires no additional data processing to complete.
• It is applicable for the limited sample size used in this study (i.e., 4 to 10 climate models).

The climate variables demonstrating the most robust trends were identified using the following two-step process:

Step 1. A paired t-test was applied to all projections for each station location. The climate variables that show a statistically significant change from the baseline at the 95% confidence level (p<0.05) were identified.²⁵

Step 2. The climate variables that demonstrate a statistically significant change (as determined in Step 1) at all five station locations were identified. For purposes of this study, all other climate variables were not considered to demonstrate robust statistically significant differences.

This test helps identify which of the climate projections show a significant amount of change. The paired t-test identifies whether the projected case is significantly different from the baseline case. This statistical test assumes that the distribution of differences between climate model averages of projected simulations and baseline simulations is a Gaussian distribution (this is not saying the temperature or precipitation variable itself follows a Gaussian distribution, but is saying the distribution of climate model mean differences of a temperature or precipitation variable is Gaussian). The hypothesis is that the true mean difference between the baseline case and projected case for a given climate variable is zero (this is rejected if the difference between the baseline case and projected case for a given climate variable is significant, thereby suggesting this variable should be considered in the climate change vulnerability/risk assessments). This test assumes the climate models represent a sample or subset of the entire population of climate model simulations (e.g., changes in initialization, small physical variations within models, etc.).

The algorithm for the paired t-test is:

where is the mean of the baseline case (e.g., A1 emission scenario, 1980-2009), is the mean of the projected case (e.g., A1 emission scenario, 2040-2069 time period), Sd is the sample standard deviation of the differences between the baseline and projected cases, and n is the total number of climate models (i.e., 10 climate models for A1 and B2 emission scenario, and 4 climate models for the A1Fi emission scenario).

The hypothesis can be rejected with 95% confidence if:

where K is the 97.5th percentile of a t distribution with n – 1 degrees of freedom, i.e. 2.262 for 10 climate models and 3.182 for 4 climate models. This is a two-tailed test because temperature variables get warmer for all emission scenarios, but precipitation may be projected to get drier or wetter.

For an example, the table below shows the application of the paired t-test of the annual total precipitation for the B1 emission scenario for the Mobile station location. The baseline case is the average total annual precipitation across 1980 to 2009, and the projected case is the average total annual precipitation across 2010 to 2039.

Table 29: Application of the Paired T-Test of the Annual Total Precipitation for the B1 Emission Scenario for the Mobile Station Location

Total Annual Precipitation	B1
Climate Model	1980-2009	2010-2039	Diff	Sqr Diff
BCM2	67.93	59.47	-8.46	71.57
CCSM3	64.7	71.02	6.32	39.94
CGCM3-T47	62.78	74.55	11.77	138.5
CGCM3-T63	63.13	63.9	0.77	0.593
CNRM	64.32	68.28	3.96	15.68
ECHAM5	66.53	78.56	12.03	144.7
GFDL-CM2.0	64.91	70.2	5.29	27.98
GFDL-CM2.1	62.04	68.6	6.56	43.03
HadCM3	61.97	60.45	-1.52	2.31
PCM	65.87	70.3	4.43	19.62
Mean of the climate ensemble	64.42	68.53	4.12
Std dev of the climate ensemble	1.98	5.94	6.10
Numerator of t test				4.12
Std dev of the t test				7.48
Reject?				No

The steps in applying the paired t-test are as follows (note that some rounding errors may be evident in the numbers):

1. Calculate the climate model ensemble mean for the baseline case and for the projected case. In this example, the values are 64.42 inches and 68.53 inches. The absolute value of the difference, 4.12 inches, between these ensemble means provides the left hand side of equation above (second equation).
2. Calculate the differences between the projected case and baseline case for each climate model. In this example, the difference between the projected and baseline case for the HadCM3 is a reduction in precipitation of 1.52 inches.
3. Calculate the standard deviation of the differences across all climate models.
4. Square the standard deviation of the differences and divide by the number of climate models. In this example, there are 10 climate models.
5. Take the square root of the value calculated in (4) and multiply by 2.262 as 10 climate models were used (or 3.182 if 4 climate models were used). In this example, the value is 4.36 inches. This step provides the value of the right hand side of equation.
6. Check if the value of (1) is greater than the value of (5). If so, then the projected case is statistically significant from the baseline case and this variable should be considered in assessments.

The paired t-test is applied to all temperature and precipitation variables across all five station locations, three emission scenarios, and three projected time periods. This test accounts for not only the amount of change projected by the mean averaged across all the climate models but also the variability across the climate models.

C.3.3. Determination of Reporting Format

As this study produced voluminous amounts of projected data, it was important to present the findings in a way that will be useful for the vulnerability assessment to be conducted in later stages of the Gulf Coast Study. Though projections are available for the five station locations from the downscaling process, the vulnerability assessment can initially draw from regional climate projections that are averaged across all five locations. Appendices 5.7 and 5.8 provide a complete database of the climate projections by emission scenario, location, and time period.

Comparing results that have been averaged for each emission scenario at each future time period illustrates some of the uncertainty that is associated with future development pathways and resulting greenhouse gas emissions. As discussed in Appendix C.2 , the climate model ensemble average is considered the most robust design by the scientific modeling community for informing assessments. As a result, the climate projections for each emission scenario are provided as an average across all downscaled climate models. The range across the climate models is also provided, describing an important component of the model uncertainty. During the vulnerability assessment, changes will be assessed to determine which have a magnitude large enough to impact the asset.

Due to the large number of climate variables and projections, only those variables with a significant change projected (based on the paired t-test described above) were illustrated and discussed. This helped to focus the illustrations and discussion on variables that should be considered in the climate change impact assessment of Task 3.

The paired t-test was applied to all provided USGS climate projections. The results of the paired t-test along with the magnitude and direction of the projected change are presented in tables in Appendix C.5 and Appendix C.6.

• Shaded cells with the letter "Y" indicate projections that exhibit a significant change.
• Cells with grayed-out font and the letter "N" indicate projections that do not exhibit a significant change. These projections are not considered different from baseline conditions.

The projected changes in environmental variables are also illustrated by box plots. Each box on the plot shows the mean (represented by the line between the two colors) and variability (represented by the height of the box) of climate projections for each time period and emission scenario, averaged across all five stations and climate model ensemble. The variability of projections under the low (B1) and moderately-high (A2) emission scenarios is estimated as one standard deviation from the mean. The variability of projections under the high (A1FI) emission scenario is estimated as the full range across all climate models at all five stations. The "model baseline" is the average daily temperature from 1980 to 2009, as modeled by all climate models and averaged across emission scenarios. There is negligible difference between the baseline projections across emission scenarios, and this line is functionally equivalent to the observed average temperature over that time period. The simulated baseline is used to determine projected changes, as this helps correct for any preexisting biases in the climate models.

C.4. Comparing End-of-Century Temperature and Precipitation Projections by Climate Model

As discussed in Appendix C.2 , climate models project varying levels of temperature and precipitation change. This appendix provides a high-level investigation of how the end-of-century projections vary by climate model and emission scenario. The findings are helpful in loosely guiding the interpretation of the temperature and precipitation projections in this report.

The scatterplot shown in Figure 90 explores the relationship between the change in total annual precipitation and the change in mean annual temperature at end-of-century, relative to baseline conditions. The projections are provided by climate model and emission scenario, averaged across all five station locations.

Figure 90 : Projected Changes in Temperature and Precipitation by Climate Model and Emission Scenario, Changes by End-of-Century (2070-2099) Relative to Baseline (1980-2009)

Figure 90 suggests that the models show an inverse relationship, where increasing temperatures are associated with decreasing precipitation. Drawing upon this qualitative observation, the Pearson's correlation coefficient and the Spearman rank correlation coefficient were computed where a result of -1 would indicate a one-to-one relationship between increasing temperature and decreasing precipitation. These computations indicate that the relationship is not particularly strong, with a Pearson's correlation coefficient of -0.4 and a Spearman rank correlation coefficient of -0.3. However, both coefficients suggest a tendency towards an inverse relationship, supporting the qualitative observation. Additional analysis with 'finer-tuned' temperature and precipitation variables (e.g., with extreme outliers removed) could reveal a stronger relationship.

Figure 90 also illustrates some interesting patterns by emission scenario. Climate model simulations driven by the low (B1) emission scenario project the smallest increase in annual mean temperature but the greatest increase in total annual precipitation. As described in Section 2.5.2, the simulations driven by the low (B1) emission scenario uniquely project a statistically significant change in precipitation compared to the baseline. Climate model simulations driven by the moderately-high (A2) emission scenario project a warmer world, but do not project any statically significant change in precipitation. In fact, almost as many climate models project an increase in precipitation as a decrease in precipitation. Climate model simulations driven by the high (A1FI) emission scenario suggest the greatest increase in temperature, but also do not project a statistically significant change in precipitation. Note that comparing the projections associated with the high (A1FI) emission scenario to the climate projections associated with the moderately-high (A2) and low (B1) emission scenarios may be misleading, as the high (A1FI) simulations are only informed by four climate models.

Some climate models exhibit a tendency towards wetter or warmer projections. For example, the CCSM3 model consistently projects a warmer and wetter climate than the climate model ensemble mean. The climate projections in this report are provided using the climate model ensemble mean with the variability across climate models provided for each emission scenario and averaged across all five station locations.

C.5. Summary Tables for Projected Temperature Analysis

This appendix contains summary tables corresponding to the projected temperature analysis described in Section 2.4.2. Please note that shaded cells with the letter "Y" indicate statistically significant changes. Cells with grayed-out font and the letter "N" indicate projections that do not exhibit a statistically significant change. These projections are not considered different from baseline conditions. Though this test accounts for both the amount of change projected by the mean averaged across all the climate models and the variability across the climate models, the table only provides the change in ensemble mean and does not describe the change in variability (see associated figures for an illustrative description of the change in mean and variability). The following tables are included in this appendix:

• Table 30: Projected Change in the Average Annual Temperatures (°F) from the Model Baseline (1980-2009), Averaged Across All Five Stations
• Table 31: Projected Change in the Average Seasonal Temperatures (°F) Relative to the Model Baseline (1980-2009), Averaged Across All Five Stations
• Table 32: Projected Change in Average Monthly Temperatures (°F) Relative to Model Baseline (1980-2009), Averaged Across All Five Stations
• Table 33: Increase in Projected Heat Events Relative to Model Baseline (1980-2009), Averaged Across All Five Stations
• Table 34: Projected Change in Seasonal Heat Events Relative to Model Baseline (1980-2009), Averaged Across All Five Stations
• Table 35: Projected Change in Extreme Heat Events Relative to Model Baseline (1980-2009), Averaged Across All Five Stations
• Table 36: Projected Change in Extreme Cold Events Compared to Model Baseline (1980-2009), Averaged Across All Five Stations

Table 30 : Projected Change in the Average Annual Temperatures (°F) from the Model Baseline (1980-2009), Averaged Across All Five Stations.

Projections representing a significant change are highlighted and marked with a "Y".

	1980-2009 (°F)	2010-2039 (Δ°F)			2040-2069 (Δ°F)			2070-2099 (Δ°F)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Average Annual Mean Temperature	66.6	1.4 (Y)	1.4 (Y)	1.7 (Y)	2.4 (Y)	3.5 (Y)	4.6 (Y)	3.2 (Y)	6.6 (Y)	7.7 (Y)
Average Annual Minimum Temperature	56.2	1.5 (Y)	1.6 (Y)	2.0 (Y)	2.6 (Y)	3.9 (Y)	5.5 (Y)	3.5 (Y)	7.5 (Y)	9.2 (Y)
Average Annual Maximum Temperature	77.0	1.3 (Y)	1.3 (Y)	1.3 (Y)	2.2 (Y)	3.1 (Y)	3.8 (Y)	2.9 (Y)	5.8 (Y)	6.3 (Y)

Table 31 : Projected Change in the Average Seasonal Temperatures (°F) Relative to the Model Baseline (1980-2009), Averaged Across All Five Stations

Projections representing a significant change are highlighted and marked with a "Y".

	1980-2009 (°F)	2010-2039 (Δ°F)			2040-2069 (Δ°F)			2070-2099 (Δ°F)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Average Seasonal Mean Temperature
Winter	52.0	1.0 (Y)	1.3 (Y)	1.2 (Y)	2.2 (Y)	3.0 (Y)	3.3 (Y)	2.9 (Y)	5.7 (Y)	5.7 (Y)
Spring	66.3	1.3 (Y)	1.0 (Y)	1.7 (Y)	2.1 (Y)	3.0 (Y)	4.7 (Y)	2.8 (Y)	6.0 (Y)	7.8 (Y)
Summer	80.4	1.5 (Y)	1.5 (Y)	1.8 (Y)	2.4 (Y)	3.8 (Y)	5.5 (Y)	3.0 (Y)	6.9 (Y)	9.0 (Y)
Fall	68.0	1.8 (Y)	1.9 (Y)	1.9 (Y)	2.9 (Y)	4.2 (Y)	5.1 (Y)	4.2 (Y)	7.9 (Y)	8.5 (Y)
Average Seasonal Maximum Temperature
Winter	62.7	1.0 (Y)	1.3 (Y)	1.1 (Y)	2.2 (Y)	3.0 (Y)	3.2 (Y)	3.0 (Y)	5.7 (Y)	5.6 (Y)
Spring	77.2	1.1 (Y)	0.9 (Y)	1.5 (Y)	1.8 (Y)	2.6 (Y)	4.2 (Y)	2.5 (Y)	5.3 (Y)	6.6 (Y)
Summer	89.7	1.3 (Y)	1.3 (Y)	1.3 (Y)	2.0 (Y)	3.2 (Y)	4.2 (Y)	2.5 (Y)	5.7 (Y)	6.7 (Y)
Fall	78.7	1.7 (Y)	1.6 (Y)	1.4 (Y)	2.6 (Y)	3.5 (Y)	3.8 (Y)	3.7 (Y)	6.5 (Y)	6.4 (Y)
Average Seasonal Minimum Temperature
Winter	41.3	0.9 (Y)	1.3 (Y)	1.4 (Y)	2.2 (Y)	3.2 (Y)	3.3 (Y)	2.8 (Y)	5.8 (Y)	5.7 (Y)
Spring	55.4	1.5 (Y)	1.2 (Y)	1.9 (Y)	2.4 (Y)	3.3 (Y)	5.3 (Y)	3.2 (Y)	6.6 (Y)	8.9 (Y)
Summer	71.2	1.7 (Y)	1.7 (Y)	2.3 (Y)	2.7 (Y)	4.3 (Y)	6.9 (Y)	3.5 (Y)	8.1 (Y)	11.4 (Y)
Fall	57.3	1.9 (Y)	2.2 (Y)	2.5 (Y)	3.2 (Y)	4.9 (Y)	6.4 (Y)	4.7 (Y)	9.3 (Y)	10.5 (Y)

Table 32: Projected Change in Average Monthly Temperatures (°F) Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

Projections representing a significant change are highlighted and marked with a "Y"

	1980-2009 (°F)	2010-2039 (Δ°F)			2040-2069 (Δ°F)			2070-2099 (Δ°F)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Average Monthly Mean Temperature
Jan	50.4	1.0 (Y)	1.6 (Y)	1.3 (N)	2.0 (Y)	3.7 (Y)	3.0 (Y)	2.8 (Y)	6.5 (Y)	5.5 (Y)
Feb	53.5	0.6 (N)	0.8 (N)	0.8 (Y)	1.8 (Y)	2.1 (Y)	3.3 (Y)	2.0 (Y)	4.9 (Y)	5.2 (Y)
Mar	59.7	1.6 (Y)	1.0 (Y)	1.6 (N)	2.0 (Y)	2.6 (Y)	4.2 (Y)	2.7 (Y)	5.5 (Y)	6.8 (Y)
Apr	65.7	1.1 (Y)	1.1 (Y)	1.5 (Y)	2.1 (Y)	2.9 (Y)	4.6 (Y)	2.8 (Y)	6.0 (Y)	7.6 (Y)
May	73.4	1.3 (Y)	1.1 (Y)	2.0 (Y)	2.2 (Y)	3.4 (Y)	5.3 (Y)	3.0 (Y)	6.8 (Y)	8.9 (Y)
Jun	79.1	1.4 (Y)	1.3 (Y)	1.7 (Y)	2.2 (Y)	3.5 (Y)	5.3 (Y)	2.7 (Y)	6.6 (Y)	8.7 (Y)
Jul	81.3	1.5 (Y)	1.5 (Y)	1.9 (Y)	2.3 (Y)	3.9 (Y)	5.7 (Y)	3.0 (Y)	7.1 (Y)	9.1 (Y)
Aug	81.0	1.6 (Y)	1.8 (Y)	2.0 (Y)	2.6 (Y)	4.0 (Y)	5.6 (Y)	3.3 (Y)	7.3 (Y)	9.3 (Y)
Sept	77.1	1.8 (Y)	1.8 (Y)	2.1 (Y)	2.7 (Y)	4.3 (Y)	5.3 (Y)	4.0 (Y)	7.6 (Y)	9.1 (Y)
Oct	67.9	1.8 (Y)	2.3 (Y)	2.2 (Y)	2.8 (Y)	4.7 (Y)	5.8 (Y)	4.3 (Y)	8.9 (Y)	9.4 (Y)
Nov	59.2	1.7 (Y)	1.5 (Y)	1.5 (Y)	3.3 (Y)	3.6 (Y)	4.1 (Y)	4.2 (Y)	7.3 (Y)	6.8 (Y)
Dec	52.3	1.3 (Y)	1.6 (Y)	1.6 (Y)	2.7 (Y)	3.5 (Y)	3.5 (Y)	3.8 (Y)	6.3 (Y)	6.3 (Y)
Average Monthly Maximum Temperature
Jan	61.0	0.9 (Y)	1.5 (Y)	1.0 (N)	2.0 (Y)	3.4 (Y)	3.0 (Y)	2.8 (Y)	6.1 (Y)	5.5 (Y)
Feb	64.5	0.6 (N)	0.8 (N)	0.8 (Y)	1.7 (Y)	2.0 (Y)	3.3 (Y)	2.2 (Y)	4.9 (Y)	5.3 (Y)
Mar	70.9	1.3 (Y)	1.0 (Y)	1.5 (N)	1.8 (Y)	2.5 (Y)	3.9 (Y)	2.5 (Y)	5.2 (Y)	6.2 (Y)
Apr	76.9	1.0 (Y)	0.9 (Y)	1.5 (N)	1.7 (Y)	2.5 (Y)	4.1 (Y)	2.5 (Y)	5.1 (Y)	6.6 (Y)
May	83.8	1.0 (Y)	0.8 (Y)	1.4 (Y)	1.8 (Y)	2.8 (Y)	4.4 (Y)	2.4 (Y)	5.6 (Y)	7.1 (Y)
Jun	88.5	1.1 (Y)	1.0 (Y)	1.3 (N)	1.8 (Y)	2.8 (Y)	4.2 (Y)	2.1 (Y)	5.1 (Y)	6.7 (Y)
Jul	90.3	1.3 (Y)	1.3 (Y)	1.4 (Y)	2.0 (Y)	3.3 (Y)	4.2 (Y)	2.6 (Y)	5.8 (Y)	6.6 (Y)
Aug	90.1	1.4 (Y)	1.6 (Y)	1.3 (N)	2.1 (Y)	3.3 (Y)	4.0 (Y)	2.8 (Y)	6.0 (Y)	6.7 (Y)
Sept	86.8	1.6 (Y)	1.4 (Y)	1.3 (Y)	2.2 (Y)	3.4 (Y)	3.5 (Y)	3.3 (Y)	6.0 (Y)	6.4 (Y)
Oct	78.9	1.7 (Y)	1.8 (Y)	1.6 (Y)	2.6 (Y)	3.8 (Y)	4.2 (Y)	3.7 (Y)	7.1 (Y)	6.8 (Y)
Nov	70.4	1.7 (Y)	1.5 (Y)	1.3 (Y)	3.1 (Y)	3.3 (Y)	3.6 (Y)	4.0 (Y)	6.5 (Y)	6.1 (Y)
Dec	62.9	1.4 (Y)	1.7 (Y)	1.4 (Y)	2.7 (Y)	3.5 (Y)	3.4 (Y)	4.0 (Y)	6.0 (Y)	6.0 (Y)
Average Monthly Minimum Temperature
Jan	39.8	1.0 (Y)	1.6 (Y)	1.5 (N)	2.0 (Y)	3.7 (Y)	3.1 (Y)	2.7 (Y)	6.3 (Y)	5.5 (Y)
Feb	42.6	0.5 (N)	0.6 (N)	0.8 (Y)	1.8 (Y)	2.2 (Y)	3.2 (Y)	1.9 (Y)	4.7 (Y)	5.2 (Y)
Mar	48.5	1.8 (Y)	1.1 (Y)	1.7 (N)	2.2 (Y)	2.8 (Y)	4.5 (Y)	2.9 (Y)	5.6 (Y)	7.3 (Y)
Apr	54.6	1.2 (Y)	1.2 (Y)	1.6 (Y)	2.4 (Y)	3.2 (Y)	5.1 (Y)	3.1 (Y)	6.6 (Y)	8.7 (Y)
May	63.0	1.6 (Y)	1.3 (Y)	2.5 (Y)	2.6 (Y)	3.9 (Y)	6.2 (Y)	3.7 (Y)	7.7 (Y)	10.8 (Y)
Jun	69.7	1.6 (Y)	1.5 (Y)	2.0 (Y)	2.6 (Y)	3.9 (Y)	6.4 (Y)	3.2 (Y)	7.5 (Y)	10.7 (Y)
Jul	72.2	1.7 (Y)	1.6 (Y)	2.4 (Y)	2.6 (Y)	4.5 (Y)	7.1 (Y)	3.5 (Y)	8.3 (Y)	11.6 (Y)
Aug	71.8	1.8 (Y)	1.9 (Y)	2.6 (Y)	3.0 (Y)	4.7 (Y)	7.2 (Y)	3.8 (Y)	8.6 (Y)	11.9 (Y)
Sept	67.4	2.1 (Y)	2.2 (Y)	2.9 (Y)	3.3 (Y)	5.1 (Y)	7.0 (Y)	4.7 (Y)	9.3 (Y)	11.8 (Y)
Oct	57.0	1.9 (Y)	2.8 (Y)	2.7 (Y)	3.0 (Y)	5.7 (Y)	7.3 (Y)	4.9 (Y)	10.6 (Y)	12.0 (Y)
Nov	47.9	1.8 (N)	1.6 (Y)	1.8 (Y)	3.4 (Y)	3.9 (Y)	4.7 (Y)	4.4 (Y)	7.9 (Y)	7.6 (Y)
Dec	41.8	1.2 (Y)	1.6 (Y)	1.7 (Y)	2.7 (Y)	3.5 (Y)	3.5 (Y)	3.6 (Y)	6.2 (Y)	6.5 (Y)

Table 33 : Increase in Projected Heat Events Relative to Model Baseline (1980-2009), Averaged Across All Five Stations.

Projections representing a significant change are highlighted and marked with a "Y"

	1980-2009 (days/°F)	2010-2039 (Δ/Δ°F)			2040-2069 (Δ/Δ°F)			2070-2099 (Δ/Δ°F)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
# days per year above 95°F	9.6	8 (Y)	8 (Y)	9 (Y)	14 (Y)	28 (Y)	38 (Y)	21 (Y)	64 (Y)	76 (Y)
# days per year above 100°F	0.6	1 (N)	0 (N)	0 (N)	1 (N)	4 (Y)	6 (Y)	3 (N)	18 (Y)	20 (Y)
# days per year above 105°F	0.0	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	2 (N)	1 (N)
# days per year above 110°F	0.0	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)
Longest # of consecutive days above 95°F	3.9	3 (Y)	3 (Y)	2 (N)	6 (Y)	12 (Y)	15 (N)	9 (Y)	30 (Y)	34 (N)
Longest # of consecutive days above 100°F	0.4	0 (Y)	0 (N)	0 (N)	1 (N)	2 (Y)	3 (Y)	2 (N)	8 (Y)	8 (Y)
Longest # of consecutive days above 105°F	0.0	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	2 (N)	1 (Y)
Longest # of consecutive days above 110°F	0.0	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)	0 (N)
Hottest Day of the Year (°F)
Mean	97.0	1.4 (Y)	1.4 (Y)	1.4 (Y)	2.2 (Y)	3.8 (Y)	4.4 (Y)	3.0 (Y)	6.9 (Y)	7.0 (Y)
50th Percentile	96.8	1.4 (Y)	1.6 (Y)	1.5 (Y)	2.3 (Y)	4.0 (Y)	4.3 (Y)	3.1 (Y)	7.1 (Y)	6.8 (Y)
95th Percentile	101.3	1.5 (Y)	1.1 (N)	0.8 (N)	1.6 (Y)	3.9 (Y)	4.3 (Y)	2.7 (Y)	7.2 (Y)	7.0 (Y)
Maximum	102.8	2.2 (Y)	0.8 (N)	0.6 (N)	1.7 (Y)	3.9 (Y)	4.5 (Y)	2.9 (Y)	7.2 (Y)	7.9 (Y)

Table 34 : Projected Change in Seasonal Heat Events Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

Projections representing a significant change are highlighted and marked with a "Y"

	1980-2009 (days)	2010-2039 (Δ)			2040-2069 (Δ)			2070-2099 (Δ)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Number of Days Above 95°F
Winter	0.0	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.1 (N)	0.0 (N)
Spring	0.0	0.3 (N)	0.2 (N)	0.5 (N)	0.5 (N)	1.4 (N)	2.7 (N)	0.6 (N)	5.3 (Y)	6.9 (N)
Summer	9.0	6.7 (Y)	6.2 (Y)	8.2 (Y)	11.9 (Y)	22.8 (Y)	31.5 (Y)	16.5 (Y)	45.0 (Y)	56.3 (Y)
Fall	0.8	1.2 (N)	1.2 (Y)	0.7 (N)	2.2 (N)	4.2 (Y)	3.6 (N)	4.5 (N)	13.4 (Y)	13.0 (Y)
Number of Days Above 100°F
Winter	0.0	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)
Spring	0.0	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.1 (N)	0.3 (N)	0.0 (N)	0.6 (N)	0.9 (N)
Summer	0.6	0.7 (N)	0.4 (N)	0.4 (N)	1.2 (N)	3.5 (Y)	4.9 (N)	2.0 (N)	13.9 (Y)	17.0 (Y)
Fall	0.0	0.1 (N)	0.1 (N)	0.0 (N)	0.2 (N)	0.2 (N)	0.4 (N)	0.8 (N)	2.8 (N)	2.3 (Y)
Longest Number of Consecutive Days Above 95°F
Winter	0.0	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.1 (N)	0.0 (N)
Spring	0.1	0.2 (N)	0.1 (N)	0.3 (N)	0.3 (N)	0.9 (Y)	1.9 (N)	0.4 (Y)	3.6 (Y)	4.9 (N)
Summer	3.8	3.0 (Y)	2.6 (Y)	2.5 (Y)	5.2 (Y)	10.8 (Y)	14.7 (Y)	7.4 (Y)	25.7 (Y)	31.2 (Y)
Fall	0.5	0.8 (N)	0.8 (Y)	0.5 (N)	1.3 (N)	2.6 (Y)	2.4 (N)	2.9 (N)	8.4 (Y)	8.2 (Y)
Longest Number of Consecutive Days Above 100°F
Winter	0.0	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)
Spring	0.0	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.1 (N)	0.2 (N)	0.0 (N)	0.6 (N)	0.6 (N)
Summer	0.4	0.4 (Y)	0.3 (N)	0.3 (N)	0.7 (N)	2.0 (Y)	2.7 (Y)	1.2 (Y)	7.6 (Y)	8.1 (Y)
Fall	0.0	0.1 (N)	0.1 (N)	0.1 (N)	0.2 (N)	0.2 (N)	0.4 (N)	0.7 (N)	1.8 (N)	1.6 (Y)

Table 35 : Projected Change in Extreme Heat Events Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

Projections representing a significant change are highlighted and marked with a "Y"

	1980-2009 (°F)	2010-2039 (Δ°F)			2040-2069 (Δ°F)			2070-2099 (Δ°F)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Hottest Week of the Year
Mean	94.4	1.4 (Y)	1.3 (Y)	1.3 (Y)	2.1 (Y)	3.6 (Y)	4.3 (Y)	3.0 (Y)	6.7 (Y)	6.8 (Y)
50th Percentile	94.2	1.3 (Y)	1.4 (Y)	1.4 (Y)	2.2 (Y)	3.8 (Y)	4.2 (Y)	3.0 (Y)	6.6 (Y)	6.7 (Y)
90th percentile	97.2	1.4 (Y)	1.1 (Y)	1.2 (N)	1.6 (Y)	3.7 (Y)	4.4 (Y)	2.4 (Y)	6.9 (Y)	6.9 (Y)
95th Percentile	98.5	1.5 (Y)	1.2 (Y)	0.8 (N)	1.7 (Y)	3.7 (Y)	4.4 (Y)	2.8 (Y)	7.0 (Y)	7.1 (Y)
99th percentile	99.7	1.8 (Y)	1.2 (Y)	0.7 (N)	1.6 (Y)	3.7 (Y)	4.4 (Y)	3.0 (Y)	7.1 (Y)	7.8 (Y)
Warmest Four Days in Summer
Mean	84.1	1.3 (Y)	1.3 (Y)	1.3 (Y)	2.0 (Y)	3.2 (Y)	4.1 (Y)	2.5 (Y)	5.6 (Y)	6.6 (Y)
5th percentile	87.6	1.3 (Y)	1.5 (Y)	1.4 (N)	2.3 (Y)	3.1 (Y)	4.6 (Y)	2.9 (Y)	5.5 (Y)	7.0 (Y)
25th percentile	89.7	1.3 (Y)	1.4 (Y)	1.4 (Y)	1.9 (Y)	2.9 (Y)	4.0 (Y)	2.4 (Y)	5.2 (Y)	6.4 (Y)
50th percentile	91.7	1.2 (Y)	1.3 (Y)	1.4 (Y)	1.9 (Y)	3.0 (Y)	3.9 (Y)	2.4 (Y)	5.4 (Y)	6.5 (Y)
75th percentile	95.0	1.2 (Y)	1.3 (Y)	1.6 (Y)	2.0 (Y)	3.4 (Y)	4.3 (Y)	2.5 (Y)	5.9 (Y)	6.7 (Y)
95th percentile	89.7	1.2 (Y)	1.1 (Y)	1.4 (Y)	1.9 (Y)	3.5 (Y)	4.2 (Y)	2.5 (Y)	6.5 (Y)	6.7 (Y)
Warmest summer in 30 years	100.8	1.8 (Y)	0.9 (N)	0.2 (N)	1.5 (Y)	3.5 (Y)	3.7 (Y)	2.7 (Y)	7.0 (Y)	7.6 (Y)

Table 36 : Projected Change in Extreme Cold Events Compared to Model Baseline (1980-2009), Averaged Across All Five Stations

Projections representing a significant change are highlighted and marked with a "Y"

	1980-2009 (°F)	2010-2039 (Δ°F)			2040-2069 (Δ°F)			2070-2099 (Δ°F)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Coldest Day of the Year
Mean	18.9	1.2 (N)	1.6 (Y)	2.7 (Y)	2.8 (Y)	3.3 (Y)	4.7 (Y)	2.7 (Y)	6.1 (Y)	6.8 (Y)
1st percentile	4.2	1.2 (N)	3.9 (N)	5.7 (Y)	4.0 (N)	4.1 (N)	8.2 (N)	2.9 (N)	8.6 (Y)	9.2 (Y)
5th percentile	7.9	2.4 (Y)	3.0 (N)	5.6 (Y)	4.0 (Y)	5.2 (Y)	8.5 (Y)	3.1 (Y)	9.0 (Y)	11.0 (Y)
10th percentile	8.9	3.4 (Y)	2.2 (Y)	5.6 (Y)	4.9 (Y)	4.9 (Y)	8.0 (N)	4.4 (Y)	7.7 (Y)	11.0 (Y)
50th percentile	20.3	0.7 (N)	1.0 (N)	1.9 (N)	2.3 (Y)	2.7 (Y)	3.8 (Y)	2.2 (Y)	5.6 (Y)	6.0 (Y)
Coldest Four Days in Winter
5th percentile	28.1	1.0 (Y)	0.9 (Y)	1.3 (N)	1.9 (Y)	2.7 (Y)	3.1 (Y)	2.3 (Y)	5.0 (Y)	5.3 (Y)
25th percentile	35.1	0.9 (Y)	1.2 (Y)	1.1 (Y)	2.0 (Y)	2.9 (Y)	2.9 (Y)	2.6 (Y)	5.4 (Y)	5.2 (Y)
50th percentile	40.9	0.8 (Y)	1.4 (Y)	1.3 (Y)	2.2 (Y)	3.2 (Y)	3.3 (Y)	2.8 (Y)	5.8 (Y)	5.9 (Y)
75th percentile	47.4	0.8 (N)	1.4 (Y)	1.2 (Y)	2.3 (Y)	3.3 (Y)	3.2 (Y)	3.0 (Y)	6.2 (Y)	6.0 (Y)
95th percentile	56.2	1.0 (N)	1.5 (Y)	1.8 (N)	2.5 (Y)	3.5 (Y)	3.4 (Y)	3.0 (Y)	6.3 (Y)	6.6 (Y)
Mean	41.3	0.9 (Y)	1.3 (Y)	1.4 (Y)	2.2 (Y)	3.2 (Y)	3.2 (Y)	2.8 (Y)	5.8 (Y)	5.8 (Y)
Coldest winter in 30 years	12.6	1.0 (N)	2.3 (N)	2.7 (N)	4.2 (Y)	3.6 (N)	6.7 (N)	2.5 (N)	8.9 (Y)	7.8 (N)

C.6. Summary Tables for Projected Precipitation Analysis

This appendix contains summary tables corresponding to the projected precipitation analysis described in Section 2.5.2. Please note that shaded cells with the letter "Y" indicate statistically significant changes. Cells with grayed-out font and the letter "N" indicate projections that do not exhibit a statistically significant change. These projections are not considered different from baseline conditions. The following tables are included in this appendix:

• Table 37: Projected Change in Total Annual Precipitation (inches) Relative to Model Baseline (1980-2009), Averaged Across All Five Stations
• Table 38: Projected Change in Total Seasonal and Monthly Precipitation (inches)
• Table 39: Projected Change in Maximum Three-Day Precipitation Totals (inches)
• Table 40: Projected Change in the Magnitude of 24-Hour Storm Events (inches)
• Table 41: Change in the Probability of Current Storms (1980-2009) Occurring in the Future
• Table 42: Projected Change in Precipitation Events (inches) by Exceedance Probability

Table 37 : Projected Change in Total Annual Precipitation (inches) Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

Projections representing a significant change are highlighted and marked with a "Y"

	1980-2009 (in.)	2010-2039 (Δ in.)			2040-2069 (Δ in.)			2070-2099 (Δ in.)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Total Annual Precipitation	65.4	3.4 (N)	3.5 (N)	4.4 (N)	6.9 (Y)	3.3 (N)	3.5 (N)	8.4 (Y)	2.0 (N)	0.6 (N)

Table 38 : Projected Change in Total Seasonal and Monthly Precipitation (inches) Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

Projections representing a significant change are highlighted and marked with a "Y"

	1980-2009 (in.)	2010-2039 (Δ in.)			2040-2069 (Δ in.)			2070-2099 (Δ in.)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Total Seasonal Precipitation
Winter	15.3	1.6 (Y)	0.9 (N)	1.7 (N)	1.7 (Y)	1.3 (N)	0.6 (N)	2.0 (N)	1.8 (N)	-0.7 (N)
Spring	15.7	-0.1 (N)	0.6 (N)	0.7 (N)	-0.2 (N)	-0.3 (N)	-0.4 (N)	1.1 (N)	-0.8 (N)	-0.7 (N)
Summer	20.2	0.9 (N)	0.8 (N)	0.0 (N)	3.2 (N)	0.8 (N)	-0.5 (N)	2.7 (N)	-1.0 (N)	-1.9 (N)
Fall	14.2	1.0 (N)	1.2 (N)	2.0 (N)	2.2 (Y)	1.6 (N)	3.8 (N)	2.6 (N)	2.0 (N)	3.9 (N)
Total Monthly Precipitation
Jan	5.5	0.7 (N)	0.3 (N)	0.0 (N)	0.6 (N)	0.4 (N)	0.1 (N)	0.8 (N)	0.9 (N)	-0.5 (N)
Feb	5.1	0.3 (N)	0.2 (N)	0.5 (N)	0.4 (N)	-0.1 (N)	0.1 (N)	0.4 (N)	0.2 (N)	0.0 (N)
Mar	5.9	0.0 (N)	-0.1 (N)	0.4 (N)	-0.1 (N)	0.0 (N)	0.2 (N)	0.4 (N)	0.0 (N)	-0.1 (N)
Apr	4.8	0.2 (N)	0.5 (N)	-0.1 (N)	0.0 (N)	-0.4 (N)	0.0 (N)	0.2 (N)	-0.4 (N)	-0.4 (N)
May	5.0	-0.3 (N)	0.2 (N)	0.4 (N)	0.0 (N)	0.0 (N)	-0.6 (N)	0.4 (N)	-0.4 (N)	-0.1 (N)
Jun	6.1	0.2 (N)	0.3 (N)	-0.7 (N)	0.6 (N)	-0.3 (N)	-0.8 (N)	0.6 (N)	-1.0 (N)	-0.4 (N)
Jul	7.7	0.5 (N)	-0.4 (N)	-1.2 (N)	1.5 (N)	0.6 (N)	-1.7 (N)	1.3 (N)	-0.3 (N)	-2.4 (N)
Aug	6.4	0.3 (N)	0.9 (N)	1.8 (N)	1.1 (N)	0.4 (N)	2.0 (N)	0.8 (N)	0.4 (N)	0.9 (N)
Sept	5.5	0.2 (N)	0.8 (N)	1.3 (Y)	1.2 (N)	0.7 (N)	2.0 (N)	1.0 (N)	1.5 (N)	2.4 (N)
Oct	3.9	0.5 (N)	0.2 (N)	0.3 (N)	0.6 (Y)	0.5 (N)	1.8 (N)	0.9 (N)	0.0 (N)	1.8 (N)
Nov	4.8	0.2 (N)	0.3 (N)	0.4 (N)	0.4 (N)	0.4 (N)	0.0 (N)	0.7 (N)	0.4 (N)	-0.3 (N)
Dec	4.7	0.6 (N)	0.5 (N)	1.2 (N)	0.7 (N)	0.9 (N)	0.4 (N)	0.8 (N)	0.7 (N)	-0.2 (N)

Table 39 : Projected Change in Maximum Three-Day Precipitation Totals (inches) Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

	1980-2009 (in.)	2010-2039 (Δ in.)			2040-2069 (Δ in.)			2070-2099 (Δ in.)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Maximum Three-Day Precipitation
Winter	3.7	0.9 (Y)	0.4 (N)	0.6 (N)	0.9 (Y)	0.7 (N)	0.8 (N)	1.1 (Y)	1.3 (Y)	0.5 (N)
Spring	4.8	0.4 (N)	0.5 (N)	0.3 (N)	0.4 (N)	0.5 (N)	0.4 (N)	0.7 (N)	0.6 (N)	0.4 (N)
Summer	4.9	0.6 (N)	0.6 (N)	0.6 (N)	1.2 (Y)	0.8 (N)	0.6 (N)	1.0 (N)	0.2 (N)	0.1 (N)
Fall	4.7	0.4 (N)	0.5 (N)	1.1 (N)	0.9 (Y)	0.7 (N)	1.2 (N)	1.2 (Y)	0.6 (N)	1.2 (N)

Table 40 : Projected Change in the Magnitude of 24-Hour Storm Events (inches) Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

	1980-2009 (in)	2010-2039 (Δ in.)			2040-2069 (Δ in.)			2070-2099 (Δ in.)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
24-Hour Precipitation Events
500-year storm	13.5	6.1 (Y)	5.2 (Y)	5.4 (N)	6.3 (Y)	6.4 (Y)	4.2 (N)	8.0 (N)	7.8 (Y)	4.3 (N)
100-year storm	13.5	4.7 (Y)	4.0 (Y)	4.2 (N)	4.9 (Y)	4.9 (Y)	3.3 (N)	6.2 (N)	6.0 (Y)	3.3 (N)
50-year storm	12.5	4.1 (Y)	3.5 (Y)	3.7 (N)	4.3 (Y)	4.3 (N)	2.9 (N)	5.4 (N)	5.2 (Y)	2.9 (N)
20-year storm	9.5	3.3 (N)	2.8 (Y)	3.0 (N)	3.5 (N)	3.5 (Y)	2.3 (N)	4.4 (N)	4.2 (Y)	2.4 (N)
10-year storm	8.5	2.7 (Y)	2.3 (Y)	2.5 (N)	2.9 (Y)	2.9 (Y)	1.9 (N)	3.6 (Y)	3.4 (Y)	1.9 (N)
5-year storm	7.1	2.0 (N)	1.8 (Y)	1.9 (N)	2.3 (Y)	2.2 (N)	1.5 (N)	2.7 (Y)	2.6 (Y)	1.5 (N)
2-year storm	4.8	1.1 (N)	0.9 (Y)	1.1 (N)	1.3 (N)	1.2 (Y)	0.8 (N)	1.5 (Y)	1.4 (Y)	0.8 (N)

Table 41 : Change in the Probability of Current Storms (1980-2009) Occurring in the Future

	1980-2009 (%)	2010-2039 (Δ%)			2040-2069 (Δ%)			2070-2099 (Δ%)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
24-Hour Precipitation Events – Change in Probability of Occurrence of Baseline Storm Event
500-year storm	0.2%	3.6% (Y)	2.6% (Y)	3.1% (N)	3.7% (Y)	3.0% (Y)	1.4% (N)	5.0% (Y)	4.5% (Y)	1.8% (N)
100-year storm	1.0%	6.7% (Y)	5.7% (Y)	5.8% (N)	7.4% (Y)	6.8% (Y)	3.7% (N)	9.5% (Y)	9.0% (Y)	4.1% (N)
50-year storm	2.0%	8.8% (Y)	7.9% (Y)	7.5% (N)	9.8% (Y)	9.5% (Y)	5.4% (N)	12.2% (N)	12.0% (Y)	5.8% (N)
20-year storm	5.0%	12.2% (Y)	11.7% (Y)	10.3% (N)	14.0% (Y)	14.0% (Y)	8.7% (N)	16.7% (Y)	16.9% (Y)	8.8% (N)
10-year storm	10.0%	14.8% (Y)	15.0% (Y)	12.7% (N)	17.4% (Y)	17.9% (Y)	11.9% (N)	20.0% (Y)	20.7% (Y)	11.4% (N)
5-year storm	20.0%	16.4% (Y)	17.4% (Y)	14.7% (N)	20.0% (Y)	20.7% (Y)	14.6% (N)	21.9% (Y)	23.0% (Y)	13.3% (N)
2-year storm	50.0%	11.9% (N)	13.6% (Y)	12.7% (N)	16.1% (Y)	16.3% (Y)	12.6% (N)	16.2% (Y)	16.8% (Y)	10.5% (N)

Table 42 : Projected Change in Precipitation Events (inches) by Exceedance Probability Relative to Model Baseline (1980-2009), Averaged Across All Five Stations

	1980-2009 (in)	2010-2039 (Δ in.)			2040-2069 (Δ in.)			2070-2099 (Δ in.)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Maximum Four-Day Precipitation
0.2%	11.3	3.7 (N)	3.9 (Y)	2.4 (N)	4.8 (Y)	4.6 (N)	3.7 (N)	4.8 (Y)	5.1 (Y)	3.0 (N)
1%	6.9	1.2 (N)	0.9 (N)	1.1 (N)	1.5 (Y)	1.3 (Y)	1.4 (Y)	2.0 (Y)	1.7 (Y)	1.2 (Y)
2%	5.3	0.5 (N)	0.5 (N)	0.7 (N)	1.0 (Y)	0.8 (Y)	1.0 (Y)	1.2 (Y)	1.0 (Y)	0.9 (Y)
5%	3.7	0.2 (N)	0.3 (Y)	0.3 (N)	0.5 (Y)	0.4 (Y)	0.4 (N)	0.6 (Y)	0.5 (Y)	0.4 (Y)
10%	2.7	0.1 (N)	0.2 (N)	0.1 (N)	0.3 (Y)	0.2 (Y)	0.2 (N)	0.3 (Y)	0.2 (N)	0.2 (N)
20%	1.7	0.1 (N)	0.1 (N)	0.1 (N)	0.1 (N)	0.1 (N)	0.1 (N)	0.2 (Y)	0.1 (N)	0.1 (N)
50%	0.7	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.1 (N)	0.0 (N)	0.0 (N)
Maximum Two-Day Precipitation
0.2%	9.3	3.6 (N)	3.3 (Y)	3.0 (N)	3.9 (Y)	4.2 (N)	3.8 (N)	4.6 (Y)	4.8 (Y)	2.4 (N)
1%	5.5	0.7 (N)	0.6 (N)	0.9 (N)	1.1 (Y)	1.0 (Y)	1.0 (Y)	1.4 (Y)	1.3 (Y)	0.8 (N)
2%	4.1	0.3 (N)	0.3 (N)	0.6 (N)	0.7 (Y)	0.6 (Y)	0.6 (Y)	0.8 (Y)	0.7 (Y)	0.5 (Y)
5%	2.8	0.1 (N)	0.1 (N)	0.1 (N)	0.3 (N)	0.2 (Y)	0.3 (N)	0.3 (Y)	0.2 (N)	0.3 (N)
10%	2.0	0.1 (N)	0.1 (N)	0.1 (N)	0.2 (Y)	0.1 (Y)	0.1 (N)	0.2 (Y)	0.1 (N)	0.1 (N)
20%	1.3	0.0 (N)	0.1 (N)	0.0 (N)	0.1 (N)	0.1 (Y)	0.0 (N)	0.1 (Y)	0.0 (N)	0.1 (N)
50%	0.4	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)	0.0 (N)

C.7. Methodology to Select Stream Gages for Historical Streamflow Analysis

This appendix discusses how five stream gages were selected for this analysis of historical streamflow, runoff, and flooding in Mobile.

Initially, the following three USGS networks of streamflow and runoff data were considered.

• USGS Water Watch program of historical and current runoff data²⁶
• USGS Hydro-Climatic Data Network
• USGS Surface-Water database of stream gage data

The USGS Water Watch program data are aggregated values across the stream gage sites located within each basin or hydrologic unit. Unlike individual stream gage data, this hydrologic unit runoff data set tends to provide long-term historical records. However, this runoff data set was not ideal for the analysis because the stream gage sites used to inform the basin values are not uniform across the period of record, thereby artificially affecting the long-term trend.

The USGS Hydro-Climatic Data Network is a subset of the USGS Surface-Water database of stream gages that have been identified to be largely unaffected by human disturbance. As such, these stream gages can be used to inform climatic analysis. This database, however, does not provide stream gages in Mobile County.

Ultimately, the USGS Surface-Water database of stream gage data was used to provide streamflow data for this analysis.

Criteria were developed to select stream gages from this database that provide a representative range of basin characteristics and stream sizes. This allows the analysis of this smaller sample set to be representative of the various streams in the study region. The selection criteria include the following requirements:

• Located in Mobile County
• USGS regional flood-frequency analysis²⁷ available
• Description of the explanatory basin characteristics of greatest relevance to predict peak flows available
• Data sets of streamflow properties (peak streamflow, annual mean discharge, monthly mean discharge) available
• Full period of record (≥20 years) available

USGS has conducted analyses determining the explanatory basin characteristics for stream gages in Alabama and summarized these findings in a series of reports.²⁸ The basin characteristics initially considered are provided in the textbox titled, "Basin Characteristics." The USGS analysis for urban streams in Alabama found that the key characteristics needed to inform estimates of peak flow include the contributing drainage area and the percentage of development in the drainage area.²⁹ A similar analysis conducted for small rural streams in Alabama found the key characteristics needed to inform estimates of peak flow include the contributing drainage area, the main channel slope, and the percentage of forest cover in the drainage area.³⁰

The application of the selection criteria outlined above began with over 390 stream gage sites in Alabama, of which approximately 14 were operating in Mobile County.³¹ Figure 91 outlines the stream gage selection process.

The selection criteria identified three stream gage sites for the analysis: Chickasaw Creek (site 02471001), Crooked Creek (site 02479980), and Hamilton Creek (site 02480002). Two sites did not meet the criteria but were included due to their unique location, size, and basin characteristics: Mobile River (site 02470630) and Fowl River (site 02471078). The Mobile River site provides only peak streamflow data while the Fowl River site period of record does not begin until 1995.

An additional check was conducted to ensure that the five sites are near high-density areas of transportation assets. Detailed information on the data available at each of these stream gage sites is shown in Table 43.

Figure 91 : Flowchart Describing the Stream Gage Selection Process

Table 43 : Streamflow and Discharge Data Available for Selected Mobile County Stream Gage Stations

Site	Site Number	Characteristics	Annual Peak Streamflow		Monthly Mean Discharge		Annual Mean Discharge
			Start	End	Start	End	Monthly	Annual
Chickasaw Creek32	02471001	Large stream; 125 mi2 (325 km2) drainage area	5/1952	5/2010	10/1951	9/2010	1952	2010
Mobile River33	02470630	Large river; 44,000 mi2 (114,400 km2) drainage area	4/1951	2/2004	X	X	X	X
Fowl River34	02471078	Urban stream; 16.5 mi2 (42.9 km2) drainage area	4/1995	1/2010	3/1995	9/2010	1995	2010
Crooked Creek35	02479980	Small rural stream; 8 mi2 (21 km2) drainage area	1/1991	1/2010	6/1990	9/2010	1990	2010
Hamilton Creek36	02480002	Small urban stream; 8 mi2 (21 km2) drainage area	5/1991	1/2010	6/1990	9/2010	1990	2010

Figure 93 shows a map of Mobile County, the individual stream gage stations used in this study (defined by green diamonds), and key highways in the region. Major rivers in the area are shown in Figure 92. Three of the four basins in Mobile County are represented by the selected stream gage stations (there are no stream gage sites selected in the southern coastal basin (number 03170009)).

Figure 92 : Map of the Mobile-Alabama-Coosa River

Source: Evans

Figure 93 : Selected Stream Gage Sites in the Mobile Region

C.8. Detailed Streamflow Projections Methodology

This Appendix describes the methodology that was used to develop projections of future streamflow in the Mobile region. Monthly discharge projections were developed for an artificial basin using the USGS's modified Thornwaite monthly water balance model (WBM) driven by Mobile-specific information.³⁷ This model estimates monthly runoff which can be converted to stream-specific discharge, evapotranspiration, and soil moisture within a basin or sub-basin using user-provided monthly precipitation and temperature data.

C.8.1. Model Assumptions

The model assumes:

• A portion of precipitation immediately becomes runoff (termed the "direct runoff factor"). This portion is determined by the model user.
• A portion of precipitation infiltrates into the soil and is stored (termed the "soil moisture"). This portion is determined by the model user.
• A portion of precipitation that evaporates back into the atmosphere from land surface, water surface, or plants (termed "evapotranspiration"). This portion is a function of temperature.
• Any remaining precipitation is multiplied by a user-defined runoff factor which determines how much of the remaining precipitation becomes additional runoff and how much is considered surplus that gets carried over to the next month. The monthly runoff (mm) results from WBM can be converted to monthly discharge (cubic feet per second) using the basin area.

Hay and McCabe (2002) tested the performance of a monthly water balance model at a set of diverse physiographic and climatic basins across the United States and concluded that "WB models can be used reliably to estimate monthly runoff in the eastern U.S., mountainous areas of the western U.S., and the Pacific Northwest." This study suggests it is acceptable to use a WBM for the Mobile region for modeling monthly runoff.

Optimum values for the user-defined parameters were determined for Mobile using runoff data from three stream gage sites and meteorological data averaged across the observation station data from Coden and Mobile (Coden and Mobile were chosen because they are located in Mobile County).³⁸ Stream gage sites were selected that provided monthly stream gage discharge for 1990 to 2010. The discharge was first converted to monthly runoff (millimeters) and then compared against the WBM results. The optimum user-defined parameters were the same as those described in the Hay and McCabe (2002) study.

Once calibrated, the WBM was run with the climate model baseline simulations and compared against the stream gage runoff values. Then the WBM was run with projected temperature and precipitation simulations to provide projected runoff and evapotranspiration for each emission scenario and time period.

Soil parameters tend to vary seasonally and interannually (by year). The WBM, however, assumes a steady-state value across each thirty-year climate period. An important question is whether these values can be assumed to remain constant in the future.

To answer this question, a literature review was conducted and available historical data sets were reviewed. One study investigated how soil moisture changes with temperature using records from over 600 global stations that provided a minimum of 6 years of information (most with more than 15 years). The study found: "in contrast to predictions of summer desiccation with increasing temperatures, for the stations with the longest records summer soil moisture in the top 1 meter has increased while temperatures have risen. The increased trend in precipitation more than compensated for the enhanced evaporation."³⁹ This suggests soil moisture may increase with projected increasing temperatures; however, a quantitative relationship was not provided. A sensitivity analysis could be conducted to see how sensitive estimated monthly runoff may be to changes in soil moisture.

C.8.2. Model Calibration

To calibrate the model, modeled runoff data was calibrated using observed stream gage data. More specifically, monthly runoff data for 1990 to 2009 produced by the WBM and driven by monthly temperature and precipitation data observed at Mobile and Coden stations were calibrated against the stream gage data.

Figure 94 illustrates that the WBM time series of monthly runoff appears to be similar to the monthly runoff measured at the stream gage sites. The WBM does not capture the extreme peaks in runoff, underestimating periods of low precipitation such as the fall of 1990, 1993, and 1996 and overestimating periods of high precipitation such as the fall of 2002 and summer of 2003. Overall, the model has the most difficulty accurately portraying fall monthly runoff. It is unclear what environmental reason can explain this seasonal signal.

Figure 94 : Monthly Runoff by Stream Gage Station (mm), 1990-2009

Table 44 shows how well the WBM represents monthly runoff for each stream gage site. For the 1990 to 2009 time period, the WBM captures much of the variability at the Chickasaw Creek stream gage with a coefficient of determination (R2) of 0.74 (where an R2 of 1.0 would suggest the WBM explains all variability observed at the stream gage site). The WBM underestimates the period's average monthly runoff by only 1%. The WBM does a less accurate job replicating runoff for Hamilton and Crooked Creeks.

Table 44 : The WBM and Goodness-of-fit Parameters for Each Stream Gage Site

Stream Gage site	Observation Stations for Meteorological Data	WBM Parameters	Average Monthly Runoff (mm) (% diff compared to WBM)	Standard error	R2
Chickasaw Creek	Coden, Mobile	Runoff Factor of 44%; Direct Runoff of 5%; Soil Moisture at 145 mm	63 (-1%)	5	0.74
Hamilton Creek			76 (-18%)	16	0.53
Crooked Creek			59 (+6%)	15	0.62

As the WBM will be used to project monthly runoff based on projections of temperature and precipitation, the WBM was run with baseline modeled conditions and compared against the stream gage sites. These runs show that the monthly runoff for baseline conditions driven by climate model data underestimates the observed stream gage monthly runoff data. The WBM does a reasonable job across baselines from all emission scenarios, with an R2 of approximately 0.8, for Chickasaw Creek. However, Hamilton and Crooked Creeks are below an R2 of 0.5. This suggests the projected runoff may be most able to represent changes at Chickasaw Creek.

C.9. Summary Tables and Figures for Projected Streamflow Analysis

This appendix contains summary tables and figures corresponding to the projected streamflow analysis described in Section 9.2 of the main report. The following tables and figures are included in this appendix:

• Figure 95: Modeled Baseline and Projected Monthly Streamflow Discharge (ft3/sec) for Chickasaw Creek and Actual Evapotranspiration (mm) by Time Period and Emission Scenario
• Table 45: Monthly Streamflow Discharge (ft3/sec) for Chickasaw Creek and Evapotranspiration (mm), Change from Baseline (1980-2009)
• Figure 96: Modeled Baseline and Projected Monthly Streamflow Discharge (ft3/sec) by Time Period and Emission Scenario for Crooked Creek and Hamilton Creek

Figure 95: Modeled Baseline and Projected Monthly Streamflow Discharge (ft3/sec) for Chickasaw Creek and Actual Evapotranspiration (mm) by Time Period and Emission Scenario

Table 45: Monthly Streamflow Discharge (ft3/sec) for Chickasaw Creek and Evapotranspiration (mm), Change from Baseline (1980-2009)

	1980-2009	2010-2039 (Δ)			2040-2069 (Δ)			2070-2099 (Δ)
Variable	Observed	B1	A2	A1FI	B1	A2	A1FI	B1	A2	A1FI
Monthly Streamflow Discharge (ft3/sec)
Jan	390.4	121.7	103.7	77.7	135.3	123.8	111.4	160.5	141.7	23.9
Feb	337.5	76.8	62.7	69.2	83.7	52.9	65.7	101.6	70.9	0.9
Mar	386.8	42.3	25.2	49.5	35.0	12.8	43.1	66.2	14.9	-15.4
Apr	301.5	25.2	32.4	8.5	2.6	-26.0	-13.7	24.8	-47.0	-90.5
May	271.0	12.0	18.4	7.3	2.6	-12.8	-10.7	17.1	-26.0	-49.5
Jun	214.6	8.5	11.1	-0.9	3.4	-9.0	-8.5	10.2	-19.6	-30.7
Jul	243.2	6.4	-3.0	-13.7	8.1	-6.4	-20.9	11.1	-15.4	-37.6
Aug	189.1	-26.5	3.4	-21.3	56.3	-56.8	-23.5	30.7	-58.9	-37.6
Sept	237.0	-14.5	24.8	9.4	65.7	-57.2	-65.3	35.9	-82.8	-67.9
Oct	172.7	6.8	8.5	1.7	53.8	-32.4	-52.9	42.3	-107.6	-56.3
Nov	204.1	16.2	14.1	19.6	46.1	-12.4	-66.2	49.9	-163.1	-150.7
Dec	297.7	36.3	26.9	65.7	63.2	32.9	-28.2	63.2	-82.0	-175.4
Monthly Evapotranspiration (mm)
Jan	-	1.0	1.7	1.4	2.1	4.0	3.3	3.0	7.3	6.2
Feb	-	0.7	1.0	0.9	2.2	2.6	4.0	2.5	6.3	6.6
Mar	-	3.0	2.0	3.0	3.9	5.3	8.3	5.4	11.5	14.1
Apr	-	3.0	3.1	4.3	5.9	8.3	13.7	8.1	18.3	23.6
May	-	5.4	4.4	8.5	9.3	14.5	23.8	13.0	31.2	42.3
Jun	-	6.2	6.5	6.0	11.3	14.4	10.7	13.8	16.5	23.1
Jul	-	8.4	7.8	5.4	13.0	22.1	-1.6	17.0	19.5	-11.2
Aug	-	7.9	8.7	9.7	12.9	20.4	29.1	16.7	37.1	24.0
Sept	-	6.9	6.8	8.0	10.6	16.6	20.3	15.4	30.9	36.9
Oct	-	4.4	5.7	5.2	7.0	11.9	14.9	10.9	23.9	25.4
Nov	-	2.6	2.3	2.1	4.9	5.5	6.1	6.4	11.9	10.7
Dec	-	1.4	1.8	1.6	3.0	4.0	3.8	4.4	7.5	7.2

Figure 96: Modeled Baseline and Projected Monthly Streamflow Discharge (ft3/sec) by Time Period and Emission Scenario for Crooked Creek and Hamilton Creek

---

1 Knutti et al., 2011; Mote et al. 2011

2 IPCC, 2000; IPCC, 2007a; IPCC, 2007b; Knutti et al., 2011

3 The A1FI scenario group describes a convergent world of low population growth, very rapid economic growth, and rapid introduction of new and more efficient technologies. However, the A1FI world is one of less concern for environmental sustainability compared to the B1 storyline, and the direction of technological change in A1FI is fossil intensive.

4 The B1 scenario family describes a convergent world in which regional per capita income gap substantially decreases. The scenario is characterized by low population growth, rapid changes in economic structures toward a service and information economy, with reductions in material intensity, and the introduction of clean and resource-efficient technologies.

5 The A2 scenario family describes a heterogeneous world in which economic growth is uneven and the income gap between now-industrialized and developing parts of the world does not narrow. The scenario is characterized by high population growth, slow economic development, and slow technological change.

6 Climate models utilize well-understood physical principles and have been demonstrated to reproduce observed features of recent and past climate changes. (IPCC, 2007a)

7 IPCC, 2008; USGCRP, 2009

8 IPCC, 2007a

9 For each time period, the approximate percent-contribution are estimated as an average of each percent-contribution noted for the end-points. These values presented here are only intended for qualitative and illustrative purposes.

10 Hawkins and Sutton, 2009

11 IPCC, 2007a

12 Mote et al, 2011; Knutti et al., 2011

13 Climate sensitivity is defined as the temperature change resulting from a doubling of atmospheric carbon dioxide concentrations relative to pre-industrial times (IPCC, 2007a).

14 USCCSP, 2008c

15 Hayhoe and Stoner, 2012

16 USCCSP, 2008c

17 Hayhoe and Stoner, 2012

18 See Hayhoe and Stoner (2012) for description of methodology of statistically downscaling the daily data.

19 Hayhoe and Stoner, 2012

20 Hayhoe and Stoner, 2012

21 Since there are ten GCMs providing results for the A2 and B1 emission scenarios, the uncertainty estimates include ranges of one standard deviation from the mean based on the set of all relevant climate model simulations. Since only four GCMs provide results for the A1FI emission scenario, the uncertainty estimates are a coarser range of model results described by the minimum and maximum GCM values.

22 The daily precipitation data was used as a substitute for 24-hour precipitation. This study does not apply a conversion factor of 1.143 for converting 1-day to 24-hr rainfall (see Durrans and Brown).

23 Goodness of fit tests (i.e., how well the theoretical distribution fits the data) were applied for each station, scenario, climate model, and 30 year period. The model was reasonable at the 5% level for about 60% of the cases and was reasonable at the 1% level for all of the cases.

24 Hyndam and Fan, 1996

25 Variations of statistical significance were considered for this study (e.g., 90%, 95%, and 99% confidence). It was decided that, though interesting, this level of effort and detail was likely more than what was needed for the Task 3 assessment (e.g., how would a statistical difference at the 99% level be treated differently than a 95% level). Therefore, a streamlined description of what is significant at the 95% level was agreed as adequate for purposes of this study.

26 Historical runoff data for hydrologic units provided by the USGS Water Watch program (http://waterwatch.usgs.gov/new/index.php?id=romap3). A hydrologic unit is, in theory, equivalent to a basin.

27 The U.S. Geological Survey (USGS), in cooperation with Alabama Department of Transportation (ALDOT) updated flood-frequency estimates for a number of steam gages. The purpose of updating these estimates was to accurately inform the design drainage structures for highways in Alabama (USGS, 2004; USGS, 2010a).

28 USGS, 2004; USGS, 2007; USGS, 2010a

29 USGS, 2010a

30 USGS, 2004

31 These numbers are an estimate as the number of stream gage stations change as new gages become operational and operational gages are retired. The USGS website provides the most up-to-date information (http://waterdata.usgs.gov/nwis/annual/?referred_module=sw).

32 USGS, 2011a

33 USGS, 2011d

34 USGS, 2011b

35 USGS, 2011c

36 USGS, 2011e

37 http://wi.water.usgs.gov/Soil_Water_Balance/index.html

38A sensitivity analysis was conducted in which additional station location data were included. This analysis revealed Mobile and Coden provided the best data.

39 Robock et al., 1999