UNCERTAINTIES IN CHARACTERIZING DROUGHT WITH CLIMATE MODELS

(1)

U

NCERTAINTIES IN

C

HARACTERIZING

D

ROUGHT WITH

C

LIMATE

M

ODELS

Lauren Schmeisser

Student Number: 10407278

MSc Earth Sciences – Geo-‐Ecological Dynamics Track Master’s Thesis – 45 EC

February 2014 – August 2014 Examiner: Dr. John van Boxel Co-‐assessor: Dr. Erik Cammeraat

ABSTRACT

Extreme climatic events like drought can have detrimental effects on the environment, ecosystems, and socioeconomic activity. As such, there is great incentive to be able to model and characterize drought in order to better predict and mitigate negative impacts of drought. Unfortunately, many uncertainties hinder the ability to accurately model and forecast drought events. Uncertainties in characterizing drought-‐ specifically those related to definitions of drought, physical mechanisms behind drought, timescale of drought, and modeling drought-‐ are explored here. A comparative analysis of drought indices found that the Standardized Precipitation Index is the most appropriate for drought analysis in the regions of interest, the Great Plains and Southwest U.S. The sensitivity of drought characteristics to model resolution found that model grid size generally does not have a significant influence on drought model output. Drought time scale was revealed to be an important factor in drought analysis-‐ short-‐term drought tends to occur more often in the midst of long-‐term droughts, even though different physical mechanisms may be at the root of each. Finally, a perturbed physics ensemble reveals that model parameterizations do indeed influence

drought characterization.

Photo Credit: Bert Kaufmann

Photo Credit: Robert Couse-‐Baker

(2)

List of Figures

FIGURE 1. APPROXIMATE LOCATIONS OF GREAT PLAINS AND SOUTHWEST REGIONS WITHIN THE UNITED STATES ... 8

FIGURE 2. MONTHLY PRECIPITATION CLIMATOLOGY FOR THE GREAT PLAINS AND SOUTHWEST U.S. ... 9

FIGURE 3. PRECIPITATION ANOMALIES FOR (A) THE GREAT PLAINS AND (B) SOUTHWEST U.S. USING HISTORICAL PRECIPITATION OBSERVATIONS FROM YEARS 1900-‐2010 ... 10

FIGURE 4. FOUR PHASES OF THE THESIS PROJECT ... 15

FIGURE 5. GREAT PLAINS TOTAL ANNUAL PRECIPITATION (MM) FROM (A) GPCC OBSERVATIONS AND (B) TRMM OBSERVATIONS ... 19

FIGURE 6. SOUTHWEST U.S. TOTAL ANNUAL PRECIPITATION (MM) FROM (A) GPCC OBSERVATIONS AND (B) TRMM OBSERVATIONS ... 20

FIGURE 7. SEASONAL PRECIPITATION CLIMATOLOGY IN MM/DAY FOR OBSERVATIONAL DATA, 0.5°, 1°, AND 2° CCSM4 CONTROL RUNS IN THE GREAT PLAINS ... 28

FIGURE 8. SEASONAL PRECIPITATION CLIMATOLOGY IN MM/DAY FOR OBSERVATIONAL DATA, 0.5°, 1°, AND 2° CCSM4 CONTROL RUNS IN THE SOUTHWEST U.S. ... 29

FIGURE 9. SEASONAL PRECIPITATION ANOMALIES (IN MM/DAY) DURING FLASH DROUGHTS IN THE GREAT PLAINS ... 38

FIGURE 10. SEASONAL PRECIPITATION ANOMALIES (IN MM/DAY) DURING LONG-‐TERM DROUGHTS IN THE GREAT PLAINS ... 39

FIGURE 11. SEASONAL PRECIPITATION ANOMALIES (IN MM/DAY) DURING FLASH DROUGHTS IN THE SOUTHWEST U.S. ... 40

FIGURE 12. SEASONAL PRECIPITATION ANOMALIES (IN MM/DAY) DURING LONG-‐TERM DROUGHTS IN THE SOUTHWEST U.S. ... 41

FIGURE 13. SEASONAL SOIL MOISTURE ANOMALIES (IN KG/M2₎_{DURING
FLASH
DROUGHTS
IN
THE}_G_REAT_P_LAINS_{...
43}

LONG}_-‐_{TERM
DROUGHTS
IN
THE}_G_REAT_P_LAINS_{...
44}

FLASH
DROUGHTS
IN
THE}_S_OUTHWEST_{U.S.
...
45}

LONG}_-‐_{TERM
DROUGHTS
IN
THE}_S_OUTHWEST_{U.S.
...
47}

FIGURE 17. SUMMER (JJA) SEA SURFACE TEMPERATURE ANOMALIES DURING FLASH DROUGHT AND LONG-‐TERM DROUGHT IN THE GREAT PLAINS ... 48

FIGURE 18. SUMMER (JJA) SEA SURFACE TEMPERATURE ANOMALIES DURING FLASH DROUGHTS AND LONG-‐TERM DROUGHTS IN THE SOUTHWEST U.S. ... 50

FIGURE 19. HISTOGRAMS OF NUMBER OF DROUGHT MONTHS PER 3 YEARS IN THE GREAT PLAINS AND SOUTHWEST U.S. FOR PERTURBED PHYSICS RUNS AND 2 DEGREE CCSM4 CONTROL RUN ... 52

FIGURE 20. TIME SERIES OF SPI VALUES FOR OBSERVATIONAL DATA IN THE GREAT PLAINS, POSITIVE SPI VALUES IN BLUE AND NEGATIVE SPI VALUES IN RED, WITH A 5-‐YEAR RUNNING AVERAGE SMOOTHER ADDED IN BLACK ... 60

FIGURE 21. TIME SERIES OF SPI VALUES FOR 0.5 DEGREE DATA IN THE GREAT PLAINS, WITH 5-‐YEAR RUNNING AVERAGE SMOOTHER ADDED ... 60

FIGURE 22. TIME SERIES OF SPI VALUES FOR 1 DEGREE MODEL IN THE GREAT PLAINS, WITH 5-‐YEAR RUNNING AVERAGE SMOOTHER LINE ADDED IN BLACK ... 61

FIGURE 23. TIME SERIES OF SPI VALUES FOR 2 DEGREE MODEL IN THE GREAT PLAINS, WITH 5-‐YEAR RUNNING AVERAGE SMOOTHER LINE ADDED IN BLACK ... 62

FIGURE 24. TIME SERIES OF SPI VALUES IN THE SOUTHWEST U.S., WITH 5-‐YEAR RUNNING AVERAGE SMOOTHER LINE ADDED IN BLACK . 63 FIGURE 25. TIME SERIES OF SPI VALUES FOR 0.5 DEGREE MODEL IN THE SOUTHWEST U.S., WITH 5-‐YEAR RUNNING AVERAGE SMOOTHER LINE IN BLACK ... 63

FIGURE 26. TIME SERIES OF SPI VALUES FOR 1 DEGREE MODEL IN THE SOUTHWEST U.S., WITH 5-‐YEAR RUNNING AVERAGE SMOOTHER LINE ADDED IN BLACK ... 64

FIGURE 27. TIME SERIES OF SPI VALUES FOR 2 DEGREE MODEL IN SOUTHWEST U.S., WITH 5-‐YEAR RUNNING AVERAGE SMOOTHER LINE ADDED IN BLACK ... 65

(4)

List of Tables

TABLE 1. COMPARISON OF DROUGHT INDICES ... 25 TABLE 2. SPI THRESHOLD VALUES FROM LITERATURE ... 26 TABLE 3. DROUGHT CHARACTERIZATION FOR 1-‐, 3-‐, 24-‐, AND 48-‐MONTH SPI DROUGHTS IN THE SOUTHWEST U.S. AND GREAT PLAINS 30 TABLE 4. KOLMOGOROV-‐SMIRNOV STATISTICAL TEST P-‐VALUE RESULTS FOR THE GREAT PLAINS ... 33 TABLE 5. KOLMOGOROV-‐SMIRNOV STATISTICAL TEST RESULTS FOR SOUTHWEST U.S. ... 34 TABLE 6. PERCENTAGES OF FLASH DROUGHTS DURING LONG-‐TERM DROUGHTS AND FLASH DROUGHT DENSITIES FOR OBSERVATIONS, 0.5°,

1°, AND 2° CONTROL RUNS IN BOTH THE GREAT PLAINS AND SOUTHWEST U.S. ... 36 TABLE 7. PARTIAL REGRESSION COEFFICIENTS AND STANDARDIZED REGRESSION COEFFICIENTS FROM MULTI-‐LINEAR REGRESSION

ANALYSIS OF DROUGHT AND PHYSICS PARAMETERIZATIONS IN BOTH THE GREAT PLAINS AND SOUTHWEST U.S. ... 53

(5)

Glossary

TERM DEFINITION

CLIMATE MODEL Large set of mathematical equations that

represent the processes in Earth’s atmosphere, ocean, land, and sea ice components, used to recreate and predict long-‐term global weather conditions COMMUNITY ATMOSPHERE MODEL V.5.0 (CAM5) The fifth version of a global atmosphere

model and the atmospheric component of the coupled community earth system model COMMUNITY CLIMATE SYSTEM MODEL V.4.0 (CCSM4) The older version of the NCAR fully-‐coupled

global climate model (see CESM1)

COMMUNITY EARTH SYSTEM MODEL V.1.0 (CESM1) The latest NCAR fully-‐coupled global climate model that is used to provide updated recreations of the past and present climate and predictions of future climate

DROUGHT A sustained, but temporary, period of time

characterized by abnormally low

precipitation. Specific definitions of drought can vary greatly depending on context

FULLY-‐COUPLED Integration of both atmosphere and ocean

models into a climate model

GREAT PLAINS A region in the central United States bounded

by 30°-‐50°N and 95°-‐105°W and

characterized by flat landscape and heavy agricultural practices

PARAMETERIZATION Mathematical specification of a process that

occurs on a scale too small to be physically represented in a climate model with limited grid size, and therefore must be simplified using parameters

PERTURBED PHYSICS ENSEMBLE (PPE) Climate model runs that alter parameter values in order to evaluate the impact of parameter uncertainty on accuracy of climate model predictions

RESOLUTION Climate model resolution refers to the grid

sizes used when running the model

THE SOUTHWEST A desert-‐like region with complex

topography in the southwestern United States bounded by 32°-‐42°N and 106°-‐118°W STANDARDIZED PRECIPITATION INDEX (SPI) A metric used to determine presence and

severity of drought based on precipitation data

(6)

Acronyms

CCSM Community Climate System Model CESM Community Earth System Model

GPCC Global Precipitation Climatology Centre NCAR National Center for Atmospheric Research NCL NCAR Command Language

NOAA National Oceanic and Atmospheric Administration PDSI Palmer Drought Severity Index

PPE Perturbed Physics Ensemble SLP Sea Level Pressure

SPI Standardized Precipitation Index SST Sea surface temperature

Acknowledgements

A heartfelt thanks to my University of Amsterdam advisors, John van Boxel and Erik Cammeraat, for their willingness to work with me remotely from across the Atlantic Ocean. And a big thank you to Jerry Meehl for guidance, support and resources provided to me as a visiting student at NCAR. I greatly enjoyed my scientific journey with this thesis, and I am grateful for the direction and encouragement from each of you.

(7)

1 Introduction

Climate is an inextricable part of life on Earth-‐ it affects the environment, ecosystem health, and human activity. Given its direct impact on our lives, there is vested interest in studying the

mechanisms of climate and how climate might change in the future with anthropogenic influence.

Extreme climatic events, like drought, are of particular concern due to the potential future increase in frequency and intensity, or change in timing, of such events (IPCC, 2014). Intensified extreme events can produce unprecedented effects. Specifically, drought can interfere with food and drinking water supply, vegetation survival, and income generation. In order for the world’s populations to be able to prepare for or adapt to drought conditions, it is essential to understand the physical mechanisms behind drought so it is possible to better predict their occurrence, especially in light of climate modeling uncertainties.

This thesis will seek to add to the body of knowledge on drought and will do so by evaluating the uncertainties in characterizing drought in the Great Plains and Southwest United States

(henceforth referred to as ‘the Southwest U.S.’) using climate models. The outcomes of this thesis will provide insight into how model resolution, parameterization, drought type, and selection of drought index affect characterization of drought in the specified regions, with the potential to also elucidate what physical mechanisms may be at play in creating drought-‐like conditions.

This thesis provides an overview of background topics essential to the study, the objective of the work and research questions to be addressed, and the methods used, as well as results and a discussion of findings.

2 Background and Research Framework

The following section provides background information on major topics to be covered in the thesis study (definition and impacts of drought, drought mechanisms, modeling drought, and drought in the Great Plains and the Southwest), the societal importance of the research, and how this research fits into the work currently being done in the fields of climate modeling and drought. This will serve as a framework within which the thesis study was executed.

2.1 Drought and Its Impacts

Drought refers to a sustained period of abnormally dry environmental conditions that is

temporary rather than permanent (Dai, 2011a; NOAA, 2013). The temporary aspect of drought separates it from the more permanent state of aridity (Heim, 2002). Drought can bring a plethora of negative social, economic and environmental impacts, including damage to food supply,

drinking water supply, recreational activities, ecosystems, and hydropower generation, and are incredibly expensive natural disasters (Heim, 2002; Cayan et al., 2010). For example, the limited water supply during droughts means that agricultural production decreases, causing financial hardship to farms and increasing food prices (Wilhite et al., 2007). Furthermore, depending on resiliency of flora and fauna populations within a region, the lack of water during drought

(8)

negatively impacts biodiversity by decreasing ecosystem productivity and increasing mortality rate (Archaux & Wolters, 2006).

Droughts are intimately linked with the Earth’s climatic processes, and thus are subject to change in intensity and frequency with changing climate (IPCC, 2012; Seager et al., 2007; Dai, 2012). In fact, the models show that drying in the subtropics is forthcoming or already observed, and is unlike anything seen in the instrumental record (Seager et al., 2007). As such, it is important now more than ever to better understand the mechanisms of drought and the potential to predict drought conditions in an attempt to mitigate the potentially amplified negative environmental and socioeconomic impacts of future drought.

Though drought is a recurring phenomenon that has been the topic of studies for decades, it remains difficult to define both qualitatively and quantitatively (Heim, 2002; Kallis, 2008). There are four generally accepted qualitative categories of drought: meteorological, agricultural,

hydrological, and socioeconomic. Meteorological drought refers simply to a reduction in

atmospheric precipitation from the average or expected precipitation values in a region, while an agricultural drought occurs when the soil moisture levels do not satisfy the needs of regional crops. Hydrological drought refers to conditions in which surface and subsurface water supplies are affected by long-‐term reductions in precipitations, and socioeconomic drought defines the type of drought that begins to affect the supply and demand balance between environmental water supply and water demands by human activities (Heim, 2002; NOAA, 2013). In order to limit the scope of this analysis, the paper will focus specifically on meteorological drought.

Quantitatively, drought is measured and defined by drought indices-‐ numerical standards that take into account physical parameters to produce a number on a scale that indicates drought severity. These drought indices are needed to monitor progression of drought over time and to compare different drought events from separate regions (Kallis, 2008). Examples of popular drought indices include the Palmer Drought Severity Index, the Standardized Precipitation Index, precipitation deciles, and the percent of normal. Although no drought index can be considered the best, some indices are more useful than others depending on the type of drought and context in which information on drought is needed (Guttman, 1998; Alley, 1984). Consequently, a

comparative analysis of meteorological drought indices is performed in section 5.1.1, in order to choose the best drought metric for the analyses in this thesis.

Despite the existing qualitative and quantitative descriptions of drought, there is still not one agreed upon threshold for drought. In other words, even given a decided upon drought index, it is subjective what number on that index scale would indicate drought. For example, does decile 1 in the precipitation deciles index (meaning precipitation amount is not exceeded by 10% of

occurrences) signify drought? Or does decile 0.5 indicate drought? It is here where uncertainty in drought definition still lies. A sampling of definitions in recent drought publications characterize drought as: years where soil moisture falls below the 10th_{percentile
over
a
long-‐term}

measurement period (Cayan et al., 2010), a period when supply demand drought index less than -‐ 2.0 (Findell & Delworth, 2010), and time when precipitation in the lowest quartile of a 100-‐yr period (McCabe et al., 2004). In part, this thesis will provide a comparative analysis of drought indices and drought definitions in hopes of finding the index and threshold that is the best fit for characterizing drought in the areas of interest-‐ the Great Plains and Southwest.

(9)

Despite the lack of agreement on drought description, implementing definitions of drought, either qualitatively or quantitatively, is necessary for continued research of drought, its causes, and its impacts. In being able to identify drought occurrences by their intensity and duration, and track the drought over time, it is possible to begin to make scientific progress towards understanding drought mechanisms. A deeper understanding of drought mechanisms will be instrumental in improving drought forecasts, which in turn will help society better prepare for and adapt to current and future drought conditions.

2.2 Drought in the Great Plains and Southwest U.S.

Some regions in particular are more sensitive to drought than others. Within the United States, two regions of interest are the Great Plains and the Southwest, which both have unique reasons why they are vulnerable to the effects of drought. Figure 1 displays the approximate locations of the Great Plains and Southwest U.S. regions on a map of the United States.

Figure 1. Approximate locations of Great Plains and Southwest regions within the United States

The Great Plains region of the United States is defined as the area between 30°-‐50°N and 95°-‐ 105°W, and is characterized by grasslands that support large agricultural production zones (McCrary, 2010; Capotondi & Alexander, 2010). The Great Plains are particularly vulnerable to drought conditions, given the high water demand from local agriculture and expanding

populations, as well as the unsustainable use of groundwater from the regional Ogallala Aquifer (McCrary & Randall, 2010). The drought sensitivity of the area is apparent in the historical drought records, which show long periods of reduced precipitation in the region (Clark et al., 2002; Woodhouse & Overpeck, 1998). One of the worst and most notable periods of drought is that of the 1930s, commonly referred to as the Dust Bowl. During this time, most crops perished and dust storms disrupted the lives of thousands in the region (Schubert et al., 2004b; Clark et al.,

Southwest

U.S.

Great

Plains

(10)

2002; Capotondi & Alexander, 2010). Given that this area of the central United States is a valuable agricultural center, drought prediction and preparation is extremely useful in avoiding

socioeconomic devastation.

Monthly precipitation climatology for the Great Plains, shown in the left plot of Figure 2, illustrate a seasonal precipitation cycle in the region in which winter has precipitation minimums and summer has precipitation maximums. Given the region’s dependency on agriculture, droughts in the spring and summer (crop growing months) are especially devastating, even though these are times of the highest regional precipitation. Figure 3(a) shows historical precipitation anomalies trends for the Great Plains. Anomalies were calculated by subtracting long-‐term precipitation rates from average precipitation rates for any given month. The precipitation anomaly time series for the Great Plains indicates the Dust Bowl decadal drought in the 1930s, as well as other decadal scale droughts throughout the region’s history, as described in the literature mentioned above. This further supports the region’s tendencies for extended periods of drought conditions.

Figure 2. Monthly Precipitation Climatology for the Great Plains and Southwest U.S.

The Southwest is defined as the area between 32°-‐42°N and 106°-‐118°W, and is characterized by low annual precipitation, warm annual temperatures, clear skies, strong interannual hydrological variability, and populations that rely heavily on water supply from the Colorado River (Sheppard et al., 2002; Meehl & Hu, 2006; Cayan et al., 2010). Most of the region shows characteristics of a desert. Since 2000, the region has experienced widespread drought, and many lakes and

reservoirs are at diminished capacity (MacDonald et al., 2008). As for the future of Southwestern U.S. droughts, climate models predict that the region will get drier and warmer, and experience more severe droughts (Cayan et al., 2010). Due to the growing population and inherent water shortages from increasing demands within the region, the Southwest would benefit greatly from better drought predictions that allow further preparation for incoming droughts.

Monthly precipitation climatology for the Southwest U.S., shown in the right plot of Figure 2 illustrate a slight seasonal precipitation cycle in the region in which summer has precipitation minimums and winter has precipitation maximums. Figure 3(b) shows historical precipitation anomalies trends for the Southwest U.S. Anomalies were calculated by subtracting long-‐term from average precipitation for any given month. The precipitation anomaly time series for the

(11)

do not appear as large as those in the Great Plains. In part, this is due to the fact that the

Southwest U.S. does not receive as much precipitation as the Great Plains, and thus cannot have negative anomalies as large in magnitude as those in the Great Plains. Similarly, the Southwest U.S. appears to have more positive precipitation anomalies than negative anomalies, but this is again due to the extremely low overall precipitation that the region receives.

Figure 3. Precipitation anomalies for (a) the Great Plains and (b) Southwest U.S. using historical precipitation observations from years 1900-‐2010

In comparing the two regions of interest, it is of note that the Southwest U.S. generally receives substantially less precipitation than the Great Plains region in all months. Additionally, the precipitation cycles in each season are offset such that the Southwest U.S. receives the least precipitation in the summer, when the Great Plains receives its maximum precipitation. On the other hand, the Southwest U.S. has a precipitation high in winter, while the Great Plains receives less moisture in the winter. This goes to show that although these regions are relatively close to each other on a global scale, their precipitation patterns are substantially different, thus providing motivation for a regionalized drought analysis as opposed to an analysis of a larger area, such as the entirety of the United States.

Due to the vulnerability of the Southwest U.S. and the Great Plains to periods of drought, it is of particular importance to continue to study drought mechanisms and improve future drought predictions for these areas as a means of mitigating local degradation to the environment and socioeconomic activities. It is for these reasons that this thesis project will focus on these two regions.

2.3 Uncertainties in Characterizing Drought

It is clear now the potential devastating nature of droughts, and reasons why research to learn more about them is essential. However, the research needed to further characterize drought is laden with uncertainty. Already mentioned was the uncertainty in simply defining drought; however, there are many more types of uncertainty impeding satisfactory and confident scientific

(12)

progress in the field. Uncertainty is also inherent in the physical mechanisms inducing drought, the types of drought selected for analysis, and how those droughts are modeled in a climate model. Each type of uncertainty will be introduced here, and explored in depth throughout the phases of the thesis.

2.3.1 Drought Mechanisms

Though droughts are naturally occurring phenomenon that can occur stochastically by

atmospheric processes on an interdecadal timescale, there is also evidence for external forcings affecting occurrence of drought. One of the ongoing scientific struggles in the fields of drought and climate is to determine what portion of extreme drought variability is attributable to internally generated conditions, and what portion is ascribed to externally generated forcings (Solomon et al., 2011; Lei et al., 2011).

Much work has been done speculating what specific physical mechanisms or combination thereof are responsible for creation of drought conditions. Some studies have found that sea surface temperature (SST) variability-‐ both in the Atlantic and the Pacific Oceans-‐ drives precipitation anomalies. Atlantic SST anomalies have been shown to influence precipitation variability over North America (Hoerling & Kumar, 2003). Specifically, abnormally warm SSTs in the Atlantic are associated with reduced precipitation in areas of North America like the Southwest and the Great Plains (Kushnir et al., 2010). Other studies claim that Pacific SST anomalies or cycles of El Niño and La Niña events at least partially drive precipitation variability, where negative Pacific SST anomalies are associated with low precipitation anomalies over North America (Capotondi & Alexander, 2010). Seager et al. (2005) describe the physical mechanism behind the link between Pacific SST variability and North American drought-‐ the SST variability changes the subtropical jet stream, which alters propagation of eddies and influences mean meridional circulation, which enhances subsidence over North America and encourages drought. Furthermore, there is research that shows Atlantic and Pacific SSTs together both play a role in drought forcing, though maybe not in equal proportion (Findell & Delworth, 2010; Mo et al., 2009; McCabe et al., 2004; Pegion & Kumar, 2010; Schubert et al., 2009; Ruiz-‐Barradas & Nigam, 2010). Findell & Delworth (2010) show that drought intensity in North America (particularly over the United States) is highest when SST anomalies in the Pacific and Atlantic are of opposite signs, i.e., the Pacific is colder and the Atlantic is warmer than average.

Land-‐atmosphere interactions involving soil moisture and rainfall feedbacks also might contribute as a mechanism for drought (McCrary & Randall, 2010; Schubert et al., 2004a; Taylor et al., 2013). In the Great Plains, specifically, these feedbacks regulate evapotranspiration, and also interact with stability and boundary layer characteristics, all of which influence precipitation (McCrary & Randall, 2010). Climate models that included soil moisture feedback showed about five times more variance in precipitation than models run without the feedback (Schubert et al., 2004a), thus indicating that these land-‐atmosphere interactions make some contribution to drought conditions.

Apart from naturally varying physical mechanisms, drought is also attributed to anthropogenic forcings. Since human activity has been deemed responsible for rising surface temperatures due to increased greenhouse gases, it can be concluded that human activity is also contributing to drying over land (Dai, 2011a). Barnett et al. (2008) found that up to 60% of the observed trends in hydrology in the Western United States from years 1950-‐1999 can be attributed to human activities, and Hidalgo et al. (2009) found that changes in stream flow since 1950 in the western

(13)

U.S. are significantly different from natural variability and attributable to anthropogenic emissions and changing land use. These studies indicate that indeed human activity can contribute to

hydrology and drought in some way.

What is clear in the literature is the lack of certainty regarding physical mechanisms of drought. Though evidence exists for contribution by SST variability, land-‐atmosphere feedbacks and anthropogenic forcings, it is not clear to what extent these mechanisms contribute to drought variability, especially compared to contribution by internal and stochastic variability. This thesis aims to contribute to furthering understanding of what physical mechanisms are at play in causing drought.

2.3.2 Drought Timescale: Flash Drought versus Long-term Drought

One aspect of drought’s complexity is that the timescale on which droughts are measured changes depending on the drought impact with which one is concerned. For example, short-‐term ‘flash’ droughts are of great interest to farmers who may be concerned with crops suffering just a few weeks without enough precipitation, while long-‐term droughts may be of interest to water managers who plan on the annual and decadal time scale how to care for water resources (Heim, 2002).

The term ‘flash drought’ refers to a very short-‐term drought event that occurs on the scale of

weeks to months. Although there is no widely available and broadly accepted definition of drought, a few studies try pinning down a useful flash drought definition. Senay et al. (2008) refer to a flash drought as a severe short-‐term event with negative moisture anomalies and high temperatures. Mozny et al. (2012) define flash drought as a very rapid decline in soil moisture in a 3-‐week period. The media has recently picked up on the term flash drought as a means of conveying an intense dry event that comes and goes quickly, as would a flash flood. Given the lack of peer-‐reviewed literature on the topic of flash drought, there is much room for interpretation of the term and development of a qualitative definition and quantitative flash drought measure.

Long-‐term drought, as used in this study, refers to decadal scale droughts. Long-‐term drought can have an effect on stream flow, agriculture, groundwater, socioeconomic activities and basin-‐wide hydrology (Belayneh et al., 2013). Long-‐term decadal scale droughts are better documented and more thoroughly studied than flash-‐droughts, both in the Great Plains and Southwest U.S (Clark et al., 2002; Cayan et al., 2010; Belayneh et al., 2013).

What is not immediately clear from the literature is how short-‐term and long-‐term droughts differ from each other in length, in cause, and in impact. This thesis will attempt to add knowledge to this scientific gap.

2.3.3 Modeling Drought

Climate models are essential tools for being able to understand Earth’s processes and how they impact climate. They also are the only way to look at how climate may change in the future with changing radiative forcing, and they provide insight into what climatic changes may be due to natural internal variability of Earth’s systems, anthropogenic impacts, or combinations of the two. In being able to predict climatic variables like temperature, precipitation, and soil moisture values,

(14)

global climate models are able to shed light on what drought occurrence might look like in the future. As it stands, climate models still have considerable uncertainty (Murphy et al., 2004). It is in the interest of science and society to reduce this uncertainty and improve the prediction capabilities of climate models so that they can more accurately forecast climate and climatic events like droughts.

Uncertainty in climate modeling can come from model resolution (grid size), parameterizations, future socioeconomic scenarios, and natural stochastic processes. This thesis will narrow the focus and concentrate solely on model resolution and parameterizations.

2.3.3.1 Climate Model Resolution

Climate model resolution in this context refers to spatial resolution, or the size of the model grid in degrees latitude and longitude. If a model has increased spatial resolution (e.g., 0.5°), it has more grid cells, while models with lower resolutions (e.g., 2°) have fewer grid cells. Climate models with increased spatial resolution and more grid cells increase the necessary computing time

substantially. In general, increasing model resolution by a factor of two equates to ten times as much computing power needed, or will take ten times as long to run on the same computer (“Climate Modeling”, UCAR, 2011, http://scied.ucar.edu/longcontent/climate-‐modeling).

Consequently, there is motivation to run lower resolution models if the fewer number of grid cells will not affect the analysis of the model output of interest. For example, if drought characterization proves to be similar across all climate model control run resolutions, it is desirable to run those necessary model runs at the lowest resolution to save computing resources.

Past studies have found that although broad climate predictions at large, continental scales often remain similar across resolutions, as climate model grid size decreases, regional climate

predictions generally get more detailed. This is especially true for areas that have complex

topography (like mountainous areas) or coastlines (Giorgi & Marinucci, 1996). The consideration of model resolution for climate models has various facets. For one, higher climate model

resolution may better represent large-‐scale circulation and atmospheric processes, thus better capturing the big processes that affect climate on a smaller scale. Secondly, increased resolution may better simulate topographical interactions with the atmosphere and better indicate localized processes. And finally, any choice of model resolutions may have an impact on parameterizations and how processes are resolved at the model’s grid size, which will be discussed more in the next section 2.3.3.2 (Giorgi & Marinucci, 1996).

Understanding differences in model resolutions, especially when focusing on specific regions like the Great Plains and Southwest, will allow for more informed characterizations of drought. Knowing how model resolution affects drought characterization will also aid in comparison of droughts across models, as will be done in this thesis project.

2.3.3.2 Climate Model Parameterizations

With respect to parameterizations, climate model uncertainty can be evaluated with multi-‐model ensembles, like in the Coupled Model Intercomparison Project, or with perturbed physics

ensembles (PPEs) (Bellprat et al., 2012). The focus here will be on PPEs. PPEs take a single climate model and alter the physics settings so that unconfined model parameters vary and allow

(15)

what parameterizations are important to modeling a specific climatic event (Fischer et al., 2007)-‐ in this case, drought.

Although model parameterizations will be explored separately from model resolution in this project, it is important to note that they are not unrelated. As mentioned briefly in the previous section, model parameterizations may behave differently with different model resolutions. This specific aspect of model parameterizations will not be analyzed here, though the topic can be explored further in Giorgi & Marinucci (1996).

3 Objective and Research Questions

Given the problematic and inherent uncertainty in characterizing drought, there is a need for working towards better understanding of this uncertainty and its sources. The objective of the thesis study is to better understand uncertainty in portraying drought, as it relates specifically to drought definition, model resolution, drought type, physical mechanisms of drought and climate model physics parameterizations.

To achieve the objectives of the study, the following research questions will be answered: 1. Which drought index is most appropriate for the analyses?

2. How does climate model resolution affect drought characterization? 3. How do flash drought and long-‐term drought characterizations differ?

4. What physical mechanisms may be contributing to drought in the Great Plains and Southwest U.S.?

5. What climate model parameterizations may be affecting uncertainty in drought characterization?

4 Methods and Materials

4.1 Methodology

The thesis study is separated into four main project phases-‐ (1) a comparative analysis of drought indices, (2) a sensitivity analysis of drought characterization to model resolution, (3) a

comparison of flash and long-‐term droughts, and (4) an investigation of a perturbed physics ensemble to estimate sensitivity of drought characterization to model parameterizations. The four phases of the project are presented visually in Figure 4.

(16)

Figure 4. Four phases of the thesis project

In the following sections, methodologies for each project phase are described in detail.

4.1.1 Phase I. Comparative Analysis of Drought Indices

The comparative analysis of drought indices uses information from literature to compare and contrast various drought metrics, looking specifically at pros and cons of each index. This thesis looks specifically at meteorological drought; therefore, only indices that evaluate meteorological drought are considered. The drought metric will be chosen based on its applicability to both regions of interest, as well as its flexibility in evaluating drought at multiple time scales. Moreover, the drought index will also be evaluated based on criteria presented in the results section.

Based on findings from the literature, one drought index that applies well to this analysis project in the Great Plains and Southwest U.S. is chosen to complete the rest of the drought data analysis throughout the project. Drought periods for the entirety of the analyses are determined and

defined by the specified drought metric, as decided in the comparative analysis of drought indices.

4.1.2 Phase II. Model Resolution Sensitivity Analysis

Three control runs of different spatial resolutions (0.5°, 1° and 2° latitude by longitude) from the Community Climate System Model version 4.0 (CCSM4) are utilized for the model resolution sensitivity analysis in Phase II. CCSM4 model control runs are used for the resolution sensitivity analysis rather than the more recent Community Earth System Model version 1.0 (CESM1)

because long control runs for 0.5°, 1° and 2° resolutions of CESM1 are not available for this project. An observational dataset of precipitation (see Materials section for more details) is also utilized as a basis to which drought characteristics in each model resolution is compared.

First, for a qualitative look at sensitivity of drought to model resolution, contour plots of

precipitation rates are created for all model resolutions and observations. The model resolution contour plots are compared to the contour plots of the observations to get a sense for how well the

PHASE I:

Comparative Analysis of Drought Indices

PHASE II:

Sensitivity Analysis of Drought Characteristics

to Model Resolution

PHASE III:

Analysis of Flash Droughts versus Long-‐

term Droughts

PHASE IV:

Analysis of Perturbed Physics Ensemble

(17)

models are capturing spatial nuances of regional precipitation, as well as distributed precipitation intensities.

For each spatial resolution and for observational data, drought characteristics are calculated and compared using the metric decided upon for a more quantitative analysis. The drought

characteristics compared over control runs and observations include: raw number of drought months observed, average length of droughts, and number of droughts per decade. Drought periods are determined based on drought indices determined in Phase I.

The raw number of drought months is calculated by summing the number of months that meet drought criteria according to the selected drought index. This number is not directly comparable across resolutions and observations, because run lengths differ. To resolve this, standardized characteristics –drought months per decade and length of droughts-‐ are also analyzed. Length of drought is determined by calculating length of drought (how many consecutive months meet drought conditions), and averaging the length of each individual drought episode over the entirety of the run. The number of droughts per decade is calculated by finding the number of droughts over each decade in the run, and averaging those numbers over the entirety of the run. Since the length of drought and average number of droughts per decade are independent of run length, they are drought characteristics that are comparable across all model runs and observations.

Distributions of drought characteristics are then compared using a non-‐parametric Kolmogorov-‐ Smirnov statistical test to see if differences between the drought characteristic distributions are significantly different between each model resolution and the observational data.

4.1.3 Phase III. Analysis of Flash Droughts versus Long-term Droughts

For the analysis of flash droughts versus long-‐term droughts, 1-‐month Standardized Precipitation Index (SPI) values are used to calculate flash droughts and 24-‐month SPI values are used to calculate long-‐term droughts. A drought period is defined by SPI values that are negative and reach an intensity of at most -‐1.3 that ends when the SPI values become positive (see Section 5.1.5 Drought Index Selection). The research questions for the analysis of flash droughts versus long-‐ term droughts are twofold-‐ (1) Do flash droughts tend to occur more often in the midst of long-‐ term decadal droughts?, and (2) Are flash droughts and long-‐term droughts correlated with different physical mechanisms?

Since this study seeks to explore the relationships between flash and long-‐term droughts, the first step in Phase III is to see if flash droughts have a tendency to occur more often in the midst of long-‐term droughts. The raw percentage of flash droughts that occur in the midst of long-‐term drought is calculated by taking the raw number of flash droughts that occur in long-‐term droughts, divided by the total number of flash droughts that occur during that precipitation record.

Additionally, flash drought ‘densities’ (a numeric created by the author for this study) are calculated. Flash drought densities represent how much of long-‐term drought months are also occupied by flash droughts, or how many months without long-‐term drought are occupied by flash drought. This number gives a better sense of flash drought tendencies in the midst of long-‐term droughts if long-‐term droughts don’t happen often throughout the time series.