Transcript Slide 1
Twentieth Century Drought in the Conterminous United States: An Application of Severity-Area-Duration Analysis Konstantinos M. Andreadis1, Elizabeth A. Clark1, and Dennis P. Lettenmaier1 1. Department of Civil and Environmental Engineering, Box 352700, University of Washington, Seattle, WA 98195 AMS Annual Meeting, 19th Conference on Hydrology 9-13 Jan 2005, San Diego, CA ABSTRACT Drought characterization typically involves severity, frequency and duration; however, the often neglected spatial extent of a drought also influences its impact on water resources management. In the past, station data have been used to calculate drought severity for individual climate divisions across the United States, but a basis for the spatial characterization of drought on a nationwide scale has been lacking. Spatially distributed hydrologic models provide a means for simulating both agricultural drought (related to soil moisture) and hydrological drought (related to runoff) over a grid mesh. The output of such models can be used to identify the spatial extent of drought. Depth-area-duration analysis, widely used in characterization of precipitation extremes for specification so-called design storms, provides a basis for evaluation of drought severity when storm depth is replaced by an appropriate measure of drought severity. Precipitation and temperature data over the continental U.S. for the entire period of observational record are now available in electronic form from the National Climatic Data Center, and these recently available extended data greatly facilitate the ability to reconstruct U.S. drought history. We used these data, starting with 1920 (prior to which station density was too sparse to allow production of meaningful simulations) as input to the physically-based Variable Infiltration Capacity (VIC) macroscale hydrology model, which was used to simulate soil moisture and runoff over the conterminous United States. To standardize soil moisture and runoff anomalies, we computed percentiles of each at a monthly timestep. We then used a clustering algorithm to identify individual drought events in space and time. A series of severity-area-duration curves for all drought events were constructed, to relate the spatial extent of each drought to its severity. From these curves, an envelope curve of the worst drought events in the conterminous U.S. during the 20th century, was derived. These events included the 1930s, 1950s, late 1980s, and the current western U.S. drought, all of which have been previously cited as extreme droughts. 1 3 The overall impact of a drought event depends on several factors, including severity, frequency, area, and duration (Dracup et al., 1980). Several drought indices have been defined for the characterization of drought severity over time. These indices typically describe one of four drought types: agricultural, hydrologic, meteorological, and socioeconomic. Generally, agricultural drought relates to soil moisture, hydrologic to runoff and streamflow, meteorologic to precipitation and temperature, and socioeconomic to the disparity between supply and demand for water-related goods (Wilhite and Glantz, 1985). Severity We use simulated monthly soil moisture anomalies as a measure of agricultural drought and simulated monthly runoff anomalies as a measure of hydrologic drought. Because the absolute magnitude of these departures varies with climate region, we measure the departure from normal in terms of percentiles (Weibull estimate) based on the 1920 to 2003 climatology. Area The algorithm used to identify areal extent of droughts begins with spatial smoothing using a 3-by-3 median filter to minimize discontinuities in the original data. Based on the threshold level, pixels that are experiencing drought at a given monthly time step are identified and are classified using a simple clustering algorithm. The first pixel "under drought" is assigned to the first class. Then the 3-by-3 neighborhood of this pixel is searched for pixels "under drought" which are classified in the same drought cluster. This procedure is repeated until no pixels in the neighboring pixels of the current cluster are "under drought" and a new cluster is created for the next pixel below the drought threshold. After the initial partitioning step, a final classification step excludes all clusters with areas below 10 half-degree pixels from subsequent calculations. At the end of this step, Figure 4. Spatio-temporal drought the remaining clusters are defined as identification: three consecutive separate drought events for the current timesteps, with two drought areas time step. merging (A1 and A2); drought area B2 breaking up into two drought areas, B2 Duration Drought events are assigned a and B3, with the latter merging with temporal extent based on the another drought area (B1). The continuation of drought conditions algorithm classifies each smaller (20th percentile or for lower soil drought area to a larger drought event (A and B). moisture, 20th percentile or lower for Background Drought is among the most costly natural disasters in the United States. In 1995, the Federal Emergency Management Agency (FEMA, 1995) estimated that the annual cost of U.S. droughts was in the range of $6-8B. To assess the potential impacts of current and future drought, water managers often compare current or potential drought severity with historic drought severities. This process, however, often overlooks the effects of areal extent on drought intensity. The purpose of this study is to create a context for comparison of extreme drought events based not only on an event’s severity, but also on its areal extent and duration. NOAA's National Climate Data Center (NCDC) recently released digitized Cooperative Observer Network (Co-op) station precipitation and temperature data for much of the 20th century. This data set makes the extension of model simulations of hydrologic conditions over most of the 20th Century possible. Such simulations provide a spatially and temporally continuous data set. They also allow us to investigate historical droughts in new ways. 2 Hydrologic Model and Validation The Variable Infiltration Capacity (VIC) macroscale hydrologic model (Liang et al., 1994) is applied to a half degree resolution grid mesh across the conterminous United States for the 1920 to 2003 period. Soil properties for a 3-layer soil column are taken from Maurer et al. (2002) and aggregated from 1/8o to 1/2o resolution. Vegetation fractions are based on the University of Maryland's 1-km global land cover classification (Hansen et al., 2000). The meteorological forcings used include observed precipitation and temperature from NCDC Cooperative station data and wind from NCEP/NCAR reanalysis. The VIC model essentially solves an energy and water balance, and accounts for sub-grid scale variability in soil, vegetation, precipitation, and topography (Figure 1). The vertical dimension of each cell is divided into three soil layers, in which the soil moisture storage capacity is treated as a spatial probability distribution. Baseflow from deepest soil layer is produced according to the Arno model nonlinear baseflow formulation (Francini and Pacciani, 1996). Evapotranspiration, surface runoff and baseflow are calculated for each Figure 1. Schematic vegetation class, and the weighted sum of each, based on vegetation fraction, is assigned to representation of VIC the grid cell. Once generated, surface runoff and baseflow are released from the grid cell, hydrologic model. though they can be routed to simulate streamflow using the model developed by Lohmann et al. (1998). Validation The VIC model has been tested at one-eighth degree over the continental U.S. for the period from 1950 to 2000 (Maurer et al., 2002). Runoff was routed through a stream network to simulate streamflow for continental scale river basins with reasonable results, showing good agreement of seasonal cycle, low flows, and peak flows between simulated streamflow and USGS observed streamflow (Figure 2). Maurer et al. (2002) also compare VIC simulated soil moisture with Illinois observed soil moisture (Figure 3). The seasonal variability of soil moisture flux (soil moisture storage change), and its coefficient of variability and temporal persistence, were all reasonably well-simulated. Figure 2. Six monthly hydrographs, aggregated from daily flows. Numbers correspond to location of hydrograph flow. Observed flows from USGS gaging stations. Simulated flows from VIC model (Maurer et al., 2002) Obs. Model Figure 3. Comparison of soil moisture at 19 observing stations in Illinois and soil moisture from the 17 1/8º modeled grid cells that contain the observation points. 5 Drought Characterization runoff) in at least one cell between monthly time steps. If multiple clusters merge to form a larger drought in later time steps, or if a single drought splits into multiple smaller droughts, the smaller droughts are considered part of the larger drought. In these special cases, the spatial extents and severities are calculated separately based on a contiguity constraint (Figure 4). 4 Envelope Curves Soil Moisture Runoff This envelope curve represents four major hydrologic From the envelope curve, we can see that there were drought events between 1920 and 2003. These include four major agricultural drought events between 1920 1932 to 1938, 1950 to1957, 1975 to 1978, and 1999 to and 2003. These occurred from 1928 to 1932, 1932 to 2003 (current). The 1970s drought was most severe 1938, 1950 to 1957, and 1998 to 2003 (current). The over the 3-month duration at very small areas. The current drought dominates at smaller areas for durations current drought dominates over a greater range of areas from 3 months to 2 years. As expected, the 1930s for durations from 3-months to 2-years when defined by drought was most severe at large spatial extents for runoff shortages. As we can see in the accompanying short durations. During the “Dust Bowl” severe drought spatial maps, the current drought effectively covers the covered areas larger than 5 million km2 for durations entire western U.S., suggesting that there were minimal from 3-months to 1-year. the 1950s drought appears to recovery periods for the entire region, amplifying be the most severe at an increasing number of area severity in terms of drought impacts. The 1930s drought values as durations become longer, especially for 1was most severe for the 3-month duration over areas year durations and longer covering the entire envelope from 2 million km2 to 6 million km2. The 1950s drought curve for the 6-year duration. Our analysis technique was most severe for larger areas and longer durations; prevented the 30s drought from continuing during in fact, it is always the most severe drought for the 61932, due to a one month recovery and subsequent year duration. However, if the current drought, which relocation, the drought was split into two events, one in has a maximum duration of 4 years in this analysis, 1928-1932 and one in 1932-1938. The effect of this is continues, it could surpass the 1950s drought in severity most evident in the longest duration results. at the 6-year duration. 6 Soil Moisture and Runoff Drought Differences Even though the current drought has a relatively high soil moisturebased severity, it occupies a much larger portion of the runoff-based envelope curve. If we examine the spatially averaged correlation coefficient between soil moisture and runoff for the periods of the three major drought events (1930s, 1950s and current), we find that precipitation has a much higher correlation with runoff than with soil moisture. The autocorrelation is higher for the current drought, which suggests that dry (or wet) spells during the current drought were longer than during the other two droughts. Because runoff responds faster to precipitation, parts of the areas under hydrological drought in the 30s and 50s recovered, thus decreasing the average severity. Severity-Area-Duration Analysis Depth-area-duration analysis, widely used in characterization of precipitation extremes for specification of so-called design storms, provides a basis for evaluation of drought severity when storm depth is replaced by an appropriate measure of drought severity. For the purposes of this paper, drought severity (S) is defined as S=(1-ΣP/t), where ΣP is the percentile of soil moisture or runoff summed over duration t (in months). We adapted the computational method of WMO (1960) to calculate the average severity corresponding to each standard area for each drought identified. In this study, we examine durations of 3, 6, 9, 12, 24, 36, 48, and 72 months and areas from 10 grid cells, or approximately 25,000 km2, to the maximum drought extent, in increments of 20 grid cells, or approximately 50,000 km2. Figure 6 shows a flow diagram of the entire methodology. Threshold VIC model output Total column soil moisture Runoff 20th Weibull percentiles percentile and lower soil moisture 30th percentile and lower runoff Spatial contiguity Initial Temporal drought contiguity classification Final drought and subdrought classification Severity-AreaDuration SAD curves for each event Highest severities Figure 6. Overview of methodology from initial input to envelope curves. Envelope curve for each duration Construction of SAD Curves: 1. Rank cells by severity & identify potential drought centers 2. Search 3x3 neighborhood of drought center 3. Average severities & add areas 4. Output severity and area at specified area intervals Figure 5. Process of computing area and area-averaged severities for SAD analysis. (1) Start with maximum severity grid cell, or “drought center”. (2) Add area and average severity with most severe grid cell that is contiguous with drought center. (3) Continue adding next most severe grid cells until prescribed areal extent is reached.