Introduction
Drought is a natural phenomenon that can result in significant economic loss and grave damages to society. Drought and heat waves resulted economic loss between 1980-2007 tolls up to $171 billion (NOAA NCDC report).Number of drought declarationsmade across United Statessince 1967 is 46 with the most recent drought declared on July 31, 2007 over South-East United States (FEMA). Occurrence of drought is not unknown to WashingtonState. In the last decades the state has experienced two major statewide drought in 2001 and 2005 both of which resulted in enormous economic losses of $300 million and $500 million (Fontaine and Steinemann, 2007). Additionally future of water supply in the state is anticipate to decline owing to global warming and resulted decrease in snowpack (Barnett et al. 2008, Mote et al. 2005, 2006). Repeated occurrence of drought and its expected continuance emphasizes the need of shift in the drought management approach to a more proactive risk based (Wilhite 2000). Timely and accurate information about the development of drought conditions and the water supply outlook is needed to be prepared and trigger attempts to mitigate losses (Hayes, 2005, book). The drought indicators and triggers are important to detect and monitor drought conditions; to determine the timing and level of drought responses; and to characterize and compare drought events (Steinemann et al).
Significant progress has been made towards development of national level drought monitoring and prediction portals. National Integrated Drought Information System (NIDIS) is such a drought monitoring and prediction system which providesassessment of drought exhibited across the nation, based on the various portals including United States Drought Monitor (USDM) (Svoboda et al 2000), United States Geological Survey hydrological drought portal and North American Land Data Assimilation System (NLDAS) project Land Surface Model (LSM) based drought monitoring systems. USDM is based on the blend of multiple indices and the human expertise to provide drought assessment and the associated impacts. The drought assessment in the USDM is done by merging various information such as observed streamflow, reservoir storage, drought indices, vegetation cover map and the modeled soil moisture. Drought characterization and delineation in USDM is thus achieved by either using the direct measures of droughts such streamflow characterization of drought events is essential to initiate the relevant drought management plans. Climate indices used in USDM such as PDSI and SPI are the indicator of meteorological drought only.
The weight given to different meteorological drought indices and their time-averaged for different time periods varies for short term and the long term drought analysis. In additions to this a different formulation is applied for Western United States to take into account of the time lag occurred because of snowpack by including the snow water content and the water supply indices. It also incorporates observed streamflow and uses long term averaged SPI as the surrogate for reservoir information. The one of the main reason why there is more weight to the drought indices based on climate only is because precipitation and temperature are the major driver of the drought and also precipitation and temperature data are easily available at more spatial density and for the longer time period. Soil moisture and runoff data are indicator of agricultural and hydrological drought. Given the dearth of the long term soil moisture and streamflow data across the United States the data sets produced by North American Land Data Assimilation System (NLDAS) models become valuable alternatives for drought monitoring and prediction (Kingste, 2008). Owing to the physical basis of these models they encompass all the physical process which leads to any drought events. They are well capable of mimicking the hydrological characteristics of a basin and thus can simulate the time lag when a meteorological drought actually results into soil moisture and runoff deficiency. In addition to this use of the LSM is also promising since the output can be obtained at a finer resolution which can be easily aggregated to any geographical area (such as Counties, Watershed, Hydro-ClimateZoneetc.)
LSM derived soil moisture and runoff data have already been utilized to reconstruct and characterize drought events. Soil moisture percentile was used to perform a retrospective drought analysis for United-states (Scheffield et al. 2004; Andreadis et al. 2005, 2006; Kingste, 2008) and trend analysis of drought events during 20th century (Andreadis and Lettenmier 2006) showed to perform well as an indicator for the vegetative growth hence the agricultural drought can be represented by total soil moisture percentile (Scheffield et al, 2004;) As the moisture deficiency period elongates it can result into reduced streamflow and hence inflow to reservoir, ponds and lakes which is defined as to be a hydrological drought. Lack of streamflow is the indicator of hydrological drought. Model derived runoff percentile (Scheffield et al. 2004; Andreadis et al. 2005) and Standardized Runoff Index (SRI) incorporate the seasonal lag between in the influence of climate information on streamflow (Shukla and Wood, 2008; Kingste, 2008) thus are indicator of hydrological drought.
In the present study we describe the implementation of a land surface model based drought monitoring and prediction system for WashingtonState. Observed precipitation data along with the simulated soil moisture and runoff data are used to reconstruct and characterize the historical drought events of 2001 and 2005. Model based drought characterization is compared with the other independent drought measures such as observed streamflow and reservoir conditions.
Furthermore we also demonstrate how drought prediction skill can be derived from the hydrologic initial conditions alone when merged with the ensemble of weather scenarios taken from the past. Through various numerical experiments done previously it has been established that soil moisture feedback is important in deciding the duration of anomalous dry events. Inertia of hydrologic conditions namely soil moisture makes it possible to anticipate the future outlook. We evaluate the applicability of such prediction system and its usefulness for 2001 and 2005 drought events and diagnose how well this system does in detecting the onset and termination of different forms of drought.
SRI is a measure of hydrological drought which in turn is associated with the streamflow, municipal water supply and reservoir storage. Therefore drought assessment based on SRI is potentially useful for various water management decisions such as imposing restrictionsonthe water rights, initiation of water conservation practices and raising water rates. Water management and fulfilling only the true water needs during a drought event can not only relieve the effects of drought but also help offset the adverse impacts of long-term shortages (Vickers, 2005, book). The drought indices includingSRI provide information on the level of drought situation however information like how long the current water supply is going to last in a no-rain condition can not be easily interpreted from these indices. For the enforcement of water conservation measures during a drought event it is necessary to know reasonably well how the future might look like. Drought outlook under given climatological conditions is thus needed. One such system is Climate Prediction Center (CPC) Seasonal Drought Outlook (SDO) which issues a drought outlook map monthly. The accuracy of the information provided by this system is limited by the short-term climate forecast. The use of hydrological state in conjunction with the climate forecast for drought prediction is
We propose two new drought metrics “Days of supply remaining” and “Days before recovery” can be used in conjunction with the other drought indices. They can be directly applicable for the water management decisions such as imposing restrictions on the water uses and initialization of conservation practices. We used Variable Infiltration Capacity (VIC) model for the probabilistic estimation of the “Days of supply remaining” and the “Days before recovery”. At first the model was forced with the observed precipitation and temperature forcings to generate the current hydrologic state. To estimate the “Days of supply remaining” and “Days before recovery” the model was then forced with the ensembles of forcings. “Days of supply remaining” was estimated for the little/no precipitation scenarios, which is basically the number of days after which the current soil moisture/runoff will hit the bottom threshold. On the other hand “Days of recovery” which is when the current soil moisture/runoff values approach the normal value is estimated for each of the other rather favorable precipitation and temperature scenarios.
2.0 Study domain:
This study was performed for the WashingtonState. Given the hydrological dissimilarities between the western and eastern part of thestate it provides the opportunity to assess the performance of model based drought analysis on both the snow-melt and rainfall driven water resources. The annual average precipitation in WashingtonStatevaries from less than 20 inches to more than 150 inches. The water supply is critical in the state for various purposes namely agricultural production, hydropower, municipal uses and the fish production. The state is divided in the 62 Water resources inventory areas (WRIAs) which represent all the major watersheds. The state has experienced 24 major drought events during the last and current centuries. The further details about the drought in WashingtonState are given in the section 2.4.
2.1 Hydrology model description
The macro-scale physically based, semi-distributed VIC model (Liang et al. 1994) was used in the present study to derive the soil-moisture and runoff data over the domain of WashingtonState. The VIC model balances both surface energy and water over each grid cell (1/16th degree, approximately 6km for this study). VIC accounts for the feedback of vegetation on land-atmosphere moisture and energy fluxes like other soil-vegetation-atmosphere transfer schemes, VIC as it represents the sub-grid variability of the soil, topography and vegetation. This feature of VIC allows the representation of the observed non-linear dependence of soil-moisture in the partitioning of precipitation into infiltration and direct runoff. The VIC model has three soil layers. The first layer is 10 cm deep and responses quickly to changes in surface conditions and precipitation. Moisture movement between the soil-layers is governed by gravity and diffusion to the upper layer is allowed in unsaturated conditions. Water from the second layer drains to the third layer and the base flow is a non-linear function of the moisture content in the third soil-layer. The depths of the mid and bottom layers, infiltration shape parameter (binf) and the base flow parameters (Dmax, Ds and Ws) were adjusted during the calibration process of simulated streamflow to match with observed naturalized flow.
VIC has been successfully used in numerous studies to simulate streamflow over many large river basins in the United States and other parts of the world (Nijssen et al. 1997, 2001; Wood et al. 1997; Maurer et al. 2002). Maurer et al 2002, have compared the VIC simulated soil moisture values and flux with the observed values recorded over the 17 stations uniformly distributed over Illinois, and concluded that although VIC underestimates the soil-moisture values, it simulates the average monthly flux accurately, which influences the model’s water balance.
2.2 Retrospective Simulation:
A retrospective run of VIC was done for the period 1915-2005 at 1/16th degree resolution and daily time step over WashingtonState in the water balance mode only. There are 5282, 1/16th degree cells in this domain. Daily precipitation, maximum Temperature (Tmax), minimum Temperature (Tmin) data were acquired from the Applied Climatic information System and were used to generate the meteorological input forcings. Data from 196 stations in Washington, Montana, Oregon and Montana states were gridded at 1/16th deg, using the method described in Maurer et al. (2002). Precipitation data were lapsed based on the difference in the station elevation and grid elevation. Both precipitation and temperature were than rescaled to match the long-term average of the parameter-elevation regressions on independent slopes model (PRISM) climatology (Daly et al. 1994, 1997). This climatology is the monthly mean of the precipitation, Tmax and Tmin over the 30 year period (1971-2000).
2.3 Model based drought indices:
2.3.1 Soil moisture percentile: Soil moisture is an indicator of agricultural drought since it is crucial for the plant growth. For the present study daily soil moisture data were aggregated to the monthly values. The 91 years monthly soil moisture data for each grid cell serves as the climatological distributionbased on which absolute soil moisture values were converted into the percentiles. The method adopted is essential the same as in Andrieadis et al 2005.
2.3.2 Standardized runoff index (SRI): SRI is an indicator of hydrologic drought (Shukla and Wood, 2008; Moe 2008;) Model derived runoff data (overland plus baseflow) are utilized for the estimation of SRI. It is done on the same analogy as Standardized precipitation index (SPI) (Mackee et al, 1997). The daily runoff data were converted into monthly values and then log-normal distribution was fitted on the data. The percentile values thus obtained were then converted to the standard normal values. For the details about the SRI refer to (Shukla and Wood 2008)
2.3.3 Probabilistic estimation of drought recovery period:
Estimation of the time before recovery from the ongoing drought is necessary besides providing drought severity to help in the water management decisions like water use restrictions. In the present study we also demonstrate how the skill in the estimation of the drought period can be made possible based on the initial hydrologic conditions only. . Owing to the inertia in the initial hydrologic state skill in the prediction of the drought conditions can be derived from the initial hydrologic conditions. This method of probabilistic estimation of drought recovery is essentially the same as the popular method of streamflow forecast which is Ensemble Streamflow Forecast (1977) which is done by merging the information on initial hydrologic condition with the ensembles of the weather elements taken from the past. It is assumed that the variability in the climate can be represented by the set of the ensembles chosen from the past. To demonstrate the applicability of such probabilistic estimation we picked a few cases of ongoing drought of different severity and the beginning from the true hydrologic conditions the model run were done out to one year. The ensemble of soil moisture and runoff data were then converted to the percentile values using the climatological distribution. The recovery from the drought is considered when the percentile values reach back to the 75% of the normal. This criterion was kept a little conservative to allow some time lag after the drought is over.