Exploring mechanisms and predictability of U.S. drought using remote sensing, hydrological modeling, reanalysis and the NCEP Climate Forecast System

Understanding the mechanisms of drought initiation, maintenance and recovery is a crucial part of drought prediction. We use a synthesis of remote sensing observations, observation-driven land surface hydrologic simulations, reanalysis and seasonal climate forecasts to diagnose the occurrence of U.S. drought over the past 30 years and its predictability. Large scale drought events are idenitifed using long-term (1915-present), observation-driven simulations with the VIC and Noah land surface models and compared to that derived from the North American Regional Reanalysis (NARR) for 1979 onwards. Drought is characterized in terms of spatially contiguous soil moisture deficits as defined by runs of monthly percentile values below a given threshold (20th percentile). U.S. averaged time series of soil moisture percentiles and contiguous area in drought indicate that the three datasets co-vary reasonably well. However, large regional differences exist, in particular the NARR fails to pick up the western drought in the mid-2000s. There are many other differences in the magnitue and timing of drought events between the datasets, especially in the wetter east where high variability in drought events (spatially extensive but short duration) is seen in the VIC and Noah datasets but not in the NARR.

The mechanisms of drought development are governed by interactions between all components of the local land-atmosphere water budgets and the remote source and strength of atmospheric moisture. Despite using very similar sources for precipitation (which is assimilated in the NARR), it is partitioned into evaporation and runoff quite differently in the NARR versus the off-line simulations, which impacts the depiction of drought. This is especially the case in mountainous and snow-dominated regions, where large underestimation of snow by the NARR (despite ingestion of snow obervations) leads to differences in the timing and magnitude of peak runoff. For the atmospheric budgets we identify and inter-compare the seasonal sources of atmospheric moisture for a number of regions from reanalysis and satellite derived products (NVAP, AIRS), which compare well in terms of annual variation but diverge in the West in the winter and southwest in the summer. For the identified drought events, the land-atmosphere water budgets and moisture advection are analyzed to see how they differ from their climatological mean.

We evaluated how well seasonal hindcasts of the NCEP Climate Forecast System (CFS) are able to replicate the mechanisms for drought initiation and development in terms of water budget anomalies, advected moisture sources and the rates of precipitation recycling. This is based on hindcasts back to 1981 but with a detailed focus on 2007 using recent operational forecasts and in particular the drought in the southeast. Hydrologic forecasts driven by the bias corrected and downscaled CFS precipitation and temperature show reasonable skill in predicting the 2007 drought conditions. The CFS hindcasts (which are run in coupled mode) depend, in part, on how well they replicate observed SST anomalies, particularly for the southwest and other regions affected strongly by SSTs. The analysis is repeated for AMIP-type simulations, which use specified SST boundary conditions, to help diagnose the source of skill, whether from the SST anomalies, the atmospheric response or the land response.