Water Resources Research Act Program

Details for Project ID 2014AR352B

Improved ensemble forecast model for drought conditions in Arkansas using residual re-sampling method

Institute: Arkansas
Year Established: 2014 Start Date: 2014-03-01 End Date: 2015-02-28
Total Federal Funds: $8,846 Total Non-Federal Funds: $18,852

Principal Investigators: Yeonsang Hwang

Project Summary: Increasing water use for agriculture activities, power generation, and municipal growth has added concerns to water resources sustainability in the state of Arkansas. Rapid growth of agricultural activity in the Northeast Arkansas is also putting the entire aquifer system under great stress. Although effective use of water resources would greatly help to reduce the current stress on our aquifer systems, further improvement would be extremely difficult without a good understanding of temporal and spatial surface water availability. Because of the possible shift of climate regime, the tendency of increased extreme climate events in the US would impact the short-term variability of drought in Arkansas. Prediction of hydro-climate variables is challenging, but very important in water management and planning in the region. The proposed research idea is a part of an effort to address the need for robust and flexible forecast methodologies for hydro-climate variables and comprehensive water management plans in Arkansas. This proposal focuses on regional stochastic drought forecast model using Palmer Drought Severity Index (PDSI). Climate and hydrologic time series including drought indices are non-linear random processes that can be effectively described by stochastic models. However, we hypothesize that we will be able to utilize a stochastic model to forecast drought conditions for the climate divisions of Arkansas. In order to test this hypothesis, we will test a data-driven stochastic model to construct a reliable forecast model for hydrologic and climate variables. Although the PDSI time series is very well known for its auto-regressive nature, a non-parametric regression technique will enhance the predictability of our model under highly non-linear climate conditions including extreme meteorological events. In order to provide probabilistic forecasts, an ensemble technique wil be used to provide sound statistical representation drought conditions. A non-parametric forecast ensemble generation technique will look up the record similar to our scenario (or given climate change projection) in historical data (e.g., precipitation, temperature, PDSI, etc) to produce reliable and accurate probabilistic representation of historical drought conditions. Our long-term projection of drought states will also be tested through a similar approach.