Short-term Prediction of Drought Evolution using SMAP

Prolonged drought has devastating societal impacts on aspects ranging from recreation to water and food supply and can ultimately result in significant financial losses. In a warming world, the frequency of seasonal- and multi-year drought episodes is likely to increase. Those in the poorest countries who rely on domestic ecosystem services (including food production) are the most vulnerable. Accurate drought prediction has the potential to reduce negative impacts and support societal resiliency by informing sustainable water-management and planning practices. Such predictions require a thorough understanding of the physical mechanisms that lead to the onset, persistence, and recovery of drought and the ability to replicate these processes in forecast models at appropriate time scales (weeks to months). To do this requires better observations of key processes that drive drought evolution. While synoptic drivers play an important role in drought development, there is increasing evidence that the land surface plays a key role in the skill of sub-seasonal to seasonal forecasts. However, there is still a lot unknown about the direct impact of local processes, such Land-Atmosphere (L-A) interactions, on drought evolution due to low signals relative to the noise of natural variability and insufficient observational data sets. The aim of this work is to produce and analyze statistical based predictions of drought evolution based on the persistence of L-A interactions using a variety of techniques and assess the importance of including direct measurements of soil moisture using remotely sense observations from SMAP. This will be done by producing short-term (30 day) predictions of precipitation and the Coupling Drought Index (CDI) and comparing them with observations from reanalysis over a period of 2015-2021. Two different statistical models will be used in this analysis. First, the Coupling Stochastic Model (CSM), which is a Markov-Chain model based on the persistence in L-A coupling regime. The second model will use deep learning to create a spatially consistent prediction of the evolution of the L-A coupling regime. To isolate the impact of soil moisture on the short-term predictions, each of these models will be run with and without incorporating SMAP observations. The strengths and limitation of each prediction model will be discussed as well as future directions for improving drought forecasting through observations.