Examine the potential of spatial downscaling of TRMM precipitation with environmental variables: An evaluation for the Ohio River Basin

Friday, 19 December 2014: 9:30 AM
Yeosang Yoon and Edward Beighley II, Northeastern University, Department of Civil and Environmental Engineering, Boston, MA, United States
Accurately quantifying precipitation in both space and time is a central challenge in hydrologic modelling. Data products from the Tropical Rainfall Measuring Mission (TRMM) are commonly used as precipitation forcings in many models. TRMM provides 3-hr precipitation estimates at a near-global scale (-50 S to 50N) with a 0.25 degree spatial resolution. However, when applied in regional scale hydrologic models, the spatial resolution of the TRMM is often too coarse limiting our ability to simulate relevant hydrologic processes.This study focuses on addressing the science question: can we improve the spatial resolution of the TRMM using statistical downscaling with environmental variables derived from finer scale remote sensing data? The goal is to downscale the TRMM resolution from 0.25 degrees (25 km) to 0.05 degrees (about 5 km). In our approach, we first identify environmental variables (i.e., vegetation cover, topography, and temperature) that are related to the formation of or result from precipitation by exploring their statistical relationships with TRMM precipitation at varying temporal scales (i.e., daily, monthly, and yearly) using an analysis of variance in multiple regression. The MODIS vegetation index, MODIS leaf area index, and SPOT vegetation are examined as a proxy for vegetation. To represent the topography, the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) is used. MODIS land surface temperature are used for temperature. Second, we characterize a residual component, which cannot be explained by the statistical relationship between precipitation and environmental variables, to improve the accuracy of the downscaling results. For example, recent studies have shown that approximately 30-40% of the variability in annual precipitation cannot be explained by vegetation and elevation characteristics. According for this unexplained variability in statistical downscaling methods is a significant challenge. Here, we use a data assimilation technique to interpolate the residual component and generate the downscaled precipitation. Results are presented for the Ohio River Basin. The final downscaled TRMM is evaluated by comparing with the National Center for Environmental Predication (NCEP) Stage VI precipitation and rainfall gauge data.