A space-time error model for satellite precipitation products

Abstract:

Satellite multi-sensor precipitation datasets are a valuable resource for users around the globe seeking to manage water resources and model processes which depend on rainfall, especially in regions with limited ground-based precipitation data. NASA’s 30-minute, 0.1º IMERG Early product is publicly available at a 4-hour latency and is used broadly in applications such as flood monitoring, landslide hazard assessment, and land use modeling. A continuing challenge to accounting for satellite rainfall uncertainty is the need to replicate correlation of errors in time and space. As satellite precipitation products advance to higher resolutions, accounting for the spatiotemporal correlation of errors is increasingly important.

This work presents a framework for generating IMERG-based spatiotemporally correlated precipitation ensembles which incorporate IMERG uncertainty. The spatial structure and anisotropy of IMERG at a given time is used to generate correlated noise fields and large-scale wind processes are used to model the evolution of error fields over time. Spatiotemporally correlated noise fields are then paired with an IMERG error model to generate fields of possible true rainfall. The resulting ensemble reflects possible true realizations of rainfall based on IMERG estimates and known error properties of IMERG.

Rainfall ensemble members produced using this methodology over the central U.S are able to better replicate the characteristics of ground-reference products (NEXRAD Stage IV) and exhibit reduced RMSE and bias relative to IMERG. These ensembles can be used as input to a hydrological model or land surface model to account for IMERG uncertainty without any adjustment to the model. The ensembles produced by this space-time error model can help users to better understand the impacts of rainfall uncertainty and the benefits of accounting for IMERG uncertainty.