Precipitation exhibits variability over a wide range of spatial and temporal scales. Numerous studies have underlined the fact that, for basins of any size, the hydrologic response to precipitation…

The University of California, Irvine received a $459,443 research grant from a civilian federal agency's Shared Services Center to advance satellite-based precipitation measurement and monitoring. The award, issued February 4, 2022, with completion targeted for February 3, 2025, supports research addressing the complex spatio-temporal variability of precipitation across multiple scales relevant to climate, weather, and hydrologic applications. The research encompasses two primary objectives: developing a comprehensive framework for space-time satellite precipitation error diagnostics, uncertainty quantification, and error modeling; and improving and developing operational retrieval algorithms to better capture the space-time dynamics of precipitation systems. This work is particularly significant because extreme hydrologic responses, including catastrophic flooding, are often driven by the spatial and temporal structure of storm systems rather than total precipitation amounts alone, making accurate multi-scale precipitation measurement critical for both scientific understanding and operational forecasting.

UCI's research will employ spectral representations and advanced validation methodologies to assess satellite precipitation products across relevant space-time scales. The project addresses a fundamental challenge in earth observation: existing precipitation measurement systems have inherent limitations in spatial and temporal sampling and instrument resolution, resulting in scale-dependent performance that varies across different precipitation modes. By developing improved error diagnostics and enhanced retrieval algorithms, this research will support better precipitation characterization for climate modeling, weather prediction, and hydrologic forecasting applications. No set-aside designation was applied to this award.

Precipitation exhibits variability over a wide range of spatial and temporal scales. Numerous studies have underlined the fact that, for basins of any size, the hydrologic response to precipitation depends not only on the precipitation average over the basin but is also strongly dependent on the spatio-temporal structure of the precipitation systems. In particular, several case studies of catastrophic flood events report that some extreme hydrologic responses can be linked to the particular spatio-temporal dynamics of a storm rather than to an unusual total precipitation amount. The spatio-temporal structure of precipitation is also an important aspect of the global weather and climate system, with the organization of convection playing a particularly significant role in the earth s energy balance. In addition to causing complex hydrologic responses and complex atmospheric energy flux, the multi-scale variability of precipitation renders its measurement particularly challenging. Indeed, any observation system of precipitation has inherent limitations associated with the spatial and temporal sampling and instrument resolutions. Moreover, a given observation system may show varying performance in capturing different modes of precipitation variability, ultimately leading to a scale-dependent retrieval performance. The proposed research has two main objectives: (1) develop and implement a framework for space-time satellite precipitation error diagnostics, uncertainty quantification and error modeling across space-time scales relevant to climate, weather and hydrologic applications and (2) improve existing operational retrieval algorithms and develop new algorithms specifically geared toward enhanced representation of the space-time dynamics of precipitation. Regarding objective 1, we propose to develop a methodology for product assessment and validation focused on the multiscale spacetime dynamics of precipitation, relying in particular on spectral representations. Within this multiscale spectral framework, we will develop a quantitative error model for computing sensor-dependent uncertainty estimates. The uncertainty quantification is not only important for the end user who often needs information on the accuracy of precipitation estimates but IT is also a necessary step for optimal merging of multi-sensor observations. The developed methodologies will first be tested in the continental US using mrms as the reference product, followed by analysis in other parts of the world for which ground observations are missing and other innovative indirect assessment methodologies have to be developed. Regarding objective 2, we will work on modifying existing operational retrieval and merging algorithms and on developing new algorithms for enhanced representation of the space-time dynamics in multi-sensor precipitation estimates. For this purpose, we will explore the use of mixed-scale deep convolutional neural networks which offer the advantage of efficiently learning space-time features and keep in memory the whole pathway of learning for improved performance. We will also focus on defining appropriate objective functions that go beyond the pixel-wise penalty metrics to explicitly incorporate the space-time dynamics of the target variable in machine learning algorithms.