The challenge of rainfall estimation and prediction across scales: learning from patterns
Abstract
The need to observe precipitation globally at high spatial and temporal resolutions and with high fidelity has never been more imperative, driven by the urgency to understand the global water cycle and how it might change in the future, as well as to improve climate and earth system models. Observations from space offer a unique vantage point, but face immense challenges in developing algorithms that transform indirect measurements from different platforms to surface precipitation with sufficient resolution to inform decision making and research applications. At the same time, seasonal precipitation prediction using a combination of observations and models is advancing, but still lacking the accuracy and reliability needed to guide critical water resource management decisions around flood control, drought resilience and water supply. We explore a new direction that more explicitly acknowledges and leverages "patterns", namely, the organized structure of storm systems and climate variables, to enhance estimation and prediction. In satellite estimation from microwave sensors, we propose to move away from the pixel-wise retrieval approach, and learn from the neighborhood of each pixel by deriving storm-wide context information and nonlocal radiometric attributes to enrich the spectral information and increase retrieval accuracy. In seasonal prediction, we optimally combine the space-time structure of oceanic and atmospheric variables learned from climate models with advanced graph-based estimation techniques to unravel the most explanatory predictors and interpret them physically. These advances hold promise for hydrologic applications that rely on accurate precipitation input.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H24A..01F
- Keywords:
-
- 1899 General or miscellaneous;
- HYDROLOGY