Uncertainty Aware Machine Learning based Quantitative Precipitation Estimation from Geostationary Satellites.
Abstract
High frequency quantitative precipitation estimation (QPE) is a challenging yet crucial process in the management of hydrometeorological catastrophes such as flash floods or flash droughts. Early detection of extreme weather events with ground- or space-based radar data plays an important role in saving human lives and property. However, ground radars are constrained by limited spatial coverage, while satellite-based precipitation products often cannot handle convective events well because of limitations in applied parameter retrieval algorithms. We propose a machine learning (ML) based framework for QPE, with uncertainty quantification (UQ), from NOAAs GOES-16/17 Advanced Baseline Imaging (ABI) sensors, which observe 16 spectral bands at high spatiotemporal frequency. We use the GeoNEX L1G dataset provided by the NASA Earth eXchange (NEX), which are geocorrected, regridded, and solar-corrected. We will retrieve precipitation rate maps from geostationary satellites by incorporating the gridded ground radar precipitation measurements from the NEXRAD Multi-Radar Multi-Sensor (MRMS) System. The MRMS precipitation measurements are available at a spatial resolution of 1-km and temporal resolution of 2-minutes. A training dataset of GOES-16/17 and MRMS is generated to develop uncertainty-aware ML-based QPE from satellites. Our first hypothesis is that GOES-16/17, with higher spatiotemporal resolution and 16-spectral bands (including 10 thermal bands), can improve QPE over previous generation satellites (GOES 14/15) with 5 spectral bands. Our second hypothesis is that satellite-based QPE with UQ can be further developed for hydrological predictions and water resources risk assessments. We will examine the generalizability of our approaches across a diversity of sensors, as well as the possibility of transfer-learning from data-rich to data-poor regions. This can be a first step toward assessing the value of the latest and state-of-the-art precipitation measurements from satellites in terms of addressing scientific and stakeholder problems in hydrology and water resources.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.A35F1705D