Application of Deep Neural Networks and Geographical Information for Improving the Near Real-time Precipitation Estimation Products
Abstract
Accurate precipitation estimates at high spatial and temporal resolutions are essential as input for many hydro-meteorological and water resources applications and studies. In addition, providing reliable near real-time precipitation estimates is essential for monitoring and managing of natural disasters such as floods. Quality of input data and capability of the retrieval algorithm are two important aspects for developing a satellite-based precipitation dataset. Most retrieval algorithms utilize infrared (IR) information as their input due to its fine spatiotemporal resolution and near-instantaneous availability. However, their sole reliance on IR cloud-top temperature limits their capability to learn different mechanisms of precipitation during training, resulting in less accurate estimates. Moreover, recent advances in the field of deep neural network (DNN) offer attractive opportunities to improve the precipitation retrieval algorithms. This study investigates the effectiveness of adding geographical information (i.e. latitude and longitude) to the IR information as inputs for improving the accuracy of precipitation retrieval algorithms. Furthermore, the study explores the application of a U-Net-based convolutional neural network (CNN), one of the most efficient DNN frameworks. Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks (PERSIANN)-Cloud Classification System (CCS), which is an operational satellite-based product, is used as a baseline. Results demonstrate that a CNN-based model—leveraging both geographical and IR information—shows a higher estimation accuracy than a model that only utilizes IR information. Furthermore, the model which applies U-Net CNN architecture on geographical and IR information (referred to as PERSIANN-CNN) shows better agreement with Stage IV radar data compared to PERSIANN-CCS in terms of detecting rain/no rain and estimating the intensity of precipitation over the CONUS. This research suggests that applying an appropriate CNN architecture on geographical and IR information provides an opportunity to improve the current satellite-based precipitation products.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMH020...03S
- Keywords:
-
- 3354 Precipitation;
- ATMOSPHERIC PROCESSES;
- 1817 Extreme events;
- HYDROLOGY;
- 1854 Precipitation;
- HYDROLOGY;
- 1855 Remote sensing;
- HYDROLOGY