Precipitation Estimation from Remotely Sensed Data Using Deep Neural Networks

Abstract:

This research develops a precipitation estimation system from remotely-sensed observations using state-of-the-art machine learning algorithms. Compared to ground-based precipitation measurements, satellite-based precipitation estimation products have advantages of temporal resolution and spatial coverage. Also, the massive amount of satellite data contains various measures related to precipitation formation and development. On the other hand, deep learning algorithms were newly developed in the area of machine learning, which was a breakthrough to deal with large and complex dataset, especially to image data.

Here, we attempts to engage deep learning techniques to provide hourly precipitation estimation from long wave infrared data from operational geostationary weather satellites. The brightness temperature data from infrared data is considered to contain cloud information. Radar stage IV dataset is used as ground measurement for parameter calibration. Denoising stacked auto-encoders (DSAE) is applied here to build a 4-layer neural network with 1000 hidden nodes for each layer. DSAE involves two major steps: (1) greedily pre-training each layer as an auto-encoder using the outputs of previous trained hidden layer output starting from visible layer to initialize parameters; (2) fine-tuning the whole deep neural network with supervised criteria.

Rain/No-rain classification is dealt as the first step of precipitation estimation in this research. Our experiments show that deep neural networks outperform the classic approach originally used in developing the PERSIANN-CCS (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks Cloud Classification System). In an experiment over a 3-month summer period focusing on the central U.S with hourly data, the proposed approach’s Probability of Detection (POD) increased to 0.433 as compared to PERSIANN-CCS value of 0.403 and decreased the False Alarm Ratio (FAR) to 0.606 as compared to 0.633 for the PERSIANN-CCS.