Precipitation Retrieval From Fengyun-3D MWHTS and MWRI Data Using Deep Learning

Abstract

In this paper, two multitask deep learning models, Multi-layer Perceptron (MLP) and Convolutional Neural Networks (CNN) are constructed to detect precipitation flags and retrieve precipitation rates simultaneously over the Northwest Pacific area. The retrieval results are verified by Integrated Multisatellite Retrievals for GPM (IMERG). Compared with using a single payload, the results show the retrieval advantages of incorporating two microwave spaceborne payloads (MWHTS and MWRI). In addition, each separated task in multitasking deep learning models can outperform the single-task models. The MAE values of MLP and CNN are decreased by 0.05 mm/hr and 0.04 mm/hr, respectively. Nevertheless, it is still a challenge to extend the application of these models, which demands extracting the features learned from a certain area to other areas characterized by different precipitation features. A flexible network framework is developed with two Transfer Learning (TL) methods (freeze and fine-tuning) to tackle this problem. The pre-trained models (MLP and CNN) are trained from the Northwest Pacific area, whereas the transferred models are applied to the Northeast Pacific area. The encouraging results show that the performance of the TL model using fine-tune method has exceeded that of the comparison model without TL, which is trained by 4 times more data. The main advantage of using TL is the time reduction and efficiency improvements.

Gap Filling of the ESA CCI Soil Moisture Data Using a Spatiotemporal Attention-Based Residual Deep Network

“…Transfer learning is a commonly used deep-learning strategy for limited training samples, which usually transfers the similar knowledge from the source domains to target domains through pretraining and fine tuning, thereby alleviating the overfitting problem to a certain extent and further improving model performance [33]. It has been widely applied in image classification [40], soil organic content prediction [41], and meteorological forecasting [42], [43]. The feasibility of applying transfer learning for SM predicting has also been demonstrated by Li et al [33], and this article further examined the potential of utilizing transfer learning to impute SM gaps.…”