Data and code underlying the PhD thesis: Deep Learning and Earth Observation for the Study of West African Rainfall

The PhD thesis "Deep Learning and Earth Observation for the Study of West African Rainfall" develops a Deep Learning-based satellite rainfall retrieval model for West Africa, called "RainRunner". RainRunner classifies 3-hour sequences of Meteosat Second Generation (MSG) WV and TIR images in rain/no-rain. After being trained in Northern Ghana, RainRunner is applied to a wider area in West Africa (the Sudanian Savana), to evaluate generalization capability and understand better the rainfall mechanisms in the wider area. This dataset allows to do a full performance evaluation of the model by downloading and processing MSG data to create the test dataset, applying the model and evaluating the results. More information about the exact goal of each script can be found the README.txt file.

博士论文《面向西非降雨研究的深度学习与地球观测》研发了一款面向西非区域的基于深度学习的卫星降雨反演模型，命名为“RainRunner”。该模型可对欧洲静止气象卫星二代（Meteosat Second Generation, MSG）的水汽（Water Vapor, WV）与热红外（Thermal Infrared, TIR）影像的3小时时序序列进行降雨与非降雨分类。该模型在加纳北部完成训练后，被应用至西非更广范围的苏丹稀树草原（Sudanian Savana）区域，以评估其泛化能力，并深入理解该区域的降雨形成机制。本数据集支持通过下载并处理MSG影像以构建测试集、部署模型并评估结果的完整流程，从而对该模型开展全面的性能验证。各脚本的具体功能详情，可查阅README.txt文件获取。