DESCRIPTION
We implement an application of Convolutional Long-Short Term Memory neural network proposed by Shi et. al, 2015 in 3D precipitation nowcasting in Kobe city, Japan. We show a significant improvement using this method, in comparision with an optical-flow based method of Otsuka et. al, 2016
Contact me if you need source code of this application.
Architecture of ConvLSTM
ConvLSTM
Conditional ConvLSTM: combine with optical-flow based forecasting
Comparision in term of 3 metrics: MSE, B-MSE, threat score CSI
-
Quantitative evaluation
-
Horizotal & Vertical evolution of rainfall
