Fig. 5: Information decomposition of internal signals during the simulation of representative flood events. | npj Climate and Atmospheric Science

Fig. 5: Information decomposition of internal signals during the simulation of representative flood events.

From: Shifted dominant flood drivers of an alpine glacierized catchment in the Tianshan region revealed through interpretable deep learning

Fig. 5

a Information decomposition from the 10th fold of the LSTM-SG model during the simulation of the AMF on July 1, 2016. b Information decomposition from the 10th fold of the LSTM-DDF model during the simulation of the AMF on July 1, 2016. c Information decomposition from the 10th fold of the LSTM-SG model during the simulation of the AMFSp on May 14, 2018. d Information decomposition from the 10th fold of the LSTM-DDF model during the simulation of the AMFSp on May 14, 2018. Subplots (i) to (iii) show the snowmelt, glacier meltwater, and rainfall amounts in the previous 180 days used to simulate floods. The multiple lines in subplots (iv) to (ix) denote signals in the respective hidden units. Subplots (x) to (xii) correspond to the information at every time step extracted from the second column.

Back to article page