Complexity to Forecast Flood: Problem Definition and Spatiotemporal Attention LSTM Solution

Table 3

Comparison experiments on Changhua dataset to test effectiveness of attention module. Note that bold and underline texts indicate best and secondary best performance among comparative methods, respectively. We define unit for M as %.

Methods	RT+3	RT+6	RT+9	RA	Rσ	MT+3	MT+6	MT+9	MA	Mσ	DT+3	DT+6	DT+9	DA	Dσ
With spatial	54.51	79.77	95.26	76.51	1.42	13.59	19.93	23.81	19.11	0.34	0.84	0.69	0.60	0.71	0.02
With temporal	54.33	78.46	94.61	75.80	0.71	13.55	19.64	23.67	18.95	0.18	0.84	0.71	0.61	0.72	0.01
STA-LSTM	52.37	78.36	94.55	75.09	—	13.09	19.59	23.64	18.77	—	0.85	0.71	0.62	0.73	—