Convolutional Residual-Attention: A Deep Learning Approach for Precipitation Nowcasting

Table 4

The effect of the number of heads in multihead attention on the experimental results.

Number of heads	6	8	9	10	12	13	14
RMSE (no res) ↓	9.57	9.07	8.57	8.62	8.43	8.46	8.55
RMSE (residual) ↓	8.15	8.09	8.13	8.10	7.90	8.04	8.24
EVS (no res) ↑	0.70	0.69	0.72	0.70	0.70	0.70	0.70
EVS (residual) ↑	0.75	0.764	0.765	0.76	0.77	0.75	0.74