I have two 3-dimension arrays actual.shape = [333,71,57] and predicted_cnn.shape = [333,71,57] [time, latitude, longitude]. Keeping the time fixed I can plot those two arrays as a 2-D map.
When I compute the RMSE root mean squared error between two arrays I use: sqrt(mean_squared_error(actual,predicted_cnn)) and I obtain a scalar.
Would be possible to compute the same quantity without having a scalar but an RMSE value for each point [latitute, loongitude](time fixed)? I would like to obtain the spatial distribution of the RMSE.
I would need spatial_rmse=sqrt(mean_squared_error(actual,predicted_cnn)) with spatial_rmse.shape=[0,:,:]
thanks