How to train data with NARX model in MATLAB

I have sets of 20 years' rainfall and water level data, and I would like to do the prediction using NARX model, I know it needs to train the rainfall data first, but how exactly do I need to train the data? Should I separate it into 10 years each, one set of data used to train, and another set of data used to test the performance? and how can I test it?

and also how to determine the training, validation, testing percentage?

and I wonder whether I have done the prediction correctly, below are my steps in MATLAB:

Open nnstart -> enter NARX model -> define the parameters needed (no. of hidden Neuron and delays)-> train the 1st 10 years --> recorded down the parameters which is closest to water level (least error) --> run the NARX model again with the next 10 years, with the recorded parameters -->result