How do I make R-R prediction with NARX in MATLAB?

I have 20 years data of rainfall and water level, and I would like to use NARX model to test the ability of it to forecast the water level.

I have gone through some of the teaching, whereby I need to train and test the data, but I am not sure I got the correct direction, and I am confused about the separation of data into training, validation and testing.

For example, I put the first 10 years of rainfall data (1996-2006) into NARX, trained it with selected parameters, and obtained the least error (record down the parameters) [testing stage]. Then I put the following 10 years (2006-2016) into the model again with that particular parameters [testing stage].?

OR inside the nntool GUI, there is a setting of which I can select the percentage of training, validation, and testing. For example (70-15-15), then I input all 20 years of data into the model at once, and the output shown is the final answer?

I am really confused on how to make a prediction and the data separation. I hope somebody out there can help me, in case anyone need any more information I can provide more in comment. Thank you in advance.