The only real way to test for model performance is by comparisson against some "true value". In simple terms, measured values (but even these are only our best approximation to reality).

You could also assume your "true value" to be the result of some well known, well established modell. But this is of course a risky approach.

The second issue is what metrics to use; mean values, extreme values, relative or absolute differences and so on.This is a matter of choice and interest. Maybe the most well known model performance measure is the Nash-Suttclife model efficiency coeficient, but there are others as you may see in the paper below. But no matter what, you still need a reference value (true values).

And finally you can apply this for any time scale, yearly, seasonal, monthly or any other time scale you desire. In your specific case and from what I understand about your question, I would then also compare performance ammong seasons.

Hope it helps.

Hello,

Rafael provides a key document to evaluate the modelling results, specially in hydrological context. Thanks. The following forums will help you regarding the issue.

https://www.researchgate.net/post/what_is_the_best_metric_to_evaluate_the_quality_of_a_prediction

https://www.researchgate.net/post/How_to_compare_Hydrological_Model_Results

You could start with checking the seasonal (temporal) parameter sensitivities of your model as we suggested here:

http://onlinelibrary.wiley.com/doi/10.1002/hyp.10764/full

Best,

Martin