Hello Everyone!!
im looking for the what should be the threshold value for performance criteria such as coefficient of correlation, Nash coeff. p value, error, etc..
i understand i cannot put all the criteria on a single class but im wondering how one should decide the threshold for model to be best.
for the many modelling i have seen 0.3 R^2 was considered to be best. but there should be i feel criteria to the performance measuring criteria to be evaluated for depicting the best.
i have read R^2>= 2/sqrt(n) should justify the linearity in the predictors and predictant.
stil looking for the best rule to say best or worst.
thank you.
Hi, Dr. Rathod,
Only one statistical paradigm was designed specifically to obtain the most accurate solution for any application. Here is an open-source introductory review:
- https://odajournal.com/2017/04/18/what-is-optimal-data-analysis
I was suffering from the same scenario what you are having now. Measure of performance indicator like R2, NS, PBIAS etc are not enough to evaluate the model. After having best results of performance please do not forget to check water balance of your model. There are plenty of cases where model is giving very excellent performance indicator but there is no Water Balance indeed.
Binod Bhatta sir , you are Absolutely true. currently im working on the scour models and their performance. R^2 and other performance indicator wont go hand in hand. there should be something which gives consistent results to indicate the relative best amongst the group.
Praveen Rathod ,
I do have previous experience with SWAT in relation to the above question you raised. My model showed good statistical indicators for daily simulations (Eg: R^2 and N.S.E. were above 0.70) but the annual average values of water balance did not agree well (compared the proportions with previous studies),
Later I performed calibration in monthly basis which showed N.S.E and R^2 above 0.75, but daily values were between 0.40 and 0.50 for R^2 and N.S.E.. But still my water balance matched when considering the climatic conditions and other factors. So I would say that R^2 and N.S.E. are not solely the indicators of a good model.
Regards
Miyuru
- Badges
- Science topic
- Similar topics
- Mathematics