Reply to comments on 'Evaporation estimation using artificial neural networks and adaptive neurofuzzy inference system techniques'

摘要

The comments from Ozgur Kisi on our paper [4] are very useful and of general interests to the readers in this field. Here are the replies to the points raised:rn(1) This comment is valid. We attempted to illustrate and explore the application of the Gamma Test for evaporation estimation so that readers can try this new tool in their evaporation modelling tasks. Clearly, more work is still needed in this area. At the moment, we are working on a comparison between the Gamma Test and other model input selection approaches (such as Information Entropy, Cross validation and Akaike Information Criterion). Principal Component Analysis (PCA) is suitable for linear systems and it may not work well for nonlinear systems (again, it may be worthwhile to try PCA and compare it with other approaches). In addition. Nonlinear Principle Component Analysis (NPCA) could be explored. As to the suggested inputs of W, T, RH, Ed from the discusser, we believe this combination will produce less accurate modelling results due to the redundant information from T, RH, Ed (i.e., any one of the three variables could be fully derived from the other two, hence one of them is redundant). If there are redundant variables in linear systems, they will cause multi-collinearity problem and make matrix inversion impossible. For nonlinear systems, the effect of redundant variables is more difficult to predict. However, based on Occam's razor, it may not be a good idea to include redundant variables for any mathematical models. Nevertheless, it could be useful to build ANN and ANFIS models for both input combinations to prove the point.