Modeling the physical dynamics of daily dew point temperature using soft computing techniques

The objective of this study is to develop soft computing models, including Generalized Regression Neural Networks (GRNN) and Multilayer Perceptron (MLP), for modeling daily dew point temperature. For the data from U.C. Riverside and Durham stations in California, USA, the best input combinations (1-, 2-, 3-, and 4-input) were identified using GRNN. The performance evaluation and scatter diagrams of GRNN indicated that the average soil Temperature (Ts) produced the best results among 1-input combinations for both stations. Adding other input variables to the best combinations improved the performance of GRNN. MLP was used to estimate daily dew point temperature using the best input combinations (1-, 2-, 3-, and 4-input) identified by GRNN. Adding other input variables to the best input combinations also improved the performance of MLP. Comparison indicated that results of GRNN were better than those of MLP for both stations. A Multiple Linear Regression Model (MLRM), one of the conventional statistical models, was also compared with GRNN and MLP, and the soft computing models were found to estimate daily dew point temperature more accurately.