Evaluation of Bayesian Network and Support Vector Machine Models ‎in Estimation of Reference Evapotranspiration (Case Study: ‎Khorramabad)‎

Around the world, the Penman-Monteithe-FAO model is used as a reference method to estimate reference ‎evapotranspiration. This method requires a lot of input data, which in many cases are difficult to access, so ‎it is necessary to replace simpler models with low inputs and good accuracy. Therefore, the purpose of this ‎study was to evaluate the accuracy and capability of Bayesian Network and Support Vector Machine ‎models in estimating reference evapotranspiration and comparing it with the Penman-Monteithe-FAO ‎model. For input data, monthly data of Khoramabad synoptic station including: maximum and minimum ‎temperature, maximum and minimum relative humidity, solar radiation and wind speed in period 1990-‎‎2016 (420 months) were used. Based on the effect of input parameters on output, six input patterns were ‎determined for modeling. 70% of data were used for training and 30% for model validation. The results ‎showed that pattern number 5 includes: maximum Temperature, wind speed, solar radiation, minimum ‎temperature and minimum relative humidity in has the best accuracy all models. This model in test phase, ‎has R2 = 0.97 and RMSE = 0.93 in the Bayesian network and 8.9 R2 = 0 and RMSE = 0.41 in support ‎Vector Machine with radial basis functions kernel. Comparison of the performance of the models showed ‎the superiority of the vector machine model over the other models with AARE of 0.0525 and MR of 0.005.‎