Development Of Reference Evapotranspiration Equations Using An Artificial Intelligence-Based Function Discovery Method Under The Humid Climate Of Northeast India

The present investigation aims to develop symbolic regression based substitute to the FAO56 Penman-Monteith (FAO56-P-M) equation for calculating reference evapotranspiration (ET0) using meteorological data of thirty sites in northeast India. Further, five different input combination models were developed using a symbolic regression (SR) technique. Extraterrestrial radiation(R-a), altitude (Alt), maximum temperature (T-max), and minimum temperature (T-min) were used as default input variables in all the developed models. Temperature (T) based model showed high values of performance indices in all the selected stations and is highly dominated by the R-a and T-min. Temperature and relative humidity (Rh) based (T-Rh) model had reduced error metrics and improved correlation than the T based model, suggesting that T-Rh models have better prediction capability. Temperature and sunshine hour (n) T-n based model had the lowest error metrics and highest correlation among all combinations, suggesting that T-n based model is the best suitable practical alternative to FAO56-P-M. Temperature, sunshine hour, and Rh (T-n-Rh) based model had almost similar prediction ability as found in T-n based model; hence Rh is a reluctant input variable in ET0 modeling. Temperature, sunshine hour, and wind speed (T-n-V-w) based model had the lowest error metrics and highest correlation among all combinations tested, suggesting that T-n-V-w based model is the best suitable alternative to FAO56-P-M. Sensitivity analysis also showed that the T-n-V-w combination has a cent percent positive contribution in controlling the ET0 process. The performance of the T-n-V-w model on T-n based model was found almost similar accuracy, which could be due to minimal influence of wind speed.