Supplanting missing climatic inputs in classical and random forest models for estimating reference evapotranspiration in humid coastal areas of Iran

Although evapotranspiration might not be a limiting factor for water consumption in humid regions, it is a very important parameter in irrigation scheduling and agricultural water management in those regions, and should be estimated with high accuracy. The reference evapotranspiration (ETo) concept is a way for retrieving the actual crop water requirements using a crop coefficient. The standard procedure for determining ETo values in literature is the Penman-FAO-Monteith (PFM) equation, which needs considerable meteorological inputs. In practice, when all necessary data of PFM are not available, the consideration of estimated meteorological inputs for this equation might be a suitable approach for deriving accurate ETo values. Further, the empirical ETo equations and machine learning techniques relying on fewer meteorological variables are usually employed as alternatives to the PFM equation. The present paper aimed at assessing the effect of considering calculated missing meteorological variables on the performance accuracy of the standard PFM and some commonly used empirical equations using daily data from six humid coastal locations in Iran covering a period of 10 years. Moreover, the machine learning bootstrap aggregating-based random forest (RF) technique was applied under the same conditions. Based on the obtained results, when replacing missing inputs by calculated ones, the combination-based models (relying on estimated wind speed values) provided the most accurate results, while using estimated solar radiation values reduced the model's performance accuracy.