Application of machine learning approach for Chickpea yield estimation based on Agroclimatological Indices (Case Study: Kermanshah region)

This study aims to estimate the yield of spring chickpea using machine learning methods of linear regression models in Kermanshah region, west of Iran. The meteorological variables, agrometeorology and remotely-sensed based indices as predictor variables and yield data of Agricultural Jihad Organization of Kermanshah as a response variable were used for four growth stages during 1990-1991 to 2017-2018. Twenty four and three years data were used for training and model validation, respectively. The results revealed that among the linear models, Lasso model with a coefficient of determination of 67% and a standard error of 59.8 kg.ha-1 was chosen as a best model for crop yield estimation in the emergence to 50% of flowering stages. This model has relative deviations of 0.4, -0.3 and 3.5 for the years 1997-1998, 2005-2006 and 2010-2011, respectively.