Use of regression techniques for rice yield estimation in the North-Western province of Sri Lanka

Abstract

One of the aspects in the agriculture sector beneficial to farmers and all other stakeholders is the prior knowledge on the yield of crops expected from an agricultural season. In this paper, several regression techniques have been used to model the relationship between climatic factors and paddy production in the North-Western province of Sri Lanka that makes a significant contribution to the total harvest of the country. Nearly two decades of rice yield data from 2000 to 2018 and several climatic factors in the two agricultural seasons of Yala (May-August) and Maha (September-March) were considered in the analysis. Monthly mean climatic data of temperature, evaporation, sunshine, and wind speed were applied along with the overall rainfall in four regression techniques viz. Support Vector Machine Regression, Multiple Linear regression, Power Regression, and the Robust Regression on MATLAB and R software. The performance of the models developed on those techniques was evaluated in terms of the Mean Absolute Error and the Coefficient of Determination. It was found that the Support Vector Machine Regression produces the best correlation between actual and predicted yields in both administrative districts of the Province, which can be used for yield estimation under normal climatic conditions.

Keywords

• regression, coefficient of determination, mean absolute error, yield estimation, north-western province