Assessing accuracy of barley yield forecasting with integration of climate variables and support vector regression

Laleh Parviz

doi:10.17951/c.2018.73.1.19-30

Abstract

<p>Investigations of the relation between crop yield and climate variables are crucial for agricultural studies and decision making related to crop monitoring. Multiple linear regression (MLR) and support vector regression (SVR) are used to identify and model the impact of climate variables on barley yield. The climate variables of 36 years (1982–2017) are gathered from three provinces of Iran with different climate: Yazd (arid), Zanjan (semi-arid), Gilan (very humid). Air temperature by high correlation coefficient with barley yield was introduced as the dominant climate variable. According to evaluation criteria, SVR provided accurate estimation of crop yield in comparison with MLR. The diversity of climate impressed the estimated yield in which UI, decreasing from Gilan to Yazd provinces, was 47.77%. Support vector machine (SVM) with capturing the nonlinearity of time series, could improve barley yield estimation, with the minimum UI for Yazd province. Also, the minimum correlation coefficient between the observed and simulated yield was found in Gilan province. Based on GMER calculations, SVM forecasts were underestimated in three provinces. All findings show that SVM is able to have high efficiency to model the climate effect on crop yield.</p>

Highlights

Forecasting of crop yield has a growing importance to ensure food security, optimize agromanagement practices and resource use
In order to compare the eficiency of support vector regression (SVR) and Multiple linear regression (MLR) models, GMER, MAE, RRMSE, UI were used as evolution criteria, whose structure is brought in equations 5 to 8
M Time series of precipitation, mean air temperature, minimum air temperature, maximum air temperature and wind speed during 1982–2017 were used as climate variables which were the average of stations in each province as climate variables of provinces

Summary

LALEH PARVIZ

Multiple linear regression (MLR) and support vector regression (SVR) are used to identify and model the impact of climate variables on barley yield. Air temperature by high correlation coeficient with barley yield was introduced as the dominant climate variable. SVR provided accurate estimation of crop yield in comparison with MLR. The diversity of climate impressed the estimated yield in which UI, decreasing from Gilan to yazd provinces, was 47.77%. Support vector machine (SVM) with capturing the nonlinearity of time series, could improve barley yield estimation, with the minimum UI for yazd province. The minimum correlation coeficient between the observed and simulated yield was found in Gilan province. All indings show that SVM is able to have high eficiency to model the climate effect on crop yield

INTRODUCTION

Case study

Multiple linear regressions

Support Vector Regression

Evaluation criteria

RESULTS

Correlation coeficient yazd zanjan

Evaluation of the performance of regression models

The climate changes in recent years indicate the need for comprehensive