Building and precision assessment of regression models for determining of cereals’ and legumes’ crop yield based on Earth remote sensing data and climatic characteristics

Abstract

Crop yields are strictly dependent from natural and climatic conditions of the growing region, in addition specific weather conditions in the southern part of the Far East necessitates the analysis of a large number of factors when building a predictive regression model. The article presents regression models for assessing the average productivity of the main crops in Chernigovsky district of Primorsky region: soybean, spring wheat, barley and oat. Between 2012 and 2018 the sown area of these crops ranged from 78 to 86 % of the total sown area in the Chernigovsky district. We used the indicators obtained from Earth remote sensing data (the maximum weekly NDVI per year, calculated from the mask of arable land in the Chernigovsky district) and meteorological characteristics (from 2008 to 2018): hydrothermal Selyaninov coefficient, the duration of the growing season, temperature and humidity of the upper soil layer, photosynthetically active radiation and the Budyko radiation index. Climatic characteristics of arable land, representing reanalysis data and combining ground based and remote observations, were obtained using the Vega–Science web–service. Also, we used data about sown area and gross crop in the Chernigovsky region from 2008 to 2018. It was found that average annual oat yield has the biggest coefficient of variation (31.5 %). The corresponding indicator for the remaining crops is in range from 16 to 18 %. The accuracy analysis of the obtained models showed that the average error of the model in period from 2008 to 2017 was 4.1 % for barley, 5.1 % for oat and spring wheat, and 5.4 % for soybean.