Paddy Yield Estimation Models Using Satellite Data: A Case Study in Kamburupitiya DS Division

In Sri Lanka, the crop cutting survey method is used to estimate the average paddy yield during the harvesting period, and it has failed to forecast the yield prior to harvesting. This method is time-consuming, expensive and make various errors. The yield prediction is worth to estimate yield before harvesting thousand tonnes of paddy. Satellite Remote Sensing is used to forecast crop yield using images during vegetative growth. The research was conducted in Kamburupitiya Divisional Secretariat (DS) division, Matara District, Sri Lanka, during Yala and Maha paddy growing seasons. Landsat 8 OLI/TIRS images with 30 m resolution were utilized to develop the paddy yield models using paddy yield records during 2016 to 2020. The simple linear regression models were developed using recorded average paddy yield data of the Kamburupitiya DS division and the average values of respective vegetation indices (Normalized Difference Vegetation Index (NDVI); Green Vegetation Index (GVI); Difference Vegetation Index (DVI); Infrared Percentage Vegetation Index (IPVI) and Ratio Vegetation Index (RVI) and reflectance values of three spectral bands (Green (G), Red (R) and near-infrared (NIR)). The validation process was carried out using Mean Absolute Percentage Error (MAPE) as the statistical indicator between modeled and predicted yield. When considering coefficient of determination (R₂) value, MAPE percentage, Durbin Watson statistics values, and Pearson correlation values, the yield model based on the GVI for Maha season and NDVI based model for Yala season are selected as the best models for paddy yield prediction using Landsat 8 OLI/TIRS imagery. The models will allow authorities to make effective decisions in paddy supply management.