Rice Yield Estimation Using Free Satellite and Field Data

Abstract

An effective pre-harvest rice yield estimation method is truly significant for the assessment of seasonal rice production in terms of strategic planning purposes. In Sri Lanka, a conventional method named crop-cut survey is used to estimate seasonal rice production, yet it fails to forecast rice yield before the harvest as it is conducted during the harvest. Therefore, this study is focused on identifying cultivated paddy lands and forecasting rice yield using free satellite data. Landsat 8 OLI/ TIRS images (30m spatial resolution) from Earth explorer and 8-day composite images (250m spatial resolution) from Moderate Resolution Imaging Spectro-radiometer (MODIS) sensor on board NASA EOS Terra/Aqua satellite were used from 2014 to 2017. Paddy cultivated lands were identified by land cover classification by using field training samples and Landsat 8 OLI/ TIRS data. In addition, the temporal change of Normalized Differenced Vegetation Index (NDVI) for paddy and forest was also analyzed to validate the classification. The observed minimum accuracy of the land cover classification out of the tested four (4) seasons was 99.4%, and the minimum kappa coefficient was 0.9916. The correlation coefficient between reference net harvested paddy area and paddy cultivated area identified by Landsat 8 is 0.93. Linear and exponential yield forecasting models proposed by Sirisena, et al. (2014) for Kurunagala district were validated and tested based on NDVI and EVI2 vegetation indices obtained through MODIS (MOD09Q1v006) surface reflectance image of Polonnaruwa District. The comparison of the estimated yield with national statistical records, both NDVI and EVI2 based models, provide more reliable estimations about 80 days after the transplanting of each season, but, EVI2 based model (derived at 80 days) gives more reliable estimations than NDVI based model with 86.37% of average accuracy. Therefore, seasonal rice yield can be successfully forecasted one month prior to the harvest time using EVI2 based model in the Polonnaruwa district.