Spatiotemporal analysis of LANDSAT Data for crop yield prediction

Abstract

Crop growth monitoring and yield estimation is an important process for agricultural economic return prediction and food security. Improvement in the accuracy and timeliness of the information about pre harvest prediction of crops by blending of ancillary data and remotely sensed data in the temporal domain lead to the effective and optimized decision making. Although previous studies have found strong correlation between observed and predicted yield based on Normalized Difference Vegetation Index (NDVI), yet there is a pressing need to provide more accurate and reliable yield in the small areas. The objective of the proposed model is to extract the information related to the crop yield for the sugarcane planted in the small fields of Himalayan foothills region. Relation of crop yield information with the different stages of crop growth has been considered in the proposed model. The process based on correlation analysis of spatiotemporal data to identify of the best periods for the reliable estimation of the sugarcane is presented. The best period for the crop predictability is during 210 to 270 days after plantation. Based on the historical data of 10 years, it has been found that predictability of nonlinear modelling is significant and is of the order of 0.6. By Tukey test and RMSE the best fit regression models are identified for the study area.