Estimating of rice grain yield using earth observation data.

Abstract

Reliable and timely agricultural yield estimation has become crucial for ensuring food security and effective agricultural management, as emphasised by Sustainable Development Goals (SDGs), particularly SDG 2, which focuses on zero hunger. With the rapidly increasing global population, there is mounting pressure on crop production, posing significant challenges to food security. Rice, being a staple food for three billion people worldwide, requires accurate and reliable yield estimation to inform decisions related to import/export, pricing, crop distribution, and future crop planning, particularly in the face of the continuously changing global environment. Existing approaches rely on ground surveys and traditional means, which are subjective, expensive, labour-intensive, time-consuming, and prone to errors. Thus, this study aimed at estimating rice grain yield using Earth observation data in Butaleja district. The study utilised Landsat 8 and Sentinel 2 images to map paddy rice cultivation areas in Butaleja district over seven years (2015-2021) spanning two growing seasons. The random forest classifier, a machine learning algorithm, was employed for image classification using the Google Earth Engine (GEE) platform. Three models, namely Multiple Linear Regression (MLR), Partial Least Squares Regression (PLSR), and Random Forest, were employed to estimate rice grain yield by incorporating various factors such as vegetation indices, biophysical factors, climatic factors, and historical yield data. The models were validated using rice grain yield data from the 2021-2022 season through Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) analyses. The classification results revealed that rice cultivation areas were distributed across all the wetlands in the Butaleja district, with the highest concentration observed in the northeastern parts. Over the study period, a slight increase in paddy rice cultivation areas was observed, with season one exhibiting a more extensive area than season two. The results from yield estimation revealed that all three models showed potential for estimating rice grain yield, with PLSR performing slightly better than Random Forest and MLR, exhibiting RMSE and MAE values of 64.616 Kg/ha and 53.392 Kg/ha in season one and 99.569 Kg/ha and 98.339 Kg/ha in season two, respectively. The study demonstrates the potential of leveraging earth observation data and statistical modelling techniques for accurate and timely estimation of rice grain yield. Findings suggest that remote sensing can be used for estimating rice grain yield for agricultural planning, resource allocation, and yield forecasting. Further research can focus on expanding the scope of variables encompassing soil properties and crop management factors.