Yield Forecasting Model for Maize Using Satellite Multispectral Imagery Driven Vegetation Indices

Abstract

In recent decades, the cultivation of maize (_Zea mays_ L.) has witnessed remarkable growth in Bangladesh, particularly in the northern regions. Maize, as a high-yielding grain crop with diverse applications, plays a pivotal role in the country's agricultural landscape. Traditional methods of yield prediction involve time-consuming and subjective on-site field visits, resulting in significant errors and delayed information dissemination to government authorities and decision-makers. This study explores the potential of remote sensing technology to predict maize yields before harvest, thereby enhancing agricultural decision-making processes. The research utilizes 16-day (~30 m) Landsat 8 and 10-day (~10 m) Sentinel 2A imagery of two years November 2018 to February 2019 and November 2019 to February 2020 to forecast maize yields in the Kaharole upazila of the Dinajpur district, Bangladesh. Four cloud-free images, representing the maximum normalized difference vegetation index (NDVI) for each maize growing season, are selected from the Landsat 8 and Sentinel 2A data. Regression models are established to relate NDVI values to the maize yields across 20 individual farmers' fields. The results reveal that the prediction models based on mean NDVI values for the combined growing seasons outperform those based on single growing seasons and the finer spatial resolution of Sentinel 2A contributes to its superior performance in comparison to Landsat 8, offering valuable insights for improved agricultural management and food security.