Yield Estimation of Wheat Crop using Remote Sensing and Machine Learning Techniques

Abstract

The backbone of Pakistan is its agricultural sector that faces multiple challenges each year that results in inconsistent yield of crop resulting in heavy reliance on imports for major crops to fulfill the country’s needs. To address this researchers have turned to remote sensing techniques for early yield estimation. This approach shows the importance of incorporation of smart farming techniques in our agricultural sector. Focusing on Bahawalpur city, a vital wheat producing region of Pakistan, open-source satellite imagery of 30-meter is utilised to monitor and detect the winter Rabi crop i.e Wheat using Global Food-and-Water Security-support Analysis Data (GFSAD1km) layer and incorporation of Crop Reporting Services Pakistan data, Normal Difference Vegetative Indices (NDVI) maps are also generated to monitor crop health. In this research, vegetation index values are utilised for a single farm location vs 100 farms locations and their time series charts are generated during the growing period i.e Dec-April for each year to show the changes in crop cycle period using the Google Earth Engine (GEE) platform for the year 2019-2023. Analysis of the changes in vegetation maps for Bahawalpur city were generated spanning 2019 to 2023 and future NDVI maps for the years 2025 were generated using an architecture of neural networks. Yield was estimated using the maximum vegetation indices values for the farms detected using Stack ensemble models that uses different models to generates results and then feeds them into a Meta model for better results, accuracy assessment done using Root mean square and mean absolute error indicated more than 90% of all the predicted yields were predicted accurately. The highest accuracy was observed for the 2020 year where the mean absolute error was 1.1%. The research demonstrates that, for analysis of each city or area having access to globally accessible dataset is fundamental. Moreover, for the xiv List of Figures yield estimation the availability of historical yield is critical for accuracy assessments. The developed dataset is beneficial for understanding the dynamics of crop growth cycle and changes during that period, these values can be helpful in creating policies that helps maintain the estimated yield each year. Moreover, the simplification of remote sensing techniques proves that these methods can be incorporated in our farming policies and methods which can aid in the advancement of smart farming in Pakistan.