Data-Driven Modeling for Crop Mapping and Yield Estimation

Abstract

AbstractIn the context of climate change, crop mapping and yield estimation are critical for improving crop management and ensuring food security. Data-driven methods have succeeded in crop mapping and yield estimation, using a combination of multi-source training data and the application of advanced machine learning technologies. This chapter reviews various data-driven approaches that have been applied in crop mapping and yield estimation, with particular attention given to deep learning algorithms. A typical workflow for data analysis is systematically summarized; it includes multi-source data collection, data preprocessing, model development, model evaluation, and model improvement. The advanced data-driven technologies employed for model development in crop mapping and yield estimation are described and synthesized. This chapter introduces threshold-based methods to illustrate the challenges of intra-class variability and inter-class similarity in crop mapping. Various machine learning and deep learning algorithms are further highlighted in multi-temporal crop mapping analysis, with the consideration of multi-source remote sensing data. For crop yield estimation, typical mechanism models are introduced to clarify the challenges in the context of widely distributed yield heterogeneities across different regions. The data-driven statistical models in yield estimation are thoroughly summarized, with a focus on the utilization of remote sensing and environmental data. Challenges and possible future improvements in data-driven approaches are presented.KeywordsData-driven modelingCrop mappingYield estimationSatellite remote sensingData science