Accurate and timely crop type classification is critical for agricultural management and food security monitoring. Most existing approaches rely on the spectral response time series obtained from satellite imagery, primarily focusing on end-of-season classification. By providing accurate crop maps early in the season, policymakers, irrigation managers, and producer organizations can optimize water allocation, forecast production, and stabilize food prices. This study introduces a novel AI framework that integrates remote sensing data with ancillary information, such as cloud cover percentage, previous-year crop type, and irrigation type, for early crop type classification. The framework is validated on a dataset of approximately 130,000 fields in the region of Lleida, Catalonia (Spain), a representative Mediterranean agricultural landscape. Leveraging remote sensing data from the previous growing season enhanced early-stage classification performance, achieving an accuracy of approximately 70% at the season’s onset—a significant improvement over traditional approaches relying solely on current-season data. Ancillary data played a key role in mitigating challenges such as cloud-induced data gaps and limited spectral information and better predicting crop types in fields with permanent crops or no crop rotation. Incorporating ancillary data boosted accuracy by 10%, resulting in an overall accuracy of 80% early in the season, which increased to 88% by the season’s end. Despite these advancements, underrepresented crop classes remained difficult to classify accurately due to class imbalances, underscoring the need for improved dataset balancing techniques. The findings underscore the potential of integrating multi-source data to improve crop type classification, especially early in the season.
Gené-Mola et al. (Wed,) studied this question.