The study emphasises that SMTs outperformed Machine Learning methods in crop type mapping, particularly where ground data is limited.
In a recent effort to harness satellite data for countries with limited ground reference information, the International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) conducted a study using Senegal as a model to test various algorithms. While numerous remote sensing products exist for land cover mapping, accurately identifying crop types requires machine learning algorithms to process satellite data.
This study compares Machine Learning methods with Spectral Matching Techniques (SMTs) and demonstrates that SMTs are more effective than Machine Learning methods in scenarios where ground data are scarce. Although Machine Learning algorithms are valuable for processing satellite images to identify crop types, they require large volumes of high-quality training data. The study emphasizes that SMTs outperformed Machine Learning methods in crop type mapping, particularly where ground data is limited.
Dr Murali Gumma, the study’s lead author, stated, “Among the four approaches tested, Spectral Matching Techniques achieved the highest accuracy, exceeding 76 per cent. In contrast, machine learning methods—such as CART (Classification and Regression Tree), SVM (Support Vector Machines), and RF (Random Forest)—which are used for molecular machine learning and predicting compound properties, only reached accuracies between 40 per cent and 55 per cent.”
Traditional methods have faced challenges in mapping crop types in small, irregularly shaped fields and regions with inter-cropping practices. This paper highlights that high-resolution satellite imagery when combined with semi-automated algorithms like SMTs is ideally suited for dryland regions in Africa, making it a valuable tool for enhancing agricultural mapping in these challenging environments.
“In terms of the study’s applicability, there is significant potential in West African countries such as Mali and Niger. Improved insights into crop type mapping can be instrumental in calculating agricultural yield and, when combined with weather, climate, and socioeconomic data, can help pinpoint regions where productivity is impacted,” highlighted Dr Stanford Blade, Deputy Director General – Research, ICRISAT.
