Classification of agroforestry systems by photo-interpretation of satellite imagery

Effective and large-scale monitoring of agroforestry (AF) systems is essential to assess the environmental benefits of agroforestry and support sustainable land management strategies. However, a standardized method for classifying these systems using satellite imagery is still missing. Here, we present a novel operational framework to classify agroforestry systems into three categories—Alley cropping, Scattered agroforestry, and Hedgerows—and to distinguish these systems from Cropland without trees, Natural trees, and Orchards. The proposed procedure relies on a classification tree based on photo-interpretation of satellite imagery. The accuracy and robustness of this classification tree were evaluated by five interpreters across 300 agroforestry and non-agroforestry plots spanning all continents. Results show that the classification tree accurately distinguishes agroforestry categories from one another and from non-agroforestry systems, with an overall accuracy ranging from 0.75 to 0.81 depending on the interpreter. After eliminating the interpreters’ errors, the potential classification accuracy increases to 0.86. While hedgerows were accurately classified in most cases with an omission error of 2% and no commission error (0%), the study revealed challenges in differentiating between Alley cropping and Orchards which were frequently confounded. Similarly, plots with Scattered agroforestry were also misclassified as Natural trees leading to a commission error of 19% for this class. Despite these limitations, the proposed classification tree represents a valuable tool for large-scale monitoring of agroforestry systems. Future adaptations of this framework could address regional specificities, further improving its applicability and accuracy.