Identification of coffee agroforestry systems using remote sensing data: a review of methods and sensor data

Abstract

Coffee is one of the most important agricultural commodities. Agroforestry systems (AFS) are increasingly used in coffee cultivation because of environmental benefits, adaptability of the systems, and economic profits. However, identifying the spatial distribution of AFS through remote sensing continues to be challenging. The current systematic review focuses on the accuracies obtained and the computational methods and satellite data used in mapping coffee AFS between 2000 and 2020. To facilitate the analysis, we ordered the mapped AFS into five classes according to their density and species composition of shade trees. The Kruskal-Wallis test was applied to evaluate significative differences among classes. Both shade-tree densities and species composition affected the accuracy level. The worst results were obtained in AFS retaining many woody species from the original forest and high tree density (user accuracy < 0.5). About the methods, maximum likelihood was the most widely used with very variable results; some non-parametric methods such as CART, ISODATA, RF, SMA, and SVM presented consistently high accuracy ( > 0.75). High spatial resolution multispectral imagery was suitable for mapping AFS; very few studies were found with radar imagery, so it would be desirable to increase its use combined with optical data.