In the Colombian Andes, agroforestry is a traditional form of agriculture, characterized by a heterogeneous and often diversified composition of trees and crops. This form of land use provides important ecosystem services, such as carbon sequestration, reduction of soil erosion and the maintenance of biodiversity by providing a structural complex habitat. Satellite remote sensing is widely used for studying land use patterns and forest cover, however the discrimination between agroforestry systems and forests is still a challenge, especially in heterogeneous landscapes and in rough terrain. Here, we aim to advance the remote sensing of agroforestry systems using field measurements of vegetation structure in combination with Sentinel-2 images. We use spectral and textural variables derived from Sentinel-2 imagery to predict above ground biomass (AGB), leaf area index (LAI) and canopy closure (CC). The relationship between predicted and observed values obtained from Random Forest regression models showed good fits: for AGB with an R² = 0.92 and relative RMSE = 34%; for LAI with an R² = 0.91 and relative RMSE = 19%; and for CC an R² = 0.89 and relative RMSE = 9%. This allowed us to map these important ecosystem variables at landscape scale and establish empirical thresholds, with which a discrimination of agroforestry systems from forests was possible with an accuracy of 94%. Our results suggest that the relationship between vegetation structure and the spectral information obtained by Sentinel-2 can contribute to the detection and characterization of agroforestry systems and thus help quantifying the ecosystem services and biodiversity conservation potential provided by this type of tropical agriculture.

在哥伦比亚的安第斯山脉，农林业是一种传统的农业形式，其特征是树木和农作物的组成成分往往是异质的，而且往往是多种多样的。这种土地利用形式提供了重要的生态系统服务，例如碳固存，减少土壤侵蚀和通过提供结构复杂的栖息地来维持生物多样性。卫星遥感被广泛用于研究土地利用方式和森林覆盖率，然而，农林业体系和森林之间的区别仍然是一个挑战，特别是在异质景观和崎rough地形中。在这里，我们的目标是结合Sentinel-2影像，利用植被结构的实地测量来推进农林业系统的遥感。我们使用源自Sentinel-2影像的光谱和纹理变量来预测地上生物量（AGB），叶面积指数（LAI）和树冠闭合（CC）。从随机森林回归模型获得的预测值与观测值之间的关系显示出很好的拟合度：对于AGBR ²  = 0.92，相对RMSE = 34％；对于RAI为R ²  = 0.91和相对RMSE = 19％的情况；对于CC，R ²  = 0.89，相对RMSE = 9％。这使我们能够在景观尺度上绘制这些重要的生态系统变量，并建立经验阈值，利用该阈值可以将农林业系统与森林区分开，准确度达到94％。我们的结果表明，Sentinel-2获得的植被结构与光谱信息之间的关系可以有助于农林业系统的检测和表征，从而有助于量化这种热带农业提供的生态系统服务和生物多样性保护潜力。