Synergizing GEDI, Sentinel-1, and Sentinel-2 for tree crops mapping
Esmaeel Adrah, Kent State University, eadrah@kent.edu (Presenter)
He Yin, Kent State University, hyin3@kent.edu (Presenter)
Jesse Wong, Kent State University, jwong10@kent.edu (Presenter)
Combining multi-sensors data is increasingly used for crop mapping in heterogeneous landscapes, such as tree crop, which is often confused with other woody vegetation. Multi-sensors data offers higher spectral and temporal resolutions; however, the benefits of utilizing vegetation structural information for tree crop mapping are rarely tapped. Moreover, the lack of ground reference remains a challenge to train classification models. In this study, we explored the potential of combining GEDI with Sentinel-1 and Sentinel-2 to map tree crops in Syria, which comprises heterogeneous landscapes from the Mediterranean environment to very-dry arid climate. In the first step, we generated samples that separate trees and woody vegetation from other vegetation using GEDI relative heights and plant area index. In step 2, to separate trees from forests and shrubs, we combined monthly information from Sentinel-1 VV and VH at GEDI shots’ level, collected labels by visual interpretation of very high-resolution images, and trained a random forest model to classify GEDI shots. Finally, in step 3, we used the labeled GEDI shots as inputs to map tree crops using combined Sentinel-2 and Sentinel-1 composites. We found that the inclusion of GEDI improved tree crop mapping at GEDI shots’ level from 79% overall accuracy to 83% (step 2). Our preliminary tree crop map had an overall accuracy of 86% whereas the F1 score for tree crops was 85% (step 3). The results were consistent with an independent classified map based on separate training data and the same classification input variables, yet our map required fewer training samples by using GEDI. This study contributes toward the operationalization of multi-sensor approaches in regions where ground reference is scarce. Additionally, it shows the potential of utilizing GEDI metrics to scale up tree crop mapping, thereby supporting efforts aimed at enhancing food security and livelihoods.
Associated Project(s):
Poster Location ID: 3-6
Presentation Type: Poster
Session: Poster Session 3
Session Date: Thu (May 11) 3:00-5:00 PM
CCE Program: LCLUC