Mapping Individual olive trees and irrigation to inform the sustainable expansion of perennial crops in the era of the Green Morocco Plan
Zhenong Jin, University of Minnesota, jinzn@umn.edu (Presenter)
Chenxi Lin, University of Minnesota
Elinor Benami, Virginia Tech
Rachid Bouabid, National School of Agriculture of Meknes
David Mulla, University of Minnesota
Travis Lybbert, University of California, Davis
For centuries, the olive tree has been of great importance in the Mediterranean region, influencing culture, environment, and economy. As a representative Southern Mediterranean nation, Morocco has significantly invested in olive production over the past three decades, primarily through the Green Morocco Plan (GMP). This program aims to modernize the agricultural sector with a focus on olive cultivation and irrigation promotion. To date, the traditional costly and slow ground survey remains the primary method for assessing the agronomic, environmental, and socioeconomic outcomes of this large-scale investment. The opportunity of using emerging earth observatory technology for quick and low-cost evaluations. In this study, we processed sub-meter resolution imagery collected between 2018 and 2022 with the Vision Transformer (ViT) model and the UNet model to generate a regional map of individual olive trees. While the UNet model focused on identifying each individual tree, the ViT model leveraged the spatial texture information to predict land use types at the patch level and mask out non-olive trees. We evaluated the accuracy of the olive tree map using more than 500 fields of ground labels and found F1 scores exceeding 0.87 for both olives and non-olives, as well as overall accuracy. The ViT model was particularly useful in accurately classifying spatial regions with limited spatial information. Experimentation showed that using a smaller region as a spatial constraint for the UNet model improved accuracy by up to 10%. Additionally, we combined multi-modal ML to extract spatial texture from images and temporal patterns of greenness and rainfall variability to separate rainfed vs irrigated olives, and our preliminary overall accuracy reached 0.81. This individual olive tree map is the first of its kind in Morocco and can serve as a reference to monitor the progress of the GMP, as well as provide baseline information for other objectives, such as carbon stock analysis.
Associated Project(s):
Presentation Type: Poster
Session: Poster Session 1
Session Date: Tue (May 9) 5:00-7:00 PM