| Popis: |
On November 4th, 2020 a deadly civil war broke out in Tigray, Ethiopia displacing close to 2 million people internally and more than 48,000 refugees in neighboring Sudan by August 2021[1]. Given agriculture is the livelihood of millions of people in Tigray, evaluation of the conflict’s impact on cultivated land and the consequent crop production is critical for government and non-government disaster relief institutions. Unfortunately, such evaluation is extremely challenging as the conflict is characterized by communication blackout leaving the region without access to cellphone or internet.In this study, we used Sentinel-2 and Planet satellite imagery data to map loss of well cultivated land in 2021 due to the war. We developed multiple cultivation detection criteria based on the peak and falling limb characteristics of Normalized Difference Vegetation Index (NDVI) time series, validated using limited field observations of fallow and cultivated plots from the wet season in 2021 and 2022. We employed object detection machine learning model to identify harvest piles as an additional parameter to detect farming activity.Our predicted change in cultivation map from 2019/20 to 2021 showed that the density of conflict incidents was positively correlated to the mean net loss of well cultivated land with R2 of 0.7 in Tigray highlands (elevation > 1200 m). Sub-regions with high estimated net loss of cultivated land due to abandonment of reported internally displaced people also resulted in high predicted loss of well cultivated land using NDVI based criteria in our study. In the absence of extensive in situ data, we demonstrate how satellite imagery along with good understanding of local farming practices can provide timely and useful information to assist humanitarian management efforts in times of crisis and recovery phase.[1] Annys, Sofie, Tim Vanden Bempt, Emnet Negash, Lars De Sloover, Robin Ghekiere, Kiara Haegeman, Daan Temmerman, and Jan Nyssen. Tigray: Atlas of the Humanitarian Situation (version 2.2). Zenodo, 2021. https://doi.org/10.5281/zenodo.5805687. |
