AI - BASED SUPER RESOLUTION IN CLIMATE CRISIS CONTEXT
Weak governance and agricultural policies have created social, economic and political inequalities that feed conflicts. The same factors undermine community resilience to climate change, particularly among marginalized groups.
This analysis aims at bringing innovation to the climate crisis sector by enhancing the remote sensing-based technological tools through the application of Artificial Intelligence (AI) algorithms on security domain challenges. The integration of Earth Observation (EO) satellite data in the usage of new Super Resolution (SR) AI-based applications is crucial for better information gathering and assessments in conflict or crisis affected areas otherwise largely inaccessible to effectively contribute to global stability and peace. The conflict areas on which this analysis concentrates are Ukraine and Mali.
The ongoing war between the Russian Federation and Ukraine lead to several issues in different domains. Since most of the population left the country, many activities have been slowed down or interrupted. Agriculture is one of the most important sectors for the Ukrainian economy with a global relevance in terms of food security. In this context, many fields have been abandoned due to the lack of manpower.
The objective of the application is to use vegetation temporal evolution as a proxy for the agricultural activity, by means of a combination of satellite imagery from Sentinel-1 and Sentinel-2. The analysis involves the use of Sentinel-2 vegetation indexes through time to follow and estimate crop activity. Techniques are already known to perform such activity as it can be seen in [1], a method used to extract quite efficiently the abandoned fields and therefore estimate the activity over the country. Vegetation indicators that will be used are on one side NDVI (Normalized Difference Vegetation Index) and on another side LAI (Leaf Area Index). Since Sentinel-2 optical data are strongly impacted by cloud coverage, Sentinel-1 radar imagery was introduced, exploiting the correlation between the sensors concerning the vegetation monitoring as seen in [2, 3, 4]. This correlation is an opportunity for developing a machine learning solution with the objective to retrieve vegetation estimation on Sentinel-1 using Sentinel-2 vegetation indexes as reference. Moreover the integration of super resolution inside the methodology could permit to get a better spatial definition of the fields. The pipeline for the methodology is presented in Figure 1. The outcomes will be, on one hand a location detection of the defined abandoned fields, on another hand a complete statistical analysis of the vegetation evolution over the fields.
Figure 1: Pipeline from raw data to sensor combination abandoned fields detection in Ukraine
North and Central Mali entail the country’s highest prevalence of vulnerable communes siting at the intersection of chronic vulnerability, food insecurity, and armed conflicts. Humanitarian and development programs are met with a range of complex and varied challenges as they work to address the underlying causes and consequences of vulnerabilities among affected populations. The goal of this application is to provide data-driven insights and critical information to improve situational awareness for organizations operating in the area. By gaining a deeper understanding of conflict and climate-change dynamics by closely monitoring the movement of internally displaced people (IDP) and its environmental impact, we can improve our understanding of ongoing humanitarian situations and plan appropriately.
A processing chain is implemented to provide data to every needed location in Mali. The process uses the NRGB Bands of Sentinel-2 on cloud-free acquisitions. The images at spatial resolution of 10m will be enhanced to 5m by using the AI - Super
Resolution [5] tool. In the second step the AI-based Classification tool for buildings and settlements will allow identifying the covered area inside the scene. The extracted area of vegetation and water surfaces (by NDVI and NDWI) from the scene will be added to the data. To ensure undisturbed measurements, Sentinel-5P Aerosol data is used to filter the selection of scenes which were influenced by dust and sandstorms.
The development of NDVI (Normalized Difference Vegetation Index) is an important indicator which can be linked to overgrazing of livestock brought by IDPs or new created farmland to cover increased needs. NDWI (Normalized Difference Water Index) is an indicator to water access but also linked to flooding which was the main reason of refugees in Mali besides the prevailing conflict. Analysis over several years allows distinguishing between seasonal dynamics and single events.
Figure 2: Analysis of the area of settlements in Tombouctou region
REFERENCES
[1] Boudinaud, L., et S. A. Orenstein. «assessing cropland abandonment from violent conflict in central Mali with Sentinel-2 and google Earth Engine ». The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences XLVI-4/W2-2021 (19 août 2021): 9‑15. https://doi.org/10.5194/isprs-archives-XLVI-4-W2-2021-9-2021.
[2] Filgueiras, Roberto, Everardo Chartuni Mantovani, Daniel Althoff, et Elpídio Inácio Fernandes Filho. « Crop NDVI Monitoring Based on Sentinel 1 », 2019, 21.
[3] Veloso, Amanda, Stéphane Mermoz, Alexandre Bouvet, Thuy Le Toan, Milena Planells, Jean-François Dejoux, et Eric Ceschia. « Understanding the Temporal Behavior of Crops Using Sentinel-1 and Sentinel-2-like Data for Agricultural Applications ». Remote Sensing of Environment 199 (septembre 2017): 415‑26. https://doi.org/10.1016/j.rse.2017.07.015.
[4] Vreugdenhil, Mariette, Claudio Navacchi, Bernhard Bauer-Marschallinger, Sebastian Hahn, Susan Steele-Dunne, Isabella Pfeil, Wouter Dorigo, et Wolfgang Wagner. « Sentinel-1 Cross Ratio and Vegetation Optical Depth: A Comparison over Europe ». Remote Sensing 12, nᵒ 20 (janvier 2020): 3404. https://doi.org/10.3390/rs12203404.
[4] Vreugdenhil, Mariette, Claudio Navacchi, Bernhard Bauer-Marschallinger, Sebastian Hahn, Susan Steele-Dunne, Isabella Pfeil, Wouter Dorigo, et Wolfgang Wagner. « Sentinel-1 Cross Ratio and Vegetation Optical Depth: A Comparison over Europe ». Remote Sensing 12, nᵒ 20 (janvier 2020): 3404. https://doi.org/10.3390/rs12203404.
[5] M.Houël, L.Stärker, and S.Natali, AI-based Super resolution and change detection to enforce Sentinel-2 systematic usage, Medium, https://medium.com/@sistema_gmbh/ai-based-super-resolution-and-change-detection-to-enforce-sentinel-2-systematic-usage-65aa37d0365, Vienna, Dec 6, 2021
