This poster was presented at Computer Graphics & Visual Computing (CGVC24) at City St George's, University of London.
Mansi Agrawal
In the face of worsening climatic conditions, the success of carbon offset projects has gained heightened significance. Our study addresses this imperative by developing a wildfire risk assessment model that can be seamlessly integrated into the carbon offset frameworks, filling a critical research gap within carbon markets. We extract texture features from high-resolution remote sensing images, employ dimensionality reduction techniques to select features crucial for specific risk classes, leverage unsupervised learning through clustering-based undersampling for handling imbalanced datasets, and iteratively enhance the dataset by modifying labels. Utilizing convolutional neural networks, we develop robust wildfire risk classification models, achieving recall rates up to 99% for the higher fire risk classes – a critical milestone for facilitating informed decision- making. Additionally, we (a) establish a framework for on-demand acquisition of remote sensing images to assess wildfire risk in forest offset projects, (b) evaluate the effectiveness of current permanence strategies in carbon markets for addressing unavoidable risks, and (c) consider how our findings enhance these strategies. Ultimately, this research contributes to the resilience of carbon offset projects and establishes a foundation for grounded policymaking in carbon markets, crucial for navigating the complexities of the contemporary environmental landscape. Keywords: Wildfire Risk Classification, Carbon Offset Projects, Remote Sensing Images, Clustering-Based Undersampling, Neural Network Optimization.