Publicação
A Study on ChatGPT-4.0's Performance in Cloud Masking in Landsat-8 Satellite Images
| Resumo: | This thesis analyzes cloud masking and removal performance in Landsat-8 satellite imagery using ChatGPT-4.0 and a traditional brightness thresholdbased method. The study explores how ChatGPT can be applied to geospatial tasks through prompt engineering, evaluating its effectiveness in detecting and removing cloud cover. A data set of 150 cloud-free Landsat-8 images from Portugal was used, with artificial clouds introduced to simulate realworld conditions. The research evaluates the accuracy of ChatGPT in cloud masking and compares its results to a brightness threshold-based method, which includes an inpainting step for cloud removal. Key findings indicate that ChatGPT performs well in cloud removing, but struggles with cloud segmentation. The evaluation showed that ChatGPT performed best with low density clouds, while the traditional method achieved higher accuracy with high density clouds. Prompt engineering also played a significant role in optimizing ChatGPT’s performance. These findings highlight both the potential and limitations of AI-driven cloud masking, emphasizing the need for hybrid approaches that combine LLMs with traditional remote sensing techniques. This research contributes to the ongoing exploration of large language models in satellite image analysis and provides information on the future integration of AI with geospatial data processing tools. |
|---|---|
| Autores principais: | Cardoso, Carolina Ribeiro Carneiro Fernandes |
| Assunto: | Cloud Masking ChatGPT Satellite Imagery Remote Sensing Artificial Intelligence Prompt Engineering SDG 13 - Climate action |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | This thesis analyzes cloud masking and removal performance in Landsat-8 satellite imagery using ChatGPT-4.0 and a traditional brightness thresholdbased method. The study explores how ChatGPT can be applied to geospatial tasks through prompt engineering, evaluating its effectiveness in detecting and removing cloud cover. A data set of 150 cloud-free Landsat-8 images from Portugal was used, with artificial clouds introduced to simulate realworld conditions. The research evaluates the accuracy of ChatGPT in cloud masking and compares its results to a brightness threshold-based method, which includes an inpainting step for cloud removal. Key findings indicate that ChatGPT performs well in cloud removing, but struggles with cloud segmentation. The evaluation showed that ChatGPT performed best with low density clouds, while the traditional method achieved higher accuracy with high density clouds. Prompt engineering also played a significant role in optimizing ChatGPT’s performance. These findings highlight both the potential and limitations of AI-driven cloud masking, emphasizing the need for hybrid approaches that combine LLMs with traditional remote sensing techniques. This research contributes to the ongoing exploration of large language models in satellite image analysis and provides information on the future integration of AI with geospatial data processing tools. |
|---|