Scholarlink

<p>This study evaluates the potential of Geo-Enabled Large Language Models (Geo-LLMs) for integrated flood and landslide risk assessment using a recent rainfall-triggered landslide in Kishtwar district, Jammu &amp; Kashmir, India. The proposed framework integrates multi-source geospatial datasets including satellite imagery, terrain derivatives, rainfall data, and infrastructure layers with textual disaster reports through spatial–textual multimodal inference. The model generates probabilistic landslide susceptibility surfaces and spatial uncertainty estimates. Results show strong spatial agreement between predicted and observed landslide extents (IoU = 0.70–0.80), with improved predictive performance compared with conventional GIS and machine-learning approaches. The framework also reduced analytical turnaround time by approximately 40–50%. The study demonstrates the potential of geographically constrained multimodal AI models to support uncertainty-aware geospatial analysis and decision-oriented disaster risk management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Geo-enabled large language models for landslide risk assessment in Kishtwar, India

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Vinay Kumar Gaddam,
  • Sravya Vemparala,
  • Tejsai Moturu,
  • Ahmed Abdul,
  • Jaswanth Kasturi

摘要

This study evaluates the potential of Geo-Enabled Large Language Models (Geo-LLMs) for integrated flood and landslide risk assessment using a recent rainfall-triggered landslide in Kishtwar district, Jammu & Kashmir, India. The proposed framework integrates multi-source geospatial datasets including satellite imagery, terrain derivatives, rainfall data, and infrastructure layers with textual disaster reports through spatial–textual multimodal inference. The model generates probabilistic landslide susceptibility surfaces and spatial uncertainty estimates. Results show strong spatial agreement between predicted and observed landslide extents (IoU = 0.70–0.80), with improved predictive performance compared with conventional GIS and machine-learning approaches. The framework also reduced analytical turnaround time by approximately 40–50%. The study demonstrates the potential of geographically constrained multimodal AI models to support uncertainty-aware geospatial analysis and decision-oriented disaster risk management.