Theme #2 of 5:
Mitigating humanitarian loss in disasters : Geospatial AI copilot
New video demos uploaded
Theme #2 of 5:
Mitigating humanitarian loss in disasters : Geospatial AI copilot
This is a supplementary website for the paper titled "Creative use of AI for humanitarian causes: HADR AI toolkit"
Note this paper contributes on 5 HADR themes.
This webpage presents Theme #2. For other 4 contributions of the paper, please click here..
Recommended viewing: This comprehensive video covers demos of all 5 themes in 20 minutes
Video Demo: Recommended for review
Contribution Theme #2 : AI copilot solution for mitigating Humanitarian loss in disasters
Humanitarian loss in disasters
HADR challenge: The key idea here is to learn from history of disaster events (from news articles/literature), and then understand the phenomenon and the identify reasons for humanitarian losses, and then use that understanding to identify future disaster risks by tapping into years of collected planetary scale geospatial satellite data. For example, learning about causes of a glacier lake burst after a rainfall from news & literature & the reported loss of lives, can AI analyze temporal satellite data to find patterns, and the suggest future risk areas across the globe.
This paper's Contribution: HADR-AI solution: By applying langchain & GPT3.5, we identify patterns & insights from historic disasters, and then use datasets from Microsoft Planetary Computer to look back in time on possible spatial patterns in past disasters. Then we use those spatial patterns and the learnt insights to mitigate risk elsewhere. We share the early stage geospatial AI prototype and the source code on our project website.
AI challenge: Fusion of geospatial & non-geospatial data for rescue planning & disaster mitigation. Analysis of temporal changes in spatial data over a period of time such as river swelling in upstream before a reported flood disaster can offer insights, which can be used to reduce future risks. The idea is to employ the power of ChatGPT to scan the web to learn about historic disasters from news & literature, and then use that insights to study spatial-temporal changes from satellite datasets during that year, and then discover patterns. These discovered patterns can be used to identify other locations on the globe, and can be used to reduce loss of lives during future disaster.
Video demo of solution: Enclosed