GeoFaham: Domain-Specialized Multi-Agent Geospatial Analytics for Disaster Response

Muhammad Sohail Danish; Gilles Q. Hacheme; Akram Zaytar; Satej S. Soman; Inbal Becker-Reshef; Juan Lavista Ferres; Salman Khan; Girmaw Abebe Tadesse

cs.MANov 3, 2026

GeoFaham: Domain-Specialized Multi-Agent Geospatial Analytics for Disaster Response

Authors: Muhammad Sohail Danish, Gilles Q. Hacheme, Akram Zaytar, Satej S. Soman, Inbal Becker-Reshef, Juan Lavista Ferres, Salman Khan, Girmaw Abebe Tadesse

Organizations: Work done while the author was an AI Resident at the Microsoft AI for Good Lab, Nairobi, Kenya. · Microsoft AI for Good Lab · Mohamed bin Zayed University of Artificial Intelligence

Abstract

Real-world geospatial decision support requires more than single-step retrieval or code generation. Disaster-response questions often involve multi-source workflows that combine spatial databases, map-derived features, satellite imagery, raster products, and reusable geospatial artifacts, where correctness depends on preserving spatial scope, temporal context, and intermediate execution state. While large language models and coding agents offer a promising interface for such workflows, they struggle with geospatial-specific failure modes, including handling large artifacts, selecting the wrong evidence source, losing spatial or temporal metadata, and applying invalid geometric operations. We introduce GeoFaham, a domain-specialized multi-agent system for geospatial disaster analytics. GeoFaham decomposes workflows into specialist agents for vector operations, map feature retrieval, imagery discovery, and raster analysis, coordinated by an Orchestrator with a dual-ledger control loop for dependency-aware planning, progress tracking, validation, and replanning. To support artifact-centric collaboration, GeoFaham uses a Response Store that persists full geospatial outputs and typed metadata outside the LLM context window while agents exchange compact identifiers and summaries. We also introduce GeoFaham-Bench, a 40-task benchmark grounded in 17 real-world disaster records, designed to evaluate multi-step geospatial reasoning, cross-source integration, raster-vector computation, artifact generation, and recovery under non-rigid ground truth. We evaluate GeoFaham on GeoFaham-Bench against strong general-purpose coding-agent baselines, including Codex, Claude Code, and OpenCode. GeoFaham achieves the best overall performance, scoring 0.940 versus the next-best 0.879, while using 7.5 times fewer tokens. Our code and benchmark are publicly available at https://github.com/microsoft/GeoFaham.

Figures & tables

Figure 1: GeoFaham overview. An Orchestrator coordinates four domain-specialized agents (Vector/PostGIS, Maps/OSM, STAC imagery retrieval, and Raster code generation) using nested control loops. The outer loop produces a task ledger and dependency-aware plan, while the inner loop tracks execution in a progress ledger with validation and replanning. Agents persist large geospatial outputs to a Response Store and exchange only compact identifiers and typed metadata through the Orchestrator.

Figure 2: GeoFaham’s dual-ledger orchestration loop. The task ledger supports fact extraction and dependency-aware planning, while the progress ledger tracks execution, validates artifacts, and triggers replanning when needed.

Figure 3: Artifact-aware collaboration across the LLM context boundary. Agents exchange identifiers and typed metadata, while the execution layer resolves referenced artifacts for deterministic spatial operations.

Figure 4: Cross-system comparison on GeoFaham-Bench. (a) Overall score versus token usage per task. (b) Selected dimension scores, showing the largest gains in data selection and spatial–temporal reasoning.

System	$\mathbf{S}_{\mathrm{out}}$	$\mathbf{S}_{\mathrm{meth}}$	$\mathbf{S}_{\mathrm{repr}}$	$\mathbf{S}_{\mathrm{comp}}$	$\mathbf{S}_{\mathrm{rob}}$	$\mathbf{S}_{\mathrm{c}}$
GeoFaham	0.897	0.951	0.982	0.999	0.933	0.940
Codex	0.794	0.890	0.974	0.986	0.896	0.879
Claude Code	0.738	0.863	0.902	0.974	0.873	0.837
OpenCode	0.622	0.828	0.865	1.000	0.838	0.780

Table 1: Cross-system performance on GeoFaham-Bench. GeoFaham achieves the highest overall score among Codex OpenAI, 2025 , Claude Code Anthropic, 2025 , and OpenCode OpenCode Contributors, 2025 in overall score, with consistently strong performance across task outcome, geospatial methodology, reproducibility, compliance, and robustness.

Config	$\mathbf{S}_{\mathrm{plan}}$	$\mathbf{S}_{\mathrm{exec}}$	$\mathbf{S}_{\mathrm{comp}}$	$\mathbf{S}_{\mathrm{out}}$	$\mathbf{S}_{\mathrm{i}}$
C1	0.849	0.832	0.825	0.680	0.808 $\pm$ 0.12
C2	0.944	0.906	0.912	0.885	0.914 $\pm$ 0.07
C3	0.983	0.958	0.961	0.952	0.964 $\pm$ 0.03
C4	0.980	0.957	0.958	0.934	0.959 $\pm$ 0.04

Table 2: Performance across GeoFaham model configurations. C3 achieves the highest overall score (

S_{\mathrm{i}}=0.964

), showing that 9selectively assigning stronger models to reasoning-intensive roles matches or slightly exceeds the all-GPT-5.1 configuration (C4). Overall scores are reported as mean

\pm

standard deviation across tasks; category columns report means.

Figure 5: Error occurrence and recovery across model configurations. C1 and C2 have the same error rate, while C3 and C4 reduce error occurrence.

System	Duration sec.	Avg. Tokens Per Task	Median Tokens Per Task	Total Tokens (M)
GeoFaham	169.64	106,344	86,974	4.29995
Codex	214.92	795,352	741,656	31.81
ClaudeCode	254.23	285,100	227,596	11.40
OpenCode	128.37	376,042	317,846	15.04

Table 3: Cross-system runtime and token efficiency. GeoFaham uses substantially fewer tokens than all general-purpose coding-agent baselines while maintaining competitive runtime.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

ID	Type	Location	Country	Date	Available Data
1	Tornado	Saint Louis	USA	2025-05-16	Building damage
2	Hurricane	Black River	Jamaica	2025-10-29	Building damage
3	Hurricane	Montego Bay	Jamaica	2025-10-31	Building damage
4	Flood	Porto Alegre	Brazil	2024-05-09	Building damage
5	Cyclone	Mamoudzou	Mayotte	2024-12-18	Building damage
6	Flood	Valencia	Spain	2024-10-31	Flood map

Appendix

Table 4: Disaster records used in GeoFaham-Bench. The benchmark covers 17 records from 2023 to 2025, with building damage assessments, flood maps, or both. IDs 12–14 correspond to the same Palisades Fire recorded on three different dates.

Task Category	Tasks (#)	Share (%)
Point-of-interest impact analysis	11	27.5%
Infrastructure analysis	8	20.0%
Clustering and aggregation	7	17.5%
Building-damage assessment	4	10.0%
Comparative analysis	4	10.0%
Flood analysis	2	5.0%

Appendix

Table 5: Distribution of tasks across the 10 GeoFaham-Bench task categories. The benchmark emphasizes multi-source impact and infrastructure analysis while retaining coverage of raster-, imagery-, and event-comparison workflows.

Dimension	GeoFaham	Codex	Claude	OpenCode
Task Outcome
Category Score	0.897	0.794	0.738	0.622
Answer Accuracy	0.876	0.766	0.669	0.586
Answer Completeness	0.918	0.821	0.806	0.657
Geospatial Method
Category Score	0.951	0.890	0.863	0.828

Appendix

Table 6: Detailed rubric scores across systems and evaluation dimensions.

Dimension	C1	C2	C3	C4
Planning
Plan Correctness	0.861	0.948	0.994	0.992
Plan Completeness	0.833	0.951	0.995	0.988
Execution
Tool Selection	0.924	0.978	0.998	0.998
Parameter Accuracy	0.797	0.879	0.942	0.935

Appendix

Table 7: Per-dimension evaluation scores across configurations.

Config	Errors	Recovered Runs	Replan	Replan OK	Slow	Redundant
C1	61.3%	55.0%	27.5%	22.5%	13.8%	38.8%
C2	61.3%	61.3%	31.2%	28.7%	18.8%	42.5%
C3	33.8%	33.8%	3.8%	3.8%	1.2%	16.2%
C4	28.8%	28.8%	5.0%	5.0%	1.2%	17.5%

Appendix

Table 8: LLM-judged behavioral indicators across model configurations. Values report the percentage of all evaluated runs containing execution errors, successful recovery, replanning, successful replanning, slow execution, or redundant operations.

Config	Mean	Median	Prompt	Completion
C1	124,606	80,232	122,567	2,039
C2	129,259	93,800	127,156	2,103
C3	104,146	92,844	99,606	4,540
C4	105,668	76,938	99,738	5,930

Appendix

Table 9: Token usage across model configurations. Values report the mean and median total token consumption across evaluated tasks, together with the average prompt and completion tokens.

Figure 6: Example GeoFaham session for multi-step flood-impact analysis. The conversation panel presents user queries, system responses, tables, and artifact notifications, while the map workspace displays the flood extent, retrieved points of interest, satellite imagery, and generated vector outputs.

Figure 7: Example GeoFaham session for imagery-based burn-severity analysis. The user defines an area of interest, after which the system retrieves pre-fire and post-fire imagery and generates the corresponding burn-severity output in the map workspace.

Geospatial Reasoning

Geospatial reasoning studies how AI interprets locations, maps, spatial relationships, and Earth-observation data to answer questions and make decisions.

See all 79 papers

GeoDisaster: Benchmarking Orchestrated Agents for Operational Disaster Geo-Intelligence
Jun 15, 2026Maram Hasan, Aman Verma, Savitra Roy +5Disaster Damage Assessment RL for Language Model Reasoning
Indian Institute of Technology Bombay · Mohamed bin Zayed University of Artificial Intelligence
Can LLM Agents Respond to Disasters? Benchmarking Heterogeneous Geospatial Reasoning in Emergency Operations
May 12, 2026Junjue Wang, Weihao Xuan, Heli Qi +7LLM Agent Evaluation Geospatial Reasoning
The University of Tokyo · RIKEN AIP · Waseda University +1
Google Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic Evaluation
Oct 11, 2026Noam Slonim, Aviv Slobodkin, Gil Fidel +34Geospatial QA Geospatial Reasoning
Google Research · Google Geo · Spatial Bytes

Jun 15, 2026J/K move · Enter open · S save

GeoDisaster: Benchmarking Orchestrated Agents for Operational Disaster Geo-Intelligence

Maram Hasan, Aman Verma, Savitra Roy, Hariseetharam Gunduboina+4

Indian Institute of Technology Bombay · Mohamed bin Zayed University of Artificial Intelligence

Disaster Damage Assessment RL for Language Model Reasoning LLM Agent Orchestration Geospatial Reasoning Tool-Using Vision-Language Agents Remote Sensing Benchmarks+1

Open paper AskPDF arXiv

Remote-sensing vision-language models (RS-VLMs) have advanced Earth-observation analysis toward visual interpretation and instruction-following, yet fall short of operational geo-intelligence, which demands tool-grounded spatial reasoning and structured, evidence-backed decisions. We introduce GeoDisaster, an operational geospatial disaster reasoning benchmark with 2,921 verified instances across 43 question types and five task families: deforestation monitoring, multi-hazard analysis, building-damage assessment, flood-safe routing, and Sentinel-1 SAR flood monitoring. Instances integrate heterogeneous EO/GIS evidence-optical and SAR imagery, raster masks, vector geometries, road networks, and exposure layers-spanning hazard detection, damage assessment, exposure estimation, and diagnostic report generation. Ground-truth answers are grounded in executable geospatial workflows and deterministic consistency checks, removing the need for language-model annotation. We further propose an orchestrated multi-agent framework with 18 disaster-oriented tools, where role-specialized agents coordinate through explicit execution contracts, aligned via Role-Contract Expectation Alignment (RCEA): failure-aware supervised fine-tuning combined with contract-grounded reinforcement learning over dense step-level signals. Experiments show that GeoDisaster challenges existing RS-VLMs and agentic systems, while RCEA improves tool use, evidence grounding, state consistency, and decision generation.

Abstract

Geospatial Reasoning

Explore similar work

GeoDisaster: Benchmarking Orchestrated Agents for Operational Disaster Geo-Intelligence

Can LLM Agents Respond to Disasters? Benchmarking Heterogeneous Geospatial Reasoning in Emergency Operations

Google Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic Evaluation