Google Earth AI is a family of geospatial AI models, together with an agent that orchestrates them. We present this agent, Google Earth AI's Geospatial Reasoning agent (GRA), a Gemini-powered research system that combines geospatial data sources, foundation models, and tools to answer questions about our planet. GRA organizes these into modular domain skills and pairs them with upfront geospatial planning to decompose complex queries into multi-stage workflows. Evaluating such systems is hard: real-world problems admit multiple valid workflows, require data discovery across large catalogs, and rely on evolving ground truth. We introduce two benchmarks and release their development splits. GeoExpertQueries poses complex analytical problems grounded in practitioner use cases and refined by geospatial experts; each is paired with a consensus golden workflow from three independent experts and scored by a human-validated, tool-agnostic autorater. GeoKTools consists of templated queries with deterministic reference solvers that systematically vary the number of required tool types. We also use two external benchmarks, CBGB and UnivEARTH; the latter we find to be saturated. To isolate the contribution of GRA's tools and its agentic orchestration, we compare four configurations, each with Gemini 3.1 Pro and 3.8 Flash: a standalone LLM, the LLM with basic tools, the LLM with GRA's skills and tools but without its upfront planning and execution guidance, and GRA itself. With at most basic tools, an advanced model recovers much of the answer on templated queries but satisfies only a small fraction of the expert workflows on open-ended problems; GRA performs well on both. The largest step comes from GRA's skills and tools; on open-ended problems, its planning and execution guidance add a further gain with Gemini 3.8 Flash, letting Flash match or exceed Pro at lower cost.
Figures & tables
Figure 1: The Geospatial Reasoning agent (GRA) and how we evaluate it. The agent plans its approach to a natural-language query, draws on geospatial capabilities across five pillars (Table 1 ), analyzes the retrieved data, and synthesizes a grounded answer. The bottom panel shows how our two benchmarks score the agent’s responses. Greyed-out items are not evaluated in this report.
Data Commons* (e.g., US Census ACS, CDC, FEMA NRI, World Bank); PDFM* (Population Dynamics embeddings)
Demographic, economic & health statistics across administrative levels; regional similarity search
POIs, Boundaries & Mobility
Google Maps Data*; Who’s On First, US Census TIGER/Line & ZCTA
Boundary resolution (countries to census tracts) & spatial filtering; POI search & counting; routes, travel-time matrices & isochrones; rooftop solar potential
Prediction & Spatial Compute
Scratch datasets with sandboxed Python; Google Earth Engine* (general-purpose analysis of any catalog dataset); TimesFM*; PDFM*; Planetary Prediction Engine ( Ma et al., 2026 ) , which combines them with AlphaEarth Foundations* embeddings and Data Commons* data
Automated data and model selection ( Ma et al., 2026 ) ; out-of-context data handling & custom Python analysis, including custom Earth Engine scripts; CRS-aware buffering, distances & spatial joins; regional statistics over gridded datasets; time-series forecasting; regional prediction & super-resolution
Table 1: Overview of the GRA capability ecosystem across five domain pillars (* denotes models or datasets developed at Google).
Figure 2: Overview of GRA. Skills stay on standby and are loaded only when a query needs them. The agent first sets out a plan, then carries it out in a reasoning loop, while intermediate datasets pass between tools outside the model’s context window.
Figure 3: The four evaluated configurations. Each adds one layer to the same Gemini model: basic tools, then GRA’s geospatial skills and tools, then GRA’s upfront planning and execution guidance.
Benchmark
Type
Queries (dev/test)
Evaluation
Primary Focus
GeoExpertQueries
Internal, open-ended
25 / 76
Golden workflows + tool-agnostic autorater
Real-world geospatial reasoning
GeoKTools
Internal, controlled
85 / 220
Ground-truth answers + autorater
Multi-tool orchestration
CBGB
External
— / 44
Deterministic answer match (original criterion)
Quantitative Earth-observation analysis
UnivEARTH
External
— / 140
Ternary answer match
Data discovery and end-to-end reasoning
Table 2: Summary of the benchmark suite.
Domain
Example query
Natural Hazards & Crisis Resilience
Using SAR Interferometry (InSAR) coherence change detection, estimate the structural damage in the ‘hard-hit’ zones of the 2022 Pakistan floods in the Sindh and Balochistan provinces. Differentiate between ‘inundated’ (water) and ‘destroyed’ (rubble). Overlay with poverty maps to prioritize reconstruction grants.
Public Health
Assess the physical accessibility of emergency obstetric care in Sierra Leone by calculating travel times over the Dynamic World land cover friction surface.
Site Selection & Urban Planning
For the city of Buenos Aires, determine the ‘Building Footprint’ area vs. available land area.
Table 3: Example queries from GeoExpertQueries, drawn from the publicly released development split. Appendix B explains each example in plain language.
Benchmark
LLM
LLM + Basic Tools
LLM + GRA Tools
GRA
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
GeoExpertQueries
0.0 ± 0.0
0.0 ± 0.0
2.3 ± 0.8
3.3 ± 0.8
50.8 ± 1.7
52.2 ± 5.4
47.6 ± 2.7
62.7 ± 2.5
GeoKTools
37.1 ± 1.2
40.5 ± 0.7
37.8 ± 1.9
40.0 ± 0.8
84.5 ± 1.4
86.0 ± 1.6
81.6 ± 1.4
85.9 ± 1.5
CBGB
14.4 ± 3.5
12.9 ± 1.3
13.6 ± 2.3
19.7 ± 2.6
84.8 ± 1.3
87.9 ± 1.3
84.8 ± 1.3
86.4 ± 2.3
UnivEARTH
86.9 ± 1.8
93.6 ± 0.7
96.7 ± 1.8
98.3 ± 0.8
97.9 ± 0.7
98.8 ± 0.4
89.5 ± 0.4
98.1 ± 0.4
Table 4: Main results (test set). Scores are 0–100, mean ± std over three runs. For CBGB, the score is the average over all 44 questions, across its three difficulty tiers. Bold marks the best result in each row and results whose ± ranges overlap with it.
Domain (n)
LLM + Basic Tools
LLM + GRA Tools
GRA
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
Crisis Resilience (35)
1.0 ± 1.6
1.0 ± 1.6
49.4 ± 1.5
45.6 ± 3.8
50.0 ± 1.4
66.0 ± 2.6
Public Health (17)
0.0 ± 0.0
0.0 ± 0.0
45.4 ± 8.6
42.5 ± 10.2
39.2 ± 2.9
55.6 ± 3.2
Site Selection (20)
3.6 ± 0.5
6.9 ± 1.3
57.2 ± 5.3
71.4 ± 5.0
48.1 ± 6.3
65.6 ± 3.4
Operational (4)
18.1 ± 4.8
19.4 ± 6.4
54.2 ± 7.2
55.6 ± 6.4
59.7 ± 4.8
50.0 ± 0.0
All (76)
2.3 ± 0.8
3.3 ± 0.8
50.8 ± 1.7
52.2 ± 5.4
47.6 ± 2.7
62.7 ± 2.5
Table 5: GeoExpertQueries results by domain (test set). Scores are 0–100, mean ± std over three runs. The standalone LLM scores 0.0 in every domain with both models and is omitted. Domains: Natural Hazards & Crisis Resilience; Public Health; Site Selection & Urban Planning; Operational & Commercial Utilities. Bold marks the best result in each row and results whose ± ranges overlap with it.
K
LLM
LLM + Basic Tools
LLM + GRA Tools
GRA
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
K=1 (30)
46.5 ± 2.5
43.7 ± 3.2
47.7 ± 5.0
51.6 ± 3.4
85.0 ± 2.1
90.4 ± 3.8
86.2 ± 7.1
91.2 ± 2.1
K=2 (110)
41.0 ± 1.7
44.0 ± 0.1
39.1 ± 2.8
41.8 ± 1.5
85.1 ± 1.9
85.3 ± 2.2
79.4 ± 1.6
85.3 ± 1.1
K=3 (50)
28.9 ± 2.0
33.9 ± 1.3
35.8 ± 2.6
35.3 ± 1.2
83.3 ± 4.1
86.2 ± 2.2
80.1 ± 3.6
82.0 ± 2.6
K=4 (30)
27.3 ± 3.8
35.6 ± 3.6
26.9 ± 2.9
29.4 ± 1.4
83.6 ± 1.1
84.0 ± 3.3
87.8 ± 1.6
89.2 ± 2.3
All (220)
37.1 ± 1.2
40.5 ± 0.7
37.8 ± 1.9
40.0 ± 0.8
84.5 ± 1.4
86.0 ± 1.6
81.6 ± 1.4
85.9 ± 1.5
Table 6: GeoKTools results by number of required tool types, K (test set; query counts in parentheses). Scores are 0–100, mean ± std over three runs. Bold marks the best result in each row and results whose ± ranges overlap with it.
Tier (n)
LLM
LLM + Basic Tools
LLM + GRA Tools
GRA
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
Easy (15)
24.4 ± 3.8
22.2 ± 3.8
22.2 ± 3.8
26.7 ± 0.0
100.0 ± 0.0
100.0 ± 0.0
95.6 ± 3.8
97.8 ± 3.8
Intermediate (20)
8.3 ± 5.8
11.7 ± 2.9
13.3 ± 5.8
21.7 ± 2.9
78.3 ± 5.8
81.7 ± 2.9
80.0 ± 0.0
81.7 ± 2.9
Difficult (9)
11.1 ± 0.0
0.0 ± 0.0
0.0 ± 0.0
3.7 ± 6.4
74.1 ± 6.4
81.5 ± 6.4
77.8 ± 0.0
77.8 ± 0.0
All (44)
14.4 ± 3.5
12.9 ± 1.3
13.6 ± 2.3
19.7 ± 2.6
84.8 ± 1.3
87.9 ± 1.3
84.8 ± 1.3
86.4 ± 2.3
Table 7: CBGB results by difficulty tier (test set; question counts in parentheses). Scores are 0–100, mean ± std over three runs. Bold marks the best result in each row and results whose ± ranges overlap with it.
Configuration
3.1P
3.8F
LLM
86.9 ± 1.8
93.6 ± 0.7
LLM + Basic Tools
96.7 ± 1.8
98.3 ± 0.8
LLM + GRA Tools
97.9 ± 0.7
98.8 ± 0.4
GRA
89.5 ± 0.4
98.1 ± 0.4
Table 8: UnivEARTH results (test set, 140 questions). Scores are 0–100, mean ± std over three runs.
Benchmark
LLM
LLM + Basic Tools
LLM + GRA Tools
GRA
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
3.1P
3.8F
GeoExpertQueries
$0.028
$0.010
$0.115
$0.044
$0.512
$0.167
$0.688
$0.199
GeoKTools
$0.024
$0.012
$0.575
$0.112
$0.410
$0.147
$0.360
$0.122
CBGB
$0.025
$0.017
$0.514
$0.132
$0.140
$0.110
$0.151
$0.093
UnivEARTH
$0.006
$0.002
$0.029
$0.010
$0.116
$0.046
$0.271
$0.079
Table 9: Estimated inference cost per query (USD).
LLM + GRA Tools
GRA
3.1P
3.8F
3.1P
3.8F
Imagery & Earth Observation
< 1%
2%
4%
3%
Weather, Climate & Hazards
18%
18%
23%
18%
Population, Health & Socioeconomics
21%
22%
23%
20%
POIs, Boundaries & Mobility
83%
57%
93%
89%
Prediction & Spatial Compute
83%
75%
93%
99%
Table 10: Capability usage on GeoExpertQueries (test set, all three runs). Top: share of queries in which at least one skill from each capability pillar (Table 1 ) is loaded. Bottom: average activity per query. General-purpose Earth Engine skills are counted under Prediction & Spatial Compute, since they serve all domains. Map-display skills are omitted.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Category
Capability
Analytical Functionality & Operations
Underlying Data Source or Model
Imagery & Earth Observation
Embedding-based surface change detection
Annual surface-similarity maps and vectorized change polygons between two years, from cosine similarity of 64-dimensional satellite embeddings
AlphaEarth Foundations* ( Brown et al., 2025 ) (in Google Earth Engine* ( Gorelick et al., 2017 ) )
Terrain slope
Zonal slope statistics (mean, min, max percent grade) from a 30 m elevation model over administrative or custom geometries
Copernicus DEM GLO-30 ( European Space Agency, 2022 )
Geospatial reasoning, i.e., computing distances, containment, and other spatial relations over real-world entities, is central to navigation and logistics, yet large language models (LLMs) struggle with the required geometric and topological computation despite storing considerable geographic knowledge. Existing benchmarks localize these failures only partially: they are synthetic or smallscale, largely monolingual, and offer limited control over geographic coverage. We introduce MultiGlobeQA, a multilingual benchmark of 46,060 question-answer pairs spanning 14 spatial-function families and 15 answer formats, with execution-based ground truth over three knowledge graphs. It covers 201 countries and territories via income- and density-stratified sampling, with parallel questions in English and 16 additional high- and low-resource languages. Across parametric, reasoning, and agentic settings, LLMs collapse on tasks requiring grid indexing and shape computation, while topological relations and directions fare best. Retrieval and tool use yield considerable gains, yet performance plateaus below two thirds even when gold facts are supplied, indicating that computation, not access to knowledge, is the bottleneck. Models also underperform on low-income regions, a gap that gold facts widen rather than close.
Martin Böckling, Elizaveta Nosova, Heiko Paulheim +1
Data and Web Science Group, University of Mannheim, Germany
Environmental scientists spend disproportionate effort on data wrangling rather than analysis. New AI agents can be a helpful tool, but no benchmark exists to evaluate AI agents that automate environmental geospatial workflows through structured tool calling against real APIs. We introduce the GeoNatureAgent Benchmark, the first benchmark for environmental analysis agents that operate via structured tool calls to a production-style geospatial API. The benchmark comprises 93 tasks across 18 categories. Tasks are evaluated against an open, self-hostable geospatial API that serves three environmental indicators across Spain and Portugal via sixteen tools. We evaluate nine frontier and open-weight LLMs, reporting capability and per-case cost as orthogonal axes. Results manifest that (1) Claude Sonnet 4 achieves the highest capability at 60.8% +/- 0.8%, followed closely by DeepSeek V3.2 at 56.3% +/- 3.1%, while no other model exceeds 51%; (2) the cost-accuracy Pareto frontier is occupied mostly by open-weight models, with DeepSeek V3.2 offering 93% of Claude's capability at 11.6x lower cost; and (3) structured tool calling against a real API provides a more discriminative measure of real-world agent capability, with mean accuracies 25-35 percentage points below those reported on general-purpose GIS benchmarks.
Gabriel Diaz-Ireland, Diego Prieto-Herráez, Mario García Peces +2
Universidad Católica de Ávila (UCAV) Ávila, Spain · Johns Hopkins University Baltimore, MD, USA · Independent Researcher Madrid, Spain +1
Benchmarking spatial reasoning in multimodal large language models (MLLMs) has attracted growing interest in computer vision due to its importance for embodied AI and other agentic systems that require precise interaction with the physical world. However, spatial reasoning on Earth imagery has lagged behind, as it uniquely involves grounding objects in georeferenced images and quantitatively reasoning about distances, directions, and topological relations using both visual cues and vector geometry coordinates (e.g., 2D bounding boxes, polylines, and polygons). Existing benchmarks for Earth imagery primarily focus on 2D spatial grounding, image captioning, and coarse spatial relations (e.g., simple directional or proximity cues). They lack support for quantitative direction and distance reasoning, systematic topological relations, and complex object geometries beyond bounding boxes. To fill this gap, we propose EarthSpatialBench, a comprehensive benchmark for evaluating spatial reasoning in MLLMs on Earth imagery. The benchmark contains over 325K question-answer pairs spanning: (1) qualitative and quantitative reasoning about spatial distance and direction; (2) systematic topological relations; (3) single-object queries, object-pair queries, and compositional aggregate group queries; and (4) object references expressed via textual descriptions, visual overlays, and explicit geometry coordinates, including 2D bounding boxes, polylines, and polygons. We conducted extensive experiments on both open-source and proprietary models to identify limitations in the spatial reasoning of MLLMs.
Zelin Xu, Yupu Zhang, Saugat Adhikari +6
Department of Computer & Information Science & Engineering, University of Florida, Gainesville, FL, USA · Department of Computer Science, Indiana University Bloomington, Bloomington, IN, USA.