QueryPlot: Generating Geological Evidence Layers using Natural Language Queries for Mineral Exploration

Meng Ye; Xiao Lin; Georgina Lukoczki; Graham W. Lederer; Yi Yao

cs.CLFeb 19, 2026

QueryPlot: Generating Geological Evidence Layers using Natural Language Queries for Mineral Exploration

Authors: Meng Ye, Xiao Lin, Georgina Lukoczki, Graham W. Lederer, Yi Yao

Organizations: SRI International, Princeton, NJ, USA · University of Kentucky, Lexington, KY, USA · U.S. Geological Survey, Reston, VA, USA

Abstract

Mineral prospectivity mapping requires synthesizing heterogeneous geological knowledge, including textual deposit models and geospatial datasets, to identify regions likely to host specific mineral deposit types. This process is traditionally manual and knowledge-intensive. We present QueryPlot, a semantic retrieval and mapping framework that integrates large-scale geological text corpora with geologic map data using modern Natural Language Processing techniques. We curate descriptive deposit models for over 120 deposit types and transform the State Geologic Map Compilation (SGMC) polygons into structured textual representations. Given a user-defined natural language query, the system encodes both queries and region descriptions using a pretrained embedding model and computes semantic similarity scores to rank and spatially visualize regions as continuous evidence layers. QueryPlot supports compositional querying over deposit characteristics, enabling aggregation of multiple similarity-derived layers for multi-criteria prospectivity analysis. In a case study on tungsten skarn deposits, we demonstrate that embedding-based retrieval achieves high recall of known occurrences and produces prospective regions that closely align with expert-defined permissive tracts. Furthermore, similarity scores can be incorporated as additional features in supervised learning pipelines, yielding measurable improvements in classification performance. QueryPlot is implemented as a web-based system supporting interactive querying, visualization, and export of GIS-compatible prospectivity layers.To support future research, we have made the source code and datasets used in this study publicly available.

Figures & tables

Figure 1. QueryPlot helps geologists accelerate mineral exploration workflow through automating document summarization, geologic maps database retrieval, and evidence layer visualization. It offloads the burden of interacting with multiple tools and managing different data products from geologists.

Figure 2. System overview of QueryPlot . Upper left: Geologic map data are processed to generate long description for each polygon. Lower left: Long documents are summarized into descriptive deposit models using LLM. Lower right: Semantic similarity between textual queries and polygon descriptions are computed and visualized as evidence layers. Upper right: A prototype system with user-friendly UI helps geologists accelerate the mineral exploration workflow.

Figure 3. Merging multiple individual polygons with same geologic signatures into one multi-polygon.

Characteristic	Description
Synonyms	Pyrometasomatic or contact metasomatic tungsten deposits.
Commodities	W, Mo, Cu, Sn, Zn
Description	Scheelite-dominant mineralization genetically associated with a skarn gangue.
Rock types	Pure and impure limestones, calcareous to carbonaceous pelites. Associated with tonalite, granodiorite, quartz monzonite and granite of both I- and S-types. W skarn-related granitoids, compared to Cu skarn- related plutonic rocks, tend to be more differentiated, more contaminated with sedimentary material, and have crystallized at a deeper structural level.
Textures	Porphyry has closely spaced phenocrysts and microaplitic quartz-feldspar groundmass.
Age range	Mainly Mesozoic, but may be any age.

Table 1. An example of selected characteristics of the descriptive model for tungsten skarn deposits from ( Day et al., 2016 ) .

Figure 4. Example prompt for summarizing documents into short and concise descriptive deposit models.

Figure 5. An evidence layer generated for national scale tungsten skarn exploration. The layer was generated from query ‘‘tonalite, granodiorite, quartz monzonite and granite.’’ , which includes the most common source rocks characteristic of tungsten skarn deposit model. Zoomed-in regions at the bottom shows the effect of different percentage threshold values, with higher threshold resulting in fewer polygons.

Figure 6. An example of finding high-potential areas adjacent to prospective regions from two evidence layers. (1) and (2) – Buffer two evidence layers by a distance of $r_{1}$ . (3) – calculate the overlapping region of the buffered layers. (4) – Buffer the intersected area by a distance of $r_{2}$ .

Figure 7. (1) The focus area of Tungsten skarn potential mapping case study. (2) Schematic representation of skarn deposits ( Lederer et al., 2021 )

Figure 8. Comparison of generated evidence layers and mineral sites (blue regions: high-ranking polygons in the host_rocks layer, red regions: high-ranking polygons in the source_rocks layer, yellow circles: mineral sites). (a)(c)(d): most sites are located near the contact regions of host and source rocks. (b): some sites do not lie on contact region but are captured by the source evidence layer.

Figure 9. Recall curve of QueryPlot , random ranking, and oracle ranking on site retrieval.

Figure 10. Comparison between QueryPlot prediction and ground truth. Purple regions: contact layer derived from high-ranking polygons in the host_rocks and source_rocks evidence layers. Orange regions (mesh): permissive tracts from ( Lederer et al., 2021 ) . (b)(c)(d): good alignment with ground truth with examples of ‘false positive’ regions. (a): missed regions that are possibly due to lack of descriptive attributes in the query with keywords consistent with the deposit model.

Figure 11. Best F1 scores with respect to different combination of thresholds for host_rocks and source_rocks evidence layers.

Approach	embedding	IOU	Precision	Recall	F1
QueryPlot + GTE ( Zhang et al., 2024b )	base-v1.5	42.74	49.03	76.92	59.89
QueryPlot + GTE ( Zhang et al., 2024b )	large-v1.5	63.14	79.89	75.07	77.40
QueryPlot + BGE ( Xiao et al., 2024 )	small-v1.5	60.63	68.45	84.14	75.49
	base-v1.5	61.18	75.01	76.83	75.91
	large-v1.5	48.53	51.97	87.99	65.34
	m3 ( Chen et al., 2024 )	52.80	68.63	69.59	69.11

Table 2. QueryPlot evaluation results on unsupervised Tungsten skarn MPM. All the numbers are best performance acquired through grid search over hyperparameters.

Data split	Metric	GFM ( Daruna et al., 2024 )	GFM ( Daruna et al., 2024 ) + QueryPlot
Validation	AUPRC	47.31	43.52
	AUC	89.28	90.45
	F1	43.13	46.91
Test	AUPRC	41.53	43.04 (+1.51)
	Bal. Acc	72.39	78.37 (+5.98)
	AUC †	91.85	92.91

Table 3. Evaluation results on supervised Tungsten skarn MPM. A geology evidence layer generated from QueryPlot is used as an additional input channel to the binary classifier. †Metrics that are not suited for measuring performance on imbalanced data .

Figure 12. The QueryPlot prototype system has a user friendly GUI that supports rich features, including flexible and customizable data pre-processing, interactive layer visualization, real-time filtering, and exporting processed layers to common formats readable by popular GIS software.

Explore similar work

MineTRACE: An Evidence-Grounded Interactive Reasoning System for Mineral Prospectivity
Sep 2, 2026Yiran Zhang, Jinwen Liu, Daniel Su +7Geospatial Reasoning Evidence-Grounded Generation
1The University of Western Australia · 2Wilfrid Laurier University
GeoCFM: Positive-Only Conditional Flow Matching for Mineral Occurrence Sampling
Sep 28, 2026Moshe Eliasof, Eldad HaberConditional Flow Matching Geospatial Reasoning
Faculty of Computer and Information Science, Ben-Gurion University of the Negev, Israel · Department of Earth, Ocean and Atmospheric Sciences, University of British Columbia, Canada
Bayesian deep learning integration of geophysical and drilling data for 3D prediction of copper mineralization and drill targeting: a case study from the Kogodai prospect, Rudny Altai
Sep 8, 2026Margarita Veshchezerova, Egor Barashov, Evgenii Gusev +4
Terra Quantum AG, Kornhausstrasse 25, 9000, St. Gallen, Switzerland · Kogodai Joint Venture LLP, Gorky Street 46, 070004, Ust-Kamenogorsk, Kazakhstan

Sep 2, 2026J/K move · Enter open · S save

MineTRACE: An Evidence-Grounded Interactive Reasoning System for Mineral Prospectivity

Yiran Zhang, Jinwen Liu, Daniel Su, Yisu Chen+6

1The University of Western Australia · 2Wilfrid Laurier University

Geospatial Reasoning Evidence-Grounded Generation

Open paper AskPDF arXiv

Mineral exploration requires integrating heterogeneous geochemical, geophysical, and geological evidence, yet existing prospectivity systems often provide only opaque scores or heatmaps. We present MineTRACE, a web-based system for evidence-grounded exploration of eight commodities: Cu, Au, Ni, W, Sn, Co, Ta, and Mn. Users can explore prospectivity maps, query locations or regions, inspect supporting evidence, and interact through natural language. A transparent expert tree, informed by geological knowledge and known deposits, combines multi-source evidence into interpretable prospectivity scores. For a new location, the conversational assistant retrieves the score and supporting evidence from the analysis pipeline and presents them in natural language. The scorer achieves spatial AUC values of up to 0.917 across different test scenarios, while end-to-end evaluation assesses query accuracy and response grounding. MineTRACE makes public geoscience data easier to access, interpret, and verify, supporting more efficient and transparent mineral exploration.

Abstract

Explore similar work

MineTRACE: An Evidence-Grounded Interactive Reasoning System for Mineral Prospectivity

GeoCFM: Positive-Only Conditional Flow Matching for Mineral Occurrence Sampling

Bayesian deep learning integration of geophysical and drilling data for 3D prediction of copper mineralization and drill targeting: a case study from the Kogodai prospect, Rudny Altai