Trustworthy GeoAI for Urban Air Quality: Engineering a Production Monitoring, Spatial-Analysis, and Forecasting Platform for Greater Bilbao
Year: 2026
Type: article
Abstract: This paper documents the engineering of an end-to-end GeoAI platform that monitors, spatially explains, and forecasts urban air quality across the Greater Bilbao metropolitan area. The system integrates an automated daily data pipeline (7 monitoring stations, ~29,000 daily records, 2015–2026), a four-notebook GIS spatial analysis layer (land-use buffers, landmark distances, DEM-derived terrain, wind-transport analysis), validated XGBoost forecasting models, SHAP-based explainability, and a tool-grounded conversational agent — all deployed as a public, continuously updated web application. Rather than presenting the platform as a finished product, this paper focuses on the engineering decisions behind it: why a strict temporal validation protocol replaced an initially inflated random-split evaluation (R² 0.84 → 0.479); why spatial context was engineered as a separate exploratory layer rather than injected into the forecasting models; why a parametric tool interface was chosen over free-form LLM data access; and which limitations remain unavoidable at n = 7 stations. The central claim is methodological: a GeoAI system is only as trustworthy as the engineering discipline behind it.
Author: Ghaziaskari Naeini, Arman
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7170060227
FWCI:
Cites:
Cited by:
Related to: