Dataset of Spatial Characteristics of Urban Functional Area Identification and Fire Risk in Gongyi City, China Based on Multi-source Spatiotemporal Data and Stacking Ensemble Learning (2020-2024)

This dataset is the core data achievement supporting the paper "Research on Fire Risk Characteristics of Urban Functional Areas Based on Spatiotemporal Data Integration Learning", with Gongyi City in Henan Province, China as the research area. The dataset aims to achieve high-precision identification of urban functional areas by integrating multi-source spatiotemporal big data with Stacking ensemble learning models, and on this basis, systematically reveal the spatial differentiation rules of fire risks in different functional areas. Data content: The dataset mainly consists of two parts: The urban functional area recognition feature dataset integrates six types of point of interest (POI) kernel density, road network density, Harvard's nighttime light intensity, population spatial distribution grid from the seventh national census, as well as 12 feature indicators extracted from Sentinel-2 remote sensing images, including NDVI, building cover, and land cover type. All data has been processed uniformly to a grid scale of 300 meters. Functional area fire risk correlation dataset: including Stacking ensemble learning models based on the above features (base learners: LightGBM, XGBoost, RF, MLP; The urban functional zoning map of Gongyi City (10 categories in total) generated by the meta learner LR. At the same time, 703 historical fire events (including location, economic losses, and burned area attributes) from the 2021-2024 national fire statistics system were integrated and spatially correlated with functional zoning to form a "functional zone fire" matching data table that includes multidimensional indicators such as fire frequency, unit area loss, and burned area. Eight typical fire risk spatial patterns were identified, including the "high-frequency medium loss low area" type in residential areas and the "low-frequency high loss medium area" type in industrial areas. Data processing: The raw data undergoes preprocessing such as spatial registration, coordinate system one, kernel density analysis, and Z-score standardization. The overall accuracy of the functional area recognition model is 82%, with a Kappa coefficient of 0.79. Data value: This dataset has achieved high-precision automatic identification of urban functional areas and deep coupling analysis of fire risk patterns at the scale of Gongyi City for the first time. It can provide direct data support and case references for urban refined fire planning, risk zoning prevention and control, and subsequent related spatiotemporal data mining research.