Multimodal Sensor Dataset for US Life Expectancy Predictors (2000–2019)

This dataset supports the research framework for "Beyond Temperature: Atmospheric Moisture Dominates Environmental Predictors of US Life Expectancy in a 20-Year Multimodal Framework." It contains 61,540 US county-year observations spanning 2000 to 2019. To create this dataset, we engineered a robust geospatial data pipeline to harmonize severe spatial and temporal resolution differences across multiple APIs. The dataset integrates an 8-modality satellite and environmental panel extracted via Google Earth Engine, paired with local public health metrics. The integrated modalities include: Satellite Imagery: High-resolution remote sensing via MODIS, Sentinel, and Landsat. Atmospheric & Climate Exposures: Including critical drivers like wet-bulb temperature and formaldehyde. Land Cover & Ecology: USGS National Land Cover Database (NLCD) and regional tree canopy metrics. Agricultural Metrics: Sourced from the FAO Gridded Livestock of the World. Soil Health Indicators. Socioeconomic Demographics: Derived from the US Census. Geospatial Boundaries: US Census TIGER/Line geographic structures. Public Health Outcomes: Regional life expectancy and mortality records. The dataset is fully pre-processed, featuring localized intra-county missing-data imputation to preserve extreme local spatial fidelity. It is engineered specifically for training tree-based machine learning pipelines (LightGBM, XGBoost, Random Forest) and conducting rigorous TreeSHAP interventional attribution to evaluate the environmental drivers of human longevity. Official Code Repository: https://github.com/albertfaiz/Multimodal-Sensor-xAI