What question did this study set out to answer?

This study aims to investigate the relationship between air pollution and cancer prevalence across Texas counties using satellite data and machine learning techniques.

March 7, 2026Open Access

Spatial analysis of air pollution and cancer prevalence in Texas: a machine learning approach using satellite remote sensing data

Key Points

This study aims to investigate the relationship between air pollution and cancer prevalence across Texas counties using satellite data and machine learning techniques.
Combined satellite measurements of eight air pollutants with cancer prevalence data
Used machine learning algorithms to analyze and model the data
Conducted spatial clustering analysis to identify regional pollution patterns
Performed ANOVA to assess the impact of air pollutants on model fit
High spatial autocorrelation indicates cancer prevalence is geographically clustered
SVM algorithm achieved the best predictive performance with RMSE = 0.347 and R² = 0.791
SO₂ was identified as the most influential predictor of cancer prevalence, followed by AER AI and CH₄

Abstract

Cancer prevalence in the world has been attributed to exposure to air pollutants. However, spatial analyses utilizing remote sensing data have been limited. This study combined satellite-derived measurements of eight air pollutants (SO2, NO2, CO, HCHO, CH4, PM1, PM10, and AER AI) with machine learning algorithms to analyze the relationship between air pollution and cancer prevalence across all 254 Texas counties in 2022. Satellite-based air pollution data were obtained from Sentinel-5 Precursor and the Copernicus Atmosphere Monitoring Service, while cancer prevalence data were sourced from the U.S. Centers for Disease Control and Prevention. The spatial clustering analysis revealed distinct regional patterns. The West Texas counties exhibited lower pollution except for elevated AER AI, the East Texas counties showed uniformly high pollutant concentrations, and the Central Texas counties displayed moderate levels. There was high spatial autocorrelation (Moran's I = 0.69), suggesting cancer prevalence was geographically clustered. Among the four machine learning algorithms tested (Random Forest, Support Vector Machine, Multi-Layer Perceptron, and Ordinary Least Squares), SVM showed the best performance (RMSE = 0.347, R2 = 0.791). ANOVA analysis confirmed that air pollutant variables significantly improved model fit (p 2 as the most influential environmental predictor, followed by AER AI, CH4, and particulate matter, each contributing 30-60% of the predictive weight relative to the top socio-demographic predictor (liquor consumption). These findings reveal significant associations between air pollution and cancer prevalence in Texas, with important implications for targeted public health interventions in high-risk regions. This study shows the value of integrating remote sensing data with machine learning algorithms for regional environmental health assessment where ground-based monitoring is unavailable.

Bookmark

View Full Paper