Evaluating stream health based environmental justice model performance at different spatial scales

- Effects of spatial resolution and dependency on environmental justice were studied.
- Stream health models were developed using Bayesian Conditional Autoregressive.
- No spatial dependency was found at the county level.
- Considering level interactions improved models' prediction.
- Spatial variations were captured at the block group and census tract.

SummaryThis study evaluated the effects of spatial resolution on environmental justice analysis concerning stream health. The Saginaw River Basin in Michigan was selected since it is an area of concern in the Great Lakes basin. Three Bayesian Conditional Autoregressive (CAR) models (ordinary regression, weighted regression and spatial) were developed for each stream health measure based on 17 socioeconomic and physiographical variables at three census levels. For all stream health measures, spatial models had better performance compared to the two non-spatial ones at the census tract and block group levels. Meanwhile no spatial dependency was found at the county level. Multilevel Bayesian CAR models were also developed to understand the spatial dependency at the three levels. Results showed that considering level interactions improved models' prediction. Residual plots also showed that models developed at the block group and census tract (in contrary to county level models) are able to capture spatial variations.