A Domain Adaptation Approach for Performance Estimation of Spatial Predictions

Gaussian Markov random fields improve ensemble predictions of daily 1 km PM2.5 and PM10 across France

“…The clustering of monitors near cities means our model is roughly weighted by population density, which may or may not be appropriate depending on the intended use for the predictions (Sarafian et al, 2020). New low-cost PM sensors might complement the existing monitoring network, particularly since our model's weaker spatial performance suggests that a few PM observations at new locations might be more useful than a long timeseries of measurements at a single location.…”