Vadose Assess
California Subsurface Hazard Surface
Every tracked cleanup case in California, spread onto a ~500 m grid by a transport kernel: load decays slowly downgradient, sharply upgradient, and as a Gaussian across flow, with the flow direction fitted locally from measured groundwater elevations. Where no gradient can be fitted, the contribution falls back to an isotropic decay at reduced weight, so unconstrained basins read as less certain rather than as confidently symmetric.
What this is worth
Scored against detections at 10,802 California public supply wells, held out by whole county, the surface reaches roughly 0.69 AUC. Counting cases within 3 km of a well — no transport model at all — reaches 0.654, so the kernel is a real but modest correction on proximity rather than the main event. Supply wells are also sited to avoid known contamination, which makes that figure a lower bound of unknown tightness.
The score ranks locations against each other. It has not been calibrated, so a cell twice as dark is not twice as likely to be impacted, and no value here should be read as a percent chance of contamination. It screens; it does not replace a site investigation.