Soil infiltration rates are underestimated by models in an urban watershed in central North Carolina, USA.

Stormwater management problems are expanding as urbanization continues and precipitation patterns are increasingly extreme. Urban soils are often more disturbed and compacted than non-urban soils, therefore, rainfall run-off estimates based on models designed for non-urban soils may not be accurate due to altered soil infiltration rates. Our objective was to quantify soil infiltration rates across an urban watershed and compare them to estimates from rainfall-runoff models commonly used in stormwater management (Horton and Green-Ampt) as well as an alternate, random-forest model created using available geospatial data. We measured infiltration rates and collected data on soil properties (texture, bulk density) and context (land use, ground cover, time since development) at 89 points across the 102 ha Walnut Creek watershed in Raleigh, North Carolina (USA). Forest land covers and forest ground covers (leaf litter) had the highest infiltration capacities; however, all of our measurements indicate that urban soils in the Walnut Creek watershed are able to absorb most precipitation events and are likely capable of infiltrating additional urban stormwater runoff. Comparisons between observations and the rainfall-runoff model estimates reveal that both underestimated urban soil infiltration rates. Despite higher than expected urban soil infiltration capacity, stormwater management remains a challenge in this urban watershed. Therefore, to reduce stormwater runoff from impervious surfaces through soil infiltration, impervious surfaces should be disconnected, especially adjacent to new development, and urban forests should be conserved. Further, because our random forest model more accurately captured watershed infiltration rates than the rainfall-runoff models, we propose this type of machine learning approach as an alternative method for informing stormwater management and prioritizing areas for impervious disconnection.