MAPPING RISK ASSOCIATED WITH HIGH WATER TABLES IN LOW-LYING COASTAL REGIONS

Septic systems in coastal areas are at risk of impairment due to rising groundwater tables driven by the ongoing changes in climate that increase precipitation events and sea levels. Understanding groundwater dynamics in this area is critical to estimating the threats to the septic systems. We established the long-term risk of impairment of septic systems using machine learning models calibrated to a 13-month record of water table elevations from twelve individual wells in our study area in Beaufort County, South Carolina. Observed head showed an average of 1 m rise in water levels between August and September 2022, reflecting rainfall intensity of about 3 inches and a regional sea level rise of about 7 cm due to tropical storm activity. The water table generally returned to what we termed the “baseline” water level between storms. Our machine learning models confirmed annual variability of 1 m in water levels over 34 years (1990-2023) using historical rainfall and tidal records. Tidal records were adjusted to 2023 mean sea level to estimate current risk, while also avoiding running the model beyond the conditions for which it was calibrated. When different rainfall scenarios were tested, there was a modest increase in the risk at a centimeter-to-millimeter scale which only impacted 17% of our wells. Overall, water tables were high enough to cause impairment of septic systems at least three times a year in 83 % of our wells, and risk increased as the depth to the baseline water table decreased. The relationship between risk and depth to baseline presents an opportunity to describe risk spatially instead of via intensive traditional groundwater modeling, if the baseline water table can be mapped spatially. We initially used an analytical solution to resolve the baseline. The analytical solution fit the individual wells very well, but the water table interacted with the land surface when the analytical solution was transferred to GIS platform, preventing its use. Alternate methods can be used to estimate the baseline water table within a GIS database platform, but we are testing the steady-state MODFLOW to estimate the new baseline. The machine learning model will also be tested against a fully transient MODFLOW model to understand the differences between the two approaches.