Lateral Connectivity as a Dominant Predictor of Groundwater Dynamics in an Agricultural Watershed of Eastern North Carolina

Abstract

This study investigates hydrologic interactions at an agricultural field site in Camden County, within the Pasquotank River Basin of eastern North Carolina. Using groundwater level, precipitation, soil water content, and surface water level data, we first applied a soil moisture metric to assess how site-specific soils influence water retention. Soil moisture analysis shows that a predominantly sandy shallow profile containing a thin finer-textured (clay loam) layer, combined with shallow groundwater levels, results in increased water retention and prolonged soil saturation. We then combined cross-correlation and machine learning approaches across multiple temporal scales to identify the dominant predictors of groundwater fluctuations at the site, assessing differences between maximum and minimum values across datasets. For maximum values, limited infiltration and near-surface saturation suppressed the rainfall signal within soil and groundwater, producing signal masking that reduced the apparent influence of precipitation. In contrast, results from the analysis of minimum values exhibited stronger seasonal associations, with a clearer signal. However, for both maximum and minimum conditions, surface water level maintained strong correlation and predictive strength with groundwater levels, reflecting lateral surface groundwater connectivity. Results further reveal that surface water level from one to two days prior exhibited the strongest association with groundwater level fluctuations and, when used in input-limited models along with precipitation, maintained strong predictive performance. Both cross-correlation and model-based feature importance analyses confirm that surface water level is the dominant predictor of groundwater fluctuations at the Camden County site, where vertical infiltration is limited and rainfall responses are muted due to near-constant soil saturation. The ability to predict groundwater levels using limited inputs demonstrates the potential for simplified, resource-efficient monitoring strategies in data-scarce agricultural settings. Such approaches offer scalable, cost-effective tools for adaptive water management in vulnerable coastal landscapes undergoing environmental change.