From Pixel to Processes in Soil Erosion Research: Advancing 4D Analysis for Spatiotemporal Process Disaggregation and Model Evaluation

Current soil erosion models are constrained by parameter uncertainty, limited empirical data, and an inability to capture the complex spatiotemporal dynamics of surface processes. This study introduces a novel image-based approach using 3D photogrammetric time series to detect and quantify both erosive and non-erosive soil processes.

Using Structure from Motion (SfM) and high-frequency time-lapse photogrammetry, we generate dense 3D surface models of bounded experimental plots during controlled rainfall simulations. These multi-site datasets capture surface evolution in detail, enabling analysis of processes such as soil compaction, aggregate breakdown, sheet erosion, and rill formation.

To address early-stage surface subsidence (e.g., due to compaction) that masks initial sediment yield measurement, we develop a correction method. Based on soil and plot properties, non-linear regression is used to derive s-shaped correction functions that estimate and adjust for these masking effects.

A hierarchical, time series-based change detection strategy supports automated process classification by analyzing full surface model sequences, moving beyond bitemporal comparisons. This enables the temporal segmentation of overlapping erosional phases, enhancing interpretation of subtle and high-magnitude changes.

The datasets also support the calibration of a runoff-driven soil erosion model using microtopographic inputs. Multiple spatiotemporal averaging techniques are assessed to identify robust calibration metrics. Results show no single metric captures all aspects of erosional behavior, highlighting the need for multi-metric evaluation.

By integrating time-lapse imaging, 4D analysis, and advanced model evaluation, a comprehensive pathway from pixel-level measurements to supporting process-level understanding is introduced to improve soil erosion modeling under dynamic environmental conditions.