Gong et al. (2025) A changepoint approach to automated estimation of soil moisture drydown parameters from time series data

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-12-02
• Authors: Mengyi Gong, Jessica Davies, Rebecca Killick, Christopher Nemeth, Shangshi Liu, John Quinton
• DOI: 10.1038/s41598-025-27067-w

Research Groups

• School of Mathematical Sciences, Lancaster University, Lancaster, UK
• Lancaster Environment Centre, Lancaster University, Lancaster, UK
• Department of Earth and Environmental Sciences, The University of Manchester, Manchester, UK
• Yale School of the Environment & Yale Center for Natural Carbon Capture, Yale University, New Haven, CT, USA

Short Summary

This study introduces an automated, changepoint-based method to analyze in-situ soil moisture time series, autonomously detecting wetting events and estimating drydown parameters. The method successfully extracts physically interpretable information, demonstrating that these parameters correlate with climatic regimes and soil texture across diverse field sites.

Objective

• To develop and apply an automated, changepoint-based method for analyzing in-situ soil moisture time series data to autonomously detect wetting events and dynamically estimate parameters describing soil moisture drydown characteristics.
• To explore if the extracted drydown characteristics and travel times correspond with soil properties and climatic regimes.
• To provide a proof-of-concept for autonomously extracting insights relevant to soils and climate from dynamic soil moisture sensor data.

Study Configuration

• Spatial Scale: Nine field sites across the United States from the National Ecological Observation Network (NEON), covering a wide range of soil, climate, and ecosystem types.
• Temporal Scale: At least 12 months (minimum 10 months for sites with winter freeze gaps) of continuous soil moisture data, collected as 30-minute averages and down-sampled to hourly time series. Study periods varied by site, generally from 2018-2022.

Methodology and Data

• Models used:
  • Changepoint detection method to identify peaks in soil moisture time series.
  • Exponential decay model for soil moisture drydown: $\theta(t) = \Delta\theta \exp \left( -\frac{t}{\omega} \right) + \thetaf$, re-parameterized as $\thetat = \alpha{0i} + \alpha{1i} \lambda^{(t-\taui)}i + \epsilon_t$.
  • Penalised exact linear time method for simultaneous estimation of changepoints and model parameters.
• Data sources:
  • In-situ continuous soil moisture (volumetric water content) time series from Sentek TriSCAN soil capacitance probes at NEON field sites (https://data.neonscience.org/data-products/DP1.00094.001).
  • Climate data (mean annual precipitation, mean annual temperature) and soil properties (soil type, sand, silt, clay content, vegetation type) for each NEON site.

Main Results

• The changepoint method successfully detected wetting events and estimated distributions of drydown parameters (asymptotic soil moisture level $\alpha_0$ and decay rate $\lambda$) for nine NEON sites.
• Distributions of $\alpha_0$ and $\lambda$ varied significantly across sites, and these features were associated with climatic regimes and soil texture.
• A strong negative correlation was found between the median and quartiles of $\alpha_0$ and mean annual temperature (MAT).
• A moderate negative correlation was observed between the quantiles of $\lambda$ and mean annual precipitation (MAP), indicating that wetter sites tend to have faster soil moisture decay rates.
• Weak negative correlations were found between sand content and $\lambda$, and weak positive correlations between silt/clay content and $\lambda$, consistent with faster drainage in coarser textured soils.
• Multi-depth analysis of travel times (infiltration speed) between soil layers showed that travel times were generally longer in deeper layers. TALL and OSBS (sandy soils, high MAP) exhibited shorter travel times compared to SRER (sandy soil, low MAP).

Contributions

• Developed and validated an automated, changepoint-based method for robustly and efficiently extracting physically interpretable parameters from high-frequency in-situ soil moisture time series.
• Demonstrated the method's ability to dynamically estimate event-specific soil moisture drydown characteristics (asymptotic level and decay rate) and travel times between soil depths.
• Provided a proof-of-concept that these autonomously extracted parameters offer valuable insights into the influence of soil properties and climatic regimes on soil moisture dynamics.
• Addressed the challenge of processing large volumes of continuous soil sensing data, paving the way for more ubiquitous, cost-effective, and real-time soil monitoring and diagnostics.
• Discussed the potential for using shifts in these drydown characteristics over time to identify changes in soil properties and processes, such as soil degradation or health improvements.

Funding

• UKRI-funded project Signals in the Soil (Grant No. NE/T012307/1)

Citation

@article{Gong2025changepoint,
  author = {Gong, Mengyi and Davies, Jessica and Killick, Rebecca and Nemeth, Christopher and Liu, Shangshi and Quinton, John},
  title = {A changepoint approach to automated estimation of soil moisture drydown parameters from time series data},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-27067-w},
  url = {https://doi.org/10.1038/s41598-025-27067-w}
}