Cao et al. (2026) Differential impacts of soil hydrothermal properties on root and leaf phenology in cropland

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2026
• Date: 2026-04-10
• Authors: Mengying Cao, Pir Mohammad, M.A. Rahman, Qihao Weng
• DOI: 10.1016/j.jag.2026.105282

Research Groups

• JC STEM Lab of Earth Observations, Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
• Research Centre for Artificial Intelligence in Geomatics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
• Research Institute for Land and Space, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

Short Summary

This study developed a hybrid deep-learning model to extract root phenology from minirhizotron imagery and examined the differential impacts of soil hydrothermal properties on root and leaf phenology in a German cropland. It found that root growth initiates earlier and ceases later than leaf phenology, with root phenology being significantly more sensitive to soil moisture than temperature.

Objective

• To extract root phenology metrics directly from minirhizotron imagery using Convolutional Neural Network (CNN) architectures.
• To identify the influence of key edaphic factors (soil moisture, soil temperature) on root phenological patterns.
• To elucidate the connections and temporal decoupling between above-ground (leaf) and under-ground (root) phenological processes.

Study Configuration

• Spatial Scale: A cropland area in Selhausen, Germany (50°52′7.8″N, 6°26′59.7″E), within the Rur River catchment area.
• Temporal Scale: Data collected from 2015 to 2021, excluding 2019. Root images were captured weekly, and Sentinel-2 satellite imagery was used for leaf phenology.

Methodology and Data

• Models used:
  • Hybrid deep-learning model (U-Net for semantic segmentation and a regression network for feature extraction) for root phenology.
  • Physical Constraints Neural Networks (PCNNs) for leaf phenology prediction.
  • Structural Equation Modeling (SEM) to evaluate relationships between phenology and soil hydrothermal properties.
  • Generalized Linear Mixed-Effects Model (GLMM) to analyze the effects of soil moisture and temperature on leaf and root phenology.
• Data sources:
  • Minirhizotron imagery (RLT system) providing in-situ root architectural traits (e.g., total root length, branching frequency, root length density) from 2015 to 2021.
  • Sentinel-2 satellite imagery (10 meter spatial resolution, 5-day composite) to derive Enhanced Vegetation Index 2 (EVI2) for leaf phenology.
  • In-situ soil sensor data for soil moisture (water potential from MPS-2 series probes, T4 pressure transducer tensiometers) and soil temperature (−40 °C to 60 °C range, 0.1 °C resolution).
  • Meteorological datasets (relative humidity, vapor pressure, global radiation, wind speed, rainfall) from nearby TERENO weather stations.
  • Crop cultivation information (Triticum aestivum L. (wheat) and Zea mays L. (maize)).

Main Results

• The hybrid deep-learning model effectively predicted root phenology, achieving high accuracy (R² = 0.80–0.92; Mean Absolute Error (MAE) = 0.17–0.90 across years, with peak performance in 2021: R² = 0.95, MAE = 0.08).
• Root growth initiation occurred approximately 3 days earlier than leaf phenology (Start of Season, SOS).
• Root cessation was delayed by 25 days under dry soil conditions compared to leaf phenology (End of Season, EOS).
• Root phenology was significantly influenced by soil moisture (standardized path coefficient β = 0.53, p < 0.001), exhibiting greater sensitivity than soil temperature (β = 0.13, p < 0.001).
• Structural Equation Modeling revealed significant direct positive relationships of crop type (β = 0.59, p < 0.001) and soil moisture (β = 0.46, p < 0.001) with leaf phenology. For Root Length Density (RLD), crop type had a significant positive effect (β = 0.47, p < 0.001).
• Indirect associations showed that deeper soil moisture positively influenced both leaf (β = 0.13, p = 0.011) and root phenology (β = 0.06, p = 0.032), while higher soil temperature had a negative indirect impact on leaf (β = -0.08, p = 0.008) and root phenology (β = -0.14, p = 0.001).

Contributions

• Developed and validated an automated, scalable, and high-resolution machine-learning-based method for extracting root phenology metrics from minirhizotron imagery, overcoming limitations of manual segmentation.
• Provided quantitative evidence of the differential impacts of soil hydrothermal properties on root and leaf phenology in cropland, highlighting the greater sensitivity of roots to soil moisture.
• Revealed a significant temporal decoupling between above-ground and under-ground phenology, with root initiation preceding leaf emergence and root cessation extending beyond leaf senescence, particularly under dry conditions.
• Emphasized the critical need to distinguish between root and leaf phenological responses in biogeochemical models and to consider soil heterogeneity as a key factor in under-ground phenological processes.
• Enhanced the mechanistic understanding of how plant phenological strategies mediate ecosystem-scale biogeochemical cycles.

Funding

• Global STEM Professorship, Hong Kong SAR Government (P0039329)
• Hong Kong RGC GRF (15300923)
• Hong Kong Polytechnic University (P0046482 and P0038446)

Citation

@article{Cao2026Differential,
  author = {Cao, Mengying and Mohammad, Pir and Rahman, M.A. and Weng, Qihao},
  title = {Differential impacts of soil hydrothermal properties on root and leaf phenology in cropland},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2026},
  doi = {10.1016/j.jag.2026.105282},
  url = {https://doi.org/10.1016/j.jag.2026.105282}
}