Abstract
This study introduces a structured, multi-step workflow for harmonizing diverse soil water content (SWC) datasets to improve their meaningful comparability while preserving their intrinsic characteristics. The workflow addresses four components: unit, temporal, vertical, and lateral spatial harmonization, and is applied to point-scale (PS) sensors, cosmic-ray neutron sensing (CRNS), and remote sensing (RS) data from seven European sites. Statistical metrics and physically based plausibility checks using pedotransfer functions (PTFs) define realistic SWC ranges. A literature review of 60 studies identified five methodological classes: general statistical, spatial statistical, machine learning (ML), triple collocation (TC), and process-based models. For unit harmonization, total water capacity conversion outperformed mean-variance matching, improving the proportion of values within the field capacity (FC)-wilting point (WP) range. Temporal harmonization showed negligible differences between daily averaging and satellite overpass selection (<1% for CRNS, ~16% for PS for both methods). Depth harmonization using weighted averaging and exponential filtering enhanced dataset alignment (ΔR = 0.15, ΔubRMSD = -0.03). For lateral spatial, while simpler methods like areal mean are easy to implement with a single PS sensor, they often fail to preserve the distinct signals. The TC method provides a more robust, error-aware approach for data-scarce sites. ML methods like Random Forest (RF) can capture non-linearity but risk overfitting and are more applicable to data fusion. More spatially faithful method, ordinary kriging, provided the highest quantitative improvements (FC-WP plausibility +16% for CRNS, +30% for RS). Optimal harmonization is context-dependent, balancing data availability, interpretability, and physical consistency. The proposed framework enhances comparability and traceability across multi-sensor SWC studies.