The physical mechanism behind TED is rooted in classical thermoelasticity. As temperature variations at the surface diffuse downward through heat conduction, they produce depth-dependent thermal strain in soil and bedrock beneath GNSS stations. This strain is transferred elastically to the ground surface and appears directly in GNSS height time series. The concept traces back to foundational work by Biot in 1956 and Berger in 1975, but its global-scale consequences for satellite geodesy have only recently become fully appreciated. Previous studies have shown that shallow subsurface thermoelastic effects contribute to seasonal height variations at some GNSS sites, and millimetre-level nonseasonal thermoelastic displacements have been detected by high-precision geodetic observations in some regions.

What makes the new dataset distinctive is its treatment of the shallow subsurface as a layered medium rather than a homogeneous one. Earlier analytical models relied on idealized elastic half-space or uniform elastic-sphere formulations with effectively depth-invariant thermophysical and elastic properties. While computationally efficient, these approaches cannot represent the attenuation and phase lag of thermal signals caused by soil-bedrock stratification, nor account for station-to-station differences in shallow subsurface conditions. The team instead implemented a full-spectrum layered finite-element model (FEM_FL) that solves the transient heat-conduction problem and computes surface displacement induced by the resulting temperature-dependent strain within a consistent numerical framework for each station independently.

The model is driven by hourly ERA5 soil-temperature forcing from the ECMWF reanalysis, using the shallowest 0–7 cm soil-temperature layer to prescribe the near-surface upper thermal boundary. Depth-dependent subsurface properties are parameterized using SoilGrids 2.0, which provides global land soil-property maps at 250 m spatial resolution across six standard depth intervals from 0 to 200 cm. Below 2 m, the model represents the medium as a homogeneous bedrock background with prescribed thermal and elastic properties. This choice reflects the depth range of available global soil data rather than an assumption that thermoelastic effects vanish at that depth, and the SoilGrids-based layering captures the main shallow subsurface contrasts relevant to the annual thermal cycle.

A key technical challenge involves temporal sampling. GNSS coordinate time series are typically provided at daily resolution, whereas near-surface temperature and the resulting thermoelastic response depend strongly on local solar time. If temperature forcing is sampled at only one fixed epoch, stations at different longitudes correspond to different local times, and part of the diurnal temperature signal may alias into longer-period components. To reduce this effect, the team sampled the hourly ERA5 record at four equally spaced epochs per day (00:00, 06:00, 12:00, and 18:00 UTC), producing four TED realizations for each station. Each realization was decomposed into annual, semiannual, and residual components, with harmonic components averaged using amplitude-phase vectors to preserve the mean magnitude and timing of the seasonal signal.

The resulting dataset reveals clear spatial patterns in thermoelastic displacement. Annual TED amplitudes are mostly within about 0.2–1.2 mm, with locally larger values exceeding 1.5 mm in regions with strong seasonal temperature contrasts. Relatively large signals are concentrated in mid-latitude continental areas, including parts of North America, Europe, East Asia, and Australia, whereas humid maritime regions and low-latitude areas generally show weaker annual responses. The annual phase shows a clear hemispheric contrast, consistent with the opposite seasonal heating cycles in the Northern and Southern Hemispheres. The percentile-based peak-to-peak variability of the full daily TED series commonly reaches 2–4 mm in continental regions and locally approaches 5–6 mm, while nonseasonal variability after removing the linear term and annual and semiannual harmonics is mostly below 1 mm but can locally reach millimetre level.

Technical evaluation of the dataset focused on three aspects: the influence of the SoilGrids-based subsurface parameterization, sensitivity to key thermal and elastic parameters, and spectral behaviour of the modelled time series. For the 2,612 stations with documented foundation-depth information, site-information-constrained solutions agree closely with SoilGrids-based solutions, with annual-amplitude differences mostly about 0.01–0.03 mm and annual-phase differences generally within 1–3 degrees. Sensitivity tests using ±10 percent perturbations in the linear thermal expansion coefficient, thermal diffusivity, and Young’s modulus indicate that the amplitude and phase of the annual TED component are stable under reasonable parameter uncertainty, with the largest sensitivity occurring mainly in dry inland regions where stronger seasonal temperature forcing also produces larger TED signals.

Spectral analysis of the TED series shows a dominant annual peak at 1 cycle per year and a weaker semiannual peak near 2 cycles per year in all latitude bands, consistent with the temperature-driven origin of the modelled displacement. Spectral power is also present at higher frequencies, indicating that the released series retain nonseasonal variability rather than representing only fitted annual and semiannual terms. The annual peak is stronger in mid- and high-latitude bands than in low-latitude bands, reflecting the stronger seasonal temperature cycle outside the tropics. Regional spectra show similar features, with North America, Europe, and East Asia generally showing stronger annual and semiannual peaks, whereas Australia and some lower-latitude or maritime regions show weaker seasonal power.

When applied alone to GNSS height time series, the TED correction produces positive weighted root-mean-square (WRMS) reductions at about 79 percent of the assessed stations, with most improvements within 3–5 percent and the largest reductions reaching about 10 percent. Some stations show negative WRMS reductions, locally around −2 to −5 percent, indicating that the TED-only correction does not improve every GNSS height series. This behaviour is expected because TED is only one part of the non-tectonic vertical displacement observed by GNSS, and its amplitude is usually smaller than hydrological loading and many local site-dependent signals. Even so, the spatial pattern suggests that TED explains part of the GNSS height variability, especially in continental regions with stronger seasonal temperature forcing.

The released product represents subsurface thermoelastic deformation only and does not include thermal expansion of above-ground structures such as concrete pillars, steel masts, rooftop monuments, or antenna-support structures. These monument-related effects can be important at individual stations and are not included because globally complete information on monument height, material, geometry, anchoring condition, and local thermal environment is not available for most GNSS stations. The TED series should therefore be interpreted as a subsurface thermoelastic correction rather than a complete correction for all temperature-related site effects. The dataset is publicly available on Zenodo and through the Ground Motion Data Service platform, with files organized by station and including date tags, vertical TED displacements in millimetres, station coordinates, record start and end dates, and dataset version information. Future updates may address snow, soil moisture, and frozen ground processes, and may consider sub-daily products after further validation with appropriate GNSS solutions.

Subject of Research: Temperature-driven thermoelastic displacement of the shallow subsurface and its effect on GNSS station height measurements

Article Title: TED: a global temperature-driven thermoelastic displacement dataset for GNSS reference stations (2000–2023)

Article References: Lu, R., Li, Z., Ye, L., Yuan, P., & Feng, Y. (2026). TED: a global temperature-driven thermoelastic displacement dataset for GNSS reference stations (2000–2023). Earth System Science Data, 18(10), 7575-7591. https://doi.org/10.5194/essd-18-7575-2026

Image Credits: AI Generated

DOI: 10.5194/essd-18-7575-2026

Keywords: GNSS, thermoelastic displacement, TED dataset, ERA5, SoilGrids, finite-element model, vertical land motion, reference frame, geodesy, soil temperature, height time series, crustal deformation

The Rila Mountains, the highest range in Southeast Europe, present a challenging environment for soil modeling due to their significant altitudinal variation and diverse climatic conditions. The study area covers approximately 263,000 hectares, with elevations ranging from below 1,000 meters to the peak of Musala at 2,925 meters. This vertical gradient creates distinct vegetation zones, from oak woodlands in the foothills to alpine meadows at the summit, each influencing soil formation processes differently. The researchers utilized this natural laboratory to test the suitability of SoilGrids for ecosystem accounting, a framework that integrates environmental data with economic analysis to inform sustainable policy decisions. By focusing on a region with limited local soil data, the study highlights the potential of global models to fill critical information gaps in national monitoring programs.

The methodological approach involved comparing SoilGrids predictions with a robust set of independent point observations collected from three distinct sources. These sources included field data from the International Co-operative Programme on Monitoring and Assessment of Air Pollution Effects on Forests, known as ICP-Forest, published scientific data, and new observations conducted specifically for this study. The team analyzed 39 sample plots distributed across different altitudes and soil types, primarily Cambisols, Umbrisols, and Luvisols. To ensure a fair comparison, the researchers focused on the top 30 centimeters of soil, a standard depth for many ecosystem assessments. They evaluated six key parameters: pH, bulk density, coarse fraction, soil organic carbon content, nitrogen content, and soil organic carbon stock. This multi-source validation strategy helps mitigate the biases that might arise from relying on a single dataset.

The results indicate that SoilGrids performs well in capturing the general dynamics of soil parameters across the altitudinal range. There was a strong agreement between the observed and predicted average values for bulk density, coarse fraction, and soil organic carbon content. Specifically, the average measured soil organic carbon stock in the top 30 centimeters was 58.54 tonnes per hectare, while the average predicted value was 55.38 tonnes per hectare. Statistical tests, including paired t-tests and Wilcoxon signed-rank tests, confirmed that there were no significant differences between the field data and model predictions for these parameters. This suggests that SoilGrids can reliably estimate the carbon stored in soils, a crucial metric for understanding the global carbon cycle and assessing the capacity of ecosystems to sequester carbon.

However, the study also revealed notable discrepancies in the prediction of nitrogen content and soil pH. The model predicted nitrogen levels that were almost twice as high as those observed in the field. Similarly, the SoilGrids data indicated soils that were less acidic than those measured in the field, with predicted pH values mostly between 5.7 and 6.6, whereas field observations reported values between 4.5 and 5.8. These differences are statistically significant and suggest that the global model may not fully capture the specific biogeochemical processes occurring in mountainous forest soils. The researchers noted that the SoilGrids model is trained on a dataset dominated by agricultural land and older sampling dates, which may not accurately represent the unique conditions of high-altitude coniferous forests where organic matter decomposition and nutrient cycling differ significantly from agricultural systems.

The discrepancies in pH and nitrogen are likely attributable to the input data used to train the SoilGrids models. The global database contains approximately 240,000 soil profiles, but more than 60% of these were collected between 1960 and 2000. In regions like the Rila Mountains, where soil acidity is driven by the weathering of carbonate-free silicate rocks and the influence of coniferous vegetation, these historical biases can lead to underestimations of acidity. The study highlights that while the model captures broad spatial trends, such as the increase in soil organic carbon with altitude, it may lack the precision required for detailed local assessments of specific chemical properties. This finding underscores the importance of validating global models with local data before using them for high-stakes policy decisions, such as reporting on greenhouse gas emissions under the United Nations Framework Convention on Climate Change.

Despite these limitations, the study concludes that SoilGrids remains a valuable tool for ecosystem reporting, particularly when local data is scarce. The model provides a spatially explicit representation of soil properties that can be integrated into ecosystem accounting frameworks, such as the System of Environmental-Economic Accounting for Ecosystems. By combining SoilGrids data with local knowledge of climate, vegetation, and soil types, researchers can improve the accuracy of their assessments. The study recommends using stratification methods to group soil measurements by climate region and soil type, ensuring that each stratum has sufficient data to minimize sampling error. This approach allows for a more nuanced understanding of soil variability and helps in deriving region-specific emission factors that are essential for accurate national greenhouse gas inventories.

The implications of this research extend beyond Bulgaria, offering insights for other regions with similar ecological characteristics. As countries strive to meet their climate commitments and monitor biodiversity loss, the need for reliable soil data becomes increasingly urgent. SoilGrids offers a cost-effective and accessible solution for generating this data, but its application must be accompanied by careful validation and contextualization. The study demonstrates that while global models can provide a solid foundation for ecosystem assessments, they are not a substitute for local field measurements. By understanding the strengths and weaknesses of these tools, scientists and policymakers can make more informed decisions about land management, conservation, and climate mitigation strategies. The Rila Mountains study serves as a case study for the broader challenge of integrating global data with local realities to achieve a comprehensive understanding of our planet’s soil resources.

Subject of Research: Comparative assessment of SoilGrids system with regional soil data for advancing ecosystem reporting in Bulgaria

Article Title: Comparative assessment of SoilGrids system with regional soil data for advancing ecosystem reporting in Bulgaria

Article References: Stoeva, L., & Zhiyanski, M. (2024). Comparative assessment of SoilGrids system with regional soil data for advancing ecosystem reporting in Bulgaria. One Ecosystem, 9, Article e133091. https://doi.org/10.3897/oneeco.9.e133091

Image Credits: AI Generated

DOI: 10.3897/oneeco.9.e133091

Keywords: Comparative, assessment, SoilGrids, system, regional, soil, data, advancing, ecosystem, reporting, Bulgaria, scientific research

The research, conducted by Peter Coaguila-Rodriguez and Alberto Franco Cerna-Cueva of the Universidad Nacional Agraria de la Selva in Tingo María, Huánuco, set out to answer a deceptively simple question: can ordinary soil chemistry measurements predict which samples will show aluminium saturation at or above 20 percent, a threshold commonly associated with high or very high toxicity risk? To do so, the team turned to an anonymized institutional soil-monitoring database containing 1,842 records from cacao-growing areas of the central Peruvian Amazon. After filtering for acidity, 1,539 samples with pH below 5.5 formed the analytical cohort, a substantial dataset for a region where such comprehensive soil records are rare.

Aluminium saturation is defined as the proportion of the soil’s exchangeable cation exchange capacity occupied by aluminium rather than by base cations such as calcium, magnesium and potassium. As soils acidify, base cations are leached away and aluminium increasingly dominates the exchange complex, which is why the metric serves as a direct chemical indicator of the intensity of acid-soil stress a plant will experience. Cacao, a crop native to the upper Amazon and economically vital to Peru, is known from prior ecophysiological work to be sensitive to soil acidity, with juvenile plants showing impaired growth and altered nutrition under acidic conditions. Yet the full exchangeable-acidity analysis needed to compute saturation requires titration procedures that many regional laboratories do not routinely offer, whereas pH, organic matter and exchangeable bases are standard fare.

The researchers built six logistic-regression models, comparing predictors drawn from routine soil tests against covariates derived from two publicly available gridded data products: PISCOp, Peru’s high-resolution interpolated rainfall dataset, and SoilGrids, a global digital soil mapping product. The outcome variable in every case was the laboratory-reported aluminium saturation value, dichotomized at the 20 percent threshold. Logistic regression, a workhorse of applied statistics, estimates the probability of a binary outcome as a function of predictor variables, making it well suited to a screening task where the goal is to flag samples for follow-up rather than to measure toxicity directly.

What distinguishes the study methodologically is the rigor of its validation design. All preprocessing steps and the selection of the decision threshold were nested inside a fivefold grouped cross-validation scheme, meaning that the data transformations and cutoff choices were re-learned within each training fold rather than tuned on the full dataset. The grouping was based on surrogate environmental signatures, a strategy designed to prevent information leakage between samples drawn from similar environments, a well-known pitfall in spatially structured ecological and soil data. This kind of careful cross-validation is essential when records may cluster by farm, soil type or microregion, because otherwise a model can appear far more accurate than it truly is when deployed on genuinely new locations.

The performance of the simplest model, built entirely from routine soil test variables, was striking. Across pooled cross-validation folds it achieved a balanced accuracy of 0.876, with a fold-level standard deviation of 0.022 and a group-bootstrap 95 percent confidence interval of 0.856 to 0.896. Balanced accuracy, the average of sensitivity and specificity, is a robust metric when the two outcome classes are unevenly represented. The model’s sensitivity, its ability to correctly flag samples that truly exceed the 20 percent threshold, was 0.868, while its specificity, the ability to correctly clear samples below the threshold, was 0.885. Perhaps most importantly for a screening application, the positive predictive value reached 0.952, meaning that when the model flags a sample as high-risk, that flag is very likely to be confirmed by the full laboratory determination.

Additional metrics reinforced the picture of a well-calibrated classifier. The area under the receiver-operating-characteristic curve, which summarizes discrimination across all possible thresholds, was 0.936, and the area under the precision-recall curve, often more informative when positive cases are the minority, was 0.968. The Brier score, a measure combining discrimination and calibration that penalizes both wrong predictions and misplaced confidence, came in at 0.082, with lower values indicating better overall probabilistic accuracy. Together these figures indicate that routine soil chemistry carries a strong, quantifiable signal about aluminium saturation, enough to triage samples with a high degree of confidence before committing resources to confirmatory analysis.

Just as revealing is what the study found when it added the gridded environmental covariates. Relative to the routine model, the difference in balanced accuracy was 0.000, with a 95 percent confidence interval of −0.010 to 0.010, when PISCOp rainfall covariates were included, and −0.007, with an interval of −0.020 to 0.005, when PISCOp and SoilGrids were combined. In other words, the authors found no evidence of a stable improvement from the remotely sensed and interpolated data layers within the spatial support available. This is a noteworthy result in a field where digital soil mapping and machine learning covariates are frequently promoted as enhancements to local prediction, and it suggests that for this specific screening task, at this spatial resolution, the chemistry already measured in routine tests contains most of the relevant information.

The authors are careful to delineate what the model can and cannot do, and these caveats matter for anyone hoping to apply it. The tool is intended to prioritize confirmatory aluminium-saturation analysis in comparable acidic cacao soils, not to measure plant toxicity directly, not to replace laboratory diagnosis and not to support continuous zoning or mapping of unsampled areas. Aluminium saturation in a soil sample is a chemical property, not a biological endpoint, and actual toxicity to a given cacao genotype depends on root architecture, cultivar-specific tolerance mechanisms and management history. The model’s value lies in triage: laboratories and extension services with limited budgets can use routine test results to decide which samples genuinely need the more elaborate exchangeable-acidity workup, concentrating scarce analytical capacity where it is most likely to change management decisions.

The practical implications extend across the tropical cacao belt, where acid soils cover vast areas and liming decisions hinge on knowing where aluminium stress is severe. Acid soils are estimated to occupy a large share of the world’s potentially arable land, and aluminium toxicity is among the principal chemical constraints on crop production in these regions. A screening approach that leverages data already flowing through regional monitoring programs could accelerate the identification of high-risk fields without new instrumentation or new sampling campaigns. The study’s data and code are available from the corresponding authors upon reasonable request, subject to the confidentiality restrictions of the institutional monitoring database, and the work received no external funding. For the farmers of the central Peruvian Amazon, and potentially for cacao producers far beyond it, the message is quietly transformative: the answers to one of tropical agriculture’s most stubborn soil problems may already be sitting in the routine test reports that laboratories produce every day.

Subject of Research: Predictive screening of high aluminium saturation in acidic cacao soils of the Peruvian Amazon using routine soil test data and logistic regression

Article Title: Routine soil tests support screening of high aluminium saturation in acidic cacao soils

Article References: Coaguila-Rodriguez, P., & Cerna-Cueva, A. F. (2026). Routine soil tests support screening of high aluminium saturation in acidic cacao soils. Environmental Monitoring and Assessment, 198(10), Article 1105. https://doi.org/10.1007/s10661-026-15945-3

Image Credits: AI Generated

DOI: 10.1007/s10661-026-15945-3

Keywords: soil acidity, aluminium saturation, Theobroma cacao, Peruvian Amazon, logistic regression, soil screening, cross-validation, digital soil mapping, SoilGrids, PISCOp, environmental monitoring, soil chemistry