Environmental and Earth Sciences

Abstract: CONTEXT Crop yield forecasting is essential for agricultural planning, food security assessment, and climate change adaptation. In Poland and other regions of Central Europe, interannual variability in early-season meteorological conditions is a key driver of yield variability, yet its systematic use in regional-scale forecasting models remains limited. OBJECTIVE This study aims to quantify the influence of early-season meteorological anomalies on yields of nine key field crops in Poland and to assess the potential for generating yield forecasts several months before harvest using satellite and reanalysis data. METHODS Linear regression models were developed using anomalies of 2-metre temperature, precipitation, reference evapotranspiration and root-zone soil moisture derived from ERA5 reanalysis and EUMETSAT H-SAF and LSA-SAF satellite products. Models were trained separately for each crop and each of the 16 Polish voivodeships using data from 2004 to 2022. A rolling time-based cross-validation scheme was applied, complemented by an independent hold-out validation for 2019–2022. RESULTS AND CONCLUSIONS Positive temperature anomalies during January–April were consistently associated with higher yields across all winter cereals, while soil moisture deficits in the upper root zone (SM1 and SM2 layers) represented the primary limiting factor. Reference evapotranspiration anomalies in May–June provided a proxy signal for radiation and atmospheric demand conditions. Preliminary forecasts were feasible up to four months before harvest for winter cereals and up to three months for spring cereals. Independent validation for the Opolskie voivodeship yielded Pearson correlation coefficients of 0.6–0.8 for most crops, with maize showing lower agreement (r ≈ 0.45). SIGNIFICANCE The proposed anomaly-based framework provides interpretable, timely and operationally relevant yield forecasts for regional agricultural planning under increasing climate variability. Its transparency and reliance on freely available satellite and reanalysis products make it suitable for implementation in early-warning systems across Central Europe.

Abstract: Flooding is the most recurrent and economically devastating natural hazard in Nigeria, yet no standard or consistent nationwide assessment method exists. Moreover, spatially explicit flood susceptibility information remains scarce—a critical gap for a country of over 220 million people with severely limited hydrometric monitoring infrastructure. This study presents one of the first nationwide, multi-model flood susceptibility mapping efforts for Nigeria, integrating four complementary approaches: Frequency Ratio (FR), Logistic Regression (LR), Random Forest (RF), and Gradient Boosting (XGBoost). The framework incorporates a bivariate FR component and Height Above Nearest Drainage (HAND) as a conditioning factor. Six conditioning factors were initially evaluated—elevation, TWI, HAND, LULC, slope, and soil type—with elevation, TWI, HAND, and LULC retained for final model development. The flood inventory was derived from a HEC-RAS 100-year floodplain simulation driven by satellite-derived discharge records from the Dartmouth Flood Observatory (DFO), yielding 973,111 binary flood/non-flood observations used as training labels for the LR, RF, and XGBoost models and as a flood-pixel count reference for FR computation. The HEC-RAS floodplain was independently verified against documented DFO historical flood reports. A three-tier accuracy assessment was conducted. For statistical accuracy using a 20% test subset, XGBoost achieved the highest AUC (0.956) and Overall Accuracy (0.892). For spatial consistency against the HEC-RAS reference, LR, RF, and XGBoost achieved substantial agreement (Kappa = 0.662–0.700), while FR achieved moderate agreement (Kappa = 0.410). Validation against the 2022 Sentinel-1 SAR flood extent showed that all four models exceeded HEC-RAS flood detection accuracy. For operational flood risk management, XGBoost is recommended due to its strong predictive performance and ability to minimize missed flood-prone areas. The resulting maps provide actionable spatial intelligence for disaster risk management, land-use planning, and early warning systems across Nigeria and other data-sparse regions.

Abstract: Remote sensing change detection is often degraded by noise-induced uncertainty, particularly in unsupervised scenarios where the absence of labeled information limits the ability to distinguish real land-cover variations from stochastic disturbances. Existing methods mainly rely on pixel-level spectral differences or clustering strategies, which are vulnerable to noise amplification and unstable decision boundaries. To address this issue, this study proposes a noise-robust unsupervised change detection framework by integrating deep image restoration with posterior probability space modeling. A channel–spatial attention aggregation network (CAANet) is first developed to recover discriminative structural information from noise-contaminated remote sensing images. Instead of directly performing change analysis in the original feature space, the restored images are transformed into posterior probability representations through fuzzy clustering and context-sensitive Bayesian inference, where change information is characterized by probability variations rather than pixel differences. Furthermore, a posterior probability vector analysis strategy is introduced to enhance the separability of unchanged and changed regions under noisy conditions. Experiments conducted on multiple remote sensing datasets with different Gaussian noise levels demonstrate that the proposed framework maintains stable detection performance and improves robustness compared with conventional unsupervised approaches. The proposed method provides a general solution for reliable unsupervised change detection under degraded observation conditions.

Abstract: Wildfire detection and characterization from space critically depend on accurate thermal infrared measurements across a wide dynamic range. OroraTech’s SAFIRE payloads feature mid-wave infrared (MWIR) and long-wave infrared (LWIR) bands optimized for this purpose, yet the volume and power constraints of many CubeSat platforms preclude the use of onboard calibration sources. This reflects a broader limitation of current Earth observation systems: the absence of robust calibration and validation methodologies at high brightness temperatures relevant to active fires. We evaluate the potential of using thermophysical models (TPM) of the Moon as a vicarious calibration reference for calibration transfer between instruments. The Moon reaches surface temperatures of up to 400 K, offering a stable target in the thermal domain that is visible from many different orbits. Previous research has established the use of TPMs for Moon observations in the infrared; however, uncertainties in surface properties remain, particularly in the mid-wave infrared. We investigate these effects using SAFIRE observations of the lunar surface acquired over a wide range of lunar phase angles (from waxing −81.5° to waning +122.2°). We constrain the TPMs using observations from the Sentinel-3 Sea and Land Surface Temperature Radiometer (SLSTR) fire channels and demonstrate their ability to uncover systematic differences between the calibration of SLSTR-A and SLSTR-B at high brightness temperatures. Applying the same methodology to observations of our SAFIRE payloads yields calibration residuals of 3% to 4% in both MWIR and LWIR bands. These results establish lunar vicarious calibration as a viable approach in the thermal domain for high-temperature applications, providing a pathway toward improved fire radiative power retrievals and enhanced global wildfire monitoring.

Abstract: Remote sensing change detection technology is widely used in land-use monitoring, urban planning, and disaster assessment. However, during imaging and transmission, bi-temporal remote sensing images are vulnerable to Gaussian noise, which makes it difficult for change detection algorithms to distinguish truly changed areas from noise-affected regions. To address this issue, this study proposes an unsupervised Gaussian-noise-robust change detection algorithm, termed FRIH-SEEDSAM. The proposed method first applies the Fast and Robust Fuzzy C-Means (FRFCM) algorithm to perform noise-resistant fuzzy clustering on bi-temporal remote sensing images. To establish reliable correspondences between the clustering results, the Integrated Region Matching (IRM) algorithm is introduced to construct weighted matching relationships while reducing the influence of abnormal memberships. The change intensity of spatially corresponding pixels is then calculated to generate a more stable change intensity map. Subsequently, the change intensity map is input into the Hybrid Conditional Random Field (HCRF) to infer pixel-level change labels, where the object potential function is constructed from the segmentation results of the Energy-Driven Sampling (SEEDS)-guided Segment Anything Model (SEEDSAM), which uses the centroids of the SEEDS superpixel regions as point prompts for SAM, thereby enhancing change-label consistency within the same changed object region. The experimental results show that the FRIH-SEEDSAM algorithm maintains stable change detection performance across different datasets and under varying Gaussian noise levels. It outperforms the comparison algorithms in terms of several accuracy evaluation indicators, including Kappa and F1. Furthermore, even when the Gaussian noise variance increases to 0.05, Kappa remains at 0.8 or above on multiple dataset images.

Abstract: Community-based blue carbon projects generate long-term monitoring records that are rarely harmonized across sites or linked to validated satellite carbon frameworks, limiting landscape-scale assessment. This study integrated eleven years of permanent-plot inventories with multi-sensor satellite imagery across the 8,212 ha Gazi–Vanga mangrove corridor, Kenya. A unified species-specific allometric framework was applied to 10,197 trees across 161 plots (2014–2025), and Gaussian Process Regression was trained on RapidEye, PlanetScope, and SuperDove satellite imagery to produce corridor-wide carbon-density estimates. The restoration plantation at Gazi Bay showed the only statistically confirmed AGB increase (+16.4 Mg ha⁻¹ yr⁻¹; p < 0.0001; ≈304 t CO₂e ha⁻¹ over eleven years); other strata showed directionally positive but non-significant trends. Sii Island consistently maintained the highest biomass (324–462 Mg ha⁻¹), and first-inventory baselines were established for Bodo (73.9 ± 9.4 Mg ha⁻¹), Shirazi (79.7–104.3 Mg ha⁻¹), and Munje (59.9–100.2 Mg ha⁻¹). The satellite framework achieved R² = 0.486 (RMSE = 61.1 Mg C ha⁻¹; n = 77) and agreed closely with field means in five of six mapped years; a 28.1% divergence in 2014 coincided with a sensor transition. Sentinel-1 C-band SAR did not improve upon optical-only predictions. Community monitoring archives can be harmonized and integrated with Earth observation data to support cost-effective, landscape-scale blue carbon assessment in community-managed mangrove ecosystems.

Abstract: Forest understory vegetation contributes to aboveground and belowground carbon budgets and affects terrestrial ecosystem function, yet quantifying understory biomass at large spatial scales is difficult. Increasingly, forest management operations are aided by unmanned aerial vehicles (UAVs) that collect imagery and airborne light detection and ranging (LiDAR) data, potentially providing a dataset for understory monitoring. In this study, we evaluated the suitability of two different remote sensing datasets (A and B) for predicting understory biomass amounts (field average of 2 Mg ha-1) using five different machine learning models. We compared (1) Acquisition A LiDAR and multispectral data, (2) Acquisition A multispectral data only, (3) Acquisition B SfM photogrammetry point cloud and multispectral data, and (4) Acquisition B multispectral data only. Remotely sensed data were temporally linked to ground plots in open post-harvest conditions, where we employed a method for quick in-field biomass estimation via calibrated photos. Our best model was a stochastic gradient boosting model built with LiDAR and multispectral data (testing R2=0.43, training R2=0.67). The multispectral-only model from Acquisition A performed similarly (testing R2=0.44, training R2=0.53), despite larger bias in shaded areas, suggesting an alternative streamlined method for creating wall-to-wall estimates. Structure-for-motion was unfit for modeling biomass under all tested conditions, indicating the importance of data acquisition and post-processing. Our study provides a novel framework for evaluating low-volume understory biomass efficiently and creating landscape scale predictions using limited ground measurements and remotely sensed data.

Abstract: In the semi-arid Jel Basin (Guercif Province, Morocco), declining surface water resources highlight the need for effective groundwater assessment to sustain agricultural activities. This study applies an integrated approach combining remote sensing, Geographic Information Systems (GIS), and two multi-criteria decision-making methods Analytic Hierarchy Process (AHP) and Multiple Influencing Factors (MIF) to map groundwater potential zones (GWPZ) in the Plaine Jel aquifer. Eight geo-environmental factors were considered, including lithology, lineament density, slope, land use/land cover, drainage density, soil, geomorphology, and precipitation. The results reveal a heterogeneous distribution of groundwater potential, with high-yield zones mainly located in the northern and northeastern areas, where favorable geological and structural conditions enhance infiltration. Model validation using ROC analysis demonstrated high accuracy, with AUC values of 91.1% (AHP) and 91.5% (MIF). Additionally, Mann-Kendall and Sen’s slope tests (2002–2023) indicate an overall stable trend in groundwater levels, despite seasonal water stress. By integrating spatial mapping with temporal trend analysis, this study provides a reliable framework for groundwater exploration and sustainable water resource management in the Jel Basin.

Abstract: Soil moisture is a key forecast variable of land surface models. Direct assimilation of microwave brightness temperature data to optimize soil moisture initial fields is an effective approach to improve simulation accuracy of soil moisture. However, most existing direct assimilation methods adopt physical radiative transfer models as observation operators, and their complex parametric errors greatly restrict the improvement of assimilation performance. This study introduces a high-precision MLP-based surrogate radiative transfer model as the observation operator. Combined with the Simplified Extended Kalman Filter (SEKF), it develops a direct radiance data assimilation system for the Common Land Model (CoLM). Assimilation experiments are conducted using brightness temperature data from the Microwave Radiation Imager (MWRI) onboard the FY-3D satellite. Their performance over China's land areas is systematically assessed through comparison with the assimilation scheme based on the Community Microwave Emission Model (CMEM). The results show that the MLP-based assimilation scheme can effectively improve soil moisture simulation accuracy, yet the improvement varies across vegetation types: grassland areas achieve the largest error reduction (10.2%), while semidesert areas present the most prominent increase in correlation coefficient (53.9%). Compared with the CMEM scheme, the MLP scheme exhibits better error stability and produces generally improved assimilation effects—specifically, in semidesert areas, the error decreases by 9.4% and the correlation coefficient increases by 62.8%. This study demonstrates that deep learning-based observation operators have strong application potential for land surface data assimilation under complex physical mechanisms.

Abstract: Accurate urban floodwater depth estimation is vital for disaster management but traditionally relies on data-intensive hydrodynamic models or supervised deep learning restricted by labeled data requirements. To address these bottlenecks, this study proposes a fully unsupervised framework for rapid flood depth estimation using post-event remote sensing imagery and Digital Terrain Models (DTMs). The methodology operates in two steps: first, a binary flood extent map is automatically delineated using a label-free unsupervised approach. Second, leveraging the hydrostatic equilibrium principle, floodwater depth is computed by integrating the extracted flood footprint with the underlying DTM. This framework was evaluated using the Inundation2Depth dataset, encompassing twelve urban and peri-urban sites in the Southeastern United States impacted by Hurricanes Matthew and Florence. Experimental results across all remote sensing sites demonstrated the framework’s viability, with segmentation F1-score and flood depth normalized-RMSE ranging from 64% to 95%, and 0.14 to 0.26, respectively. By eliminating the need for manual annotations and task-specific training, the proposed framework offers a scalable, transferable, and rapidly deployable solution for flood mapping and depth estimation in data-scarce environments, enabling efficient adaptation to new regions and disaster events without retraining or prior ground truth labels.

Abstract: The Huizhou Cultural-Ecological Reserve (HCER), China’s first nationally designated Cultural-Ecological Protection Zone, offers a distinctive setting where wetland conservation interacts with a millennia-old cultural landscape. We assemble a 26-year, 1-km grid panel (2000–2025; 14 011 grids; 364 286 grid-year observations) over the nine HCER counties, infer four-dimensional cultural ecosystem services (CES) – Aesthetic, Recreation, Heritage, Education – with Random Forest and XGBoost from a 14-variable predictor stack (mean XGBoost R² = 0.725), and apply a two-way fixed-effects panel regression with a distance-decay exposure kernel (5 km) to eight wetland protection units, using county-clustered standard errors. Reserve-wide CES-Total declines by 47% between 2000 and 2025. Once grid and year effects are absorbed, boundary cells show 4.1-percentage-points higher CES-Total than distant cells (β = +0.041, p = 0.012); Aesthetic (+8.1 pp, p = 0.029) and Recreation (+7.2 pp, p = 0.007) respond most strongly, Education positively but modestly (+0.6 pp, p = 0.001), Heritage not detectably. A halo peaks at 5–10 km rather than on the water surface. We formalise this as a Conservation Zone Externalities (CZE) framework and derive three planning levers for the HCER.

Abstract: The evaluation of Gridded Precipitation Products (GPPs) must account for the zero-inflated nature of precipitation data and the differences in spatial support between rain gauges and satellite grids. This study assesses the timing, precipitation event detection, and volume of ERA5, IMERG, CHIRPS, and TAMSAT against the Trans-African Hydro-Meteorological Observatory (TAHMO) network. We employ variance-stabilising transformations, detect rainfall events, and cluster diurnal precipitation cycles into different regimes. Our clustering results reveal spatial variability in performance, with GPP and TAHMO derived diurnal regimes differing at 57.3% of the stations. Analysis of the diurnal precipitation reveals that ERA5 satisfies its daily water budget through persistent drizzle. At the daily scale, IMERG exhibits superior event detection, timing, and volume accuracy. On the other hand, CHIRPS and TAMSAT show a wet bias. We conclude that GPP selection should consider the use-case, and future meteorological and AI-driven applications should incorporate verification metrics that account for timing, event detection and volume accuracy.

Abstract: Rapid flood mapping using optical sensors such as Sentinel-2 is frequently challenged by spectral confusion, where turbid water, urban shadows, and wet soils exhibit similar reflectance signatures that undermine single-index detectors. We present the FLOod Oriented Detection Hybrid Index (FLOOD-HI), a statistically calibrated multi-index fusion that combines six complementary spectral indices (NDWI, IMP, AWEI, TCW, NDVI, and SAVI) through a multivariate linear model. Trained against high-fidelity Directly Affected Area (DAA) ground-truth maps from the catastrophic May 2024 Rio Grande do Sul floods, the model is implemented end-to-end in Google Earth Engine (GEE). Inundation extent is mapped using a straightforward sign-based decision rule (FLOOD-HI > 0), with spectral masks incorporated as pragmatic refinements. Moving beyond traditional small-sample paradigms, FLOOD-HI is validated wall-to-wall over two municipality-scale Regions Of Interest (ROI), with metrics computed across all pixels rather than on hand-picked water and non-water samples. In the external testing domain (ROI-2), FLOOD-HI achieves an F₁ ≈ 0.80, IoU ≈ 0.66, Precision ≈ 0.73 and Recall ≈ 0.88, substantially outperforming the best-performing single index (TCW, F₁ ≈ 0.63 and IoU ≈ 0.46). This represents an approximate 26,50% relative improvement in F₁ (absolute gain ≈ 0.17) and a 44% improvement in IoU (absolute gain ≈ 0.20). The major contribution is methodological, offering a reproducible multivariate index formulation, a conservative municipality-scale framework, and an open-access GEE implementation that includes a localized calibration workflow.

Abstract: Forest inventory operations require accurate and scalable methods for estimating structural tree attributes such as diameter at breast height, tree height, and crown dimensions. Terrestrial laser scanning (TLS) and mobile laser scanning (MLS) provide detailed stem information but limited spatial coverage, whereas unmanned laser scanning (ULS) provides broad coverage but sparse internal stem structure. This study evaluates whether fused LiDAR point clouds and imputation-based correction methods can transfer accurate stem information from TLS, MLS, or fused calibration areas to larger ULS-only areas. ULS, TLS, MLS, and fused laser scanning (FLS) were analysed for 10 m and 20 m radiata pine, and 30 m eucalyptus stands. The TreeLS R package and regression-based methods were used for attribute extraction, and several imputation strategies were tested, including parametric cumulative distribution function (CDF) mapping, empirical and rank-based quantile mapping, conditional mean with residual spread restoration, and voxel-based imputation. The results show that TreeLS applied to TLS, MLS, and FLS produced substantially more reliable diameter at breast height (DBH) distributions than regression models applied directly to ULS data. For ULS-only areas, uncorrected regression estimates differed from field means by 2.0 cm for tall radiata pine, 10.4 cm for medium radiata pine, and 5.4 cm for mature eucalyptus. Using terrestrially trained voxel-based imputation reduced these differences to 0.1 cm, 0.3 cm, and 0.8 cm, respectively. Other distribution-aware imputation methods also substantially improved agreement with field-observed DBH distributions. These findings indicate that relatively small areas of high-density TLS or MLS or FLS data can be used to calibrate inventory estimates across larger ULS-only regions. The approach offers a practical compromise between the structural accuracy of terrestrial LiDAR and the spatial efficiency of aerial LiDAR, supporting scalable forest inventory estimation without requiring exhaustive terrestrial coverage.

Abstract: Predicting plantation establishment failure prior to planting remains a major operational challenge due to the strong spatial variability in post-harvest environmental conditions. This study developed a spatially explicit modelling framework that integrated pre-planting unmanned aerial vehicle (UAV)-derived environmental, structural, terrain, and operational-treatment data to predict establishment risk across plantation landscapes. Environmental, terrain, vegetation, and structural predictors were derived from pre-planting UAV imagery, while plantation establishment outcomes were quantified approximately 21 months later using an automated tree-detection and assessment framework. The datasets were integrated within a ridge-regularised logistic regression model incorporating interaction terms, multi-scale predictors, operational treatment masks, and blocked spatial cross-validation. The model achieved strong predictive performance under geographically independent validation, with moisture-related variables, vegetation condition, and structural metrics contributing most strongly to establishment-failure prediction. Predicted risk surfaces closely matched observed patterns of reduced stocking density and suppressed growth. The results demonstrate that plantation establishment risk can be predicted in advance of planting using pre-existing environmental and operational information, indicating that a substantial proportion of future plantation performance is determined by site conditions present before establishment begins.

Abstract: Coastal regions face accelerating pressure from sea-level rise, shoreline erosion, storm surge, recurrent flooding, salinization, land-use change, and degradation of blue-carbon ecosystems. This review synthesizes literature published primarily from 2020 to 2026 on artificial intelligence (AI), machine learning (ML), deep learning, remote sensing, and geospatial applications for coastal resilience and environmental sustainability. Using a PRISMA-informed selection framework, 867 records were identified, 305 duplicate records were removed, 562 titles and abstracts were screened, 109 reports were sought for retrieval, 36 full-text reports were assessed, and 30 studies were retained in the final synthesis. The review identifies five major domains: shoreline extraction and erosion monitoring; coastal flooding, storm surge, and overtopping risk; mangrove, wetland, and ecosystem monitoring; land-cover change and coastal urbanization; and GIS-based adaptation support. Common methods include Random Forest, XGBoost, convolutional neural networks, U-Net variants, recurrent models, and explainable AI. The findings show that AI is most useful when embedded in transparent, validated, and locally interpretable geospatial workflows that include ground truth, uncertainty communication, governance safeguards, and participatory planning.

Abstract: This study aims to investigate the dynamics of forest fires in Brazil, particularly in the Amazon region, motivated by the fact that approximately 60% of the Amazon rainforest lies within Brazilian territory, thus making the country central to the understan-ding and managing this critical environmental issue. Forest fires in the Brazilian forests, especially within the Amazon, represent a major environmental challenge, with significant impacts on biodiversity, atmospheric composition, and climate regulation. In recent decades, fire activity has intensified due to climate variability and growing anthropogenic pressure, raising concerns about a possible transition of the Amazon from a carbon sink to a carbon source. This study examines the spatial and temporal variability of fire activity across Brazil, with a specific focus on the Brazilian Amazon, covering the period 2001–2022. The analysis is based on satellite-derived active fire data from NASA’s Fire Information for Resource Management System (FIRMS), using observations from the Moderate Resolu-tion Imaging Spectroradiometer (MODIS) and the Visible Infrared Imaging Radiometer Suite (VIIRS). Fire Radiative Power (FRP) is employed as a proxy for fire intensity and combustion dynamics. The VIIRS sensor, characterized by improved sensitivity to small and low-intensity fires, highlights the increase in fire activity observed after 2012. Significant peaks in fire activity were detected in 2004, 2005, 2007, and again after 2019. Statistical analyses reveal marked interannual variability and cyclical patterns in FRP, associated with fluctuations in drought conditions, precipitation regimes, land-use changes, human pressure and environmental policy measures. Overall, the results emphasize the importance of integrating multi-sensor satellite observations for long-term monitoring of fire regimes in Brazil, with particular relevance for the Amazon region.

Abstract: Operational satellite monitoring of pasture biomass demands models that extrapolate beyond the range of properties on which they were calibrated. We show that the dominant failure mode of Sentinel-2 pasture-biomass models in tropical Australian rangelands is the saturation and phenological inversion of greenness-based vegetation indices across sites, and that this failure can be substantially repaired with open climate and topsoil covariates from public archives. The work builds on a hierarchical pipeline that scales clip-and-weigh ground truth (n=1120 tare-corrected samples across eleven sites on five Queensland properties) through UAV digital-surface-model imagery to Sentinel-2 predictions, using TabPFN – a pre-trained transformer foundation model for small tabular data – as the regressor at all three nested spatial scales, and a seven-class deep-learning pasture mask (overall accuracy 98.6 %) to suppress mixed-pixel noise. Under a leave-one-site-out (LOSO) cross-validation protocol on twenty site-date aggregates across nine sites, spectral-only models failed to transfer across sites (R2=−0.21, RMSE=4.79 t ha-1). Appending open climate (Open-Meteo ERA5) and soil (SoilGrids 2.0) covariates, and switching to a gradient-boosted regressor on log-transformed biomass, lifted LOSO R2 to +0.07 and reduced RMSE to 4.19 t ha-1. A leaf-nitrogen growth trajectory, predicted by the TabPFN nitrogen regressor developed in our earlier pasture chemistry work, reduced LOSO error by a further 11 % relative to greenness-only growth features. Three additional covariate classes – BARRA-R2 reanalysis climate, three independent fractional-cover products, and Sentinel-1 C-band SAR backscatter – were tested and rejected, all hitting the same RMSE floor. The symmetric negative results suggest that the residual LOSO ceiling at the current nine-property footprint is a sample-size and sensor-saturation limit rather than a feature-engineering one, and that the most tractable operational path forward is to stratify the production model by climatic zone and Queensland Land Type rather than pursue further covariates within a single global learner. Expanding UAV calibration footprints and integrating open climate, soil and plant-chemistry data are complementary, not competing, investments for operational rangeland remote sensing.

Abstract: The miombo woodlands of southeastern Democratic Republic of the Congo (DR Congo) are increasingly threatened by mining expansion and associated land-use changes. Unlike previous studies focusing separately on mining impacts or land-cover change, this study explicitly quantified the combined influence of mining roads and settlements on forest degradation gradients in an emerging mining frontier. The study was conducted in the Mutshatsha Territory, a rapidly transforming landscape within the Katangese Copperbelt. Land-cover dynamics between 1998 and 2023 were reconstructed from six Landsat image series using Random Forest classification implemented in Google Earth Engine. Landscape transformation was assessed through landscape ecology metrics, while accessibility effects were quantified using spatial gradients around a mining exploration road and four villages representing contrasting accessibility contexts. Forest cover declined from 39.9% in 1998 to 12.5% in 2023, corresponding to an annual deforestation rate of 2.75%, substantially exceeding the national average for DR Congo. Simultaneously, agricultural land, built-up areas, and open vegetation expanded, while landscape disturbance increased and forest patch cromplexity decreased. Forest cover consistently increased with distance from both villages and the mining road, demonstrating strong accessibility gradients. Distance from villages explained up to 80% of the variation in forest cover, whereas road-distance gradients accounted for 71–94% of the observed variation. Forest loss along the mining road extended up to approximately 5 km from the corridor and intensified over time. Villages located along the RN39 transportation corridor exhibited substantially greater forest depletion than the more isolated village of Mpwita. These findings demonstrate that mining roads and settlements operate synergistically to structure forest degradation patterns through accessibility-driven processes. Integrating accessibility considerations into land-use planning and conservation strategies is therefore essential for mitigating forest degradation in rapidly expanding mining frontiers.

Abstract: Mountain forest greenness varies across seasons, yet growing-season averages can mask season-specific climate associations. We assessed seasonal and elevational variation in forest greenness, temporal variation in spring climate associations, and whether nighttime light (NTL)–greenness covariation persisted after detrending in Changbai Mountain from 2000 to 2024. Analyses used MODIS normalized difference vegetation index (NDVI), 1 km temperature and precipitation grids, elevation zones, and annual NTL data, restricted to pixels with at least 70% forest cover based on the 2024 Annual China Land Cover Dataset. Forest NDVI increased significantly during the growing season (slope = 0.0006 yr⁻¹, R² = 0.166, P < 0.05), summer (slope = 0.0006 yr⁻¹, R² = 0.188, P < 0.05), and autumn (slope = 0.0019 yr⁻¹, R² = 0.280, P < 0.01), with the steepest increase in autumn. Spring showed no significant long-term trend (slope = 0.0011 yr⁻¹, R² = 0.087, P = 0.152), but its NDVI had the strongest temperature association at the regional scale (r = 0.760, P < 0.001), remaining positive across all elevation zones. In spring regressions, temperature represented 84.1–97.8% of the summed absolute standardised coefficient magnitude across elevation zones, with all variance inflation factors ≤1.22. Exploratory 5-year sliding-window correlations suggested that spring NDVI–climate correlations varied over time, but each estimate was based on only five observations. At low elevation, the raw NTL–NDVI regression was significant (R² = 0.198, P < 0.05), whereas the detrended regression was not (R² = 0.042, P = 0.324), indicating shared long-term trends. Growing-season averaging therefore obscures strong autumn greening and the distinct spring temperature association; raw NTL–NDVI covariation should not be interpreted without accounting for temporal trends.