Earth and Environmental Sciences

Abstract. Methane pockmarks and shallow gas systems are prominent geomorphological features in the Baltic Sea that act as hotspots of microbial activity. In the Gdańsk Basin, pockmarks vary in gas-seepage intensity and in the extent of freshened porewater discharge, both of which influence the archaeal community structure and the composition of internal methane biofilters. The objective of this study was to examine the effects of methane seepage and freshened porewater on the composition of archaeal communities and on the isoprenoid glycerol dialkyl glycerol tetraethers (iGDGTs), membrane lipids synthesised by these communities, across the gas systems examined. Additionally, the effects of these environmental factors on the use and interpretation of iGDGT-based proxies under conditions of gas and water seepage were assessed. The study investigates whether iGDGT patterns in these Baltic gas systems reflect methane-driven processes, including anaerobic oxidation of methane (AOM) and methanogenesis, porewater freshening, or pelagic contributions from ammonia-oxidising archaea (AOA). The results show elevated iGDGT concentrations in pockmark cores compared with reference non-pockmark cores; however, the summed iGDGT concentration varies by site. Overall, iGDGT concentrations are much higher at sites with reduced seabed pockmark activity and weak porewater freshening. Nevertheless, consistently low Methane Index values (MI < 0.09), together with low GDGT-0 / crenarchaeol (< 1) and low GDGT-2 / cren (< 0.04) ratios, indicate that the iGDGT pool lacks the typical enrichment in GDGT-1 to -3 associated with AOM, suggesting no AOM imprint on the iGDGT pool. However, 16S rRNA analysis revealed that the anaerobic methanotrophic archaeal lineages (ANME) consist of ANME-2b and ANME-3. A strong positive correlation between OH-GDGTs and crenarchaeol, together with the consistency of OH-GDGT% values with those previously reported for Baltic Sea surface sediments, suggests a thaumarchaeal source of iGDGTs in the studied pockmark and reference cores. The Branched and Isoprenoid Tetraether (BIT) index values suggest marine archaeal GDGT production. In this system, iGDGT-based proxies primarily reflect a strong pelagic presence of AOA, as indicated by the dominance of crenarchaeol. This suggests that, despite the local presence of methanogenic and ANME-related archaeal groups, methane-related AOM does not influence the iGDGT signal due to the archaeal community structure. These findings highlight the complex interplay between freshened porewater and gas seepage in shaping archaeal communities, and the role of ammonia-oxidising Nitrososphaeria in controlling iGDGT composition and the sedimentary record.

Abstract. Groundwater is a critical resource supplying nearly half of the world's drinking water. This study focuses on the Kurikka buried valley aquifer system in western Finland, characterized by complex hydrogeology dictated by the bedrock topography and sediment cover producing artesian conditions in deep aquifers. Using a multitracer approach, the study incorporates hydrogeochemical, isotopic (δ2H, δ18O, δ34S, 87Sr/86Sr) and microbial community analyses with residence time indicators (CFCs, SF6, 3H, 3H/3He). Groundwater samples collected from ten sites revealed differences in residence times, microbial diversity and community compositions, as well as large variation in the strontium and sulphur isotopic compositions. The bedrock groundwater sample revealed a more evolved water type, consistent with longer residence time, strong water-mineral interactions and typical deep subsurface bacterial members. Groundwater from the superficial unconsolidated aquifers contained a modern water component (< 60 years) whereas the deeper buried valley aquifers were characterized by older waters. The information provided by this study is crucial for groundwater management during extensive extraction for municipal water supply.

Abstract. Coastal marshes are key habitats contributing to organic carbon (OC) storage but remain understudied in Nordic regions regarding Blue Carbon processes. This study quantified OC stocks in above- and below-ground plant biomass and in the top 50 cm-soil across 12 grazed and ungrazed marshes, spanning a major environmental gradient, and assessed how biotic (plant communities, livestock grazing) and abiotic (soil properties, environmental conditions) drivers shape OC storage. Soil OC stocks accounted for ∼ 73 % of total OC in grazed sites and ∼ 63 % in ungrazed ones and was higher in grazed sites (99.7 ± 57.9 Mg ha−1) than in ungrazed sites (78.2 ± 44.2 Mg ha−1). Grazing and the large-scale environmental gradient strongly structured plant communities, partly by regulating reed (Phragmites australis), prevalent in ungrazed sites. Abiotic soil properties were major large-scale drivers of soil OC storage, while grazing affected soil OC storage indirectly through plant composition. Soil OC increased with finer textures, whereas vegetation and grazing effects were variable and locally expressed. Aboveground OC stocks were reduced by grazing, both directly through biomass removal and indirectly by reducing reed dominance. Belowground OC stocks were driven by plant community composition and indirectly by grazing effects on vegetation. Root biomass was concentrated in the top 15 cm in grazed sites and deeper (15–50 cm) in ungrazed sites, reflecting contrasting plant strategies. Overall, soil OC stocks in Nordic coastal marshes fall within the lower range of global estimates. These findings highlight the need to consider soil processes, grazing and environmental gradients in the sustainable management of Nordic coastal marshes and their carbon storage potential.

Abstract. Phytoplankton community composition is a key determinant of ocean biogeochemical cycles, yet its observation from autonomous platforms remains challenging. In this study, we assessed the potential of in situ multispectral excitation fluorescence (MXF) to predict phytoplankton community structure indices in the Northwestern Mediterranean Sea. With a view toward applications on Biogeochemical-Argo (BGC-Argo) profiling floats, we evaluated a miniaturised, three-excitation-channel fluorometer. Laboratory measurements on ten phytoplankton strains confirmed that MXF ratios at 440, 470, and 532 nm provide taxon-specific signatures, especially for picocyanobacteria and green algae. Field observations of phytoplankton pigments were clustered into four ecologically distinct phytoplankton communities across the seasonal cycle, which defined the targeted phytoplankton community structure indices. A machine learning model was then trained to classify these clusters using MXF and additional bio-optical indices. Results show that existing BGC-Argo configurations (single-wavelength fluorescence, particulate backscattering, and beam attenuation coefficients) reliably distinguish broad community structures, such as pico- versus microphytoplankton dominance, but resolving finer pigment-based differences requires the additional spectral information provided by MXF. The different excitation channels contributed unequally: 440 and 470 nm provided robust pigment sensitivity across communities, while 532 nm was particularly informative for detecting phycoerythrin-rich taxa. Overall, combining MXF with bio-optical proxies improved classification performance by integrating pigment-specific and size-structure information, demonstrating the potential of MXF to enhance autonomous monitoring of phytoplankton community dynamics and their role in ocean biogeochemical cycles.

Abstract. Anthropogenic CO2 emissions and their continuous dissolution into seawater lead to seawater pCO2 rise and ocean acidification (OA). Phytoplankton groups are known to be differentially affected by carbonate chemistry changes associated with OA in different regions of contrasting physical and chemical features. To explore responses of phytoplankton to OA in the Chinese coastal waters, we conducted a mesocosm experiment in a eutrophic bay of the southern East China Sea under ambient (410 μatm, AC) and elevated (1000 μatm, HC) pCO2 levels. The HC stimulated phytoplankton growth and primary production during the initial nutrient-replete stage, while the community diversity and evenness were reduced during this stage due to the rapid nutrient consumption and diatom blooms, and the subsequent shift from diatoms to hetero-dinoflagellates led to a decline in primary production during the mid and later phases under nutrient depletion. Such suppression of diatom-to-dinoflagellate succession occurred with enhanced remineralization of organic matter under the HC conditions, with smaller phytoplankton becoming dominant for the sustained primary production. Our findings indicate that, the impacts of OA on phytoplankton diversity in the coastal water of the southern East China Sea depend on availability of nutrients, with primary productivity and biodiversity of phytoplankton reduced in the eutrophicated coastal water.

Abstract. Root exudation, defined as labile carbon (C) allocation into soils through fine roots, is a substantial yet often overlooked pathway of the terrestrial carbon cycle. Root exudation is likely to increase under rising levels of atmospheric CO2, but the implications of the increase in this flux are poorly understood. Increased labile C availability in soils may stimulate microbial growth and increase soil carbon storage but at the same time microbial nutrient acquisition could offset this accumulation by enhanced decomposition of soil organic matter Here, we implement a dynamic representation of root exudation based on plant surplus carbon and nutrient limitation in the microbial explicit terrestrial biosphere model QUINCY-JSM (QUantifying Interactions between terrestrial Nutrient CYcles and the climate system). We evaluate the effect of elevated CO2 on root exudation and its consequences for microbial C, nitrogen (N) and phosphorus (P) cycling using observations from the Eucalyptus Free Air CO2 Enrichment (EucFACE) experiment in a soil phosphorus impoverished forest. In the experiment, more than half of additional gross primary productivity (GPP) under elevated CO2 (eCO2) could not be assigned to a measured vegetation flux. With the explicit implementation of root exudation, our model predicted that elevated CO2 caused an increase in belowground carbon flux and an increase in microbial growth, but a limited effect on soil carbon storage. Root exudation was increased to 30 %, but more than half of this additional input was directly respired by microbes. As a result, root exudation gives a possible explanation for the not measured vegetation flux and the enhanced heterotrophic respiration under eCO2 observed in the experiment. Increased C input through root exudation also enhanced microbial growth, but in order to support this growth, microbes mostly gained nutrients from decomposition and mineralization of organic matter. As a consequence, increased decomposition negated build-up of microbial necromass. Our study emphasizes the role of root exudation and microbial activity for soil carbon sequestration under elevated CO2 and guides further research regarding plant-microbe interactions.

Cyanobacterial blooms are intensifying globally due to nutrient enrichment and climate change, producing a chemically diverse suite of peptides, in addition to the well-studied microcystins. These cyanopeptides, including anabaenopeptins, cyanopeptolins, aeruginosins, and microginins, frequently co-occur in blooms across freshwater and estuarine systems and exhibit potent protease- and phosphatase-inhibitory activities at environmentally relevant concentrations. This review synthesizes emerging evidence that these compounds may profoundly influence both environmental and host-associated microbiota. Bloom-associated cyanopeptides and related environmental stressors may act as ecological filters in aquatic ecosystems, contributing to microbial dysbiosis, which is characterized by changes in community composition and sometimes reduced diversity. This also leads to the enrichment of toxin-degrading components of microbiota taxa, such as Sphingomonas and Novosphingobium, and metabolic reconfiguration toward xenobiotic degradation. Microbiota exposed to bloom-associated cyanopeptides rich conditions in aquatic ecosystems occur in free and particulate forms in the water column, and these forms often recover and adapt more rapidly than host-associated microbiomes. However, conflicting results have been observed in fish gut microbiota data responses, where some host-associated microbiomes show relatively fast recovery and others show delayed restoration. Multi-omics studies have revealed conserved mechanisms linking cyanopeptide exposure to shifts in microbial structure and metabolic pathways, which together can affect host physiology. However, most studies remain biased toward microcystin-LR, and there is a significant gap in our understanding of how other cyanopeptides alter free-living and host gut microbiota in aquatic ecosystems. Therefore, this review identifies an important next step in research, which should focus on how non-microcystin cyanopeptides affect free and host-gut microbiota, and these studies should include multiomics approaches to unravel these changes under natural field observations and controlled exposure. Recognizing microbiota as both targets and agents of cyanopeptide transformation offers a new framework for understanding bloom ecology, because this knowledge will aid in predicting ecosystem recovery and mitigating the ecological risks of these compounds.

Urbanization and greening reshape the surface thermal environment, with opposing thermal effects on urban sustainability and population health. Here, we integrate multidimensional analysis and LightGBM models to quantify how urbanization and greening alter surface temperature and thereby modulate surface urban heat island (SUHI) dynamics across 325 Chinese cities from 2000 to 2025. Our study reveals that urbanization-induced warming is consistently strongest in neighboring rural areas, whereas vegetation-driven cooling intensifies sharply in urban areas during summer. These opposing thermal effects exhibit clear seasonal and diurnal asymmetries, with the strongest responses occurring in summer and during the daytime. Climatic factors emerge as the dominant drivers of surface thermal variability across the urban-rural gradient. Urbanization-induced warming was the dominant contributor to the mean SUHI intensity of 0.39 °C, whereas urban greening represented a cooling contribution share of approximately 31.83% during summer. This contrast between persistent urbanization-induced warming and seasonally concentrated vegetation-driven cooling reveals the constrained timing and spatial dependence of effective biophysical mitigation. Our findings clarify the dual and asymmetric roles of urbanization and greening in urban thermal dynamics and provide quantitative benchmarks for differentiated, climate-resilient planning.

With global warming and rapid urbanization, short-duration summer rainstorms are becoming more intense and localized, posing growing challenges to urban flood resilience. However, their spatiotemporal characteristics, vertical structures, and environmental drivers remain poorly understood. Here, we combine multi-source remote sensing datasets and China’s new-generation satellite-borne dual-frequency precipitation radar observations to investigate summer rainstorms in Nanjing, China, during 2017–2024. Results reveal pronounced spatiotemporal heterogeneity, with higher rainfall intensities concentrated over urban and adjacent areas. During the study period, rainstorm intensity and duration increased by 7.44% and 38.63%, respectively, while the affected area decreased by 8.18%, indicating a transition toward more localized yet more intense rainfall events. Environmental analyses suggest that large-scale thermodynamic conditions and regional topographic forcing provide a favorable background for convection development, while local urban thermal effects may further modulate rainfall enhancement. Three-dimensional radar detection of an illustrative rainstorm event indicates an inverted-cone vertical structure, suggesting a mixed convective-stratiform precipitation structure involving both warm-rain and ice-phase processes. An Explainable Bayesian-Optimized XGBoost (EBOX) model further identifies near-surface air temperature and specific humidity as the primary environmental factors associated with rainstorm occurrence and development. Overall, this study highlights the value of integrating satellite remote sensing with explainable artificial intelligence to improve understanding of urban extreme rainfall and provide new insights into how climate change, topography, and urbanization jointly shape precipitation extremes in rapidly urbanizing monsoon regions.

Spatial and temporal patterns of terrestrial water storage (TWS), and their relationship with groundwater levels, were investigated with the Gravity Recovery and Climate Experiment (GRACE) satellite data, the Global Land Data Assimilation System (GLDAS) land surface model results, and climate observations for the Murray–Darling Basin (MDB). The results show that: (1) TWS displays a clear temporal variability: a negative TWS anomaly with a declining trend during 2002–2009, a positive TWS anomaly with a decreasing trend during 2010–2017, and a period of mixed positive and negative TWS anomalies being accompanied by an increasing trend from 2018 to 2025; (2) five dominant cluster patterns were identified that explain the spatial variability of temporal TWS across the MDB; (3) overall, TWS temporal variability is strongly correlated with rainfall, although it is weak at certain locations; (4) TWS is also influenced by evaporation (both actual and potential evapotranspiration, AET and PET) and runoff, and a combined model significantly improves the overall performance in explaining TWS temporal variability; and (5) TWS-derived groundwater storage changes show both similarities and differences in comparison with groundwater level observation changes, reflecting complex hydrogeological processes and the influence of human activities such as groundwater extraction. These findings provide valuable insights to support improved groundwater resource management with GRACE satellite information and land surface models.

Carbon storage dynamics in dryland and semi-arid ecosystems remain a major uncertainty in global carbon cycle assessments, particularly in regions like the Yellow River Basin (YRB). Using the Arid Ecosystem Model (AEM), we simulated the spatiotemporal evolution of four major carbon pools—total carbon (TOTC), vegetation carbon (VEGC), soil organic carbon (SOC), and litter carbon (LTRC)—from 1981 to 2060 under factorial climate scenarios. During 1981–2020, TOTC increased by 0.09 Pg C (+3.54%), driven by gains in VEGC (+0.03 Pg C, +21.43%) and SOC (+0.06 Pg C, +2.78%). LTRC showed minimal net change but was highly sensitive to interannual variability. From 2021 to 2060, under the high-emission SSP5 scenario, TOTC is projected to increase by 0.114 Pg C (+4.81%), with VEGC contributing most of the gain (+23.87%). CO2_only simulations showed similar increases, underscoring the dominant role of CO2 fertilization. In contrast, warming and precipitation alone produced weaker and more variable effects. Spatially, upper YRB regions are expected to maintain strong sink capacity, while the Loess Plateau and central-western subregions remain vulnerable to warming and moisture decline. LTRC exhibited the highest variability across scenarios (−18% to +22%), highlighting its role as a sensitive indicator of sink stability. These findings emphasize the need to account for nonlinear climate–carbon interactions and regional heterogeneity. Region-specific, adaptive strategies that integrate ecological restoration and climate adaptation will be critical to enhancing carbon sinks and supporting China’s carbon neutrality targets in the Yellow River Basin.

ABSTRACT Frequent drought‐flood abrupt alternation (DFAA) events in the Yangtze River Basin (YRB), characterized by complex formation mechanisms and limited predictability, pose a prominent challenge in climate research. A daily‐scale DFAA index is developed to enhance event identification accuracy by coupling dual signals from precipitation and soil moisture in this study. The selected 86 DFAA events from 1961–2023 are classified into two distinct spatial patterns: an Eastern type affecting the lower reaches and a Northern type influencing the northern middle reaches of YRB. The multi‐scale analysis demonstrates that DFAA alternation periods are primarily driven by 10–30‐day intraseasonal oscillations (ISOs), while the maintenance of drought and flood phases relies on 30–90‐day ISOs. However, during the alternation phase, Eastern‐type events are governed by upper‐troposphere dynamical processes, manifested through energy dispersion from the high‐latitude Rossby wave trains that drive rapid reversal of the 10–30‐day geopotential height over East Asia, accompanied by eastward‐propagating upstream westerly anomalies promoting westward extension of the jet stream over Japan, ultimately establishing strong upper‐level divergence over the key region and triggering the alternation to flood conditions. In contrast, Northern‐type events are predominantly controlled by low‐level processes, where 10–30‐day scale cyclonic circulation intensifies rapidly through lee‐wave forcing from the Qinling–Daba Mountains during its southward migration, triggering low‐level convergence and ascent that facilitate precipitation development. During the drought or flood phase, both types exhibit baroclinic structures over the key region, and the Eastern type primarily relies on eastward‐propagating 30–90‐day Rossby wave energy over the subtropics, whereas the Northern type is influenced by converging wave energy transported from both mid‐high latitudes and the subtropical region. This study provides a scientific basis for understanding the mechanisms of extreme hydrological events in the basin and improving extended‐range forecasting capabilities.

Accurate field-scale surface soil moisture (SSM) measures are important for agricultural water management. Conventional satellite SSM products remain too coarse for within-field applications. Here, we developed sequence-aware deep learning models for growing-season SSM estimation by fusing data from Sentinel-1 C-band SAR, harmonized Landsat-8/Sentinel-2 (HLS), and local ancillary datasets. We assembled a multi-source dataset on Sentinel-1 overpass time for 2016–2024 (May–September), yielding 1469 samples and 65 features per sample, including SAR and optical features, meteorological data, soil texture and bulk density, topography, crop labels, irrigation-likelihood flag, and irregular-time-step encoding. We compared long short-term memory (LSTM) and temporal convolutional neural network (TCN) architectures together with attention-augmented variants, including feature attention (FA), temporal attention (TA), and the combined feature–temporal attention (FTA). Models were trained and tested on seven years of data and were validated based on a temporal generalization using combined data of a wet year (2016) and a dry year (2023). The best model, FTA-TCN, achieved R2 = 0.851, RMSE = 0.024 m3.m−3, and MAE = 0.020 m3.m−3 on the withheld validation years, outperforming the base LSTM (R2 = 0.422; RMSE = 0.053 m3.m−3; MAE = 0.043 m3.m−3) and the base TCN (R2 = 0.746; RMSE = 0.034 m3.m−3; MAE = 0.022 m3.m−3). Shapley additive explanations (SHAP) analysis indicated that antecedent precipitation and short-term rainfall accumulations were dominant forcings, while soil texture, elevation, incidence angle, and vegetation indices modulated SSM variability. Satellite-derived features accounted for ~28.5% of aggregated SHAP importance. Overall, the results show that dual-attention temporal convolution can capture field-scale SSM dynamics across wet and dry seasons when satellite signals are coupled with local soil-meteorological-management context.

Accurate rice yield estimation is essential for food security. Two key factors affecting estimation accuracy are the long-term upward trend in yield over time and regional heterogeneity across space. Current studies predominantly employ statistical detrending methods (e.g., moving averages, linear regression) to isolate temporal trends. However, such methods rely on prior assumptions about the time–yield relationship and may introduce systematic bias when these assumptions break down. Meanwhile, the individual contributions of temporal and spatial information, and their interactive effects, have not been systematically evaluated within a unified framework. We selected 112 rice-growing counties across six U.S. states (2000–2021), using vegetation index (Normalized Difference Vegetation Index), meteorological indicators (growing degree days, killing degree days, and cumulative precipitation), and spatiotemporal variables (year, longitude, and latitude). We designed six input configurations to compare conventional detrending against direct temporal variable inclusion, testing across four model architectures (Long Short-Term Memory, Random Forest, XGBoost, and Transformer). Results showed that: (1) directly inputting year significantly outperformed detrending across all models, with the combined spatiotemporal configuration achieving the best performance (LSTM R2 = 0.61 vs. 0.54 for detrending); (2) year was the most important predictor in SHAP analysis, with spatiotemporal variables ranking higher than most meteorological and remote sensing variables; (3) spatial information consistently improved accuracy and mitigated systematic bias for extreme yield regions; (4) the combined configuration performed best across different states, years (including extreme climate events), and yield levels, achieving near-end-of-season accuracy at the grain-filling stage (1.5–2 months before harvest). This study demonstrates that integrating raw spatiotemporal data directly into deep learning models is more effective than statistical detrending, offering a simpler and more robust approach for large-scale crop yield estimation.

Accurate forecasting of high-dimensional meteorological fields remains challenging due to the complex spatio-temporal dynamics of atmospheric systems and the presence of heterogeneous training difficulty across space and lead time. Existing deep forecasting approaches usually optimize all prediction units uniformly, which may overemphasize low-benefit or weakly generalizable supervision signals. To address this issue, we propose Spatio-Temporal Selective Learning (ST-SL), an online training framework that estimates the learnability of each prediction unit by comparing the main model with a frozen reference model and computes the loss only over selected high-benefit spatio-temporal units. To provide an effective forecasting backbone, we further introduce VASTFormer, a variable-aware spatio-temporal Transformer that models cross-variable dependencies, incorporates physics-enhanced Solar Positional Encoding, and captures atmospheric trajectories with an efficient temporal translator. Experiments on the ERA5 reanalysis dataset show that VASTFormer outperforms representative spatio-temporal baselines, while ST-SL further improves accuracy without adding inference-time parameters or computational cost. Compared with the strongest baseline, VASTFormer+ST-SL reduces MSE, MAE, and RMSE by 8.84%, 6.70%, and 4.54%, respectively. Meteorological skill evaluation further shows an average ACC of 0.9801 and RMSESS of 0.8104, and percentile-based extreme-condition evaluations confirm consistent improvements across standard and high-impact forecasting scenarios. These results indicate that selective supervision can improve generalization in dense meteorological forecasting.

ABSTRACT This study explores the climatic drivers influencing the monthly Burned Area (BA) during the winter fire season (November–April) in northern Italy from 2008 to 2022 at 0.11° spatial resolution, providing an example of the current climatic dynamics affecting mountains. The results of the analysis indicated that the Alps and northern Apennines are mainly characterised by a winter fire regime. In these mountain areas, the Western Alps experienced the largest wildfire events, with return periods lower than 6 years. Parallel to BA data, 150 daily precipitation and temperature ground series were collected, converted to monthly scale, quality controlled, and gridded at the 0.11° spatial resolution. Several climatic indices were computed for precipitation, temperature, and droughts. To find the best BA predictors, we checked the correlations of BA with different temporal aggregations of climatic indices. For each pixel of the grid, we performed multilinear regression models using all possible combinations of the significant drivers. The selection of the best regression models was based on an out‐of‐sample procedure, and the model performance was tested by comparing the predicted BA with the observed data, estimating explained variance and correlation. While rising temperatures are often assumed to be the main driver of BA under climate change, our study revealed that low precipitation and water balance deficit from December to March played the most significant role in influencing BA during the winter fire season.

At 1 km resolution, NDVI projections for heterogeneous landscapes can appear spatially coherent in aggregate while concealing substantial class-level prediction weaknesses, a limitation that has received limited systematic attention in the NDVI projection literature. This study applies a four-component assessment workflow to Northeast China (NEC) for 2040 under SSP1-2.6, SSP2-4.5, and SSP5-8.5, integrating multi-stage model selection, land-cover-stratified validation, quantile-regression-based uncertainty characterization, and validation-priority ranking. Among three candidate tree-based models evaluated using spatial block cross-validation, temporal holdout validation, long-jump extrapolation, and climatic perturbation tests, LightGBM showed the most balanced and consistent performance, with spatial CV R2 = 0.654 ± 0.123, temporal holdout R2 = 0.710, and long-jump R2 = 0.671, and was therefore selected for the 2040 projection. Projected regional mean NDVI increased modestly from 0.393 in 2020 to 0.414–0.417 across scenarios, with limited divergence among SSP pathways at this near-term horizon. Class-stratified validation of the 2020 holdout prediction revealed that global model performance masked strong class-level heterogeneity, with R2 values ranging from 0.576 for Construction land to −0.886 for Unused land. Water bodies and Unused land exhibited negative R2 values, indicating weak class-level predictive support relative to a simple class-mean benchmark. Residual decomposition showed that Water bodies combined high random error with elevated systematic deviation, whereas Unused land was mainly characterized by systematic bias, suggesting different needs for class-specific model improvement. The Uncertainty Risk Index (URI), derived from 95% prediction intervals, was highest in Construction land and lowest in Cropland across all scenarios. Integrating historical residuals with future URI-identified Water bodies, Unused land, and Construction land as the highest-priority classes for future targeted validation. These priorities arise from both limited class representation and intrinsic NDVI-related complexity, including low vegetation signal, mixed-pixel effects, and heterogeneous land-surface composition. These results demonstrate that land-cover-stratified error decomposition and uncertainty-informed priority ranking reveal class-specific projection limitations that aggregate accuracy metrics can conceal.

Cloud contamination obscures ground objects, interferes with surface reflectance, and disrupts spatial continuity. In thick-cloud regions, surface structures and spectral information are often extensively missing. CNN-based cloud removal methods can recover local textures, but they are less effective at modeling global structures and color consistency over large cloud-covered areas. Transformer-based methods capture long-range dependencies; however, standard self-attention introduces high computational and memory costs for high-resolution remote sensing images. Efficient attention reduces this cost but may weaken edge and texture discriminability. SAR imagery can penetrate clouds and provide surface structural information, yet repeated SAR injections may propagate speckle noise, cross-modal misalignment, and imaging discrepancies through deep restoration layers. To address these issues, this paper proposes DST-SARNet, a dual-stage SAR structural guidance network for optical remote sensing image cloud removal. In this framework, dual-stage refers to two explicit SAR-guidance positions: early structural skeleton guidance at the input side and late high-frequency modulation near the output. The Texture-Aware Asymmetric Retrieval module is placed between these two stages as a bottleneck memory retrieval operation rather than as a third dense SAR injection stage. With this design, SAR provides structural skeletons, readable texture memory, and terminal detail compensation, while the optical branch remains responsible for color, semantics, and spectral appearance recovery. Experiments on the SMILE-CR and SEN12MS-CR datasets show that DST-SARNet effectively restores cloud-contaminated imagery with a compact model scale, demonstrating its potential for efficient SAR-assisted optical cloud removal.

Hyperspectral image (HSI) classification remains a challenging task due to high spectral redundancy, complex spectral–spatial correlations, and the limited availability of labeled samples. To address these issues, this paper proposes a novel framework termed M3-Mamba, which integrates language-guided, multi-level, and Mamba-based spectral modeling for hyperspectral image classification. The proposed M3-Mamba leverages high-level semantic priors derived from multimodal representations to guide discriminative spectral modeling, enabling effective interaction between semantic information and fine-grained spectral features. In addition, a frequency-aware Mamba-based state space module is introduced to efficiently capture long-range spectral dependencies while avoiding the quadratic computational complexity of conventional attention mechanisms. Meanwhile, a text-guided modulation strategy is designed to adaptively reweight spectral responses under semantic guidance, suppressing redundant or noisy bands and enhancing class-relevant spectral responses without compromising spectral fidelity. This semantic-to-spectral modulation allows M3-Mamba to better cope with spectral variability and inter-class confusion. Extensive experiments conducted on four widely used benchmark datasets, including Indian Pines, Pavia University, Salinas, and Houston datasets, demonstrate thatM3-Mamba achieves competitive overall accuracy, average accuracy, and Kappa coefficient under the adopted benchmark settings. Ablation studies further validate the effectiveness of each key component, confirming that the proposed framework demonstrates promising effectiveness for hyperspectral image classification.

Accurate three-dimensional representations of tree structure are essential for fire modeling, radiative transfer simulation, synthetic data generation, and digital twins of forests, yet detailed 3D structure is rarely available at required scales. Current approaches approximate crowns with smooth geometric primitives, discarding the clumping, gaps, and irregular branching present in real trees. We present TreeFlow, a conditional flow matching model that generates realistic 3D tree point clouds from species, acquisition platform, and height. The model uses a transformer trained on real laser scanning data from the FOR-species20K benchmark to learn a velocity field transporting samples from a Gaussian distribution to the source data distribution. We evaluate generation quality by comparing conditioning and distributional fidelity metrics to scans of real trees. Generated trees match or approach the intra-class baseline on five of six metrics, with a Chamfer distance of 0.581 m versus 0.559 m for real trees of the same genus and height class. Performance is strongest below 25 m and degrades with increasing height. TreeFlow generates individual-tree point clouds conditioned on scalar inventory attributes using a model trained entirely on real laser scanning data.

Detecting weak radar targets in complex sea conditions is inherently challenging due to non-stationary sea clutter and sea spikes. Furthermore, traditional dictionary learning algorithms for clutter suppression suffer from high computational complexity. To address these issues, this paper proposes an efficient sea clutter suppression method cascading Block Coordinate Descent (BCD)-accelerated dictionary learning with Tunable Q-factor Wavelet Transform (TQWT) denoising. During dictionary learning, a BCD strategy replaces global Singular Value Decomposition (SVD) with analytical optimization. Combined with an adaptive soft-thresholding operator, this enables low-complexity joint optimization of dictionary atoms and sparse coefficients, drastically reducing training time. Subsequently, a batch-adaptive Orthogonal Matching Pursuit (OMP) algorithm featuring Gram matrix precomputation and a dual-stop mechanism achieves efficient reconstruction and preliminary cancellation of clutter components. Finally, TQWT is applied to filter out residual non-stationary clutter and noise by leveraging its narrowband feature representation and shift invariance. Experiments on measured radar data from the IPIX database and datasets published by the Journal of Radars demonstrate that the proposed method significantly outperforms traditional K-SVD-based algorithms. Specifically, it improves the average signal-to-clutter-plus-noise ratio (SCNR) by 17.48 dB and requires a total execution time of only 7.99 s, achieving a highly favorable trade-off between suppression performance and computational efficiency.

Spaceborne imaging spectroscopy has created new opportunities for monitoring soil properties at regional scales. Its use for predicting soil heavy metal concentrations in mountainous environments, however, remains insufficiently tested, especially when EMIT hyperspectral data are used. In this study, EMIT Level-2A surface reflectance data were integrated with DEM-derived terrain variables to estimate soil arsenic (As), copper (Cu), and zinc (Zn) concentrations in Renhuai, Guizhou Province, Southwest China. Only soil samples falling within valid EMIT coverage were used for element-specific modeling, resulting in 139 samples for As, 136 for Cu, and 130 for Zn. To reduce redundancy among predictors, EMIT spectral variables and terrain factors were screened before model construction. Random forest and XGBoost models were then tested using repeated spatial cross-validation. The best-performing model for As combined EMIT predictors with elevation and achieved a validation R2 of 0.460. Model performance was considerably weaker for Cu, with a validation R2 of 0.188. For Zn, the model failed to outperform the mean-based benchmark, producing a negative validation R2 of −0.028. The spatial prediction maps and residual patterns suggested that the EMIT-based prediction showed moderate potential for As, limited predictive value for Cu, and exploratory rather than reliable mapping capability for Zn under the current sample and predictor conditions.

Small object detection is fundamentally constrained by the lack of discriminative fine-grained features. Although introducing higher resolution detection scales can improve performance, it also amplifies background noise. In addition, the independently decoupled design of conventional detection heads is insufficient to address the persistent challenges of missed detections and false positives for small objects. To this end, we propose MSIA-YOLO, a YOLOv11-based detector with multi-scale semantic interaction and alignment, optimized from three complementary perspectives: feature modeling, high resolution semantic compensation, and task coordinated alignment. First, Receptive Field Attention Convolution (RFAConv) is integrated into the backbone to enhance critical local details, such as edge and texture cues, via receptive field aware attention. Second, to alleviate fine detail attenuation caused by repeated downsampling, we construct a CHSP-P2 small object detection framework with an additional P2 branch. A scale sequence fusion mechanism is further introduced to perform high resolution semantic compensation through cross scale hybrid inputs. Finally, we design a DTIA-Head (Dynamic Task Interaction and Alignment Head), which promotes joint optimization of classification and localization through dynamic task interaction and spatial alignment. Extensive experiments on the public datasets VisDrone, TinyPerson, and RSOD show that, compared with the YOLOv11n baseline, MSIA-YOLO improves mAP50 by 7.7%, 10.3%, and 1.0%, respectively, while also outperforming several advanced detectors. These results demonstrate the effectiveness and generalization capability of the proposed method in small object, dense object, and complex scene object detection scenarios.