Monitoring vegetation dynamics and optical proxies of photosynthetic activity is essential for understanding ecosystem functioning and forest responses to climate change. Optical remote sensing indices such as Normalized Difference Vegetation Index (NDVI) and Near Infrared of Vegetation Index (NIRv) are widely used as proxies for vegetation health and photosynthetic capacity, but their application is limited by cloud-induced temporal gaps. While SAR data offer weather-independent observations, NIRv reconstruction from SAR has rarely been investigated, particularly in forest ecosystems. In this study, we introduce two novel SAR-based indices, the Radar Volume Vegetation Index (RVVI) and the Radar Structure Vegetation Index (RSVI), and assess their relationship with NIRv as an established optical proxy for photosynthetic capacity. Using Sentinel-1 SAR and Sentinel-2 optical data, we analyzed species-specific relationships between a collinearity-reduced set of seven SAR vegetation indices and NIRv, and subsequently evaluated the predictive performance of the complete set of 15 SAR vegetation indices using a Foreward Feature Selection based Random Forest model. RVVI and RSVI showed low collinearity with existing SAR indices and the strongest correlations with NIRv, reaching up to |r| = 0.65 for deciduous species, while correlations were lower for conifers (|r| = 0.26). The cross-species Random Forest model achieved high predictive accuracy with an overall R2 of 0.85 and MAE of 0.03. However, performance differed between forest types, with higher predictive accuracy for deciduous species (R2 = 0.86 – 0.88) than for conifers (R2 = 0.77 – 0.79). RVVI and RSVI were identified as the most important SAR predictors. These findings demonstrate that the proposed indices provide complementary structural information that can improve the reconstruction of NIRv-like variability and show the potential to support more temporally continuous, cloud-independent monitoring of this optical proxy across forest ecosystems. However, the lower performance observed for coniferous species highlights the need for further refinement to fully capture species-specific NIRv variability.
Coastal ecosystems are critical components of the global carbon cycle, yet the reliability of satellite-derived gross primary productivity (GPP) products in these regions remains poorly understood. Compared with inland ecosystems, coastal ecosystems are exposed to distinctive environmental stresses, including tidal inundation and salinity stress, which challenge the applicability of existing GPP models primarily parameterized for terrestrial systems. Here, we conducted a comprehensive assessment of MODIS and GLASS GPP products across China’s coastal zones during 2001–2020. Product performance was evaluated from multiple perspectives, including spatiotemporal continuity analysis, inter-product comparison, consistency with vegetation indices, and validation against eddy covariance observations. Particular attention was given to variations across offshore distance gradients and vegetation types. Results revealed that GLASS outperforms MODIS in data completeness and spatiotemporal continuity. The two products exhibited generally moderate to high agreement (R = 0.60–0.95), varying substantially with season, offshore distance, and vegetation type. The weakest agreement occurred in summer and in evergreen forests. Correlations between GPP and three vegetation indices (NDVI, EVI and NIRv) generally strengthened with offshore distance, and GLASS exhibited stronger correlations with the three vegetation indices than MODIS. Validation against EC observations showed that GLASS generally achieved higher correlations at larger pixel windows and consistently lower RMSE and bias. MODIS and GLASS GPP showed widespread significant increasing trends, accounting for 43.9% and 50.8% of valid pixels, respectively, broadly consistent with the greening patterns indicated by NDVI, EVI, and NIRv, which showed significant increases across 59.1%–60.4% of valid pixels. Our findings highlight substantial and spatially heterogeneous uncertainties in satellite-derived GPP products in coastal ecosystems and underscore the need for coastal-specific model parameterization and improved representation of environmental stressors.
Land surface temperature (LST) plays a critical role in understanding the interactions between the land and atmosphere. Particularly, hourly LST data holds significant potential for a variety of applications, such as improving the understanding of daily fluctuations in surface heat fluxes, enhancing the accuracy of weather and climate forecasting, and supporting ecosystem management. The reconstruction of all-weather hourly LST is both essential and valuable. Currently, thermal infrared (TIR) LST data is often limited by infrequent observations and adverse weather conditions such as clouds and precipitation. These factors result in the generation of LST data at daily or even coarser temporal resolutions, with hourly LST datasets remaining scarce. Recently, advances in sub-hourly observations from geostationary satellites (e.g., Himawari-8), minute-by-minute ground truth data from automated meteorological stations, and hourly reanalysis datasets have made considerable progress. This raises the question: is it feasible to generate a gapless hourly LST product by integrating high-frequency observational multi-source data? In this study, to generate all-weather hourly LST, we applied an optimized Long Short-Term Memory (LSTM) model (referred to as ILSTM) to fuse multi-source data and produce hourly LST under cloud-free conditions. For cloud-covered conditions, a Dynamic Diurnal Temperature Cycle (DDTC) model was proposed to reconstruct hourly LST across China, which integrates real-time physical covariates (cloud optical depth, downwelling radiation, land cover, etc.) and uses Gradient-Boosted Regression Trees (GBRT) to dynamically parameterize diurnal LST trends. When validated against in-situ radiation flux stations, the results showed an RMSE of 2.876 K for cloud-free LST and 3.314 K for cloud-covered LST. Compared with MODIS LST, the reconstructed all-weather hourly LST data displayed consistent spatial coverage without significant boundary reconstruction anomalies. The findings of this study demonstrate the effectiveness and reliability of the proposed method for generating all-weather hourly LST data.
Optical Earth observation satellites provide scientists and coastal managers with multiple information on the seafloor, water column, environmental parameters, natural or human-induced pressures, ecosystem services or biodiversity. However, the use of remote sensing is strongly constrained by the presence of liquid water which attenuates the light signal depending on the wavelength and depth. All applications that require perceiving the bottom through the water column, such as habitat mapping and bathymetry estimation are thus affected by a limit of about 20 meters beyond which the light signal becomes too weak and noisy to provide accurate information. This study uses hypertemporal (HT) data cubes constituted by long time series of optical images from the Sentinel-2 satellites to improve data quality and significantly extend this commonly accepted limit to depths reaching 70 to 80 m. The generated multi-temporal composite images reveal details of the seabed in remote areas of the world's oceans, never before perceived from space. Satellite-derived bathymetric (SDB) estimates using this new data layers and Random Forest and XGBoost machine learning algorithms show very good correlations with reference data over the extended depth range at 100 m spatial resolution. At the study sites located in the Mediterranean Sea as well as in the Atlantic, Indian, and Pacific Oceans, where the waters are clearest, good estimates were obtained across the entire 0-80 m bathymetric range, with Mean Absolute Errors (MAE) of 3.21 m in Martinique, 2.62 m in Corsica, 1.88 m in La Reunion, and 3.17 m in New Caledonia. The two more turbid sites in the North Atlantic also maintained low MAE levels, even when the bottom signal was no longer visible, thanks to the water column properties information provided by the HT images: 4.74 m in Ouessant and 2.10 m in Glenan. Analyses of cloud cover, time series length, temporal statistics, geographical areas and water types specify the influence of these parameters on data quality as well as the limits of application of the models. The processing chains developed here, using freely accessible Sentinel-2 satellite data available worldwide, enables operational implementation for fast and large-scale data production.
This study develops a reproducible multi-sensor machine learning-based framework for mapping forest structure, biomass-related patterns, carbon stocks indicators, and ecological vulnerability in tropical forests of the West Usambara (WUSA) Mountains, Tanzania. Sparse but accurate ICESat-2 LiDAR canopy-height observations were integrated with Sentinel-1, Sentinel-2, and topographic data within a machine learning-based framework, in which several regression algorithms were compared and the optimized Random Forest approach was selected to generate a wall-to-wall 10 m resolution forest canopy height (FCH) map that represents broad spatial heterogeneity across the study area. Building on this, the resulting continuous FCH layer was incorporated as a structural predictor in RF models for estimating above-ground biomass (AGB) and above-ground carbon (AGC), which were calibrated and evaluated using available field inventory data collected across multiple forest reserves in the WUSA region. Results indicate that incorporating FCH provides modest relative improvements in AGB and AGC estimation compared to models based solely on optical and radar predictors, although absolute predictive accuracy remains moderate. Furthermore, feature-importance analyses showed that Sentinel-2 spectral bands dominated model performance (≈45–60%), while FCH provided consistent complementary structural information. The integrated FCH, AGB, and AGC layers were further used in a geospatial vulnerability assessment to identify spatial patterns of degradation risk and restoration potential, providing spatially explicit references to support conservation planning and management. To further strengthen the analysis, Monte Carlo-based uncertainty and sensitivity assessments were implemented to quantify uncertainty in AGB and AGC predictions and to evaluate the propagation of these uncertainties into the derived vulnerability maps. Results show that the vulnerability estimates were most sensitive to FCH and slope weights, whereas AGC exhibited comparatively low influence. The source code for this work is publicly available at https://github.com/EhAI4EO/TANZEO-BioStock.
Dengue fever has been a significant public health issue in Peru due to its highly localized and cyclical nature and to changes in the environment, demography, and climate. While many existing models can assess dengue risk at the country level, they often fail to account for the fine-scale landscape variation that drives varying transmission rates. We developed a spatiotemporal dengue risk model at a 1 km spatial resolution by integrating high-resolution climatic, land-cover, and demographic variables across 20 years (2003-2022) of dengue surveillance data. Tree-based approaches achieved higher predictive accuracy than linear models, with XGBoost achieving the highest R2 value of 0.68. Additionally, model performance increased when spatially and temporally lagged predictors were included. Land surface temperature and precipitation showed strong neighborhood effects on dengue suitability. Specifically, there exists a strong spatial dependency in model validation across 14 ecoregions, with Amazon moist forest and transitional Yungas yielding high predictive accuracy (R2 = 0.78), while high-altitude Andean Puna and coastal mangrove zones had poor predictive accuracy, performing worse than the mean prediction, suggesting non-climatic factors such as human mobility and imported transmission also play a role in dengue infection and transmissions. Overall, the predictability of dengue is strongly dependent on the ecoregion where it occurs, with national-level averages masking local dynamics. Our research demonstrates the potential for scalable high-resolution risk mapping when combined with ecoregion-specific performance metrics to predict and inform targeted vector control and public health interventions for climate-sensitive vector-borne diseases.
Wildfire simultaneously disrupts ecosystem carbon, water, and soil processes, yet these coupled effects and post-fire recovery trajectories cannot be reliably assessed from spectral vegetation indices alone. Here, we developed a fine-scale, multi-dimensional framework to quantify wildfire impacts and post-fire ecohydrological recovery, using the 2016 megafire in Great Smoky Mountains National Park in the eastern United States as a testbed. This framework integrated gap-filled 30 m NASA Harmonized Landsat and Sentinel-2 surface reflectance with an expanded diagnostic ecosystem model (Coupled Carbon and Water Model) to generate spatially consistent estimation of gross primary productivity (GPP), evapotranspiration (ET), water yield (WY), and soil erosion (SE) from 2014 to 2024 within Google Earth Engine. Fire effects were isolated from climate variability using pixel-level counterfactual simulations that produced no-fire baselines under identical meteorological forcing. We found that GPP declined by 20.7%, ET by 20.0%, WY increased by 19.2%, and soil erosion increased 12.5 times in the first post-fire year. Recovery trajectories varied strongly with burn severity, with high-severity patches retaining persistent modeled functional deficits (−13% GPP, −9% ET, +9% WY, and 5× SE) after eight years despite full spectral recovery (NDVI +2%). Our findings reveal a distinct decoupling between spectral recovery and the modeled carbon, water, and erosion responses, driven by incomplete ecological succession, where high-severity areas remain in herb- and shrub-dominated stages rather than recovering to forest. The severity-stratified diagnostic framework we provide offers a directly applicable tool for post-fire vegetation assessment, hydrological response monitoring, and long-term restoration planning in fire-affected ecosystems.
Terrestrial water storage anomaly (TWSA) is one of the key indicators reflecting the temporal variations of terrestrial water storage (TWS). Although traditional methods such as runoff monitoring achieve high precision, their spatial coverage limitations impede the comprehensive capture of large-scale hydrological patterns. The global navigation satellite system (GNSS) provides an innovative method for TWSA inversion. However, existing studies predominantly adopt a spatial resolution of 0.25° × 0.25° for GNSS-based TWSA inversion, which falls short of the resolution required for refined water resource analysis. Although this resolution adequately captures regional-scale TWSA variation patterns, it remains limited in characterizing local aquifer dynamics, sub-basin-scale water storage fluctuations, and hydrological response processes in complex topography areas. To address this gap, this study investigates the optimal trade-off between spatial resolution and accuracy in GNSS-based TWSA inversion through a threefold methodology. First, we integrate Pearson’s correlation coefficient (PCC) and root mean square error (RMSE) to formulate the spatial correlation accuracy (SCA), which underpins the development of the Multi-resolution Green function-based Terrestrial water storage anomaly Inversion Method (MGTIM). Second, the MGTIM-derived TWSA was systematically compared with traditional GNSS inversion results, GLDAS, and GRACE datasets, and the optimal spatial resolution of 0.05° × 0.05° was determined through SCA metric evaluation. Third, we apply the optimized MGTIM to invert and analyze the spatiotemporal evolution of TWSA in Southwest China (2011∼2022), integrating precipitation data for mechanistic interpretation. The results demonstrate that the PCCs between the optimal MGTIM and traditional GNSS inversion, GRACE/GRACE-FO, and GLDAS are 0.99, 0.88, and 0.83, respectively. The corresponding RMSEs are 7.87 mm, 12.57 mm, and 13.56 mm, confirming the effectiveness of MGTIM in TWSA inversion. In addition, the spatiotemporal consistency between TWSA and precipitation data obtained based on the optimal MGTIM is notable. MGTIM overcomes the limitations of traditional GNSS inversion methods and offers a novel approach for TWSA inversion with higher spatial resolution and accuracy.
Accurate monitoring natural and flooded water body areas is critical for natural resource management and disaster emergency response. Extracting water bodies over complex geographical scenes often struggles to balance spatial integrity and local detail accuracy, leading to problems like fragmentation, boundary loss, and omission of various water bodies, thereby reducing the reliability of the results. We propose the novel SWENet, a feature-collaborative convolutional neural network that utilizes transformer-assisted feature extraction under a multi-feature strategy. The framework synergistically strengthens linear and planar features of water bodies through edge, spectral, and texture features. A Transformer branch is introduced to capture supplementary contextual information, while a Bidirectional Gated Feature Fusion Module (BGFM) facilitates cross-feature interaction. A redesigned ASPP module fuses multi-scale features more effectively to strengthen the perception of global and local information. Comparative evaluations on SWF and FAS datasets demonstrate SWENet's performance against six leading SOTA segmentation models, with module-level functionality ablation experiments further verified via the XT dataset. The results indicate that SWENet can achieve outstanding efficiency in fragmented small water bodies and those affected by flood inundation in large-scale geographical context, demonstrating its comprehensive capabilities in various water body extraction tasks. This study can provide technical support for the monitoring of natural and flood-affected water bodies in areas with variable and complex hydrological conditions.
Accurate extraction of island coastlines from remote sensing imagery is critical for coastline dynamics analysis and coastal zone management. However, complex background interference, such as cloud cover and shallow water reefs, significantly hinders the segmentation of fine island details. To address these challenges and enable long-term dynamic monitoring, this study proposes a robust segmentation model, TransUNet-PCASAB. The model integrates a Position-Channel Attention Block (PCA-Block) into the encoder to capture image-specific positional and channel dependencies, effectively suppressing background noise. Additionally, Semantic Alignment Blocks (SABs) are introduced in the decoder to bridge the semantic gap and enhance boundary reconstruction. The TransUNet-PCASAB model was compared against other models, including UNet, DeepLabv3, PSPNet, DANet, SegFormer, Swin Transformer, BEFUNet, Mask2Former, and CFFormer. Results demonstrate that the TransUNet-PCASAB model achieved the best overall performance across all evaluation metrics and achieved superior segmentation performance in both global structures and local details. Furthermore, to validate the model’s capability for spatio-temporal Shoreline analysis, we applied TransUNet-PCASAB to monitor the morphological evolution of Barque Canada Reef from 2020 to 2025. The analysis revealed a 4.85% reduction in land area, quantitatively identifying substantial morphological changes that may be associated with anthropogenic activities. These results indicate the potential of the proposed model for large-scale, long-term shoreline monitoring in complex marine environments.
This study focuses on the 2020 M5.9 Dingri earthquake in Tibet, which occurred within a monthly M5.5 ± 0.2 hazard zone predicted by the China Earthquake Administration using NOAA OLR data. It retrospectively examines OLR, tidal force, GRACE-FO gravity, and GPS data across a broad region (17°N–55°N, 72°E−135°E) covering the pre-, co-, and post-seismic phases. The earthquake occurred exclusively during the trough phase of cycle B among three regional tidal cycles (A, B, C). A unique, localized NE–SW oriented OLR anomaly appeared near the epicenter during this cycle, displaying a sequence—slight increase → increase → minor attenuation → increase → quiescence—consistent with the infrared emission stages of progressive rock fracture. GRACE gravity images (June 2019–May 2020) revealed significant local gravity changes spatially adjacent and morphologically similar to the OLR anomaly. The gravity anomaly showed pre-seismic enhancement and rapid post-seismic decay, preceding and outlasting the transient OLR change. This spatiotemporally coupled sequence indicates a transition from deep stress accumulation (gravity signal) to shallow crustal failure (OLR signal) at a critical threshold. Integrated GPS analysis suggests a causative mechanism: northward compression from the Indian Block induced stress accumulation southwest of the epicenter, evidenced by co-located gravity and OLR anomalies. As stress was transferred into the fragmented Tibetan Plateau, it transformed into near-east-west tensional stress. Tidal triggering then initiated a normal-faulting rupture along a near-north-south structure (marked by an OLR surge), culminating in the M5.9 event.
Accurate, high-resolution forest canopy height information is critical for monitoring forest structure, estimating aboveground biomass, and informing land management and climate resilience strategies. In particular, remotely-sensed canopy height estimates can provide an estimate of fuel volume for wildland fire effects analyses, including in the Wildland Urban Interface where fuels in the home ignition zone are influential. This study presents a scalable machine learning framework to estimate canopy height across California by integrating Global Ecosystem Dynamics Investigation (GEDI) LiDAR Relative Height measurements with a suite of multi-sensor remote sensing data. The model was trained on more than 5.6 million rigorously filtered GEDI footprints from California and Arizona to minimize noise and ensure robust performance across ecologically diverse and topographically complex landscapes. Predictor variables included optical (Sentinel-2), radar (Sentinel-1, PALSAR-2), topographic, land cover, and a range of derived spectral indices and spatial-textural features. Model validation using an independent GEDI hold-out set demonstrated strong performance, with an R2 of 0.72, RMSE of 6.05 m, MAE of 4.31 m, and a slight positive bias of 0.80 m, closely matching the training metrics, suggesting a well-calibrated model. Compared to existing global canopy height products, the proposed regional approach shows improved agreement with GEDI observations in complex terrain and dense vegetation areas within California, while reflecting differences in training scope and objectives between regional and global models. A key outcome of this study is a 10-m resolution biennial canopy height time series for California (2016, 2018, 2020, 2022, and 2024) suitable for stand-scale and multi-epoch structural assessment. The resulting 10-m, biennial California canopy-height series for 2016–2024 is used to assess regrowth trajectories of canopy height and biomass in four major wildfire areas.
Agroforestry systems are highly heterogeneous and often underrepresented in global biomass products, creating challenges for carbon stock estimation. Accurate mapping of Above-Ground Biomass (AGB) and canopy structure is therefore essential for improving monitoring and reporting. In this study, we evaluate the use of GEDI-derived canopy structural variables, namely canopy cover (CC), 95th percentile height (RH95), and foliage height diversity (FHD) as intermediate predictors for AGB estimation across smallholder agroforestry landscapes. Using Sentinel-1, Sentinel-2, ALOS-2 PALSAR-2 and ancillary data inputs, we compared early, intermediate, and late multimodal fusion strategies within a U-Net architecture. Intermediate fusion achieved the best performance for structure retrieval, particularly for RH95 (r = 0.86 against airborne LiDAR). Incorporating FHD into an allometry-informed AGB model increased accuracy from an R2 value of 0.64 to 0.74, reducing the RMSE from 8.23 to 7.07 ton/ha, and decreased the bias from −1.27 to −0.56 ton/ha. These results demonstrate that structural diversity captured by FHD provides additional information beyond height and cover, and that fusion design influenced retrieval performance. The framework uses GEDI-derived structural traits as intermediate supervision to improve AGB estimation in data-scarce agroforestry systems with an allometry-informed and interpretable AGB model. The approach offers a scalable way for integrating LiDAR missions with multisensor satellite data for AGB monitoring and agroforestry planning.
Over the last decade, substantial progress has been made in the development of seamless all-sky Land Surface Temperature (LST) products. However, many agricultural and environmental applications require spatial resolutions finer than the currently available (≥1km). A practical solution to this limitation is to downscale existing all-sky datasets by extending methods originally developed for clear-sky data to also account for cloudy-sky conditions. In this study, we propose such an approach. Our method first generates a synthetic LST image representing average clear-sky conditions at the target spatial resolution, that then adjusts using downscaled residuals derived from the difference between the all-sky and synthetic LST. This approach offers two key advantages: it ensures physically realistic, fine-scale LST patterns by using the synthetic LST image as an initial guess, and amplifies the local weather effects in the all-sky LST, making them easier for the model to learn. To evaluate the method, we use half-hourly all-sky LST from the Spinning Enhanced Visible and Infrared Imager (SEVIRI), which we downscale from ∼5km to ∼1km and validate against independent in-situ and satellite LST. The results demonstrate improved performance relative to the original data across a range of land covers, topographies, and climate conditions, while also highlighting the potential of a single random forest (RF) model to predict the LST under both clear and cloudy conditions.
The Gravity Recovery and Climate Experiment (GRACE) and its Follow-On mission (G-FO) are promising tools as satellite-derived monitoring of terrestrial water storage (TWS) and groundwater storage (GWS) changes globally. However, the coarse spatial resolution of these datasets (ranging from 0.25° to 1°) limits their utility decision-relevant characterization of water resource changes. Herein, we used geospatial modeling approaches including machine learning (ML) (random forest (RF) and hybrid geographically weighted random forest (hybrid RFgw)) and spatial regression (geographically weighted regression (GWR)) in geospatial downscaling of three different GRACE datasets (JPL-SH, JPL-M, and CSR-M), to generate 1 km2 downscaled-GRACE data for purpose of quantifying water resources changes. Specifically, we quantified TWS anomalies (TWSA) and GWS anomalies (GWSA) across natural and managed landscapes in different catchments of the Indus Basin from 2002 to 2020. We evaluated the effectiveness of downscaled results against in-situ GWSA over 1238 monitoring wells. Results indicate that: (1) the hybrid RFgw model outperformed GWR and global RF models in predicting high-resolution TWSA and GWSA; (2) downscaled-GWS anomalies from the hybrid RFgw model agreed well with in-situ observational data, with higher R2 values for the downscaled JPL-SH (0.60 - 0.86), JPL-M (0.54 - 0.75), and CSR-M (0.54 - 0.85); (3) the highest TWS (−133.49 mm/year) and GWS (−150.38 mm/year) decline rates were detected in catchments dominated by managed landscapes; and (4) croplands experienced the largest storage losses (62-85 km3) followed by urban areas (8-15 km3) despite their much smaller spatial extent. The paper demonstrates the effectiveness of 1-km2 downscaled GRACE/G-FO datasets to provide localized information for water resource management across landscapes facing data-scarcity and water-food security challenges.
Leaf Area Index (LAI) retrieval using unmanned aerial vehicles (UAVs) is crucial for precision agriculture. Spatial window size (SWS) selection is a key step in LAI retrieval for average spectral reflectance extraction, which determines the optimal spatial neighborhood used to balance noise reduction and information preservation. Hybrid models combining physical and machine learning (ML) techniques have exhibited superior LAI estimation performance; however, the rational SWS selection specifically optimized for hybrid inversion frameworks remains understudied. To address this issue, this study proposed an SWS selection-based hybrid inversion framework for winter wheat LAI retrieval. Specifically, sensitive PROSAIL parameters were first determined through local sensitivity analysis to generate simulated spectral datasets. Subsequently, both simulated spectra and multi-scale UAV aggregated reflectance under diverse SWS were fused to construct the hybrid training dataset. Three ML-based nonlinear regression algorithms were further embedded into the framework to establish hybrid retrieval models with varying SWS settings. The performance of those models was evaluated using the coefficient of determination (R2) and relative root mean squared error (RRMSE), and significance tests were adopted to screen the optimal SWS. Our results demonstrated that with the increase in SWS, the LAI retrieval accuracy of both hybrid models and empirical models initially improved (R2 increase, RRMSE decrease) and subsequently reached a saturation state. Accordingly, the optimal SWS for LAI retrieval was determined to be 51×51 pixels at the jointing stage and 41×41 pixels at the heading-flowering stage. Overall, the hybrid model employing optimal SWS configurations (R2 = 0.63–0.65, RRMSE = 18.51%–19.68%) outperformed the empirical model under the same growth stages and optimal SWS (R2 = 0.27–0.53, RRMSE = 14.75%–21.09%) in cross-year validation. Collectively, our hybrid method demonstrates significant potential for providing practical reference for optimal SWS selection in winter wheat LAI retrieval.
Collection of polarimetric information in tandem with ranging data can greatly enhance the utility of lidar remote sensing. The three-dimensional (3D) data, coupled with measures of the polarization of received light, opens new application space and high-confidence discrimination for natural resource management, land cover mapping, object detection, and other uses. Polarimetric lidar sensing using Geiger-mode Avalanche Photodiode (Gm-APD) arrays opens yet more application space by leveraging large-area collection rates (ACR) and high altitudes made possible by the single-photon sensitivity and large formats of the focal planes. Whether polarization information collected by Geiger-mode Avalanche Photodiode (Gm-APD) array lidar systems has utility in material identification or characterization has not yet been determined and is complicated by the additional signal processing required compared to traditional APD-based lidar systems. A properly calibrated and configured polarized lidar system will not only record multiple channels of polarimetric information but also quantify the Stokes vectors and degrees of polarization of the returning signals. This paper demonstrates that there are visually interpretable, consistent quantitative differences in Gm-APD probabilities of detection (PDETs) across material type, outgoing polarization, and incoming polarization. Polarimetric Gm-APD array lidar systems should therefore be considered for applications where 3D data and a single intensity channel may not be sufficient for high-confidence classification
Inland water bodies are increasingly monitored from space, yet water level (H), surface area (A), storage volume (V), and water quality (Q) have historically been investigated through separate disciplinary traditions and sensor systems. Here, Q refers primarily to optically active constituents, including water transparency, chlorophyll-a (Chl-a), chromophoric dissolved organic matter (CDOM), and total suspended matter (TSM). This review reframes inland-water monitoring as a coupled, uncertainty-aware inference problem and proposes a unified H–A–V–Q framework. In this framework, satellite altimetry from nadir, SAR-mode, laser, and Surface Water and Ocean Topography (SWOT) Ka-band Radar Interferometer (KaRIn) missions, optical/SAR water mapping, and multi-/hyperspectral water-quality retrievals are jointly used to constrain the four state variables. Across a systematic-review corpus retained following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol, comprising 400 papers published between 2015 and 2025 from Web of Science, Scopus, and Google Scholar, two-dimensional water-extent retrieval reaches overall accuracy (OA) of approximately 74–99%. Water-level root-mean-square error (RMSE) ranges from a few centimetres for large and favourable lakes to several decimetres for small or morphologically complex water bodies, while Q retrievals achieve coefficient of determination (R2) values of approximately 0.4–0.8 for Chl-a, CDOM, TSM, and transparency. Storage uncertainty, denoted as σV, where σ represents standard-deviation- or RMSE-equivalent uncertainty, is dominated by area–elevation (A–E) and bathymetric errors rather than by A or H retrieval accuracy alone. Representative anchor values include A–E RMSE = 0.60 m and bathymetric RMSE ≈1.37 m. Explicit propagation of σ from H and A through V to Q allows the dominant error source to be diagnosed for individual basins, enabling targeted improvement. SWOT-anchored joint A–H reconstruction further reduces temporal-mismatch uncertainty and can lower σV by nearly an order of magnitude relative to pre-SWOT multi-sensor compositing, shifting the limiting factor toward the A–E–V relationship. The proposed H–A–V–Q framework provides a transparent scaffold for benchmarking and a practical foundation for operational, uncertainty-aware inland-water monitoring systems.
Wildfires alter surface soil moisture by consuming vegetation and exposing the soil to increased solar radiation, increased precipitation throughfall, and reduced infiltration from soil hydrophobicity. This can affect the water balance within the watershed and, in turn, influence post-fire nutrient cycling and ecosystem succession. Modeling surface soil moisture over large areas using satellite remote sensing can aid monitoring of soil moisture patterns and dynamics. We trained a random forest regression model using an in situ soil moisture network and examined pre- and post-fire soil moisture dynamics for three wildfires that burned in 2016 and 2017 in coastal California, U.S.A. Model inputs included Sentinel-1A synthetic aperture radar (SAR) and Landsat thermal infrared imagery (TIR), which produced a high resolution (30 m) soil moisture map time series with 18–20 estimates across 2 years. An unbiased root mean square error of 0.065 and an r2 of 0.624 indicates a good fit between the model and reference data. We found a significant interaction between treatment (wildfire or unburned) and period (before or after wildfire) by examining the modeled surface soil moisture in high-severity burned areas. Before-after control-impact metrics showed that modeled surface soil moisture was lower after each fire within wildfire areas compared to unburned controls, and overall changes in soil moisture were greater in the wildfire perimeters than unburned areas. These results demonstrate the potential of Sentinel-1 SAR and Landsat TIR data for the development of high-resolution soil moisture time series products to support monitoring and assessment of post-fire soil hydrologic conditions.
