Surface nitrogen loading can accumulate in the vadose zone, posing a critical long-term threat to groundwater quality. In the Daqinghe (DQH) Basin, decades of intensive groundwater extraction have deepened the water table and substantially increased vadose-zone thickness. This study employs SWAT and MODFLOW6 models, incorporating the Unsaturated Zone Transport (UZT) package, to simulate nitrate transport processes across the root zone–vadose zone–aquifer continuum and quantify basin-scale groundwater nitrate contamination. Model evaluation showed moderate agreement with observations, with R2 values of 0.581 for annual streamflow, 0.663 for groundwater depth, and 0.671 for groundwater nitrate concentrations. By explicitly representing nitrate storage and delayed transport within a dynamically changing thick vadose zone, the framework links long-term groundwater-level change to nitrate delivery to the aquifer. From 1996 to 2016, progressive water-table decline increased simulated vadose-zone nitrate storage by 331%, confirming its role as both a temporary buffer and a legacy contaminant reservoir. Under the managed-recovery scenario, the proportion of the plain area exceeding the groundwater nitrate standard is projected to increase from 28.14% in 2020 to 72.3% in 2035. Comparisons among future pumping scenarios indicate that this deterioration is driven largely by the delayed arrival of legacy nitrate, while groundwater recovery may further alter its transfer to the aquifer. Groundwater-level restoration may therefore precede water-quality improvement by decades. Sustainable management should integrate groundwater recovery, nitrate monitoring, and agricultural nitrogen control rather than treating water quantity and quality separately.

Ratoon rice plays a vital role in boosting land productivity and contributing to stable food supplies under the context of global climate change. This system offers these advantages by demonstrating the capacity to produce an additional grain yield of 5-6 t ha-1 from the ratoon season, while simultaneously reducing the growth duration by 44-48 days compared to conventional double-season rice cultivation. However, its precise spatial distribution remains unclear under varying climatic and surface conditions, and remote sensing-based monitoring of ratoon rice has received limited attention. Existing methods often struggle to accurately distinguish ratoon rice from other morphologically similar types, such as double-season rice, and are further hampered by frequent cloud cover and rainfall, compromising optical sensing effectiveness. This study proposes a two-step ratoon rice (TSRR) mapping method using multi-source remote sensing data at the parcel scale. The TSRR method integrates Sentinel-1A synthetic aperture radar (SAR) and Sentinel-2 optical imagery, utilizing the distinct growth characteristics of main-season and ratoon rice. It employs object-based segmentation and two novel indices-the SAR-based paddy rice index (SPRI) and the SAR-based ratoon rice index (SRRI)-without relying on detailed phenological information. Results indicate that the TSRR method effectively distinguishes ratoon rice from other paddy rice types, achieving an average overall accuracy (OA) of 0.87, with particularly high performance in separating ratoon rice from double-season rice. The TSRR method demonstrates strong robustness and transferability across different regions, which can provide a reliable solution for large-scale paddy rice mapping, especially in cloud-prone areas with limited optical data availability, and offers valuable support for crop monitoring, yield estimation, and national agricultural inventory initiatives.

Urban green spaces (UGS) support health and urban resilience, yet realized access depends on both proximity and travel options, especially in transit-oriented cities. Older adults often face higher impedance from slower walking, transfer sensitivity, and first- and last-mile burdens, so walking-only or age-neutral measures may understate inequalities when trips require multimodal travel. This study assesses UGS accessibility in Hong Kong using an age-stratified Gaussian two-step floating catchment area framework (AS-G2SFCA) in the time domain. We model walking, Mass Transit Railway (MTR), and bus travel for younger adults aged 15–64 and older adults aged 65+, distinguish urban parks from country parks, and summarize distributional inequality using a disadvantage-weighted Gini while identifying low-access clusters via local indicators of spatial association. Results show that urban-park accessibility is consistently higher than country-park accessibility, with larger cohort differences for destination-oriented country parks than for neighbourhood-oriented urban parks. Within the modelling framework, incorporating public transport is associated with higher accessibility for both cohorts and lower disadvantage-weighted inequality, with weighted Gini coefficients decreasing from 0.42 (younger) and 0.46 (older) under walking-only conditions to 0.16 and 0.14, respectively, under multimodal access. Bus-based accessibility surfaces are more spatially even than MTR-based surfaces, while accessibility exhibits strong spatial clustering, with global Moran’s I values of 0.96 and 0.95 for younger and older adults, respectively. Neighbourhood disadvantage is weakly associated with accessibility at the street-block scale, highlighting mode-specific transit coverage and first- and last-mile conditions as key levers to narrow age-related inequities in green access.

The rapid evolution of satellite-borne Earth Observation (EO) systems has revolutionized terrestrial monitoring, yielding petabyte-scale archives. However, the immense computational and storage requirements for global-scale analysis often preclude widespread use, hindering planetary-scale studies. To address these barriers, we present Embedded Seamless Data (ESD), an ultra-lightweight, 30-m global Earth embedding database spanning the 25-year period from 2000 to 2024. By transforming high-dimensional, multi-sensor observations from the Landsat series (5, 7, 8, and 9) and MODIS Terra into information-dense, quantized latent vectors, ESD distills essential geophysical and semantic features into a unified latent space. Utilizing the ESDNet architecture and Finite Scalar Quantization (FSQ), the dataset achieves a transformative ~340-fold reduction in data volume compared to raw archives. This compression allows the entire global land surface for a single year to be encapsulated within approximately 2.4 TB, enabling decadal-scale global analysis on standard local workstations. Rigorous validation demonstrates high reconstructive fidelity (MAE: 0.0130; RMSE: 0.0179; CC: 0.8543). By condensing the annual phenological cycle into 12 temporal steps, the embeddings provide inherent denoising and a semantically organized space that outperforms raw reflectance in land-cover classification, achieving 79.74% accuracy (vs. 76.92% for raw fusion). With robust few-shot learning capabilities and longitudinal consistency, ESD provides a versatile foundation for democratizing planetary-scale research and advancing next-generation geospatial artificial intelligence.

Mangroves are the resistant species found in the intertidal zones, providing ecosystem services such as protection of shorelines, provision of habitats to flora and fauna, and contributing to nutrient cycling. Study of their leaf properties has always been challenging, but this has been facilitated by the advent of Hyperspectral Imaging (HSI) systems. In such a context, this study undertook the development of a hyperspectral library offering the reflectance characteristics for adaxial and abaxial surfaces of mangrove species found in Hong Kong, on the temporal scale of seven days to facilitate the species identification and monitor the leaf decay. This library contained species level data, plot level data, and decay level data. Field surveys in fifteen plots (900 m2 each) conducted in the Eastern and Western regions of Hong Kong collected hyperspectral data of five mangrove species, namely: Ceriops tagal, Kandelia obovata, Avicennia marina, Avicennia germinans, and Aegiceras corniculatum, using two different types of HSI systems i.e., Specim IQ (in-field data) and NEO Hyspex (in-lab data) hyperspectral cameras. A comparison of sensors unveiled a notably higher reflectance in field collected data than that of the lab-collected data, with a range of 11.8 % (Kandelia obovate) to 73.1 % (Aegiceras corniculatum). The Root Mean Square Error (RMSE) indicated deviation between the two sensors, i.e., 0.211 for Ceriops tagal, followed by Kandelia obovata (0.233), Avicennia marina (0.317), Avicennia germinans, and Aegiceras corniculatum (0.349). This freely available comprehensive hyperspectral library will serve as the foundation for training datasets to achieve automated classification with enhanced accuracy. This open access hyperspectral library will assist the researchers to relate the physiological and anatomical variations in leaves with the changes in hyperspectral reflectance on the temporal scale.

Ground crop yield records are typically published only at regional scales to protect farmers’ privacy. To predict regional crop yields, remote sensing and meteorological variables are commonly aggregated (e.g., averaged) to match the spatial scale of ground yield records. However, spatial aggregation of remote sensing data—such as the simple averaging of pixel-wise vegetation index (VI) time series—ignores intra-regional phenological variations. Several vegetation phenology studies unveiled a phenological bias caused by the simple average, which distort the representation of crop-specific phenological characteristics and subsequently compromising the accuracy of crop yield modeling. To address this critical limitation, we therefore proposed a novel yield prediction framework that explicitly incorporates intra-regional phenological heterogeneity. Instead of relying on aggregated VI time series, our method first applies adaptive clustering to automatically group pixels according to their growth dynamics. The resulting subregions and their respective area proportions are then integrated into a customized loss function within a convolutional neural network (CNN)-based modeling framework, termed Group-CNN. We evaluate the performance of Group-CNN in predicting soybean yields across 13 states in the U.S. Corn Belt, comparing it against multiple benchmark models. Results demonstrate that Group-CNN consistently outperforms the benchmarks, achieving a lower root mean square error (RMSE) (5.16 vs. 5.94–6.76 bu/ac), a higher coefficient of determination (R2) (0.83 vs. 0.72–0.76), and a lower mean absolute percentage error (MAPE) (9.78% vs. 11.72–12.81%). Ablation experiments confirm a notable decline in model performance when intra-regional phenological variations are excluded. This study highlights the importance of accurately representing intra-regional crop phenological characteristics—an often-overlooked factor in current yield prediction models.

Urban nighttime vitality significantly contributes to a city's economic, social, and cultural attractiveness. This study develops a dual-perspective framework that integrates annual mean nighttime light (NTL) intensity indicating stable infrastructural capacity, with a novel Nighttime Light Fluctuation Index (NLFI) capturing the dynamic pulse of human activity. Rather than measuring vitality directly, the framework extends intensity-based vitality evaluations by incorporating temporal dynamics, enabling the categorization of urban zones by distinct vitality-related patterns. Applying the framework to 25 global megacities, with detailed analyses in Los Angeles and Beijing, we identified three distinct patterns: "Bright-Fluctuating", "Bright-Steady", and "Low-Lit" zones. A consistent cross-city association was found between these patterns and urban functions: residential areas consistently exhibit low fluctuation, while commercial, industrial, and transportation hubs are hotspots of high fluctuation. Spatially, fluctuation hotspots in Los Angeles are concentrated in its central commercial and activity hubs, whereas in Beijing, they are also prominent in newly developing areas in the urban periphery, unveiling different urban dynamics and development. An XGBoost model combined with SHapley Additive exPlanations (SHAP) analysis revealed the key built-environment drivers behind these patterns. In Los Angeles, functional diversity and density (e.g., non-residential land use) are the primary drivers. In contrast, Beijing's patterns are predominantly shaped by infrastructure, with distance to the metro network being the most critical factor. These findings demonstrate the value of capturing NTL's temporal dynamics in understanding the interplay between built environment and nighttime urban dynamics, offering a powerful tool for urban planning and policy.

The integration of Landsat and Sentinel-2 observations into a virtual constellation offers unprecedented potential for near-daily medium-resolution (10-100 m) Earth monitoring. However, effective joint use of these datasets remains limited by intrinsic spectral and spatial inconsistencies arising from differences in sensor design, spectral band configurations, and spatial resolutions. To address these challenges, this study introduces the Spectral and Spatial Harmonization Model (SSHM)--a unified two-stage framework for harmonizing Landsat-7/8/9 and Sentinel-2 A/2B observations. In the first stage, SSHM estimates location-specific blur kernels for coarsening Sentinel-2 imagery to 30-m resolution, followed by cross-sensor spectral harmonization using time-series harmonic analysis and pixel-wise linear transformations. In the second stage, Landsat data are refined to 10-m resolution using an iterative optimization approach that jointly enforces temporal continuity with 10-m Sentinel-2 time series and spatial consistency with 30-m Landsat data using the pre-estimated blur kernels. Experimental results across 22 globally distributed study sites show that SSHM effectively reduces sensorinduced spectral discrepancies-by 5% between Landsat OLI and Sentinel-2 MSI, 11% between Landsat OLI and ETM+, and 22% between Sentinel-2 MSI and Landsat ETM + -outperforming existing harmonization approaches while maintaining superior robustness to observational noises. Moreover, SSHM outperforms conventional resampling and representative spatiotemporal fusion algorithms in refining Landsat to 10/20-m resolution, reducing reflectance discrepancies: 25% in the blue band, 40% in green, 43% in red, 50% in nearinfrared (NIR), 67% in shortwave infrared 1 (SWIR1), and 71% in SWIR2 (compared to nearest neighbor sampling). These advances of SSHM enable the generation of more consistent and spatiotemporally coherent 10/30m harmonized Landsat and Sentinel-2 reflectance products that are essential for monitoring fine-scale, rapid, and heterogeneous land surface changes.

Autumn leaf fall is a critical phenological event in temperate deciduous forests, with important ecological and socioeconomic implications. Traditional estimates based on daytime vegetation indices primarily capture foliage color changes rather than the actual timing of leaf fall, and are often affected by mixed-pixel effects that reduce the accuracy. This study proposes a novel workflow for detecting the autumn leaf fall date (LFD) using nighttime light (NTL) data in urban areas, validated through in-situ observations from phenological cameras and city-scale assessments. Results showed that NTL-derived LFDs closely matched in-situ observations across three cities (New York City, Boston, and Beijing), with an RMSE of around 5 days and a bias of 0.77 days. In Beijing, both interannual (2012-2024) and spatial variations (2024) in LFD were delayed by higher preseason temperature and precipitation but advanced by greater strong-wind frequency, consistent with known autumn phenological controls and supporting the reliability of the NTL-based approach. These results demonstrate that NTL data can provide more accurate and interpretable LFD estimates than traditional daytime remote sensing, enabling detailed city-scale mapping. The use of NTL-derived LFD dynamics facilitates a more comprehensive understanding of their spatiotemporal patterns in urban environments and their linkages to climate change and human activities.

Accurate mapping of forest carbon (C), nitrogen (N) and phosphorus (P) stocks is essential for advancing our understanding of global biogeochemical cycles. However, compared to carbon, large-scale quantification of forest N and P pools remains limited. We developed a two-step machine learning-based framework for estimating forest aboveground C, N and P stocks by integrating field measurements, synthetic aperture radar (SAR) and optical remote sensing data, and soil and climatic variables. We tested the method in the permafrost region of the Greater Khingan Mountains, located at the southeastern margin of the boreal forest. We first introduced the fractional vegetation cover to adjust SAR backscatter coefficient, which substantially improved aboveground biomass (AGB) estimation compared with the existing China AGB map (R & sup2; value increased from 0.22 to 0.53). We then developed a novel triangular index based on time series of vegetation indices to represent vegetation nutrient uptake and accumulation. This index, together with AGB and auxiliary predictors, was used in a Gaussian process regression model to estimate aboveground C, N and P stocks. The resulting estimates demonstrated higher accuracy than existing datasets, with R & sup2; values improving from 0.18, 0.01 and 0.44 to 0.83, 0.76 and 0.77 for C, N and P stocks, respectively. These improvements were largely attributed to the inclusion of both the triangular index and AGB as key predictors in the model. This study presents an effective approach for large-scale mapping of aboveground C, N and P stocks in boreal forest ecosystems, offering support for assessments of global carbon and nutrient cycles and for climate change research. The forest AGB, C, N, and P datasets over 2007-2010 and 2015-2023 produced using the method in this article, for the Greater Khingan permafrost region in northeastern China, are available in https://data.tpdc.ac.cn/en/data/61a8b5f9-9cd1-4e78-8f8c-f35bcce5cdba. Accurately quantifying forest aboveground carbon, nitrogen, and phosphorus stocks at large scale is fundamental to advancing understanding of global biogeochemical cycling. This study introduces a two-step framework that integrates machine learning with radar and optical remote sensing data, thereby improving the estimation of aboveground carbon, nitrogen, and phosphorus stocks in boreal forests in the permafrost region of the Greater Khingan Mountains. (sic)(sic): (sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic) (R & sup2;(sic)0.22(sic)(sic)(sic)0.53) (sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic).(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)R & sup2;(sic)(sic)(sic)0.18, 0.01(sic)0.44(sic)(sic)(sic)0.83, 0.76(sic)0.77.(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic).(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic), (sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)(sic)2007-2010(sic)(sic)2015-2023(sic)(sic)(sic)(sic)(sic)(sic)(AGB),(sic)(C),(sic)(N)(sic)(sic)(P)(sic)(sic)(sic), (sic)(sic)https://data.tpdc.ac.cn/en/data/61a8b5f9-9cd1-4e78-8f8c-f35bcce5cdba(sic)(sic).

Very-High-spatial-Resolution (VHR, <= 5 m) remote sensing products (e.g., land cover maps) are critical for providing detailed and scalable information to support broad applications (e.g., urban, agriculture, ecology, and forestry). However, accurate VHR products require imagery that is both spatially detailed and spectrally rich, which is rarely met by a single sensor. Globally available PlanetScope (eight 3-m bands) and Sentinel-2 (thirteen 10/20/60-m bands) are complementary in this regard, making them ideal for synergistic use. Conventional spatial-spectral fusion approaches, such as pansharpening and hypersharpening, however, are tailored to specific sensor modalities and thus ill-suited for blending multispectral images from these two constellations. To bridge this gap, we advance the concept of Multispectral-to-Multispectral sharpening (M2Msharpening) and propose RASSFM 2.0, an enhanced M2Msharpening model over the original Robust and Adaptive Spatial-Spectral image Fusion Model (RASSFM 1.0; Zhao and Liu, 2022) for fusing PlanetScope and Sentinel-2 imagery. RASSFM 2.0 incorporates three key improvements: (1) radiometric harmonization to align PlanetScope spectra with Sentinel-2, (2) inter-band sharpening to downscale 20-m Sentinel-2 bands to 10-m, and (3) all-band fusion to improve accuracy and efficiency. Quantitative and visual assessments across five global fusion sites with representative landscapes confirmed the superior performance of RASSFM 2.0 regarding spatial clarity, spectral fidelity, and processing time. Its practical utility was further validated through land cover classification at three larger classification sites with complex land covers, where a random forest classifier based on the RASSFM 2.0-fused image achieved the highest area-adjusted overall accuracy (90.97 +/- 0.08%), outperforming PlanetScope-only (85.04 +/- 0.10%), Sentinel-2-only (85.65 +/- 0.10%), PlanetScope-Sentinel-2 stacked (87.06 +/- 0.09%), and RASSFM 1.0-fused (88.98 +/- 0.08%) results. Notably, RASSFM 2.0 significantly reduces confusion between spectrally similar classes (e.g., sparse herbaceous and bare land) and improves the delineation of spatially complex surface structures (e.g., fragmented and small-sized objects). Furthermore, the contribution of RASSFM 2.0 to land cover classification is concrete in both pure and mixed pixels, particularly for mixed pixels. These findings demonstrate the effectiveness of RASSFM 2.0 in generating synthetic VHR imagery with rich spectra, greatly enhancing land cover classification across broad landscapes. As a transparent and physics-based model, RASSFM 2.0 serves as a robust standalone tool and can provide valuable physical priors to inform or constrain learning-based fusion methods. This work advances both upstream M2Msharpening methodology and its value in downstream applications.

The full potential of daily VIIRS nighttime light (NTL) data in characterizing human activity changes has been largely constrained due to pervasive data gaps induced by cloud contamination and poor-quality observations. While several gap-filling methods are available, their effectiveness and applicability remain limited by underdetected cloud contamination and in challenging data gap conditions. In this study, we proposed a novel spatiotemporal gap-filling framework to generate seamless daily VNP46 VIIRS NTL data. First, we refined the original VNP46 cloud mask to mitigate under-detected cloud coverage in the NTL data for subsequent gap-filling. Second, we identified pixels sharing spatiotemporal similarity to each target data gap pixel via a set of stringent search criteria. Third, based on similar pixel search outcomes, we developed a multi-strategy gap-filling approach leveraging spatiotemporal information from not only valid observations but also pixels that were data gaps at the target date. Evaluated with daily NTL images from 30 global cities, our results revealed a 39% cloud underdetection rate among pixels labelled as valid observations in the VNP46 product. Our proposed gap-filling method demonstrated an effective and robust performance in reconstructing the NTL intensity across various simulated data gap conditions, including abrupt NTL changes, cloud underdetection, and dense spatiotemporal data gap coverage (R2 = 0.80-0.85, RMSE = 2.13-10.84 nW/cm2/sr). We also found that our stringent similar pixel search criteria and multi-strategy gap-filling approach can better secure the quality of similar pixels, method applicability, and gap-filling accuracy compared to existing methods, particularly in challenging conditions. Furthermore, our method successfully reconstructed NTL images with dense data gaps during Hurricane Maria and provided seamless and consistent insights into the spatial heterogeneity of post-hurricane recovery. Our proposed gap-filling method and the resulting seamless daily NTL data will help to maximize its untapped potential by enhancing our understanding of the inherent characteristics of daily NTL data and unlocking new NTL-based applications.

Paddy rice cultivation can be implemented as single rice, double rice, and upland rice. Information on rice cropping patterns (CP) is crucial for enriching our understanding of sustainable intensification for a larger rice bowl. However, existing rice mapping efforts mainly focus on estimating planting area. It is challenging to produce annual rice CP datasets due to the difficulties of locating inundation signals from rice transplanting in complex-cropping regions. There is a lack of annual rice CP datasets in China in recent years, and there is an obvious omission error in existing rice algorithms in multiple-cropping regions. This study proposes a hierarchical Framework for the Automatic Mapping of complex Rice cropping patterns through the coupling of management practices and canopy structure. The challenge of the non-uniqueness of the "V" shaped feature in the Synthetic Aperture Radar (SAR) time series dataset was addressed by a phenology-assisted "V" shaped identification method using the Sentinel-1/2 datasets. The omission errors among multiple cropping regions were delivered using the hierarchy framework. This framework illustrated good performances among different rice cropping systems in China. This study developed the first 10-m annual rice Cropping Patterns (ChinaCP-Rice10m) maps over conterminous China from 2019 to 2024. An overall accuracy of over 90% was obtained when evaluated by ground-truth reference datasets. The upland rice cultivated areas have constantly increased, despite an overall declining trend of rice plantation areas since 2019. This study opens a new direction towards automatic complex rice cropping patterns by coupling management practices and canopy structure.

CONTEXT Climate warming is altering temperature conditions during key phenological stages of ratoon rice. However, systematic assessments of its temperature exposure risk and adaptation potential remain limited. OBJECTIVE This study aims to analyse changes in ratoon rice phenological windows, quantify climate risks during key growth periods, and evaluate the effectiveness of adjusting sowing dates as a strategy to adapt to warming. METHODS Integrating observations from 16 ratoon rice growing provinces in China with climate projections, we simulated phenological dates for the current period (2015–2024) and two future periods (2045–2054 and 2085–2094) across four SSPs (126, 245, 370, and 585) using a thermal time model. The temperature exposure risk was quantified via the Wasserstein distance, and an optimization framework was developed to determine the optimal sowing window for risk minimization. RESULTS AND CONCLUSIONS Warming advances phenology and extends the safe growth period. The current mean climate risk is 0.46, with heat stress concentrated in the Poyang Lake Plain and cold stress in southern Yunnan. Future climate risk will expand from existing hotspots. Under SSP585 for 2085–2094, the mean risk increases by 1.13, with the proportion of pixels exposed to heat stress increasing by 71.58%, far exceeding the 5.13% decrease in cold stress exposure. Compared with non-adjustment, optimized sowing reduces climate risk by 0.07–0.15, yet future risk remains higher than the current level. SIGNIFICANCE In this research, the ability of sowing date adjustment to enhance the climate resilience of ratoon rice was systematically evaluated. These findings suggest that site-specific and multidimensional adaptation strategies are needed to sustain ratoon rice production under future warming.

Change detection (CD) is an important task in Earth observation. In the past few years, significant progress has been made in supervised CD research; however, change labels are extremely expensive. The semi-supervised CD has attracted increasing attention. In semi-supervised CD, the problem of scarcity of positive samples is magnified. The imbalance of change types (e.g., disappearance and appearance), moreover, exacerbates the missing detection phenomenon. To address the above problems, we propose a semi-supervised CD method: CutMix-CD, which incorporates the change-aware CutMix augmentation into the consistency framework of CD. The semi-supervised learning framework enriches change contexts and places special emphasis on the comparative process, facilitating more robust representations of changes with improved generalization capabilities. First, mixed samples are synthesized using the change-aware CutMix operation. Then, we developed a student path and a teacher path to predict the changes in the original samples and mixed samples, respectively. Finally, the consistency loss is conducted between the two predictions to help the model learn the change information of unlabeled samples. In addition, an unsupervised feature constraint loss is proposed to further optimize the change features. Experiments on four datasets validate the effectiveness of CutMix-CD. It can effectively alleviate the overfitting problem for unbalanced types of changes and even outperforms the fully supervised methods for some challenging samples. The code will be released in https://github.com/SQD1/CutMixCD.

Recent advances in remote sensing technology have facilitated the emergence of high-quality hyperspectral satellite sensors with spatial resolutions comparable to well-established multispectral platforms like Landsat series and Sentinel-2. However, most hyperspectral satellite datasets suffer from limited temporal resolution, hindering the effective monitoring of rapid changes on the Earth's surface. To address this issue, we proposed an innovative fusion strategy named spectrotemporal fusion (SpecTF). Through SpecTF, high-frequency temporal information from multispectral images (MSIs) and narrow-band spectral information from hyperspectral images (HSIs) can be blended for applications that require high resolutions in both temporal and spectral domains. SpecTF first leverages a limited number of historical HSI-MSI pairs to learn the cross-sensor spectral mapping and then fuses this spectral mapping with broad-band time series to reconstruct narrow-band ones. The performance of SpecTF was evaluated using typical satellite datasets across six sites and a suite of field measurements. The average root mean square error (RMSE) and spectral angle of SpecTF are 0.0224 +/- 0.0142 and 3.3734 +/- 1.5476 degrees, respectively, which represent a 24.83 % and 33.23 % reduction in error compared to the second-best method. The experimental results demonstrate that the synthetic frequent narrow-band products exhibit satisfactory quality and improved accuracy of land surface parameter retrieval compared to real broad-band observations.

Human-labeled training datasets are essential for convolutional neural networks (ConvNets) in satellite image scene classification. Annotation errors are unavoidable due to the complexity of satellite images. However, the distribution of real-world human-annotated label noises on satellite images and their impact on ConvNets have not been investigated. To fill this research gap, this article, for the first time, collected real-world labels from 32 participants and explored how their annotated label noise affects three representative ConvNets (VGG16, GoogleNet, and ResNet-50) for remote sensing image scene classification. We found that 1) human-annotated label noise exhibits significant class and instance dependence; 2) an additional 1% of human-annotated label noise in training data leads to a 0.5% reduction in the overall accuracy of ConvNets classification; and 3) the error pattern of ConvNet predictions was strongly correlated with that of participant's labels. To uncover the mechanism underlying the impact of human labeling errors on ConvNets, we compared it with three types of simulated label noise: uniform noise, class-dependent noise, and instance-dependent noise. Our results show that the impact of human-annotated label noise on ConvNets significantly differs from all three types of simulated label noise, while both class dependence and instance dependence contribute to the impact of human-annotated label noise on ConvNets. Additionally, the label noise estimation algorithm (confident learning) cannot fully identify label noise. These observations necessitate a reevaluation of the handling of noisy labels, and we anticipate that our real-world label noise dataset would facilitate the future development and assessment of label-noise learning algorithms.

Nighttime light (NTL) data at daily scales presents an innovative foundation for monitoring human activities, offering vast potential across various research domains such as urban planning and management, disaster monitoring, and energy consumption. The daily moonlight-adjusted nighttime lights product (VNP46A2), sourced from Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (NPP-VIIRS), has been providing globally corrected daily NTL data since 2012. However, persistent challenges, such as fluctuations in the daily NTL series due to spatial mismatch and angular effects, as well as data holes, have significantly impacted the accuracy and comprehensiveness of extracting daily NTL changes. To address these challenges, a dataset production framework focusing on error correction, interpolation, and validation was developed. This framework led to the creation of a high-quality daily NTL (HDNTL) dataset from 2012 to 2024, which specifically targets 653 cities with populations predictably exceeding one million in 2025. A comparative analysis with the VNP46A2 dataset revealed promising results in spatial mismatch correction for two sample areas - the airport and highway (angular effect can be ignored). These areas exhibited reduced fluctuations in HDNTL time series and enhanced spatial consistency among pixels with homogeneous light sources. Furthermore, the correction of angular effects across various urban building landscapes demonstrated sound improvements, mitigating angular effects in different directions and reducing periodicity from the angular impacts. The spatiotemporal interpolation of data holes shows high similarity with the reference data, as indicated by a Pearson correlation coefficient (r) of 0.99, and it increased the valid pixels of all cities by about 2 %. The HDNTL dataset exhibited enhanced consistency with high-resolution Sustainable Development Science Satellite 1 (SDGSAT-1) NTL data regarding the NTL change rate. Also, it showed high alignment with ground truth data of power outages, showcasing superior performance in short-event detection. Overall, the HDNTL dataset effectively mitigates instability in daily series caused by spatial mismatch and angular effects observed in VNP46A2, improving data comparability across both time and space. This dataset enhances the ability of the NTL to reflect the ground events, providing a more accurate reference for daily-scale nighttime light research. Additionally, the dataset production framework facilitates easy updates from future VNP46A2 products to HDNTL. The HDNTL is openly available at 10.5281/zenodo.17079409 (Pei et al., 2025).

Land surface temperature (LST) data are crucial for global climate change research. While remote sensing data serve as a key source for LST, single-source sensor data often lack spatiotemporal continuity due to long satellite revisit intervals and cloud cover. Spatiotemporal fusion, which combines the strengths of multiple sources, can increase the available information. However, most current spatiotemporal fusion methods are designed for local-scale applications. This research proposes the Global Spatiotemporal Fusion Model (GLOSTFM) to generate global LST products. GLOSTFM, built on image pyramid principles, addresses the computational and complexity challenges of global-scale spatiotemporal fusion. Moreover, the model utilizes data from the novel Fengyun-3D satellite, which has a daily revisit capability and provides LST products separately derived from its thermal infrared (MERSI, 1 km) and microwave (MWRI, 25 km) sensors. By leveraging the cloud-penetrating capabilities of the microwave data to compensate for missing information, GLOSTFM increases the available information and reduces observational uncertainties. The results showcase high processing efficiency and enhanced spatiotemporal continuity, with an average RMSE of 2.874 K and an excellent R-2 of 0.980. The utility of the GLOSTFM model for monitoring urban heat island effects in Beijing was explored to illustrate one application among a broad range of potential applications of the proposed GLOSTFM that require global data on LST across the Earth's surface.