Although satellite remote sensing has been widely used to monitor land surface phenology, long-term, high-quality ground-based observations are essential for validating and calibrating these products, particularly for understanding agricultural ecosystem responses to climate change. However, such ground-truth datasets remain scarce, limiting the full potential of Earth system data applications. Here we present a comprehensive ground-based soybean phenology dataset for Northeast China, a region critical to national food security. The dataset integrates historical paper-based logs (1980s–1990s) with modern electronic records (2000s onwards) from 43 national agrometeorological stations operated by the China Meteorological Administration (CMA), spanning a 15° latitudinal gradient (38–53°N) across Heilongjiang, Jilin, and Liaoning provinces over 44 years (1981–2024). It provides the timing (day of year, DOY) of eight key phenophases (sowing, emergence, three-leaf, branching, flowering, pod formation, seed filling, maturity) and the durations of six intervening growth intervals. A multi-tiered quality assurance framework combining automated consistency checks, statistical outlier detection, and expert arbitration against agronomic principles and auxiliary climate data ensures high data integrity and completeness (>99%; paper era: 93.8%; electronic era: 98.1%). To facilitate reuse, we provide pre-computed kernel density estimates (probability distributions) and univariate linear regression trends for each station and phenological variable, along with 656 diagnostic plots. The dataset is openly available in the Science Data Bank under a CC BY 4.0 license as primary data tables (XLSX) and diagnostic plots (JPEG, 1200 dpi) (https://doi.org/10.57760/sciencedb.37245). As a long-term, spatially explicit, FAIR-compliant ground-truth resource, this dataset is valuable for validating satellite-derived phenology products, calibrating crop modules, and investigating climate-driven impacts on agricultural systems.
Arid regions of Central Asia (CA) depend heavily on snowmelt, but limited monitoring stations and unreliable snow depth (SD) data hinder effective snow monitoring. In this study, we applied XGBoost (XGB) and random forest (RF) machine learning models to integrate multiple SD products with covariates—including geospatial, snow-related, meteorological, land cover, and time data—along with in situ SD measurements, to enhance SD monitoring in CA. Two modeling approaches were tested: segmented modeling based on snow development stages and whole-period modeling for the entire snow season. The results showed that segmented fusion modeling outperformed whole-period modeling, with XGB-based models capturing SD variations more effectively than RF-based models and existing SD products. Notably, the XGB-based fusion models markedly improved the SD estimates for complex high-elevation terrain. Model performance analysis revealed that accuracy increased with the increasing number of integrated SD products. Moreover, the models incorporating ERA5-Land outperformed those that used the other datasets. While all covariate types enhanced the model performance, geospatial features made the most significant contributions. Ultimately, using the XGB-segmented fusion model, we developed a 0.25° daily SD dataset for CA spanning 1980–2023, providing a reliable resource for snow monitoring in the region.
As climate change continues to intensify atmospheric moisture demand across the American Southwest, predicting where wildfires cause severe vegetation mortality remains a major challenge. New thermal and optical harmonized satellite products provide emerging opportunities for evaluating how pre-fire vegetation moisture and water stress influence fire severity compared to other variables. We modeled continuous Relativized differenced Normalized Burn Ratio (RdNBR) for 45 wildfires that burned in the 2019 and 2020 fire seasons using random forest to compare Harmonized Landsat–Sentinel-2 vegetation indices, ECOSTRESS (ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station) evapotranspiration and evaporative-stress metrics, Daymet pre-fire weather variables, LANDFIRE fuels attributes, and topographic predictors across five cumulative pre-ignition windows from 1 week to 3 months before ignition. Despite exceptional drought and anomalous fire-weather conditions, low-severity fire was common over the study period. Integrated models outperformed individual product-family models, indicating that fire severity was best explained by combined antecedent weather, terrain, fuels, vegetation type, and vegetation-moisture predictors. Temperature, elevation, solar radiation, vapor pressure, and HLS Normalized Difference Moisture Index (NDMI) were among the most consistent predictors. NDMI was the most consistent HLS plant-moisture metric, but vegetation moisture alone did not explain fire severity. Canopy cover was also important in woodlands and forests. Models using data from shorter pre-ignition windows outperformed longer pre-fire window models. They varied in performance among vegetation types, with the strongest vegetation-specific performance in forests and woodlands. These findings show that emerging high-resolution satellite products can help characterize pre-fire vegetation conditions and improve fire-risk assessments, especially when integrated with short-term antecedent weather, landscape moisture, and terrain data.
Aviation emissions impose substantial pressure on the environment, necessitating the development of effective mitigation strategies. Effective emission reduction policies urgently require fine-grained aviation emissions data. The spatiotemporal resolutions of existing aviation emission datasets are largely confined to annual or monthly at the global or national scale, as shown in the EDGAR and IPCC reports, whereas fine-grained aviation emission datasets remain scarce. In this study, a technical framework is proposed to estimate instant emissions along real flight trajectories using Automatic Dependent Surveillance-Broadcast (ADS-B) data. Then, a daily aviation emission dataset in China is built that depicts carbon dioxide (CO2), nitrogen oxides (NOx), carbon monoxide (CO), and hydrocarbon (HC) emissions generated by all civil aviation activities. With the generated dataset, the emission characteristics considering the variation in aviation phases were analyzed, which were then aggregated to the corresponding airports, routes, and airlines. It is observed that average daily emissions of the civil aviation activities amount to 0.27Mt CO2, 1.34kt NOx, 0.23kt CO, and 0.02kt HC, respectively. The dataset reveals the heterogeneous spatiotemporal distribution of aviation emissions within different time periods across various regions. This moves beyond simple spatial mapping to assess airline environmental performance, manage airport air quality, and optimize flight routes.
Accurate land cover classification in mining areas is essential for environmental monitoring and resource management. However, this task is hindered by complex spectral-spatial characteristics and highly fragmented landscapes, where traditional single-source spectral analysis often fails to resolve “spectral ambiguity” among disparate land covers. To address this, we propose a Dual-Branch UNet (DBUNet), a dual-branch architecture that explicitly leverages multi-source heterogeneous features–integrating physicochemical spectral data, geometric topographic profiles (DTM), and biological vegetation indices (NDVI). The core innovations of DBUNet include: First, a dedicated branch for processing topographic and NDVI information, coupled with a Cross-branch and Cross-scale Interaction (ICMC-Trans) module and an Adaptive Fusion Module (AWS-Link), which synergistically captures discriminative elevation and vegetation priors. Second, an Edge Feature Enhancement (EFE) module employing learnable multi-scale Sobel kernels to adaptively refine complex boundaries. Moreover, we introduce a boundary-preserving loss function enhanced by the Segment Anything Model (SAM) to further supervise fine-grained segmentation. Experimental results on the Land Cover in Mining Areas (LCMA) dataset demonstrate that DBUNet achieves state-of-the-art performance with a 34.03% mIoU, outperforming recent Mamba and Transformer-based models. Furthermore, its robust generalization on the Gaofen Image Dataset (GID) underscores its transferability. The source code is available at: https://github.com/Venti666/DBUNet.
