Radiation physics processes are a core component of numerical weather and climate-Earth system models, governing the energy budget through longwave and shortwave fluxes and heating rates. However, their high computational cost and limited parallel scalability remain a major bottleneck for high-resolution integrations and large ensemble forecasts. Operational systems have reduced this burden through infrequent calls, coarse-grid treatments, and clear-sky simplifications, but such approximations can weaken cloud-radiation interactions and introduce accumulated errors and numerical instability. This review reframes these limitations as a problem of computational structure rather than incremental approximation and systematically surveys machine-learning-based radiation physics emulators by replacement scope (full-scheme, module-level, and hybrid/compound) and validation level (offline accuracy, online time-integration stability, and distributional robustness). Recent advances in deep learning architectures, physics-constrained training, three-dimensional and high-resolution effect emulation, and practical integration with Fortran-based models, GPU acceleration, and generalization strategies are synthesized to highlight key trade-offs and future directions for operational deployment.
This study examines the status and spatial characteristics of natural heritage damage under climate change by compiling damage cases from 1990 to 2025 into a database and analyzing their typological, temporal, and spatial patterns. The analysis focused on nationally designated natural heritage properties, specifically Scenic Sites and Natural Monuments, and combined statistical analysis with spatial analysis. The statistical analysis reviewed annual and monthly damage patterns, characteristics by heritage type, disaster causes, and forms of damage. The spatial analysis applied kernel density estimation to point-based damage data and calculated damage rates at the Si/Gun/Gu administrative level to identify areas of concentrated damage and relative vulnerability. The results show that a total of 338 damage cases were recorded, with Natural Monuments accounting for the larger proportion of cases. Among heritage subtypes, plant-related heritage showed particularly high damage frequency, especially old trees and village forests being the most affected. Damage cases have generally increased since the 2010s and were concentrated in August and September, indicating a close relationship with the seasonal pattern of typhoons and heavy rainfall. Storm and flood damage accounted for the largest share among disaster types, with typhoons and heavy rainfall identified as the main causes. Spatially, relatively high damage densities were observed in Jeju, the southern coastal area, and parts of the central inland region. The damage-rate analysis also identified highly vulnerable areas in places where the overall number of natural heritage properties is relatively small. These findings suggest that natural heritage damage should not be solely in terms of frequency, but as a matter of relative risk shaped by local environmental conditions and the spatial distribution of heritage. The database and analytical results presented in this study are expected to serve as a useful basis for future climate-risk assessment and preventive conservation planning for natural heritage.
This study constructed slope displacement data for the Gangwon State by applying persistent scatterer (PS) and distributed scatterer (DS) synthetic aperture radar interferometry (InSAR) time-series analysis to Sentinel-1 C-band and ALOS-2 PALSAR-2 L-band data, and evaluated the applicability of each sensor for slow-moving landslide detection in mountainous terrain dominated by dense forest cover. Following co-registration and topographic phase removal, sequential phase linking was applied to optimize interferometric phases, and line-of-sight (LOS) displacement velocity maps were generated through time-series analysis integrating both PS and DS pixels. The results revealed that Sentinel-1 observations were largely confined to exposed surfaces such as urban areas and road cuts, whereas ALOS-2 PALSAR-2 provided spatially continuous coverage extending into forested slopes. On slopes steeper than 10°, the valid pixel ratios for ALOS-2 PALSAR-2 reached 78.0%, 68.0%, and 57.0% for the 10-20°, 20-30°, and ≥30° slope ranges, respectively, compared to only 11.4%, 5.6%, and 4.7% for Sentinel-1. In the land-cover-based comparison, Sentinel-1 achieved moderate valid pixel ratios for cropland (49.3%) and grassland (37.0%) through phase optimization, but only 6.5% for forested areas, whereas ALOS-2 PALSAR-2 maintained 66.0% even in forests. In the aspect-wise comparison, ALOS-2 PALSAR-2 sustained observation densities exceeding 30% across all slope orientations, including both satellite-facing and satellite-away slopes. Furthermore, time-series analysis of two slow-moving landslide-prone sites identified from the ALOS-2 PALSAR-2 velocity map revealed mean LOS velocities of -28.2, +12.9, -39.7 mm/yr for three slope sectors, all exhibiting linear displacement trends over approximately 5 years. These results demonstrate that L-band InSAR displacement data can serve as a fundamental dataset for regional-scale early detection and long-term monitoring of slow-moving landslides in the mountainous terrain of Gangwon.
The Bakennoye deposit is one of the rare-element pegmatite deposits distributed around the Kalba igneous body of the Altai orogenic belt, which was created by the collision of the Siberian and Kazakhstan plates. Rare-element pegmatites are classified into Li-Cs-Ta (LCT) enriched type and Nb-Y-F (NYF) enriched type pegmatites. The pegmatite of the Bakennoye deposit geochemically corresponds to the LCT type pegmatite with high Li and Cs contents. Geochemical analysis results show that the Li content of the Bakennoye lithium-bearing pegmatite range from 21 to 25,700 ppm. The lithium-bearing pegmatite cuts biotite granite and metamorphosed sedimentary rocks and is distributed in the form of veins. Lithium-bearing pegmatite, which shows a zone due to internal greisen, is believed to have formed in the late magmatic differentiation and contains high concentrations of lithium along with rare elements such as cesium and tantalum. Lithium deposits in eastern Kazakhstan, including the Bakennoye prospect, are evaluated as an economically important mineral resource distribution area, and active exploration and development are expected to be conducted in the future as lithium demand increases.
Continuous ecological monitoring of coastal fish communities is essential for understanding the impacts of climate change and environmental variability on marine ecosystems. However, in remote island areas such as Ulleungdo in the East Sea of Korea, conventional dive-based surveys are constrained by weather conditions, geographical isolation, and limited observation time, making long-term continuous data acquisition challenging. This study presents a labeled underwater image dataset for marine fish object detection, constructed from continuous footage acquired via a fixed subtidal real-time video streaming system installed at a depth of 6 m at the Cheonbu Underwater Observatory, Ulleungdo, South Korea. The system employs a 4K ultra-high-resolution camera operating 24 hours a day, enabling uninterrupted day/night and seasonal recording. The system continues to collect footage on an ongoing basis; the present dataset was constructed from two periods, April to August 2024 and February to September 2025, selected to capture seasonal variation including the high water temperature period and ensure representation of thermophilic species. The final dataset consists of 6,190 high-resolution underwater images (.jpg) and 88,157 corresponding bounding box annotation files (.txt), labeled using the Roboflow platform and verified through expert review. At least 17 marine fish species identified in the coastal waters of Ulleungdo were classified into 13 object detection classes, with Chromis spp. accounting for approximately 49.1% of total annotations. For five low-frequency classes (Hyporthodus septemfasciatus, Hexagrammos spp., Semicossyphus reticulatus, Seriola spp., and Thamnaconus modestus), supplementary images acquired via SCUBA diving were incorporated to improve class-level representation. Taxonomic classification followed FishBase and the National Species List of Korea. This dataset provides a foundation for developing underwater fish recognition models and is expected to contribute to long-term monitoring of fish community changes associated with ongoing sea surface temperature rise in the East Sea.
This paper provides a whole-rock geochemical dataset (major oxides and trace elements, 57 analytes) and derived granite fertility indicators for the Precambrian Nonggeori Granite exposed in the Taebaek-Sangdong metallogenic belt (Northeastern Yeongnam Massif, South Korea). Two representative samples (NG1 and NG2) were collected from the pluton, powdered (<105 μm), and analyzed at Activation Laboratories Ltd., using lithium borate fusion or four-acid digestion followed by inductively coupled plasma optical emission spectrometry, inductively coupled plasma mass spectrometry, and fluoride ion selective electrode measurements. The samples are high-silica leucogranites (SiO2, 73.30-74.08 wt.%) with low MgO (0.17-0.38 wt.%), CaO (0.44-0.62 wt.%), and Fe2O3(T) (1.47-1.69 wt.%). Both samples are peraluminous with Al2O3/(Na2O+K2O+CaO)≈1.30. Trace-element signatures include elevated Rb (313-341 ppm), Cs (19.8-23.1 ppm), and Sn (12-19 ppm) and low Nb/Ta (3.87-4.44), collectively suggesting an evolved granite with a rare-metal affinity. When compared with 15 published criteria for Li-Sn-bearing fertile granites, NG1 satisfies 11 indicators whereas NG2 satisfies five indicators, implying significant intra-pluton heterogeneity in fractionation and fertility signals. The dataset can serve as baseline information for screening fertile zones within the Nonggeori Granite and for vectoring exploration toward rare-element (lithium-cesium-tantalum type) pegmatites, as well as for designing follow-up mineral-chemical and geochronological studies.
Habitat heterogeneity in vegetation structure and ground-surface conditions can influence biological communities by altering resource availability and microhabitat conditions. This study compared Coleoptera community structure and feeding guild composition between two sampling plots with contrasting vegetation structure on the Yeungnam University campus, South Korea: a grassland plot and a forest plot. Coleopteran insects were collected using pitfall traps from 19 April to 19 May 2022, and vegetation was surveyed monthly from April to June 2022. Collected species were classified into herbivorous, predatory, and detritivorous guilds, and differences in species richness and abundance between plots were tested using the Mann-Whitney U test. In total, 32 species and 236 individuals were collected: 21 species and 163 individuals in the grassland plot and 11 species and 73 individuals in the forest plot. Open-habitat and dry-surface-associated taxa dominated the grassland plot, whereas taxa associated with litter layers, fungi, and concealed organic substrates occurred in the forest plot. Overall species richness and abundance were significantly higher in the grassland plot. Predatory beetles showed significantly greater species richness and abundance in the grassland plot, while detritivorous beetles showed significantly greater abundance but not species richness. Herbivorous beetles did not differ significantly in trap-level species richness or abundance. These patterns suggest that differences between the two plots were reflected not only in overall richness and abundance but also in feeding guild composition. Predatory and detritivorous beetles appeared to respond to vegetation structure, ground-surface openness, and resource exposure, whereas herbivorous beetles may have been more closely related to local host plants and floral resources. Feeding guild-based analysis can help interpret resource use patterns of coleopteran communities between sampling plots with contrasting vegetation structure.
Inferring object activities from high-resolution satellite imagery is a critical task in remote sensing. However, improving activity inference performance typically requires large-scale domain-specific data acquisition and training, which incurs significant time and cost. When the application domain involves national defense and security, data access is further constrained by security classification systems, and inference performance depends on precise annotations by expert analysts rather than simple labeling. Moreover, conventional single-model static inference has structural limitations in capturing complex inter-object interactions and dynamic contexts. This study proposes an experience-based adaptive multi-agent collaboration framework that enhances accuracy and reliability solely through structural optimization at inference time and multi-agent collaboration, without domain data fine-tuning or large-scale retraining. The proposed system comprises five specialized artificial intelligence (AI) agents responsible for perception, vision-language context extraction, hypothesis generation, criticism, and supervision. Through an experience-based adaptive loop that iteratively performs hypothesis generation and critical verification, the framework progressively improves inference accuracy and reliability. Experimental results using the AI-hub satellite image object detection dataset demonstrate that the framework achieves Top-1 accuracy of 67.1%, Top-3 accuracy of 89.5%, Macro-F1 of 0.516, Brier score of 0.516, and expected calibration error (ECE) of 0.018, dominating all five baselines across every metric and reducing the calibration error by approximately one order of magnitude.
This study quantitatively estimates biomass resources in rice cultivation areas using MODIS satellite imagery and artificial intelligence, and evaluates the theoretical bioenergy potential across the Korean Peninsula. Vegetation indices were derived from MODIS data, and a machine learning model was used to estimate leaf area index, which was subsequently integrated into a remote sensing-integrated crop model to simulate biomass accumulation across different rice growth stages. Based on the estimated grain yield, conversion factors for agricultural byproducts (rice straw, 1.02; rice husk, 0.177) and heating values (rice straw, 15.3 MJ/kg; rice husk, 14.2 MJ/kg) were applied to calculate the bioenergy potential in terms of tons of oil equivalent. Geographic information system (GIS)-based spatial analysis was conducted to produce administrative-level bioenergy resource maps. The results indicate that Gimje City exhibited the highest productivity among the analyzed South Korean regions. In North Korea, Pyongyang showed lower productivity and greater interannual variability than the selected South Korean study region. This study evaluates the spatial and temporal variability of agricultural byproduct biomass resources using a satellite-based approach, providing a quantitative dataset for regional renewable energy policy planning and spatial analysis.
This study investigated the long-term changes in sediment grain size characteristics of tidal flats along the southwestern coast of Korea using datasets collected through the National Marine Ecosystem Survey from 2015 to 2025. The study area consisted of 129 survey lines and 387 monitoring stations distributed across the western and southwestern coastal regions of Korea, including the Gyeonggi-do, Incheon, Chungcheong, Jeonbuk-do, and Jeollanam-do regions. Spatial distributions and temporal variations of mean grain size were analyzed to identify regional sedimentary characteristics and long-term trends. The results showed that relatively coarse sediments were dominant in the Gyeonggi-do, Incheon, and Chungcheong regions, whereas finer sediments were generally distributed in the Jeonbuk-do and Jeollanam-do regions. These spatial differences were interpreted as the combined effects of tidal and wave energy, exposure to the open sea, terrestrial sediment supply, and geomorphological enclosure. Long-term time-series analysis revealed that although sediment grain size variations were observed in some regions and years, no clear nationwide trend of overall coarsening or fining was identified during the study period. In addition, grain-size class analysis indicated that coarse sediments showed relatively large interannual variability, whereas fine sediments exhibited comparatively stable distribution patterns. These findings suggest that tidal flats along the southwestern coast of Korea are characterized not by a simple one-way erosion system, but rather by a dynamic equilibrium environment where resuspension and redeposition continuously occur under tidal and wave influences. This study is meaningful in that it quantitatively evaluated long-term sedimentary environmental changes in Korean tidal flats based on a national-scale monitoring dataset, and the results are expected to provide fundamental information for future tidal flat conservation and coastal management policies.
The increase in international trade has accelerated the invasion of alien species. Invasive alien insects affect ecosystems through competition and hybridization with native species, disruption of ecological functions, structural damage to infrastructure, and the transmission of pathogens. The Argentine ant (Linepithema humile), one of the world’s most successful invasive species, was first recorded in South Korea in 2019 near Busan Station. To assess the status of native ant assemblages near the initial invasion site, we surveyed ant community from April to September 2024. A total of 4,099 individuals belonging to 15 species, 14 genera, and four subfamilies were collected. Among them, Tetramorium tsushimae (48.0%) and Linepithema humile (27.0%) were the dominant species, consistently accounting for more than half of the monthly catch. Notably, Linepithema humile peaked in May, declined in June, and subsequently increased steadily toward September. These findings indicate that Linepithema humile remains widely spread around Busan Station. These findings highlight the urgent need for strengthened monitoring and integrated management strategies, including early surveillance at transport hubs, ecological studies on native ant resistance, and restoration approaches centered on resilient native species.
Synthetic aperture radar (SAR) offers critical all-weather observation capabilities, yet its interpretation remains challenging due to inherent speckle noise and non-intuitive scattering characteristics. Consequently, directly applying vision-language models (VLMs) trained on natural images to the SAR domain is limited by significant modality gaps and the scarcity of high-quality SAR-text datasets. To overcome these challenges, this study proposes a two-stage framework that leverages SAR-to-optical translation to bridge the domain gap. First, we introduce a conditional Brownian Bridge Diffusion Model integrated with a SAR feature guidance module. This approach transforms SAR images into optical-like representations while preserving structural fidelity, thereby addressing the geometric distortions and hallucinations common in generative adversarial network (GAN)-based methods. Second, the translated images are analyzed by a domain-adapted VLM, utilizing the GeoRSCLIP visual encoder and a LoRA-tuned LLaVA model to generate precise semantic captions. Experimental results using Sentinel-1 and Sentinel-2 datasets demonstrate that the proposed translation model outperforms existing GAN models in terms of PSNR and SSIM. Furthermore, the framework achieves significant improvements in captioning metrics, including BLEU, ROUGE-L, and BERT-Score, compared to direct SAR interpretation. This study validates that high-fidelity modality translation can effectively extend the reasoning capabilities of pre-trained VLMs to the SAR domain without requiring extensive SAR-specific annotations.
The Geo Big Data Open Platform operated by the Korea Institute of Geoscience and Mineral Resources (KIGAM) integrates diverse geoscience resources, including geological maps, thematic maps, reports, metadata, and frequently asked questions, and provides public access through web services and open APIs. Despite this rich content, practical access through natural-language queries remains limited because most records are organized around structured metadata and domain-specific terminology. This study presents GeoBot, a conversational artificial intelligence service for the platform, and describes the construction of a structured Korean question-answer (QA) dataset designed to support semantic retrieval, geospatial filtering, and retrieval-augmented response generation. The retrieval corpus was assembled from open platform metadata, geo data, paper metadata, report metadata, and FAQ content, while the released QA records were derived from 1,200 real user queries collected during the beta service. The workflow comprised source-data collection, schema validation, metadata normalization, coordinate transformation, sentence embedding, vector indexing, similarity retrieval, and large language model-based answer generation. During preprocessing, fragmented metadata records were merged into document-level objects, coordinate strings were converted into Geo- JSON point or polygon objects in WGS84, and normalized text fields were embedded using a Korean sentence-embedding model based on KoSimCSE-RoBERTa. The resulting documents were indexed in Elasticsearch using dense_vector and geo_shape fields to enable joint evaluation of semantic similarity and spatial intersection. The final dataset contains 1,200 structured Korean QA records, including raw questions, refined questions, keywords, evidence-grounded answers, spatial information, and source identifiers. Expert evaluation showed a factual consistency rate of 96.83%, demonstrating the dataset’s reliability for geoscience conversational services.
The Sulatsay rare metal field formed through pegmatite generation associated with fertile granitic magmatism, with lithium mineralization occurring in lithium-cesiumtantalum (LCT)-type pegmatites at contacts with syenite. This study provides fundamental geochemical data to support future lithium resource development in the Sulatsay Area. Thirty samples were collected from rare metal-bearing pegmatite bodies, and chemical analyses were performed at the Institute of Mineral Resources (IMR), Uzbekistan. Lithium was detected in all samples, with concentrations ranging from 26.34 to 2,124.76 ppm. Field observations and mineralogical assemblages indicate that albitized quartz-microcline pegmatites are characteristic of the study area. Most rare metal pegmatites are interpreted to have been enriched through greisenization processes. Future work will focus on identifying the fertile granites responsible for pegmatite formation using geochronological dating. Once these granites are delineated, a detailed geological survey covering approximately 10-20 km2 will be conducted, based on the hypothesis that rare metal pegmatites occur within a 10 km radius of the fertile granite.
Reliable spatial data on wildlife occurrence are essential for ecological monitoring and environmental assessment, yet detection uncertainty often limits their interpretability. This study presents a camera-trap-based detection-non-detection dataset collected in a forested landscape, and provides grid-level occupancy estimates for water deer (Hydropotes inermis) and raccoon dog (Nyctereutes procyonoides) using an occupancy model framework. Repeated survey data were organized into detection histories, occupancy probability, and detection probability were estimated by incorporating observation variables and environmental variables. Both species exhibited higher occupancy probabilities in forest-dominated and less disturbed areas within the study extent. These results demonstrate the potential of occupancy models to quantitatively evaluate habitat suitability by accounting for detection probability, even with limited survey data, and provide baseline information for future environmental impact assessments and wildlife management.
This dataset presents mineralogical analyses of clay alteration zones from the Seongsan and Ogmaesan Mines located in the Haenam Area, southwestern Korea. A total of 15 clay-rich rock samples (10 from Seongsan, five from Ogmaesan) were collected and analyzed using X-ray diffraction (XRD) to determine mineral assemblages in different alteration zones. The host rocks belong to the Cretaceous volcanic units and have undergone varying degrees of hydrothermal alteration. Mineral zoning was categorized by dominant minerals including kaolinite, quartz, alunite, muscovite/ illite, and chlorite. The dataset includes mineral assemblage classification for each sample, zone-specific mineral content, and selected Li content values measured by bulk rock chemical analysis. This information can support future exploration and mineralogical modeling of hydrothermal clay systems.
This study predicted the potential hibernation habitats of two endangered bat species, Myotis rufoniger and Plecotus ognevi, using a Random Forest-based species distribution model (SDM) that integrated occurrence records and environmental variables. The model incorporated climatic (BIO1, BIO12), topographic (DEM), and land cover variables. Both models exhibited statistically significant predictive performance, with P. ognevi (AUC=0.8054, TSS=0.634) showing higher reliability than M. rufoniger (AUC=0.7351, TSS=0.3824). Variable importance analysis indicated that DEM, BIO1, and BIO12 were key predictors for M. rufoniger, while BIO1, DEM, and BIO12 were most influential for P. ognevi, suggesting that climatic and topographic factors predominantly shaped habitat suitability. Predicted suitability maps revealed that M. rufoniger favored lowland agricultural areas, whereas P. ognevi was primarily associated with high-elevation mountainous regions. These findings provide quantitative insights into the environmental requirements of hibernating bats and offer a robust scientific basis for developing conservation and management strategies for endangered species in Korea.
This study analyzed the fish community structure and distribution patterns in the Omcheon Stream, a tributary of the Tamjin River system, from May to November 2021. Fish surveys were conducted at a total of 24 sites along the stream, based on the weirs that were installed. A total of 11,305 individuals representing 23 species, eight families, and three orders were identified. The dominant species was Zacco temminckii (36.37%), followed by Zacco platypus (26.07%), which was the subdominant species. Two endangered species, Acheilognathus somjinensis (class I) and Coreoperca kawamebari (class II), were recorded, and the invasive alien species Micropterus salmoides was found in the lower reach of the stream (S1-S6). Ten Korean endemic species were also identified. The overall diversity index was 1.93, richness index was 2.36, evenness index was 0.62, and dominance index was 0.62, indicating a fish community structure with a distinct dominance of certain species. Species diversity and richness increased downstream, while evenness and dominance remained largely stable, except at sites S4 and S6, where the abundance of Zacco temminckii and Zacco platypus was particularly high. These results provide a quantitative baseline for understanding the ecological characteristics of fish communities in the Omcheon Stream and are expected to contribute to the effective management and conservation of regional freshwater ecosystems.
