Forest remote sensing plays a crucial role in monitoring large forested areas and analyzing forest dynamics, disasters, and ecosystems. In Korea, forest remote sensing began in the early 1970s with the first national forest resource survey using aerial photo interpretation. Since then, with the advancement of satellite technology, the methods for qualitative and quantitative interpretation of forest data have also evolved. Korea’s first satellite dedicated to forest and agricultural vegetation analysis, the Compact Advanced Satellite 500-4 (CAS500-4, Agricultural and Forestry Satellite), is set to launch in 2025. It is currently being developed with a 5-meter resolution and 5 bands (R, G, B, Red Edge [RE], Near Infrared [NIR]).

This paper aims to review major research activities in the field of forest remote sensing in Korea, in commemoration of the 40th anniversary of the Korean Society of Remote Sensing. The research scope is divided into four key areas: forest resources, forest disasters, forest ecosystems, and Forest Analysis Ready Data (F-ARD). In Chapter 1, the focus is on the evolution of qualitative and quantitative forest resource interpretation technologies through satellite imagery, with a particular emphasis on the changes in precision analysis using Light Detection And Ranging (LiDAR). Chapter 2 analyzes the development of technologies for responding to forest disasters, including wildfires, landslides, and forest pest outbreaks. Chapter 3 reviews the remote sensing-based research on Gross Primary Productivity (GPP), Leaf Area Index (LAI), and plant phenology, as well as growth stress in forest ecosystems. Finally, Chapter 3.4 discusses the introduction of F-ARD and the improvement in analytical efficiency through preprocessing simplification, while also projecting various research directions and advancements following the launch of the CAS500-4 Agricultural and Forestry Satellite.

To analyze trends in forestry remote sensing research in Korea, formal research reports and published papers were examined. Keywords were selected for each field, and data were collected through searches on Google Scholar, Web of Science, and the Research Information Sharing Service (RISS). The study focused on four main areas: forest resources, forest ecology, forest disasters, and the emerging field of F-ARD.

In the forest resources sector, 87 papers published between 1990 and 2024 were analyzed, including 33 papers from SCI(E) journals, 54 from KCI journals, and 14 research reports, covering South Korea and North Korea. The research topics included land cover and change (20 papers), forest species distribution (11 papers), stand height (17 papers), forest biomass (31 papers), and five papers specifically focused on North Korea. For the forest ecology sector, 90 papers published between 1977 and 2024 were analyzed, with 76 from SCI(E) journals and 14 from KCI journals. The research topics included vegetation phenology (22 papers), gross primary productivity (23 papers), leaf area index (34 papers), and plant growth stress (11 papers). In the forest disaster sector, 54 KCI papers published between 1998 and 2023 were reviewed, with the topics divided into wildfires (23 papers), landslides (17 papers), and pest infestations (14 papers).

In forest remote sensing, approximately 80% of the time spent on satellite data analysis is consumed by preprocessing for various satellite data (Vlach et al., 2023). To address this issue, the concept of Analysis Ready Data (ARD) has recently gained attention (Dwyer et al., 2018). ARD aims to simplify the complex preprocessing steps and fill data gaps by providing standardized satellite data in a form that is easier to analyze alongside other geospatial information, thereby enhancing satellite data usability (Dwyer et al., 2018). The Committee on Earth Observation Satellites (CEOS) is currently leading efforts to standardize ARD for surface reflectance, Synthetic Aperture Radar (SAR)-Normalized Radar Backscatter, and aquatic reflectance. This study focuses primarily on surface reflectance, which is frequently used in forest analysis. From 2018 to 2024, 24 papers on this topic were covered, with 22 published in SCI(E) journals and two in KCI journals.

3.1. Forest Resources

Remote sensing of forest resources was first used in the early 1970s as part of the ‘National Forest Status Survey,’ providing essential data for forest restoration and the regeneration of degraded forests across the country. By the 1990s, with reforestation efforts showing success, research began to focus on the quantitative evaluation of forest structure and biomass accumulation to better understand and manage the nation’s forest resources. More recently, as the role of forests in sequestering CO₂ has become critical for climate change adaptation strategies, including the ‘Nationally Determined Contribution’ (NDC) and the achievement of Net-Zero goals, various remote sensing technologies, such as high-resolution satellite imagery and ground/airborne LiDAR, are now being used to perform detailed quantitative assessments of the CO₂ stored and absorbed by forest ecosystems.

3.1.1. Land Cover

Land cover map provide essential information for interpreting forest resources using remote sensing technology (Fig. 1). In Korea, land cover classification research in the forestry sector began with the ‘1st National Forest Status Survey’ (1971–1975), during which black-and-white aerial photographs taken every 10 years across the country were visually interpreted to delineate forest boundaries for forest resource surveys, including growing stock and age-class distribution (Kim and Kim, 2015). However, the significant time and manpower required for this method, along with the long survey cycle, led to the increasing use of satellites in the 1990s for land cover classification studies. These early studies primarily used medium- and low-resolution satellite imagery, such as Landsat Multispectral Scanner (MSS) and Thematic Mapper (TM), combining field data with supervised or unsupervised classification methods to analyze small-scale, regional, and single-period forest conditions (Choung, 1994; Lee, 1994; Park, 1996).

Figure 1. The history of forest land cover analysis research using remote sensing data.

From the 2000s onwards, as awareness of forest conservation and degradation management grew, land cover change analysis became more prevalent, particularly in the context of forest area reduction and land cover changes in various regions (Cho and Oh, 2004; Jung et al., 2005; Lee et al., 2007). High-resolution satellite imagery, such as SPOT, Advanced Very-High Resolution Radiometer (AVHRR), and Korean Multi-Purpose Satellite (KOMPSAT), were merged with medium-resolution images to detect more detailed land cover changes at the regional level (Yeom et al., 2008; Choi et al., 2015). After 2008, the methodology for constructing a nationwide forest-type map on a 5-year cycle with scales of 1:5,000 and 1:25,000 was established and stabilized (Lee et al., 2008a; Ryu et al., 2011). Meanwhile, the Ministry of Environment completed the production of a large-scale land cover map in 2005 which caused a reduction in land cover analysis research. However, with the enforcement of the Paris Agreement in 2016 and the increased importance of tracking annual greenhouse gas inventories related to land use changes and forest management activities, land cover classification regained attention.

Moreover, the rapid development of Artificial Intelligence (AI) technologies since 2017 has led to a surge in studies utilizing various deep learning methods (Lee et al., 2021a). Recent studies have utilized remote sensing data with relatively high spatial resolutions, such as Sentinel-2 (10 m) and RapidEye (5 m), which are equipped with spectral bands sensitive to forest characteristics (e.g., NIR, Red Edge, Short-Wave Infrared [SWIR]). These studies have analyzed spectral values from training areas, along with vegetation indices (e.g., Normalized Difference Vegetation Index [NDVI], Enhanced Vegetation Index [EVI], Soil Adjusted Vegetation Index [SAVI]), and surface roughness characteristics using Gray-Level Co-Occurrence Matrix (GLCM). Various deep learning techniques, including Random Forest (RF), Convolutional Neural Networks (CNN), U-Net, and XGBoost, have been applied to analyze land cover changes, forest degradation, and forest management activities (Lee et al., 2019a; 2020; Piao et al., 2021; Park et al., 2021a; Pyo et al., 2022; Choi et al., 2023; Sim et al., 2024). Furthermore, Javed et al. (2023) used high-resolution satellite imagery (KOMPSAT-3, QuickBird-2, WorldView-3) combined with CNN deep learning methods to monitor forest changes in urban areas, such as Sejong and Daejeon.

Despite these technological advancements, forest land cover change analysis is still largely conducted at regional scales due to missing data from cloudy and rainy conditions between May and September—the optimal period for vegetation detection—and limitations in spatial resolution. To overcome such challenges, consistent remote sensing data at a national scale is essential. The CAS500-4, scheduled for launch, with a three-day revisit cycle and a 120-kilometer observation width, is expected to address issues of missing data and enable periodic analysis of land cover changes at a large scale. By using data that overcome the limitations of time-series changes in forests, it will be possible to identify more accurate land conversion relationships at regional and national levels and contribute to greenhouse gas inventory calculations and land conversion following the land management activities.

3.1.2. Tree Species

Korean forest-type map, which include key attributes such as tree species, forest cover, and age class, have traditionally been produced through aerial photograph interpretation, supplemented by on-site comparisons. For over 40 years, starting in the 1970s, this approach dominated forest mapping, eventually leading to studies focused on species identification using high-resolution satellite imagery. Kim (2008) was the first to conduct visual interpretation using pansharpened QuickBird satellite images, with a spatial resolution of 0.7 m, to distinguish between pine and Korean pine trees based on texture and shape. Kim et al.(2011) evaluated the performance of tree species classification using object-based segmentation and supervised classification methods, employing pansharpened IKONOS satellite imagery with a spatial resolution of 1.0 m. The research on tree species classification has transitioned from supervised classification with high-resolution satellite imagery and aerial photography (Hou and Yang, 2014) to more advanced spectroscopic approaches. Cho and Lee (2014) compared hyperspectral aerial photography and multispectral imagery, concluding that hyperspectral data offered greater advantages for differentiating tree species.

After that, researchers focused on developing tree species classification methods that integrate spectral, texture, and growth information. Lim et al. (2019a) evaluated the potential for classifying Korean pine and Japanese larch using hyperspectral satellite data (Hyperion) and multispectral satellite data (Sentinel-2). Their findings revealed no significant performance differences between the two datasets, demonstrating the feasibility of using Sentinel-2 imagery for tree species classification. Lim et al. (2019b) additionally developed a classification model incorporating spectral separation, growth information, and texture data to classify the five major tree species in Korea: pine, Korean pine, Japanese larch, fir, and oak that extended to estimating the distribution of major tree species in North Korea. A model developed using training data from South Korea and the Mt. Baekdu region in China was applied to classify tree species in Goseong-gun, North Korea. During this study, classification was performed by utilizing spectral, growth, and texture information, based on prior research (Lim et al., 2020; Fig. 2).

Figure 2. Classification results for five tree species in North and South Goseong-gun using the integrated model (Lim et al., 2020).

Since 2021, in preparation for the launch of the CAS500-4, efforts have been made to develop technology for its application by evaluating the feasibility of tree species classification using satellite imagery with specifications similar to those of the CAS500-4 (Cha et al., 2023a; 2023b; Kwon et al., 2021b). Studies compared the classification accuracy of using only spectral information from RapidEye satellite imagery, which has the same spatial resolution and a similar band configuration to the CAS500-4, with a classification that also included multi-temporal texture information. The results indicated that combining RapidEye spectral data with multi-temporal texture information enabled tree species classification with 69% accuracy (Kwon et al., 2021a). Furthermore, the potential for fusing data from RapidEye and Sentinel-2 satellites was evaluated, revealing that such data fusion improved classification accuracy when developing models (Cha et al., 2023b).

Additionally, it was found that using deep learning models further enhanced classification accuracy (Cha et al., 2023a). In recent years, there has been a growing body of research on applying deep learning techniques for tree species classification (Cha et al., 2023a; Lee et al., 2023a; 2023b), as well as attempts to classify species using LiDAR point cloud data rather than satellite imagery (Lee et al., 2023b). Examining the trends in tree species classification research in Korea, the progression from high-resolution satellite image classification—resembling aerial photo interpretation—toward multispectral image classification using spectral information is evident. More recently, research involving LiDAR data has begun to emerge. The Korea Forest Service, as part of its digital transformation initiatives for forest management, has been applying LiDAR technology alongside the development of the CAS500-4. Internationally, technologies that integrate hyperspectral imagery with LiDAR data to produce three-dimensional tree species classification map are already in use (Asner et al., 2007). To achieve effective large-scale forest management in the future, it will be crucial to obtain more precise forest resource information by integrating spectral and LiDAR data.

Moreover, beyond tree species classification, it is essential to incorporate biochemical information from forest canopies to quantitatively assess the public values of forests—such as biodiversity, gross primary productivity, and carbon storage and absorption—and evaluate their broader impacts. To support this, it will be necessary to establish a spectral library that accurately captures the spectral characteristics of different tree species, thereby enhancing the usability of upcoming hyperspectral satellites. It will also be important to build region-specific libraries to determine whether spectral characteristics differ within the same species across regions. Additionally, given Korea’s geopolitical location and the potential variations in reflectance due to latitude, it may be necessary to apply Bidirectional Reflectance Distribution Function (BRDF) corrections to account for differences caused by latitude and solar angle.

3.1.3. Stand Height

Data used for stand height analysis can be broadly divided into two categories: LiDAR (ground-based, airborne, and satellite) and imagery (stereo aerial photographs and satellite images). The choice of data is critical depending on the size of the study area (Balenović et al., 2015; Ganz et al., 2019). Typically, stand height analysis for small areas utilizes ground-based or Unmanned Aerial Vehicle (UAV) LiDAR, while for larger areas, airborne LiDAR and satellite imagery are preferred. The common principle across these methods is the generation of point clouds from either LiDAR or imagery, which are then used to create a Digital Surface Model (DSM) and a Digital Terrain Model (DTM). By subtracting the DTM from the DSM, a normalized DSM (nDSM) is produced, which allows for accurate determination of tree heights.

LiDAR is known for providing highly accurate measurements in small areas, particularly due to its ability to penetrate forest canopies and produce dense point clouds that precisely measure tree height and terrain elevation (Mcinerney et al., 2010). Airborne LiDAR was adopted earlier and became widely used in the 1990s for large-scale forest resource analysis, including terrain mapping, height, biomass, and forest structure, while Terrestrial LiDAR (TLS) was later introduced as a complementary tool for more detailed forest structure analysis (Dassot et al., 2011). While TLS provides detailed 3D forest structure information over smaller areas (Fouladinejad et al., 2019), its high cost and the need for clearing undergrowth make it less suitable for large-scale or periodic stand height monitoring.

In South Korea, airborne LiDAR-based forest structure analysis began in the early 2000s. Kwak et al. (2005) were the first to conduct quantitative analyses of forest structure, including stand height, ground height, and diameter at breast height (DBH), using airborne LiDAR data. Jang et al. (2006) further refined these methods by integrating color aerial photographs with airborne LiDAR to estimate stand height. Woo et al. (2007) extracted canopy height models (CHMs) from airborne LiDAR and used filtering techniques to estimate tree height. By the 2020s, TLS began being actively applied to forest structure analysis, as demonstrated by Ko et al. (2021). However, TLS has limitations, including the need for site preparation to remove undergrowth that can obstruct ground models, as well as high data collection costs, making it unsuitable for large-scale, periodic stand height assessments. In contrast, stereo aerial and satellite imagery offer a cost-effective alternative to LiDAR, at roughly half to one-third of the cost, while still providing similar results for large-scale forest height analysis and long-term monitoring (Lee et al., 2022). Kim et al. (2016) developed the first image-based stand height estimation technique in Korea, using stereo aerial photographs to generate point clouds similar to LiDAR. To facilitate its application in forest mapping, an automated module based on the U.S. Forest Service’s FUSION program was developed and later applied to analyze forest height in North Korea using high-resolution stereo satellite imagery (Kim, 2016; Fig. 3).

Figure 3. nDSM extraction from differencing between DSM and DTM of stereo aerial photos (Kim, 2016).

In addition, Lee et al. (2017) examined the potential use of stereo images from UAVs for localized applications, such as monitoring street trees in urban areas. By 2025, with the successful launch of the CAS500-4 and the CAS500-1 and -2, a continuous collection of 0.5 m high-resolution stereo images and 5-m resolution optical satellite imagery is expected. Considering this advancement, a joint research project between the National Institute of Forest Science and Kyungpook National University is currently developing technology to estimate stand height by integrating these two satellite datasets. Since the accuracy of stand height estimation heavily depends on the quality of surface and terrain models, it is important to consider integrating time-series data from the non-growing season for terrain models (when understory vegetation is minimal) and from the growing season for surface models (when canopy layers are fully developed) (Perko et al., 2011). Furthermore, there is potential for applying satellite-borne LiDAR systems, such as the Global Ecosystem Dynamics Investigation (GEDI), which is actively being researched abroad (Zhou et al., 2023). Research is also underway to explore synergies between the CAS500-4 and the upcoming SAR satellite (i.e., CAS500-5), which is scheduled to launch around the same time. These efforts are expected to contribute significantly to nationwide stand height estimation and forest monitoring.

3.1.4. Growing Stock and Carbon Storage

Research on growing stock in Korea primarily relied on regional estimates based on field survey data before the 2000s. The Korea Forest Service regularly published forest volume and biomass data, along with yield tables. However, conducting national-scale field surveys posed significant challenges, and with the introduction of ground-based LiDAR systems in the late 2000s, studies began focusing on smaller areas. These studies utilized CHMs generated from DSMs and DTMs to estimate DBH and forest biomass (Kwak et al., 2006). Simultaneously, k-Nearest Neighbor (k-NN) techniques were applied to combine National Forest Inventory (NFI) data with spectral information from Landsat TM and Enhanced ETM+ satellite imagery, enabling the estimation of forest biomass and volume for unsampled areas (Yim et al., 2007; 2012; 2009). To address the limitations of small-scale ground-based LiDAR studies, Lee and Ru (2012) employed airborne LiDAR data, integrated with KOMPSAT-2 satellite imagery and field survey data, to estimate forest biomass. Subsequent studies expanded on this approach by combining airborne LiDAR data with field surveys or NFI data and foest-type map to estimate stand-level biomass (Park et al., 2012). Recently, advancements in ground and airborne LiDAR platforms have enabled the integration of both types of data, leading to more precise and efficient biomass estimation at both canopy and understory levels (Lee et al., 2023d).

As forests have become increasingly recognized as critical carbon sinks for mitigating global warming and adapting to climate change, remote sensing data has emerged as a vital tool for estimating forest carbon storage and sequestration. Early estimates of carbon storage primarily utilized k-NN models based on Landsat satellite imagery, as well as ground and airborne LiDAR-based analyses of forest volume and biomass. These estimates were further refined by incorporating carbon emission and absorption coefficients developed by the National Institute of Forest Science in 2010 to calculate regional carbon storage (Jung et al., 2010; Na et al., 2010; Kim et al., 2012; Kim and Kim, 2015; Lee et al., 2013; 2015; Go et al., 2013; Fig. 4).

Figure 4. Methodology flowchart for growing stock and carbon storage analysis using satellite imagery and LiDAR data.

More recent national-scale studies have begun leveraging canopy height data from NASA GEDI, which provides global canopy height data at a 1 × 1 km spatial resolution. Shin et al.(2023) used NFI biomass data and vegetation indices (EVI, NDVI, etc.) derived from Landsat ETM+/8, along with topographical data (Shuttle Radar Topography Mission), as training inputs for an RF model to estimate national carbon storage. Jung et al. (2024) proposed a method to estimate carbon storage by interpolating GEDI canopy height data with airborne LiDAR, creating regional canopy height estimates for areas such as Gyeonggi, Incheon, and Yeoju. Additionally, Kim and Park (2024) utilized 30 × 30 m forest canopy height data, developed by Potapov et al. (2021) using GEDI and Landsat imagery, alongside forest growth models to estimate changes in carbon storage and sequestration across the Korean Peninsula. These ongoing efforts aim to expand carbon storage estimates to larger regional and national scales. To support this goal, it is essential to continue accumulating field survey data, LiDAR data, and other relevant datasets to build a comprehensive system for national carbon storage assessment.

The regular acquisition of satellite data by CAS500-4, with its high temporal frequency, broad coverage, and enhanced spectral capabilities, is expected to facilitate not only the improvement of forest resource surveys but also the broader digital transformation of forest management in Korea. Since 1972, forest-type map containing attributes such as species, DBH, age class, and size class have been updated on a five-year cycle. However, this system has several limitations, including delayed updates, data fragmentation, human error during interpretation, and inefficiencies in manual classification. The integration of CAS500-4 data with LiDAR and AI technologies could address these challenges by enabling more frequent updates, automated classification of tree species and canopy density, and enhanced quality control.

Additionally, the application of digital twin technology—creating 3D virtual environments from multidimensional real-world data (Hwang et al., 2020)—to the forestry sector is becoming increasingly feasible. The large-scale, multi-temporal satellite data collected by CAS500-4 could be integrated with key forest spatial information, such as forest type map, forest soil map, forest-water management map, and forest ecological zone map, along with field sensor data, to inform policy decisions and public services. This integration, combined with 3D data from LiDAR, would enable more precise vertical and horizontal analysis of forest carbon sequestration and emissions.

As the Fourth Industrial Revolution continues, addressing the structural challenges of labor-intensive forest management and establishing a scientific decision-making framework will require ongoing research and development in digital transformation technologies, including AI, automation, and remote sensing. At the core of this transformation will be the integration of satellite-based remote sensing technologies, providing faster and more reliable forest resource management through seamless connectivity and integration.

3.1.5. Forests in North Korea

To develop an effective inter-Korean forest cooperation policy grounded in scientific evidence, comprehensive information on North Korea’s forest conditions is essential. Given the inaccessibility of many regions in North Korea, forest research has predominantly relied on satellite imagery, leading to two primary areas of focus: 1) the assessment of forest degradation and 2) studies on the effects of climate change on North Korean forest ecosystems. In the 1990s, remote sensing-based forest research in North Korea began, with initial efforts centered on vegetation mapping, forest resource surveys, and land-use classification (Lee, 1994; Lee et al., 1998; Kim et al., 1998). Research on forest degradation in North Korea has been primarily driven by government agencies, such as the Korea Forest Service, with the goal of providing foundational data for formulating forest restoration policies in preparation for potential unification. In 1994, the National Institute of Forest Science (formerly the Korea Forest Research Institute) initiated the first satellite-based surveys of North Korean forests, using Landsat-5 TM imagery to estimate forest area, forest cover distribution, and forest volume across the country (Lee et al., 1998).

Since then, a monitoring system has been established, conducting nationwide assessments every 10 years, and more frequent, biennial assessments for key regions, enabling continuous monitoring of North Korea’s forests since the 1990s (National Institute of Forest Science, 1999; Lee et al., 2008b; Kim et al., 2013; 2014a; 2020b). The research led by the National Institute of Forest Science (1999), Lee et al. (2008b), and Kim et al. (2020a) utilized a combination of Landsat-5 TM (1999), SPOT-5 (2008), and RapidEye (2018) satellite imagery to periodically monitor changes in North Korea’s forests (Fig. 5). To facilitate swift decision-making for inter-Korean forest cooperation in response to shifting geopolitical conditions, forest status assessments were conducted in five key regions of North Korea—Pyongyang, Kaesong, Hyesan, Bongsan, and Mt. Kumgang (Kim et al., 2013). In addition, 11 major areas, including rural and urban zones, were designated as permanent monitoring regions, with forest changes observed biennially (Kim et al., 2014a; 2020b). These studies primarily utilized unsupervised classification, a technique frequently applied in regions like North Korea, where ground verification is not feasible. Furthermore, the studies employed post-classification comparison, a method that compares independently classified images from different time points, allowing for the detection of changes without needing to account for optical differences between the images.

Figure 5. Forest status map in North Korea: (a) 1999, (b) 2008, and (c) 2018 (National Institute of Forest Science, 1999; Lee et al., 2008b; Kim et al., 2020b).

Piao et al. (2021) utilized Landsat multi-temporal data combined with an RF algorithm to estimate changes in the forested area of North Korea from 2001 to 2018. Although the spatial resolution was relatively low, the consistent time-series data helped minimize interpretation errors related to spatial resolution variability during forest change analysis. Unlike South Korea, North Korea features a significant proportion of land cover in intermediate stages between forest and non-forest areas (Lee et al., 1999). These areas, such as sparsely vegetated mountain lands with few trees (a.k.a., un-stocked forest land) and terraced slopes used for agriculture, are characterized by a mixture of shrubs, grasslands, and other sparse vegetation, making satellite image analysis particularly challenging (Fig. 6). The extensive deforestation in North Korea, driven by economic difficulties and energy shortages, has led to the proliferation of these low-density forest areas, where forests remain but with reduced density or are replaced by weeds and low-lying vegetation (Lee et al., 2008b). Accurately identifying and assessing these areas is crucial, particularly because some, like un-stocked forest land, still retain some vegetation, making them potentially easier to restore compared to fully degraded lands (Kim et al., 2014a). Recognizing this, the National Institute of Forest Science has maintained consistency in its statistical monitoring of North Korean forest areas by classifying three distinct types of degraded forestlands— un-stocked forest land, reclaimed forest land, and denuded forest land—since 1999.

Figure 6. Forest degradation types in North Korea: (a) reclaimed forest land, (b) un-stocked forest land, and (c) denuded forest land (Kim et al., 2020b).

Seasonal variation plays a crucial role in misclassification. The period from June to August presents the greatest challenges for interpreting North Korean forests, as vegetation in degraded areas like un-stocked forest land, shrublands, grasslands, and crops on reclaimed slopes reaches peak vigor. This can lead to confusion between forested and degraded areas, making caution essential when interpreting growing season data. To address this issue, Kim et al. (2010) introduced texture analysis, which significantly improved the accuracy of distinguishing degraded forest areas in North Korea. One ongoing challenge in satellite image analysis is the lack of field data for validation, especially considering the timing of image acquisition (Lee et al., 1999). Since 2016, efforts have been made to build a forestry interpretation library through surveys conducted in the North Korea-China and inter-Korean border regions. This library now serves as a valuable resource for training and validating models used to analyze North Korean forests and degraded areas (Park et al., 2018).

Research on the impact of climate change on North Korea’s forest ecosystems began in the mid-2000s, though findings remain limited. Cui et al. (2014) examined how land cover changes from 1981 to 2010 affected carbon accounting, showing that forest degradation, compounded by climate change, is a major factor driving increased carbon emissions. Lim (2022) analyzed vegetation changes in North Korea from the 1980s to 2021 using Landsat imagery, assessing how recent climate shifts have influenced forest recovery. Lim et al. (2019c) also explored the connection between forest degradation and flood damage in Hoeryong, demonstrating that converting forests to agricultural land increased soil erosion and flood risk. Yeo and Lim (2022) used the InVEST model and SSP5—8.5 climate scenario data to model seasonal water supply in North Korea, showing how forest restoration could mitigate water shortages caused by climate change. The launch of the CAS500-4 in 2025 will allow forest monitoring in North Korea every three days, greatly advancing data collection efforts. However, despite the anticipated increase in satellite observations, there remains a critical lack of ground truth data needed for effective AI-based analysis. Enhancing field surveys in border regions and building comprehensive spectral information databases will be essential to improving the accuracy of forest data interpretation.

The short revisit time of CAS500-4, wide-area coverage, and advanced spectral characteristics are expected to improve forest resource monitoring and contribute to the broader digital transformation of forestry through the integration of advanced information and communication technologies (ICT). Since 1972, Korean forest-type map, which includes detailed information on tree species, DBH, and age classes, has been updated on a national scale every five years. However, this system has limitations, including delays due to the update cycle, data fragmentation, human error in interpretation, and inefficiencies from manual classification. By integrating CAS500-4 data with LiDAR and AI technologies, these issues can be addressed—shortening update cycles, automating species and canopy density classification, and improving quality control. The application of digital twin technology, which creates 3D virtual environments using real-world data (Hwang et al., 2020), is also becoming feasible in forestry. Large-scale, multi-temporal satellite data from CAS500-4 can be integrated with key forest spatial information, such as forest type map, forest soil map, water management data, and ecological zone data, as well as field sensor data. This integration is expected to support various policy decisions and public services. Combining LiDAR’s 3D information will further enable more precise analyses of carbon sequestration and emissions at the individual tree level.

As the Fourth Industrial Revolution advances, addressing the challenges of labor-intensive forest management and establishing a scientific decision-making framework will require ongoing research and development in digital transformation technologies, including AI, data analytics, and automation. Central to this transformation will be the integration of remote sensing technologies from CAS500-4, enabling more efficient and reliable forest resource management through seamless connectivity.

3.2. Forest Ecology

August 2024 set a new record as the hottest month on the Korean peninsula. Climate change intensifies extreme weather events, including droughts and heatwaves, making them stronger and more frequent. Such events are likely to adversely impact forest health, raising the need for comprehensive forest ecosystem monitoring. However, current applications or platforms for satellite-based forest eco-physiological monitoring in Korea are still in their early stages, offering room for advancement. As climate change continues, there will be growing attention to monitoring forest ecosystems using satellite data. The CAS500-4 satellite, designed for launch in 2025, is expected to play a crucial role in this effort.

3.2.1. Land Surface Phenology (LSP)

Plant phenology is closely linked to vegetation growth within terrestrial ecosystems and serves as a key indicator of land cover changes, productivity, and climate change dynamics (Sakamoto et al., 2013). Early phenological studies were conducted at the species and point level in the field, relying on manual labor. However, these studies were limited by their geographical scope, short observation periods, and the number of species covered. Consequently, recent studies have shifted to using satellite imagery to monitor LSP. LSP cannot directly detect specific phenological events using multi-spectral sensors; rather, it provides a more detailed description of vegetation dynamics at the spatial resolution of satellite images (Zeng et al., 2020). Typically, many LSP studies focus on monitoring the inter-annual life cycles of vegetation, including the start of greening or season (SOS), the peak of the growing season, the end of the season or the onset of senescence (EOS), and the length of the growing season (Reed et al., 1994, Zhang et al., 2003).

In the early 1970s, the launch of the Landsat series played a pioneering role in Earth observation, proving invaluable for monitoring global landscape changes such as land cover, vegetation dynamics, and environmental stress. However, the 16-day temporal resolution and frequent cloud contamination have limited the effectiveness of Landsat for LSP monitoring. In contrast, the Advanced Very-High Resolution Radiometer (AVHRR), with its daily revisits and global coverage, has been extensively utilized for regional and global LSP studies despite its coarse spatial resolution (1- and 8-km). Since the 2000s, Moderate Resolution Imaging Spectrometer (MODIS) and Sentinel-2 have offered time-series data with enhanced spatial and temporal resolutions, further advancing our ability to monitor and analyze these changes (Zhang et al., 2003; Zhou et al., 2019).

Moreover, other satellites including the Visible Infrared Imaging Radiometer Suite (VIIRS), Medium Resolution Imaging Spectrometer (MERIS), SPOT-VEGETATION, Geostationary Operational Environmental Satellite-R series (GOES-R), Spinning Enhanced Visible and Infrared Imager (SEVIRI), and Himawari-8 have become actively utilized as fundamental data sources for phenological monitoring (Zeng et al., 2020). To analyze shifts in vegetation phenology driven by climate change, at least 10 to 30 years of data are necessary (Kim et al., 2014b), and long-term satellite data such as AVHRR, MODIS, and Landsat are actively utilized. For instance, Park et al. (2021c) detected long-term LSP changes in the subalpine zone of Jeju Island, Korea by producing high-resolution satellite images by fusing Landsat and MODIS data.

Various satellite-based vegetation indices, including the NDVI (Wu et al., 2017; Pan et al., 2015), EVI (Cao et al., 2015; Zhang et al., 2003), two-band EVI (EVI2) (Yan et al., 2016), Leaf Area Index (LAI) (Kang et al., 2003), SAVI (Wu et al., 2014), and a fraction of Absorbed Photosynthetically Active Radiation (fPAR) (Meroni et al., 2014), have been applied to monitor LSP for the sensitive detection of plant traits, such as pigment, structure type, and water content. To reduce noise in satellite imagery (e.g., cloud contamination, sun angle, and shadow effects), various smoothing techniques, such as maximum value composite (MVC), moving window, and curve fitting, have been employed (Zhang et al., 2003; Chen et al., 2004). Choi and Jung (2014) also reduced LSP uncertainty by applying the Harmonic ANalysis of Time Series (HANTS), a smoothing algorithm.

Phenology monitoring algorithms can be categorized into three groups: 1) threshold-based methods, 2) change detection methods, and 3) machine learning models. Threshold-based methods extract phenological dates by applying fixed threshold values to specific vegetation indices (VIs) (Fischer, 1994; Cho et al., 2021). Change detection methods identify phenological dates by detecting steep changes or inflection points in VI time-series curves (Balzter et al., 2007; Dash et al., 2010). Lee et al.(2018) compared threshold-based methods with first-derivative methods, a change detection technique, for monitoring LSP in Korean forests. More recently, machine learning models (e.g., RF, Neural networks) have been increasingly used for phenology detection, outperforming rule-based methods due to their higher accuracy (Xin et al., 2020). Kim et al. (2022a) further demonstrated that the RF model outperformed multiple regression methods when modeling LSP using MODIS data (Fig. 7).

Figure 7. Flowchart of LSP detection using remote sensing and AI algorithm (Kim et al., 2022a).

The CAS500-4 is expected to provide LSP products modeled with extremely randomized trees (ERT), utilizing satellite images and meteorological data. As plant phenology is a key indicator for detecting responses to climate change, its importance, and attention are expected to grow in the coming years. Moving forward, increasingly diverse and detailed LSP data will be generated, ranging from expert visual observations to digital imagery and satellite remote sensing. Satellite-based LSP studies will also progressively shift toward detecting plant phenology at the species or individual level.

3.2.2. Gross Primary Productivity (GPP)

Vegetation is a key component of terrestrial ecosystems, with photosynthesis serving as the core link between the land surface and the atmosphere (Chen et al., 2019). GPP refers to the total carbon absorbed by vegetation through photosynthesis, playing a crucial role in carbon balance and helping to mitigate anthropogenic CO₂ emissions. However, scalable technology for directly measuring GPP beyond the leaf level is still unavailable (Ma et al., 2015). At the ecosystem scale, the eddy covariance (EC) technique is commonly used to estimate GPP by separating ecosystem respiration from Net Ecosystem Exchange (NEE) data (Aubinet et al., 2012; Du et al., 2023). Nevertheless, the spatial coverage of EC method is limited to areas ranging from tens of meters to a few kilometers. To overcome these limitations, various satellite-based models, known for their simplicity and accuracy, have been widely applied to estimate GPP at global and regional scales (Xin et al., 2015).

Satellite-based GPP models are typically divided into four categories: 1) statistical models, 2) light use efficiency (LUE) models (Wang et al., 2010a; Horn and Schulz, 2011), 3) process-based models (Li et al., 2014), and 4) machine learning-based models (Jung et al., 2019). Early models were mainly statistical, relying on vegetation indices and photosynthetically active radiation (PAR) (Myneni et al., 1995; Wu et al., 2011). Later, a variety of LUE-based models were introduced, such as MODIS GPP (Xin et al., 2017), the Carnegie-Ames-Stanford Approach (CASA) (Potter et al., 1993), EC-LUE (Zhang et al., 2015), and the Vegetation Photosynthesis Model (VPM) (Kalfas et al., 2011). At the same time, process-based models with satellite data were also advanced (Liu et al., 1997; Kato et al., 2013). Since the mid-2000s, the application of machine learning in GPP estimation has expanded, leading to the development of machine learning-based models (Lee et al., 2019a; 2019b). More recently, a GPP estimation method based on the Near-Infrared Reflectance of Vegetaion (NIRv), which multiplies near-infrared reflectance by NDVI, has been proposed (Badgley et al., 2017). This approach has shown strong correlations with in-situ GPP measurements across both spatial and temporal scales (Wang et al., 2021; Fig. 8). One advantage of this method is that it is relatively unaffected by saturation effects in areas with high leaf area index and shows low sensitivity to non-vegetation surfaces (Baldocchi et al., 2020).

Figure 8. Evaluations of (a) yearly and (b) monthly NIRv GPP using 104 flux sites (Wang et al., 2021).

The upcoming GPP products from the CAS500-4 satellite will use this intuitive NIRv-based algorithm, combined with PAR (NIRvP), to provide GPP estimates across South Korea at a spatial resolution of 30 meters every 10 days. These GPP products are expected to play a significant role in assessing climate change impacts and carbon sequestration potential (Park et al., 2021b; Kim et al., 2020a).

3.2.3. Leaf Area Index (LAI)

The LAI, defined as the one-sided leaf area per unit of the ground surface, reflects the amount of foliage in a canopy and serves as a key indicator of vegetation structure and functioning. LAI is a critical biophysical parameter essential for understanding various ecological processes, including photosynthesis, transpiration, and energy balance. LAI has been applied in numerous fields, including evapotranspiration (Kergoat et al., 2002; Wang et al., 2014), carbon cycle (Liu et al., 2018; Xie et al., 2019), land surface models (Sabater et al., 2008), crop yield estimation (Ines et al., 2013; Luo et al., 2020), and biodiversity (Skidmore et al., 2021).

In earlier research, Bunnik (1978) demonstrated that leaf area could be estimated using the ratio between red and near-infrared reflectance. Kanemasu et al. (1977) further expanded on this approach by estimating LAI using an empirical equation based on Landsat MSS bands. With the advent of satellite technology, such empirical approaches, using band ratios involving red and NIR bands or NDVI, were the primary methods for estimating LAI (Asrar et al., 1984; Peterson et al., 1987; Chen and Cihlar, 1996). However, empirical methods are often highly site- and sensor-specific, limiting their broader applicability. To overcome these limitations, researchers proposed radiative transfer model-based approaches, which offer more generalizable and robust results. MODIS was the first to provide an official LAI product, based on an algorithm that used the inversion of a three-dimensional radiative transfer model, with pre-calculated solutions stored in a look-up table (Buermann et al., 2002; Myneni et al., 2002). With the launch of satellites equipped with various sensors, different approaches for LAI estimation have been developed: empirical equations using vegetation indices (Chaurasia and Dadhwal, 2004; Maki and Homma, 2014), radiative transfer model-based methods (Atzberger and Richter, 2012; Thorp et al., 2012), machine learning techniques (Wang et al., 2017; Reisi et al., 2020; Kang et al., 2021; Lee et al., 2021b; Shen et al., 2022), data fusion methods (Verger et al., 2011; Yin et al., 2019), hybrid approaches (Wei et al., 2017; Liang et al., 2020), and multi-angle and multi-spectral approaches (Chen et al., 2003; Yang et al., 2010). In South Korea, research has tended to focus on the application of satellite-based LAI estimates for environmental analysis, rather than on the estimation process itself (Ha et al., 2008; Lee and Lee, 2017).

For the upcoming CAS500-4 mission, a machine learning-based model will be developed using field observations from the LAI network (Lee et al., 2024a; 2023e Fig. 9), which spans 33 sites across South Korea and is managed by the National Institute of Forest Science. The LAI map is expected to be produced at a 30-meter spatial resolution every 10 days.

Figure 9. Example of digital hemispherical photography of LAI network (Lee et al., 2023e).

3.2.4. Growth Stress Index (GSI)

Field surveys for forest growth assessment face significant challenges due to limitations in labor, time, and budget. This is particularly true in countries like South Korea, where complex mountainous terrain makes on-site investigations difficult. To overcome these obstacles, several studies have proposed using satellite imagery for forest ecosystem assessment and monitoring (Wang et al., 2010b; Choi et al., 2016; Barka et al., 2019). Despite uncertainties associated with satellite data—such as cloud cover, terrain effects, and resolution—satellite imagery remains a widely used tool for monitoring forest health and detecting stress changes, as it provides regular, comprehensive data on canopy vitality across large areas. For example, Barka et al. (2019) conducted a comparative study in central Europe using standardized MODIS Z-score NDVI (Z-NDVI) alongside field data (forest damage reports, tree ring data) to assess forest growth stress. They classified NDVI values to evaluate possible vitality problems in forests. Similar studies have employed satellite imagery to assess forest growth stress in various regions (Verbesselt et al., 2009; Barka et al., 2018; Puletti et al., 2019).

One notable example is the Center for Satellite Applications and Research (STAR), part of NOAA-NESDIS, which uses NDVI to monitor global vegetation health and provides datasets such as the World Vegetation Health Index. This service offers global data at 4 km resolution with weekly updates, covering vegetation health, phenology, density, productivity, and drought conditions from 1981 to the present. Similarly, the U.S. Forest Service’s ‘ForWarn’ system leverages MODIS satellite imagery to compare past and present NDVI values to detect disturbances such as wildfires, storms, insect outbreaks, and diseases (Pontius et al., 2020). In a similar vein, Choi et al. (2023) assessed tree mortality in transplanted trees in South Korea by analyzing the temporal variability of vegetation indices (e.g., NDVI, GNDVI, SAVI, and Advanced Vegetation Index [AVI]) derived from Sentinel-2 imagery.

Kim et al. (2019) used Gaussian and double logistic interpolation methods on Landsat-8 vegetation indices to create a disturbance damage map by comparing pre- and post-disturbance values (Fig. 10). Choi et al. (2016) also identified vulnerable areas in national forest parks by analyzing fluctuation patterns in MODIS EVI. While satellite-based forest stress monitoring studies are occasionally conducted in South Korea, various global service platforms for forest stress monitoring, such as the Forest Disturbance Monitor (FDM) and Operational Remote Sensing (ORS) have been operational internationally for years (Chastain et al., 2015). The upcoming CAS500-4 mission is expected to provide a forest growth stress index based on over 20 years of fused Landsat-MODIS vegetation index data (e.g., EVI) to support the protection and monitoring of South Korea’s forests.

Figure 10. Forest damage class mapping using change analysis of vegetation index (Kim et al., 2019).

Although satellite imagery has its limitations, the vast and regularly collected data from satellites has firmly established itself as an essential tool for evaluating forest health. As extreme weather events driven by climate change are expected to increase both intrinsic and extrinsic stress on vegetation, the vulnerability of forest ecosystems is likely to intensify (Cho et al., 2024). Satellite-based monitoring will be crucial in establishing an early warning system capable of quickly detecting and responding to abnormal forest stress.

As briefly mentioned before several forest ecology paragraphs, forest ecosystem monitoring and ecological variation detection have some challenges and uncertainties from remote sensing. To overcome these challenges, more sophisticated network systems are needed in the future. Due to the recent development of technology and the increase in available data, the inverse relationship between the temporal- and spatial resolution of remote sensing is gradually disappearing. Many scientists also share their field observation data over the network (e.g., FLUXNET, PEP725, NEON, etc.). If the short-term goal is to observe and monitor forest ecosystems in semi-real time using these field data, AI, and satellite images, we will move toward predicting forest ecosystem changes using remote sensing data in the future. For the preservation and management of forest ecosystems, it is effective to use satellite images that can periodically detect large areas. However, it is important to be cautious of fully trusting the remote sensing results of fragmented detection of forest ecosystems in which atmosphere-vegetation-soil is intricately intertwined.

3.3. Forest Disaster

Climate change has exacerbated the frequency and intensity of forest disasters, including wildfires, landslides, and pest infestations, resulting in increased human casualties and property damage. These events are typically large-scale and unpredictable, complicating efforts to accurately and efficiently assess damage through traditional field surveys and manual methods. In response to these limitations, satellite remote sensing has emerged as a critical tool for the monitoring and analysis of forest disasters.

3.3.1. Wildfire

In the past, wildfire damage assessments primarily relied on aerial photography to delineate affected areas, followed by field surveys to calculate the extent of the damage. However, this approach was labor-intensive and costly, making it difficult to quickly assess large areas. To address these challenges, Choi and Choi (1997) introduced a method using Landsat TM imagery to assess wildfire damage in Goseong, Gangwon Province, by comparing pre- and post-fire NDVI values. Despite its potential, the long revisit intervals of satellite imagery made immediate post-fire assessments challenging, limiting the use of remote sensing for rapid damage evaluation. With the increasing frequency of wildfires in the 2000s, remote sensing methods gained renewed attention as an efficient tool for analyzing vast regions. For example, Kim et al. (2002) analyzed wildfire damage severity using NDVI, while Won et al. (2007) assessed fire intensity in large burn areas using Landsat TM and ETM+ imagery, employing the Normalized Burn Ratio (NBR). Since the 2010s, the use of high-resolution satellite imagery has expanded. Lee et al. (2017) utilized Sentinel-2 data to address the limitations of the Landsat-NBR method, developing the Fire Burn Index (FBI) to enhance wildfire damage classification algorithms. Additionally, KOMPSAT-3 imagery, with a resolution of 0.7 m, was used to more accurately estimate changes in burn areas over time (Lee and Lee, 2020; Won et al., 2019). Lee and Jeong (2019) further advanced wildfire damage classification algorithms by using probability density functions with KOMPSAT-3A imagery. However, the temporal resolution of the KOMPSAT, which exceeds 30 days, posed limitations in disaster situations where rapid response is crucial.

In the 2020s, various studies have focused on addressing the limitations of single-satellite systems. These include the development of wildfire smoke detection algorithms (Kim et al., 2022b; Lee et al., 2024b) and the integration of spectral information with deep learning models to assess fire damage intensity (Cha et al., 2022; Lee et al., 2023c; Sim et al., 2020; Seo et al., 2023). Image fusion techniques (Kwak and Kim, 2023) have also emerged as a solution to overcome the limitations of individual satellite platforms. Notably, the CAS500-4, scheduled for launch in 2025, is expected to enable rapid post-fire damage assessments with its 5-meter high-resolution imagery and daily emergency imaging capabilities. The National Institute of Forest Science (2022) is developing RAPID MAPPING technology, which integrates AI to rapidly assess wildfire damage, including greenhouse gas emissions and biomass loss (Fig. 11). However, the optical sensor may present challenges in analyzing damage in the presence of smoke generated by wildfires. To address this, the integration of SAR or infrared imagery is necessary. Combining Sentinel-1 and Sentinel-2 data (Zhang et al., 2024) with other sensors such as MODIS and VIIRS (Luft et al., 2022) has shown promising results for wildfire monitoring. Such advancements in multi-sensor fusion research are expected to significantly enhance wildfire disaster response capabilities.

Figure 11. Example of rapid mapping technology (National Institute of Forest Science, 2022).

3.3.2. Landslide

In the early 2000s, researchers primarily focused on analyzing landslide-affected areas using Geographic Information System (GIS) techniques and optical satellite imagery. By integrating satellite imagery with field surveys, researchers identified landslide locations and extracted vulnerability factors based on data such as forest cover, soil, and digital topographic map. These datasets were used to construct various spatial models, including Digital Elevation Models (DEMs) (Jo and Jo, 2009; Kim et al., 2005; Lee et al., 2001; 2002; 2004). However, a key limitation of optical satellite sensors is that their observations can be obstructed by weather conditions, particularly in cloudy or rainy situations, which hampers accurate surface monitoring. To address this issue, microwave-based sensors such as SAR were introduced in the 2010s. SAR provides reliable surface data regardless of weather conditions, making it a valuable tool for monitoring landslide-prone areas. The launch of radar satellites like Soil Moisture Active Passive (SMAP) and Tropical Rainfall Measuring Mission (TRMM) enabled real-time monitoring of critical climatic factors, such as soil moisture and rainfall, which are essential for landslide prediction (Nam et al., 2014).

Since the 2020s, the development of landslide prediction models integrating both optical and SAR data has been actively pursued. In parallel, there has been increasing research on landslide detection and prediction using AI and deep learning technologies (Ahn et al., 2023; Seo and Lee, 2024). Lee et al.(2022) demonstrated the effectiveness of combining Landsat optical data with Sentinel-1 SAR data to create soil moisture map, highlighting the utility of this approach in landslide detection and prediction. Despite significant progress in using satellite imagery for landslide research in Korea, challenges remain in multi-sensor fusion and vulnerability monitoring technologies. For example, NASA SMAP satellite plays a key role in precisely evaluating and managing landslide-prone areas by detecting real-time changes in surface moisture (Stanley et al., 2021). However, in Korea, with its steep slopes and the prevalence of small-to-medium landslides, less than 1 ha, relying solely on single satellite data or SAR, which is sensitive to terrain effects, poses difficulties for effective landslide detection and management. Looking ahead, the CAS500-4 satellite, with its 5-meter high-resolution imagery and short three-day revisit period, is expected to enhance both the spatial and temporal resolution of landslide monitoring. The combination of different sensor data is anticipated to enable more precise management of landslide-prone areas.

3.3.3. Forest Pests and Diseases

In the process of pine tree mortality caused by pine wilt disease, a significant decline in vegetation vitality occurs, leading to a reduction in NIR reflectance. Early research on forest pests utilized Landsat TM imagery to detect large-scale damage areas by analyzing NDVI and NIR reflectance (Kim and Kim, 2008). As the need grew for more diverse satellite imagery to better capture the scale and distribution of forest pest damage, studies at the individual tree level using hyperspectral aerial and satellite imagery began in the early 2010s (Kim and Kim, 2015). By applying supervised classification techniques to time-series aerial photographs, researchers were able to classify damaged trees (Cha et al., 2017) and analyze spectral reflectance differences between infected and uninfected trees, improving the accuracy of pest detection (Kim et al., 2013).

Since 2014, research has expanded to include detection techniques that reflect the sporadic distribution characteristics of pests, incorporating high-resolution infrared and hyperspectral sensors mounted on UAVs (Kim et al., 2017), as well as automated classification using deep learning technology (Lee et al., 2019c; Kang et al., 2021). However, satellite-based pest detection research is still in its early stages. This is largely due to the lower spatial resolution of satellite imagery compared to UAVs or hyperspectral aerial imagery, making it difficult to precisely detect pests at the individual tree level (Chung et al., 2022). Furthermore, pest outbreaks are often localized and sporadic, which limits the effectiveness of satellite imagery in providing real-time detection due to its spatial and temporal resolution constraints (Kim et al., 2017).

Future research will likely focus on overcoming these limitations by integrating UAV imagery with AI-based analysis techniques. Additionally, with the upcoming launch of the CAS500-4, which will be capable of monitoring vegetation health every three days, it will be possible to detect forest health anomalies and identify pest risk areas on a larger scale. This, combined with drone imagery, will enable more proactive pest management activities.

Satellite data is an indispensable tool for the rapid and accurate detection of forest disasters, supporting all phases of forest disaster management, including mitigation, preparedness, response, and recovery. However, significant challenges remain in effectively utilizing satellite data for forest disaster research. First, accurate satellite data must be ensured through ground-truth validation to optimize its spatial and temporal resolution. This validation is essential for the effective use of satellite data in forest disaster management. Second, although satellite data provides broad coverage, its spatial and temporal resolution is often insufficient for real-time decision-making in rapidly evolving disaster scenarios. To overcome this challenge, advanced data fusion technologies that integrate multiple data sources, along with data-sharing platforms, are necessary to improve both the accuracy and timeliness of disaster response. Lastly, models that consider external factors, such as climate change, and improve adaptive capacity are needed to better predict and manage forest disaster patterns. Integrating real-time satellite data into these models will enhance the precision and operational responsiveness in disaster management. The integration of advanced satellite technologies with ground-truth validation is crucial for enhancing the efficiency and responsiveness of forest disaster management. This approach is essential for facilitating proactive and adaptive strategies in managing the escalating complexity of forest disaster challenges.

3.4. Forest Analysis Ready Data (F-ARD)

ARD research in Korea is still in its early stages. Initial efforts have focused on organizing the basic concept of ARD and reviewing international technological trends (Choi et al., 2021), with suggestions for constructing ARD for high-resolution satellites (Lee and Kim, 2021). Other studies that briefly mentioned ARD include terrain shadow exploration in satellite images (Kim et al., 2023), the development of CAS500-1/2 image utilization technology and operational systems (Yoon et al., 2020), and surface reflectance validation for KOMPSAT-3 (Kim and Lee, 2020).

In contrast, international research on ARD is more advanced. Key areas of focus include designing and developing ARD frameworks, as highlighted in CEOS ARD overview and recommendation studies (Lewis et al., 2018; Siqueira et al., 2019; Vlach et al., 2023), the production of ARD products using high-resolution satellites (Dwyer et al., 2018; Bachmann et al., 2021), and the enhanced ARD methodologies (Frantz, 2019; Bhandari et al., 2024). Additionally, ARD research has been actively integrated with Open Data Cube for cloud-based operations, with many of these studies being conducted at the national level (Killough, 2019; Killough et al., 2020). Numerous studies using ARD data have been applied to areas such as tree cover mapping (Egorov et al., 2018), vegetation mapping (Bendini et al., 2020), global land cover production (Potapov et al., 2021), and national forest inventories (Lister et al., 2020), demonstrating the applicability of ARD in forest science.

ARD research is largely centered around CEOS, with most studies based on CEOS ARD standards and data. CEOS is working to establish international ARD specifications and build a comprehensive database. Since the U.S. Geological Survey (USGS) began producing Landsat ARD in 2017, satellites like Sentinel-2, the Environmental Mapping and Analysis Program (EnMAP), and PROBA-V have also been registered with the CEOS ARD Surface Reflectance dataset. International organizations involved in building CEOS ARD include USGS, the European Space Agency (ESA), Vlaamse Instelling voor Technologisch Onderzoek (VITO), and the German Aerospace Center (DLR). In Korea, KARI is currently developing KOMPSAT-3 ARD (Table 1). The National Forest Satellite Information & Technology Center is also working on producing Forest-ARD tailored to forest ecosystems (Seoul National University, 2021), and the National Land Satellite Center is planning to develop ARD (National Land Satellite Center, 2021).

Table 1 . CEOS analysis-ready datasets of surface reflectance (Committee on Earth Observation Satellites, 2024).

CEOS Analysis-Ready Datasets (Surface reflectance only)
Product	Agency	Mission/Instruments
EnMAP	DLR	EnMap
Landsat collection2	USGS	Landsat 4,5,6,7,8,9
Landsat collection2 U.S. ARD	USGS	Landsat 4,5,6,7,8,9
L8 SR	AIR-CAS (China)	Landsat 8
PROBA-V L3 (0.1, 0.333, 1 km)	VITO/ESA	PROBA-V
Sentinel-2 L2A	ESA	Sentinel 2A,2B

Under review
Product	Agency	Mission/Instruments
DESIS L2A	DLR	DESIS-on-ISS
Gaofen-1/6 SR	AIR-CAS (China)	Gaofen-1

Under developments/Assessment
Product	Agency	Mission/Instruments
Envisat MERIS	ESA	Envisat
ERS ATSR	ESA	ERS-1, ERA-2
Fused S-2 & L-8/9 (Level-2F)	ESA	Sentinel-2A, 2B; Landsat 8,9
Harmonized S-2 & L-8/9 (Level-2H)	ESA	Sentinel-2A, 2B; Landsat 8,9
KOMPSAT-3	KARI	KOMPSAT-3
Resourcesat-2/2A	ISRO	Resourcesat-2, 2A
Sentinel-3 SYN SDR Product	ESA	Sentinel-3A, 3B
SPOT 1-7 Surface Reflectance	SANSA	SPOT 1, 2, 3, 4, 5, 6, 7
THEOS-1 Surface Reflectance	GISTDA	THEOS-1

Future
Product	Agency	Mission/Instruments
CHIME L2A	ESA	CHIME
CHIME L2H/L2F	ESA	CHIME
LSTM L2A SR	ESA	LSTM
LSTM L2H/L2F SR	ESA	LSTM

AIR: Aerospace Information Research Institute, CAS: Chinese Academy of Sciences..

Commercial satellites are also producing and distributing ARD. Planet Scope and Rapid Eye offer ARD products, while GeoEye-1 and Worldview-2 are releasing API-based ARD prototypes. By producing data that adheres to CEOS ARD guidelines—not only for KOMPSAT but also for next-generation CAS—Korea is expected to improve interoperability between domestic satellites and increase its contributions to the international satellite community. Additionally, as CEOS expands to include aquatic reflectance for aquatic environments, there is a growing need for ARD development focused on specific purposes. Given that over 60% of Korea is covered by forests, the demand for forest-specific ARD is expected to rise. While land surface products utilizing ARD have already demonstrated their utility, producing high-resolution and consistent surface reflectance remains challenging due to Korea’s complex forest terrain (Seoul National University, 2021). Preprocessing that accounts for topographic features and directional scattering of forest canopies is required, but this remains a challenge. Furthermore, improvements in cloud masks, quality flags, and pixel reconstruction for ARD are necessary for technical advancements (Vlach et al., 2023). Looking ahead, the F-ARD development that addresses the complex preprocessing, technical limitations, and spatial-temporal missing data in South Korea’s forest data has the potential to position the country as a global leader in forest satellite remote sensing.

Remote sensing in the forestry sector dates back to 1971, with the first notable application being the use of aerial photography and remote sensing technology in the 1st National Forest Inventory. In the 1990s, research projects aimed at assessing forest resources, such as the creation of foest-type map, land cover map, and the monitoring of North Korean forests, were actively conducted. Since 2017, significant advancements have been made, particularly in the use of deep learning, which is now applied across nearly all areas of remote sensing, from image preprocessing to data utilization.

More recently, the forestry sector has been focusing on solving the challenges of carbon neutrality—a major issue in the field—and on innovating precision forest management. This includes preventing forest disasters and assessing the health of forest ecosystems using satellite and remote sensing technologies. These efforts aim to provide public, economic, and cultural benefits through digital transformation, moving beyond traditional field-based, human-oriented methods by integrating cutting-edge science and technology.

To overcome future vulnerabilities in forest management, it is essential to explore development plans that harmonize advanced science and technology with traditional forestry practices. A rapid transition to an intelligent, smart forest management system based on satellite data, LiDAR, and AI could provide the necessary solutions. As science and technology continue to evolve rapidly, it is expected that by 2030, forest changes will be observable on an annual cycle and by 2070, on a daily cycle, with real-time data provided to the public. The statistical uncertainty caused by limitations in sampling points for the National Forest Inventory could be reduced to within 3% of the statistical tolerance through real-time, tree-level precision forest management, ushering in an era of comprehensive forest management across the country.

As we prepare for the next 100 years of forestry, digital technologies will play an increasingly important role in forest monitoring, planning, and management. The convergence of agricultural and forestry satellite data with AI will drive future innovations in forest management. By sharing field observation data and satellite information across sectors, from space to the ground, we can expand agricultural and forestry satellite information services, making them more accessible and user-friendly for the public. This will also pave the way for an era of internationally reliable digital forest information.

This study was carried out with the support of the National Institute of Forest Science (Project No. ‘FM0103-2021-01-2024’, ‘FM0103-2021-02-2024’, ‘FM0103-2021-04-2024’).

No potential conflict of interest relevant to this article was reported.