Development of the false alarm filtering method for GEO-KOMPSAT-2A wildfire detection product

Development of the false alarm filtering method for GEO-KOMPSAT-2A wildfire detection product | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development of the false alarm filtering method for GEO-KOMPSAT-2A wildfire detection product Seoyoung CHAE, Yong-Sang Choi, Hwayon Choi This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5749795/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 3 You are reading this latest preprint version Abstract Wildfires have caused significant damage to the economy, human health, and ecosystems over extended periods. Satellites offer advantages as wildfire detection tools because they provide near real-time, continuous observations in areas that are difficult to manage directly by humans. But in previous studies, satellite active wildfire detection products were prone to generating false alarms in areas with high reflectance, such as bare soils, urban regions, water bodies, and clouds. This issue arises because infrared channels were used for wildfire detection, but their observations include both solar and thermal radiances. In this study, we apply a calculation, based on a previous study, to wildfire detection that reduces solar reflectance radiance and propose a filtering method to eliminate false alarms occurring in cloud edge areas. South Korea, where detailed annual wildfire information is available, was selected as the study area to develop a region-specific false alarm filtering algorithm for wildfire detection product. The filtering algorithm was compared to the GEO-KOMPSAT-2A operational product using data from 62 wildfire cases that occurred in 2022. As a result, the number of false alarms was significantly reduced from 896 to 25, the False alarm ratio (FAR) decrease from 96.34–47.17% and the Probability of Detection (POD) showed a slight decrease from 53.23–45.16%. The results of this study are expected to contribute to more efficient disaster and risk management by enabling more accurate wildfire detection. Wildfire Geostationary Satellite False alarm Filter Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction Wildfires cause various types of damage to human life over an extended period, which can be categorized into economic, health, and ecological impacts. Economically, wildfires result in significant losses. Over the past decade in South Korea, 5,668 wildfires have caused financial losses totaling 4.75 trillion KRW and affected approximately 40,037 ha of 6.2% of the nation’s total forested area (Korea Forest Service 2023 ). To human health, wildfires can cause both acute and chronic health issues (Sullivan et al. 2022 ). Recent findings confirm associations between wildfire smoke exposure and respiratory health outcomes, with the clearest evidence for exacerbations of asthma (Reid and Maestas 2019 ). In particular, there is a significant association between wildfires and the exacerbation of conditions such as asthma, chronic obstructive pulmonary disease, bronchitis, and pneumonia (Cascio 2018 ). Wildfires cause irreversible damage to ecosystems. They can degrade ecosystem services such as water supply, nutrient cycling, biodiversity, and carbon storage. Additionally, wildfires affect vegetation structure on varying scales depending on fire conditions, pre- and post-disturbance climate conditions, and ecosystem type. Changes in the structure and composition of vegetation in these habitats can indirectly impact animal populations (Sullivan et al. 2022 ; Keith et al. 2009 ; Mowat et al. 2015 ; Litt and Steidl 2011 ). The ash from burned vegetation and leaf litter reduces the surface area available for evapotranspiration, decreases storage capacity for maintaining and retaining water systems, and removes obstacles that could impede surface water runoff and accelerate soil erosion (Shakesby 2011 ). Thus, wildfires are a complex issue that extend beyond mere economic losses, posing serious threats to human health and ecosystems. Satellite wildfire detection offers significant advantages by enabling real-time detection of wildfires in areas that are difficult to monitor from the ground. Nowadays with advancements in technology and a substantial reduction in satellite launch costs, attempts to utilize satellites for wildfire detection have increased, greatly contributing to wildfire response efforts (Barmpoutis et al. 2020 ). Remote sensing via satellites is one of the fastest means of obtaining essential data for research before and after disasters such as wildfires. By providing real-time data, it plays a critical role in disaster analysis and response (Voigt et al. 2016 ). In particular, during wildfire events, satellites enable the rapid collection of data that is difficult to obtain from the ground. This facilitates immediate action by practitioners and responders. Such satellite data go beyond merely detecting disasters; they play a critical role in forming evacuation plans and disaster prevention strategies by providing timely information on weather patterns and wildfire behavior (Joyce et al. 2009 ). Wildfire managers can use satellites to predict the scale and spread of wildfires, enabling them to quickly provide critical information, such as evacuation alerts, to residents. Based on this information, more accurate decisions can be made in dynamic situations. In this way, satellites offer valuable, rapid, and efficient information during disaster response. These advantages are particularly beneficial in areas where ground detection is difficult, aiding in the accumulation of wildfire data and facilitating early suppression efforts. However, satellite wildfire detection products have been criticized for not aligning well with actual wildfire data. Past studies have shown that this method tends to exhibit over-detection when compared to actual wildfire data. The false alarms in wildfire products are often caused by non-fire surfaces with high reflectance, such as bare ground, urban areas, water bodies, and clouds (Giglio et al. 2003). Ying et al. ( 2019 ) compared wildfires in Yunnan Province, southwestern China, between February 2002 and November 2015 with The Moderate Resolution Imaging Spectroradiometer (MODIS) wildfire data. They found that the MODIS wildfire product recorded significantly more wildfires than ground monitoring records and exhibited considerable omission errors. The study highlighted that detection errors might occur due to terrain and climatic variability, emphasizing the importance of ground-based data for validation and accuracy. We hypothesized that GEO-KOMPSAT-2A (GK2A) shares similar issues with MODIS, as its current wildfire detection algorithm is based on the MODIS algorithm (National Meteorological Satellite Center 2019 ). Among the various issues identified, we analyzed the most important issue of GK2A wildfire products: the high number of false alarms. Therefore, this study aims to develop a filtering method to reduce false alarms effectively. To develop a filtering algorithm capable of addressing such false detections, accurate data on the time and location of wildfire occurrences are essential to validate the algorithm South Korea was chosen as the study area due to its comprehensive records of wildfire occurrences, including precise time and location data. Using 19 wildfire cases from 2022, a foundational algorithm was created, which was then compared against 62 actual wildfire cases for evaluation. 2. GK2A forest fire product analysis The GK2A wildfire product, forest fire (FF), defined two thresholds for daytime and nighttime, determined by an 85° solar zenith angle, to reduce solar effect of 3.8 µm channel. Additionally, it checked whether the area was bare ground, urban, or beach to minimize false alarms caused by the effects of solar reflectance during the daytime. Despite these efforts, the false alarm in the GK2A FF product remained significant. The GK2A FF product recorded a total of 869 false alarms in the entire dataset used for the study. False alarms can arise from various factors, but the combination of the 3.8 µm channel and cloud conditions identified as major contributors. The 3.8 µm channel, commonly used for wildfire detection, is unique compared to other IR channels as it is located between the visible and infrared of the spectrum. This positioning makes it be affected to both solar radiance and thermal radiance. As a result, false alarms are more frequent during daytime due to excessive radiative energy from cloud scattering, which the channel interprets as high temperatures. This issue is particularly pronounced near cloud edges, where irregular scattering can accelerate the problem, leading to a higher likelihood of false alarms (Emde et al. 2021 ). In practice, an analysis of false alarm cases showed that they predominantly occurred during the daytime and in proximity to cloud edges (Table 1 ). Table 1 Timing and presence of surrounding clouds of false alarms in GK2A wildfire detection products. Total false alarms Near cloud Non near cloud 869 765 104 Daytime Nighttime 598 271 To address these false alarms, this study applied two primary filtering methods. The first method involved separating solar radiance and thermal radiance during the calculation part, utilizing only thermal radiance. This is suggested by Choi et al. ( 2007 ), using formula to calculate the cloud-reflected radiance in the 3.8 µm channel. The second method introduced an additional process to verify pixels detected as wildfires by the GK2A whether actually fires or not. This verification process utilized the physical differences between fire and cloud characteristics and comprised three different tests. These approaches aimed to correct false alarms caused by clouds and the 3.8 µm channel and filter out false alarms in wildfire detection outputs from the GK2A satellite. 3. Geostationary satellite and validation data GK2A is a geostationary satellite launched by the Republic of Korea and its data is provided by the National Meteorological Satellite Center (NMSC). We used two Level 1B (L1B) data the 3.8 µm channel and 10.5 µm channel and four Level 2 (L2) data the Cloud detection (CLD), Forest fire detection (FF), Land surface emissivity (LSE), Normalized vegetation index (NDVI). GK2A offers every 2-minutes for the Korea region so all data have 2 minutes for temporal resolution except for LSE and NDVI, which are calculated once per day (Table 2 ). Table 2 List of GK2A data and their characteristics used in the study. Product name (preprocessing level) Central wavelength (µm) Spectral band and range (µm) Spatial resolution (km) Temporal resolution (min) SW038 (1B) 3.8 B7: 3.74–3.96 2 2 IR105 (1B) 10.5 B13: 10.25–10.61 2 2 CLD (2) 2 2 FF (2) 2 2 LSR (2) 2 Once a day NDVI (2) 2 Once a day The two L1B datasets we used are 3.8 µm and 10.5 µm. The hot spot finding method using brightness temperature differences between these two channels was first proposed by Dozier et al. (1981). After this, research on the Brightness temperature difference hot spot detection method was continued and adopted operationally. When two surfaces with different temperatures exist within the same pixel, the resulting radiation fields will vary, causing the two channels to respond differently to thermal emissions. The thermal infrared channels, such as the 3.8 µm channel in GK2A, are more sensitive to thermal radiation than the 10.5 µm channel. This difference enables the detection of thermal sources in GK2A images using these two channels. The four L2 datasets we used are Cloud detection (CLD), Forest fire detection (FF), Land surface emissivity (LSE), Normalized vegetation index (NDVI). Cloud detection is used for cloud masking, which are the most significant contamination factor in satellite images. Clouds disturb satellite observations at high elevations by decreasing brightness temperatures. CLD is always used in land observation algorithms for masking data due to this characteristic. GK2A CLD data consist of the following categories: 0: Cloud (High Confidence), 1: Cloud (Low Confidence) and 2: Clear (High Confidence). In this study, we masked all areas labeled as 0 and 1 and utilized only the clear regions. The GK2A CLD algorithm was designed to minimize undetected clouds. However, the accuracy of the product tends to decrease during winter due to cooled land surfaces according to GK2A CLD Algorithm Theoretical Basis Document (ATBD). FF is used as part of the algorithm and as a comparison product. The GK2A FF algorithm detects fires by using the median field of the temperature base map and comparing the difference between the base map temperature and the raw pixel temperature against thresholds. This detection method is based on the MODIS active fire detection algorithm, with modifications and adjustments made for operational use on GK2A (National Meteorological Satellite Center 2019 ). FF data consist of 0: Non-fire and 1: Fire. The filtering algorithm was applied to GK2A fire pixels with a value of 1 in the FF data, and POD and False Alarm Rate (FAR) were compared before and after filtering. LSE is used to calculate ground albedo to separate ground-reflected radiance from the 3.8 µm channel. LSE serves the channels. Since LSE is provided differently for each channel, we used only the data corresponding to the 3.8 µm channel. NDVI is used to identify land and water. Pixels with an NDVI value of 0 were masked as water. The wildfire cases used for validation are provided by 2022 annual wildfire cases list published by Korea forest service. These data include year, month, day, and time of wildfire occurrence and extinguishment, town-level location, and damaged area. We chose 19 fires which are 1) over 10 ha damaged, 2) no clouds at that time for the frame of algorithm. The data provided refers to 3 hours, including 1 hour before and after the wildfire occurrence time, and additional cases were used as validation data if the occurrence and extinguishment times were within the 3-hour range of the available data for one case. As a result, a total of 62 wildfire cases were collected, with affected areas ranging from 0.01 ha to 16,301.98 ha. (Table 3 ). Table 3 Number of wildfire cases and distribution of damaged areas used for validation Total fire cases Fire cases over 1 ha Fire cases over 10 ha 62 38 19 4. Filtering algorithm The algorithm is composed of six components: input data, calculation, basic test, absolute fire test, time correlation test, and pop-up fire test (Fig. 1 ). To solve the two problems discussed in Section 2 , solar radiance in 3.8 µm channel and cloud edges, it is essential to focus on the calculation part and the three test parts located on the far-right side of the flowchart. This section will first provide a detailed explanation of the two parts, followed by descriptions of the remaining parts and then the steps required for the localization of the algorithm. 4.1. Calculation for reducing solar radiance from observations In the calculation part, to separate the solar radiance from observations, the method proposed by Choi et al. ( 2007 ) was applied during the data processing stage. \(\:{L}_{3.8}^{th}={L}_{obs}-{L}_{3.8}^{sr}\) ( 1 ) 3.8 µm channel thermal radiance \(\:{L}_{3.8}^{th}\) is calculated to minus from observed 3.8 µm channel radiance \(\:{L}_{obs}\) to ground-reflected radiance \(\:{L}_{3.8}^{Sr}\) . In the Choi et al. ( 2007 ), ground albedo \(\:{A}_{g}\) and \(\:{L}_{3.8}^{sr}\) when \(\:{A}_{g}=1\) are used for finding \(\:{L}_{3.8}^{Sr}\) . \(\:{L}_{3.8}^{sr}={A}_{g}{L}_{3.8}^{sr}{(A}_{g}=1)\) ( 2 ) \(\:{A}_{g}\) is calculated using GK2A LSE product with the formula below. \(\:{A}_{g}=1-land\:surface\:emissivity\) ( 3 ) \(\:{L}_{3.8}^{sr}{(A}_{g}=1)\) as created as a lookup table, calculated using Streamer, a type of radiative transfer model, for solar zenith angles ranging from 0° to 90° under the condition \(\:{A}_{g}=1\) . After using these methods to 3.8 µm channel, brightness temperature of the non-fire surface is decreasing compared to wildfires (Fig. 2 ). Calculated 3.8 µm channel solar-free radiance and observed 10.5 µm channel radiance are changed to brightness temperature through calibration table of GK2A. For the basic test, calculate the locally averaged BT value of the 3.8 µm channel \(\:\left[{BT}_{3.8,t}\right]\) , and the signal \(\:{S}_{t}\) which is temperature difference ratio of 3.8 µm and 10.5 µm channels BT. \(\:{S}_{t}=\left(\frac{{BT}_{3.8,t}}{{BT}_{10.5,t}}\right)-\left(\frac{{BT}_{3.8,t-1}}{{BT}_{10.5,t-1}}\right)\) ( 4 ) 4.2. Classification of wildfires and clouds The enter conditions for the tests are set as multiples of the channel noise \(\:Cn\) , which was calculated by accumulating \(\:{S}_{t}\) over one day and determining its standard deviation. When \(\:Cn\) was obtained separately by season and time, there was no change in the value. This suggests that the same value can be used regardless of the season, so it was fixed as 0.0011. The absolute fire test is designed to confirm pixels as wildfires that exhibit abnormally large signals. The enter condition is \(\:{S}_{t}>25Cn\) , which is set to identify wildfire pixels with significantly stronger signals compared to other tests. In this test, if no clouds are present in the 8 surrounding pixels of the test pixel, the test pixel is immediately confirmed as a wildfire. The time correlation test identifies a pixel as a wildfire if the temperature at a fixed location increases consistently over time. Unlike wildfires, clouds move over time, so we use this feature to filter out pixels that move over time as cloud while fixed pixels are identified as wildfires. The starting conditions of this test are \(\:{S}_{t}>2Cn\:and\:\:{S}_{t-1}>Cn\) . Pixels meeting these conditions are designated as a test field, which includes the pixel itself and its 8 surrounding pixels (a total of 9 pixels). Additionally, 25 fields which have 3×3 pixels around the test field from \(\:{S}_{t-1}\) are designated as compare fields. To detect movement, the correlation between the test field and the comparison field is calculated to determine whether the test field is a cloud or a forest fire. Among the calculated correlation values, if the compare field at the same position as the test field meets the following two conditions: 1) it has the highest value, and 2) the value is greater than 0.5, the pixel is determined to have not moved and is confirmed as a wildfire. The pop-up fire test is designed to identify wildfire signals that appear abruptly with high temperatures and then disappear without prior warning. The enter condition is \(\:{S}_{t}>2Cn\) . For pixels meeting the enter condition, it is first checked whether the pixel was not a cloud on \(\:{S}_{t-1}\) . Then, on \(\:{S}_{t}\) , the surrounding 24 pixels are checked to ensure there are no clouds present. This test adopts a method similar to the absolute fire test but is tailored to detect wildfires with slightly less temperature increase. The spatial resolution of GK2A is 2 km × 2 km which is significantly larger than the size of most wildfires. As a result, the signal observed from a single pixel may appear smaller than the rapid temperature change caused by the wildfire. Therefore, this test was added as a separate method from the absolute fire test (Fig. 3 ). Finally, any pixel that passed at least one of the absolute fire test, time correlation test, or pop-up fire test was confirmed as a wildfire. 4.3. Minor parts of the filtering algorithm In the input data part, check all datasets exist on the correct directories. In the basic test part, checking test pixels are satisfied to enter the time correlation test, the pop-up fire test and the absolute fire test. If the test pixel is a clear sky, detected as fire in the GK2A product, located on land and \(\:{BT}_{3.8,t}\) is bigger than \(\:\left[{BT}_{3.8,t}\right]\) , check \(\:{S}_{t}\) to enter each test or not. The calculated \(\:Cn\) was applied with different multipliers depending on the entry requirements for each test. The tests were conducted in three stages, in descending order of entry requirement magnitude. 4.4. Methods for localization values This filtering algorithm can be localized for specific regions, as each region has unique climatic characteristics and different latitudes and longitudes, making localization necessary. \(\:{L}_{3.8}^{sr}{(A}_{g}=1)\) was calculated using a radiative transfer model to create a lookup table. This table was generated based on the latitude and longitude of Seoul (37.5° N, 127° E). Therefore, the model needs to be adjusted and recalculated to match the desired region specified by the user. The \(\:Cn\:\) used as a threshold in the algorithm was calculated using the standard deviation of the time difference temperature over one day. Although this value is fixed and used in the algorithm, it was derived based on data from South Korea. Therefore, it reflects regional characteristics and may require adjustment for other locations. All these adjustments were calibrated based on the Korean Peninsula, specifically South Korea. With a total area of 100,410 ㎢, all experiments were conducted within this geographical scope. Therefore, users should keep this in mind and adjust the thresholds accordingly when applying the algorithm to other regions. 5. Fire detection results and validation The filtered GK2A FF product successfully detected 28 out of 62 wildfire cases. For wildfires with an affected area of at least 1 ha, it detected 26 out of 38 cases, and for wildfires with an affected area of at least 10 ha, it detected 18 out of 19 cases. During the entire observation period, there were 25 false alarms despite no wildfires occurring. For all wildfires, the POD was 45.16%, and the FAR was 47.17%. For wildfires with an affected area of at least 1 ha, the POD was 71.05%, and the FAR was 49.02%. For wildfires with an affected area of at least 10 ha, the POD was 94.74%, and the FAR was 58.14%. The outputs used in the analysis are of two types: 1) the GK2A wildfire detection product, and 2) the filtered output, which applies the research algorithm to the GK2A results. Each output was evaluated based on the wildfire and fire incident data provided by the Korea Forest Service. The filter uses empirical thresholds and \(\:Cn\) . POD and FAR are compared to show the performance. as multiples of \(\:Cn.\) Additionally, this study examines how detection performance varies based on the area of the wildfire, as the wildfire cases used for validation differ in size. A total of 62 wildfire cases were used for validation, of which 38 had an affected area of at least 1 ha, and 19 had an affected area of at least 10 ha. And then present the performance of each test. The GK2A FF product successfully detected 33 out of 62 wildfire cases. For wildfires with an affected area of at least 1 ha, it detected 29 out of 38 cases, and for wildfires with an affected area of at least 10 ha, it detected all 19 cases. During the entire observation period, there were 869 false alarms despite no wildfires occurring (Table 4 ). For all wildfires, the POD was 53.23%, and the FAR was 96.34%. For wildfires with an affected area of at least 1 ha, the POD was 76.32%, and the FAR was 96.77%. For wildfires with an affected area of at least 10 ha, the POD was 100%, and the FAR was 97.86%. Table 4 Contingency table of before and after filtering products. GK2A (over 10 ha) Reported Fire Non-fire Detected Fire 33 (19) 869 (869) Non-fire 29 (0) 0 (0) GK2A + the filter (over 10 ha) Reported Fire Non-fire Detected Fire 28 (18) 25 (25) Non-fire 34 (1) 844 (844) Figure 4 shows that the POD does not significantly decrease from the total cases to wildfires over 10 ha, even after applying the filter. However, the FAR decreases substantially compared to the slight drop in POD. The apparent increase in FAR for over 10 ha in the FAR graph occurs because the number of detected cases, which serves as the denominator, decreases significantly when the range shifts from over 1 ha to over 10 ha (Fig. 4 ). The filtered wildfire detection product reduced 844 false alarms compared to the GK2A wildfire detection product (Table 3 ). This corresponds to a reduction of approximately 97% of the total false alarms through filtering. In some instances, the alarm encompassed multiple tests on both sides. Consequently, the maximum number of true alarms for any individual test did not exceed 28. However, when the true alarms from all three tests were combined, the total exceeded 28. This was also observed for false alarms. The absolute fire test recorded 13 out of 28 true alarms and 3 out of 25 false alarms. The time correlation fire test reported 20 out of 28 true alarms and 8 out of 25 false alarms. Lastly, the pop-up fire test documented 21 out of 28 true alarms and 17 out of 25 false alarms. From a numerical perspective, the pop-up fire test appears to contribute significantly to false alarms, potentially rendering it seemingly less useful. However, the situation changes when examining the number of wildfires each test uniquely detected. The absolute fire test did not detect any wildfires independently, while the time correlation fire test detected 5, and the pop-up fire test independently detected 8. These results suggest that utilizing each test independently aligns with the goal of the algorithm in detecting wildfires as a critical disaster (Table 5 ). Table 5 The detection results of each test on the filtering algorithm. Absolute fire test Time correlation test Pop-up fire test True alarm False alarm True alarm False alarm True alarm False alarm Detection result 13 3 20 8 21 17 Solo detected result 0 4 8 6. Conclusion and Discussion In this study, the performance of the GK2A wildfire detection algorithm was improved by applying a filtering technique with three tests to solve the false alarms. After the filtering, the GK2A wildfire detection product had a POD of 45.16% and an FAR of 47.17% for the 62 wildfires, reducing the total false alarms to 25 —the original GK2A wildfire detection product showed a POD of 53.23% and an FAR of 96.34% for a total of 62 wildfires, with 869 false alarms recorded. Geostationary satellite-based wildfire detection still faces hardware limitations. Currently, the spatial resolution of GK2A is relatively large compared to the typical size of wildfires, making it less effective for detection. Most wildfires in South Korea affect areas smaller than 10 ha, yet the system requires wildfires larger than 10 ha to demonstrate sufficient detection capabilities. To address this issue, integrating low Earth orbit (LEO) satellites or linking detection methods with ground-based cameras may be a viable solution. Utilizing such satellites in combination with geostationary systems could enhance wildfire detection capabilities. Additionally, smoke is an important signal for wildfire detection; however, it has not been incorporated into this study. Traditionally, heat and smoke are the primary indicators used for wildfire detection, but smoke tends to lower channel temperatures, potentially delaying detection when using satellite data. Additionally, smoke from large wildfires is often misclassified as clouds, leading to the masking of wildfire-affected areas. Since smoke and wildfires are inherently linked, future research will consider this factor and incorporate it into the analysis. Declarations Funding This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2018R1A6A1A08025520). This work was supported by the Specialized University Program for Confluence Analysis of Weather and Climate Data of the Korea Meteorological Institute (KMI), funded by the Korean government (KMA). Competing Interests The authors have no relevant financial or non-financial interests to disclose. References Barmpoutis P, Papaioannou P, Dimitropoulos K, Grammalidis N (2020) A Review on Early Forest Fire Detection Systems using Optical Remote Sensing. Sensors 20(22):6442 Cascio WE (2018) Wildland Fire Smoke Hum Health Sci Total Environ 624:586–595 Choi Y, Ho C, Ahn M, Kim Y (2007) An Exploratory Study of Cloud Remote Sensing Capabilities of the Communication, Ocean and Meteorological Satellite (COMS) Imagery. Int. J Remote Sens 28(21):4715–4732 Dozier J (1981) A Method for Satellite Identification of Surface Temperature Fields of Subpixel Resolution. Remote Sens Environ 11:221–229 Emde C, Yu H, Kylling A, Van Roozendael M, Stebel K, Veihelmann B, Mayer B (2021) Impact of 3D Cloud Structures on the Atmospheric Trace Gas Products from UV-VIS Sounders–Part I: Synthetic Dataset for Validation of Trace Gas Retrieval Algorithms. Atmospheric Measurement Techniques Discussions, 2021, 1–28 Giglio L, Csiszar I, Justice CO (2006) Global Distribution and Seasonality of Active Fires as Observed with the Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) Sensors. J Geophys Research: Biogeosciences, 111(G2). Joyce KE, Wright KC, Samsonov SV, Ambrosia VG (2009) Remote Sens Disaster Manage Cycle Adv Geoscience Remote Sens 48(7):317–346 Keith H, Mackey BG, Lindenmayer DB (2009) Re-Evaluation of Forest Biomass Carbon Stocks and Lessons from the World's most Carbon-Dense Forests. Proceedings of the National Academy of Sciences, 106(28), 11635–11640 Korea Forest Service (2022) 2022 Annual Report on Wildfire Statistics. Publishing Korea Forest Service. https://www.index.go.kr/unity/potal/main/EachDtlPageDetail.do?idx_cd=1309 , Accessed Nov. 18, 2024 Korea Forest Service (2023) 2023 Annual Report on Wildfire Statistics. Publishing Korea Forest Service. https://www.index.go.kr/unity/potal/main/EachDtlPageDetail.do?idx_cd=1309 , Accessed Nov. 18, 2024 Litt AR, Steidl RJ (2011) Interactive Effects of Fire and Nonnative Plants on Small Mammals in Grasslands: Effets Interactifs Du Feu Et Des Plantes Non Indigènes Sur Les Petits Mammifères. Dans Les Prairies Wildl Monogr 176(1):1–31 Mowat EJ, Webb JK, Crowther MS (2015) Fire-mediated Niche‐separation between Two Sympatric Small Mammal Species. Austral Ecol 40(1):50–59 National Meteorological Satellite Center (2019) NMSC: GK2A AMI Algorithm Theoretical Basis Document Cloud Mask. Publishing NMSC. http://nmsc.kma.go.kr/homepage/html/base/cmm/selectPage.do?page=static.edu.atbdGk2a . Accessed 25 No-vember 2024 National Meteorological Satellite Center (2019) NMSC: GK2A AMI Algorithm Theoretical Basis Document Forst Fire. Publishing NMSC. http://nmsc.kma.go.kr/homepage/html/base/cmm/selectPage.do?page=static.edu.atbdGk2a . Accessed 25 No-vember 2024 Reid CE, Maestas MM (2019) Wildfire Smoke Exposure Under Climate Change: Impact on Respiratory Health of Affected Communities. Curr. Opin Pulm Med 25(2):179–187 Shakesby RA (2011) Post-Wildfire Soil Erosion in the Mediterranean: Review and Future Research Directions. Earth-Sci Rev 105(3–4):71–100 Sullivan A, Baker E, Kurvits T, Popescu A, Paulson AK, Christianson C, Tulloch A, Bilbao A, Mathison B, Robinson C (2022) C. Spreading Like Wildfire: The Rising Threat of Extraordinary Landscape Fires. Voigt S, Giulio-Tonolo F, Lyons J, Kučera J, Jones B, Schneiderhan T, Platzeck G, Kaku K, Hazarika MK, Czaran L (2016) Global Trends. Satellite-Based Emerg Mapp Sci 353(6296):247–252 Ying L, Shen Z, Yang M, Piao S (2019) Wildfire Detection Probability of MODIS Fire Products Under the Constraint of Environmental Factors: A Study Based on Confirmed Ground Wildfire Records. Remote Sens 11(24):3031 Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 16 Jan, 2025 Editor assigned by journal 03 Jan, 2025 First submitted to journal 01 Jan, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5749795","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":402963474,"identity":"337973c9-73c0-4e07-8910-4fb2e84baed5","order_by":0,"name":"Seoyoung CHAE","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAw0lEQVRIiWNgGAWjYJACAx4GZjl+EjQwg7UYSzaQooUBqCVxwwFiNfC39x8oeNtmzbj5dvMziY97GOT5xQholjhzmMFwbls6s9mdY2aSM54xGM6cnYBfi4FEMoMxb9thNrMbCcbGPAcYEgxuE6mFx3hG+mfjP6RokTCQyDF8zECMFqBfDAznnEs3kLiRU/iw54AEYb/wtzc+M3hTZl3fPyN9w4EfB2zk+aUJaAECNgNkWwkqBwHmB0QpGwWjYBSMgpELAH2ZQEkr3FzNAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0009-0007-6022-9148","institution":"Ewha Womans University","correspondingAuthor":true,"prefix":"","firstName":"Seoyoung","middleName":"","lastName":"CHAE","suffix":""},{"id":402963475,"identity":"bc1f08eb-8b69-4a6d-9356-b3c19e2f11de","order_by":1,"name":"Yong-Sang Choi","email":"","orcid":"","institution":"Ewha Womans University","correspondingAuthor":false,"prefix":"","firstName":"Yong-Sang","middleName":"","lastName":"Choi","suffix":""},{"id":402963476,"identity":"d98ee90d-a69d-45e0-af2b-7a6061060c14","order_by":2,"name":"Hwayon Choi","email":"","orcid":"","institution":"Ewha Womans University","correspondingAuthor":false,"prefix":"","firstName":"Hwayon","middleName":"","lastName":"Choi","suffix":""}],"badges":[],"createdAt":"2025-01-02 07:45:51","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5749795/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5749795/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":74352109,"identity":"61bda858-bf42-498e-8137-b2f20bd589f9","added_by":"auto","created_at":"2025-01-21 11:03:50","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":155870,"visible":true,"origin":"","legend":"\u003cp\u003eA flow chart for filtering algorithm.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-5749795/v1/d3f6994d055b2bacbfa3dd56.png"},{"id":74352103,"identity":"350c3346-19a5-46e6-869c-8c41808f61b6","added_by":"auto","created_at":"2025-01-21 11:03:50","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":124170,"visible":true,"origin":"","legend":"\u003cp\u003eAn example of removing solar reflected radiance and extracting only thermal radiance from a scene of the Uljin wildfire on March 5, 2022, based on Choi et al. (2007). (a) original image before calculation, (b) image after calculation. the yellow circle indicates the wildfire location.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-5749795/v1/d1f1bbea399a4971fc22ab1e.png"},{"id":74352994,"identity":"21b6eadd-fc51-426b-a70e-ef280a18600c","added_by":"auto","created_at":"2025-01-21 11:11:50","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":192884,"visible":true,"origin":"","legend":"\u003cp\u003eExample of temperature change patterns over time caused by wildfire occurrence. (a)-(d) are for the first pattern which is a slight increase in temperature over time, (e)-(h) are for the second pattern which is an extreme increase in temperature that then stops.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-5749795/v1/ef6e6abba44d4d4f94b92c4a.png"},{"id":74352107,"identity":"b288f3f4-22d8-45ff-a8d3-47c2556a4741","added_by":"auto","created_at":"2025-01-21 11:03:50","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":138077,"visible":true,"origin":"","legend":"\u003cp\u003eChanges in POD and FAR of GK2A products and filtered GK2A products based on the size of the damaged area.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-5749795/v1/fe0f73fded172d49dd457797.png"},{"id":74353117,"identity":"787b1133-d6cb-44f9-afe0-ee90a33f45c5","added_by":"auto","created_at":"2025-01-21 11:19:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1371684,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5749795/v1/49b7d951-68eb-412a-b510-9e5ba1780905.pdf"}],"financialInterests":"","formattedTitle":"Development of the false alarm filtering method for GEO-KOMPSAT-2A wildfire detection product","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eWildfires cause various types of damage to human life over an extended period, which can be categorized into economic, health, and ecological impacts. Economically, wildfires result in significant losses. Over the past decade in South Korea, 5,668 wildfires have caused financial losses totaling 4.75 trillion KRW and affected approximately 40,037 ha of 6.2% of the nation\u0026rsquo;s total forested area (Korea Forest Service \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). To human health, wildfires can cause both acute and chronic health issues (Sullivan et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Recent findings confirm associations between wildfire smoke exposure and respiratory health outcomes, with the clearest evidence for exacerbations of asthma (Reid and Maestas \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). In particular, there is a significant association between wildfires and the exacerbation of conditions such as asthma, chronic obstructive pulmonary disease, bronchitis, and pneumonia (Cascio \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Wildfires cause irreversible damage to ecosystems. They can degrade ecosystem services such as water supply, nutrient cycling, biodiversity, and carbon storage. Additionally, wildfires affect vegetation structure on varying scales depending on fire conditions, pre- and post-disturbance climate conditions, and ecosystem type. Changes in the structure and composition of vegetation in these habitats can indirectly impact animal populations (Sullivan et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Keith et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Mowat et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Litt and Steidl \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). The ash from burned vegetation and leaf litter reduces the surface area available for evapotranspiration, decreases storage capacity for maintaining and retaining water systems, and removes obstacles that could impede surface water runoff and accelerate soil erosion (Shakesby \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Thus, wildfires are a complex issue that extend beyond mere economic losses, posing serious threats to human health and ecosystems.\u003c/p\u003e \u003cp\u003eSatellite wildfire detection offers significant advantages by enabling real-time detection of wildfires in areas that are difficult to monitor from the ground. Nowadays with advancements in technology and a substantial reduction in satellite launch costs, attempts to utilize satellites for wildfire detection have increased, greatly contributing to wildfire response efforts (Barmpoutis et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Remote sensing via satellites is one of the fastest means of obtaining essential data for research before and after disasters such as wildfires. By providing real-time data, it plays a critical role in disaster analysis and response (Voigt et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). In particular, during wildfire events, satellites enable the rapid collection of data that is difficult to obtain from the ground. This facilitates immediate action by practitioners and responders. Such satellite data go beyond merely detecting disasters; they play a critical role in forming evacuation plans and disaster prevention strategies by providing timely information on weather patterns and wildfire behavior (Joyce et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Wildfire managers can use satellites to predict the scale and spread of wildfires, enabling them to quickly provide critical information, such as evacuation alerts, to residents. Based on this information, more accurate decisions can be made in dynamic situations. In this way, satellites offer valuable, rapid, and efficient information during disaster response. These advantages are particularly beneficial in areas where ground detection is difficult, aiding in the accumulation of wildfire data and facilitating early suppression efforts.\u003c/p\u003e \u003cp\u003eHowever, satellite wildfire detection products have been criticized for not aligning well with actual wildfire data. Past studies have shown that this method tends to exhibit over-detection when compared to actual wildfire data. The false alarms in wildfire products are often caused by non-fire surfaces with high reflectance, such as bare ground, urban areas, water bodies, and clouds (Giglio et al. 2003). Ying et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) compared wildfires in Yunnan Province, southwestern China, between February 2002 and November 2015 with The Moderate Resolution Imaging Spectroradiometer (MODIS) wildfire data. They found that the MODIS wildfire product recorded significantly more wildfires than ground monitoring records and exhibited considerable omission errors. The study highlighted that detection errors might occur due to terrain and climatic variability, emphasizing the importance of ground-based data for validation and accuracy.\u003c/p\u003e \u003cp\u003eWe hypothesized that GEO-KOMPSAT-2A (GK2A) shares similar issues with MODIS, as its current wildfire detection algorithm is based on the MODIS algorithm (National Meteorological Satellite Center \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Among the various issues identified, we analyzed the most important issue of GK2A wildfire products: the high number of false alarms. Therefore, this study aims to develop a filtering method to reduce false alarms effectively.\u003c/p\u003e \u003cp\u003eTo develop a filtering algorithm capable of addressing such false detections, accurate data on the time and location of wildfire occurrences are essential to validate the algorithm South Korea was chosen as the study area due to its comprehensive records of wildfire occurrences, including precise time and location data. Using 19 wildfire cases from 2022, a foundational algorithm was created, which was then compared against 62 actual wildfire cases for evaluation.\u003c/p\u003e"},{"header":"2. GK2A forest fire product analysis","content":"\u003cp\u003eThe GK2A wildfire product, forest fire (FF), defined two thresholds for daytime and nighttime, determined by an 85\u0026deg; solar zenith angle, to reduce solar effect of 3.8 \u0026micro;m channel. Additionally, it checked whether the area was bare ground, urban, or beach to minimize false alarms caused by the effects of solar reflectance during the daytime. Despite these efforts, the false alarm in the GK2A FF product remained significant. The GK2A FF product recorded a total of 869 false alarms in the entire dataset used for the study.\u003c/p\u003e \u003cp\u003eFalse alarms can arise from various factors, but the combination of the 3.8 \u0026micro;m channel and cloud conditions identified as major contributors. The 3.8 \u0026micro;m channel, commonly used for wildfire detection, is unique compared to other IR channels as it is located between the visible and infrared of the spectrum. This positioning makes it be affected to both solar radiance and thermal radiance. As a result, false alarms are more frequent during daytime due to excessive radiative energy from cloud scattering, which the channel interprets as high temperatures. This issue is particularly pronounced near cloud edges, where irregular scattering can accelerate the problem, leading to a higher likelihood of false alarms (Emde et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In practice, an analysis of false alarm cases showed that they predominantly occurred during the daytime and in proximity to cloud edges (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eTiming and presence of surrounding clouds of false alarms in GK2A wildfire detection products.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal false alarms\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNear cloud\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon near cloud\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e869\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e765\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e104\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDaytime\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNighttime\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e598\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e271\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo address these false alarms, this study applied two primary filtering methods. The first method involved separating solar radiance and thermal radiance during the calculation part, utilizing only thermal radiance. This is suggested by Choi et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2007\u003c/span\u003e), using formula to calculate the cloud-reflected radiance in the 3.8 \u0026micro;m channel. The second method introduced an additional process to verify pixels detected as wildfires by the GK2A whether actually fires or not. This verification process utilized the physical differences between fire and cloud characteristics and comprised three different tests. These approaches aimed to correct false alarms caused by clouds and the 3.8 \u0026micro;m channel and filter out false alarms in wildfire detection outputs from the GK2A satellite.\u003c/p\u003e"},{"header":"3. Geostationary satellite and validation data","content":"\u003cp\u003eGK2A is a geostationary satellite launched by the Republic of Korea and its data is provided by the National Meteorological Satellite Center (NMSC). We used two Level 1B (L1B) data the 3.8 \u0026micro;m channel and 10.5 \u0026micro;m channel and four Level 2 (L2) data the Cloud detection (CLD), Forest fire detection (FF), Land surface emissivity (LSE), Normalized vegetation index (NDVI). GK2A offers every 2-minutes for the Korea region so all data have 2 minutes for temporal resolution except for LSE and NDVI, which are calculated once per day (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eList of GK2A data and their characteristics used in the study.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProduct name\u003c/p\u003e \u003cp\u003e(preprocessing level)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCentral wavelength\u003c/p\u003e \u003cp\u003e(\u0026micro;m)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSpectral band and range\u003c/p\u003e \u003cp\u003e(\u0026micro;m)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSpatial resolution (km)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTemporal\u003c/p\u003e \u003cp\u003eresolution (min)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSW038 (1B)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eB7: 3.74\u0026ndash;3.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIR105 (1B)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eB13: 10.25\u0026ndash;10.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCLD (2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFF (2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLSR (2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOnce a day\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNDVI (2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOnce a day\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe two L1B datasets we used are 3.8 \u0026micro;m and 10.5 \u0026micro;m. The hot spot finding method using brightness temperature differences between these two channels was first proposed by Dozier et al. (1981). After this, research on the Brightness temperature difference hot spot detection method was continued and adopted operationally. When two surfaces with different temperatures exist within the same pixel, the resulting radiation fields will vary, causing the two channels to respond differently to thermal emissions. The thermal infrared channels, such as the 3.8 \u0026micro;m channel in GK2A, are more sensitive to thermal radiation than the 10.5 \u0026micro;m channel. This difference enables the detection of thermal sources in GK2A images using these two channels.\u003c/p\u003e \u003cp\u003eThe four L2 datasets we used are Cloud detection (CLD), Forest fire detection (FF), Land surface emissivity (LSE), Normalized vegetation index (NDVI). Cloud detection is used for cloud masking, which are the most significant contamination factor in satellite images. Clouds disturb satellite observations at high elevations by decreasing brightness temperatures. CLD is always used in land observation algorithms for masking data due to this characteristic. GK2A CLD data consist of the following categories: 0: Cloud (High Confidence), 1: Cloud (Low Confidence) and 2: Clear (High Confidence). In this study, we masked all areas labeled as 0 and 1 and utilized only the clear regions. The GK2A CLD algorithm was designed to minimize undetected clouds. However, the accuracy of the product tends to decrease during winter due to cooled land surfaces according to GK2A CLD Algorithm Theoretical Basis Document (ATBD). FF is used as part of the algorithm and as a comparison product. The GK2A FF algorithm detects fires by using the median field of the temperature base map and comparing the difference between the base map temperature and the raw pixel temperature against thresholds. This detection method is based on the MODIS active fire detection algorithm, with modifications and adjustments made for operational use on GK2A (National Meteorological Satellite Center \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). FF data consist of 0: Non-fire and 1: Fire. The filtering algorithm was applied to GK2A fire pixels with a value of 1 in the FF data, and POD and False Alarm Rate (FAR) were compared before and after filtering. LSE is used to calculate ground albedo to separate ground-reflected radiance from the 3.8 \u0026micro;m channel. LSE serves the channels. Since LSE is provided differently for each channel, we used only the data corresponding to the 3.8 \u0026micro;m channel. NDVI is used to identify land and water. Pixels with an NDVI value of 0 were masked as water.\u003c/p\u003e \u003cp\u003eThe wildfire cases used for validation are provided by 2022 annual wildfire cases list published by Korea forest service. These data include year, month, day, and time of wildfire occurrence and extinguishment, town-level location, and damaged area. We chose 19 fires which are 1) over 10 ha damaged, 2) no clouds at that time for the frame of algorithm. The data provided refers to 3 hours, including 1 hour before and after the wildfire occurrence time, and additional cases were used as validation data if the occurrence and extinguishment times were within the 3-hour range of the available data for one case. As a result, a total of 62 wildfire cases were collected, with affected areas ranging from 0.01 ha to 16,301.98 ha. (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eNumber of wildfire cases and distribution of damaged areas used for validation\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal fire cases\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFire cases over 1 ha\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFire cases over 10 ha\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"4. Filtering algorithm","content":"\u003cp\u003eThe algorithm is composed of six components: input data, calculation, basic test, absolute fire test, time correlation test, and pop-up fire test (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). To solve the two problems discussed in Section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, solar radiance in 3.8 \u0026micro;m channel and cloud edges, it is essential to focus on the calculation part and the three test parts located on the far-right side of the flowchart. This section will first provide a detailed explanation of the two parts, followed by descriptions of the remaining parts and then the steps required for the localization of the algorithm.\u003c/p\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Calculation for reducing solar radiance from observations\u003c/h2\u003e \u003cp\u003eIn the calculation part, to separate the solar radiance from observations, the method proposed by Choi et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) was applied during the data processing stage.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{th}={L}_{obs}-{L}_{3.8}^{sr}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e( 1 )\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e3.8 \u0026micro;m channel thermal radiance \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{th}\\)\u003c/span\u003e\u003c/span\u003e is calculated to minus from observed 3.8 \u0026micro;m channel radiance \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{obs}\\)\u003c/span\u003e\u003c/span\u003e to ground-reflected radiance \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{Sr}\\)\u003c/span\u003e\u003c/span\u003e. In the Choi et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2007\u003c/span\u003e), ground albedo \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{A}_{g}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{sr}\\)\u003c/span\u003e\u003c/span\u003e when \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{A}_{g}=1\\)\u003c/span\u003e\u003c/span\u003e are used for finding \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{Sr}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabb\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{sr}={A}_{g}{L}_{3.8}^{sr}{(A}_{g}=1)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e(\u003c/b\u003e 2 )\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{A}_{g}\\)\u003c/span\u003e \u003c/span\u003e is calculated using GK2A LSE product with the formula below.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabc\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{A}_{g}=1-land\\:surface\\:emissivity\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e( 3 )\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{sr}{(A}_{g}=1)\\)\u003c/span\u003e \u003c/span\u003e as created as a lookup table, calculated using Streamer, a type of radiative transfer model, for solar zenith angles ranging from 0\u0026deg; to 90\u0026deg; under the condition \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{A}_{g}=1\\)\u003c/span\u003e\u003c/span\u003e. After using these methods to 3.8 \u0026micro;m channel, brightness temperature of the non-fire surface is decreasing compared to wildfires (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eCalculated 3.8 \u0026micro;m channel solar-free radiance and observed 10.5 \u0026micro;m channel radiance are changed to brightness temperature through calibration table of GK2A.\u003c/p\u003e \u003cp\u003eFor the basic test, calculate the locally averaged BT value of the 3.8 \u0026micro;m channel\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left[{BT}_{3.8,t}\\right]\\)\u003c/span\u003e\u003c/span\u003e, and the signal \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}\\)\u003c/span\u003e\u003c/span\u003e which is temperature difference ratio of 3.8 \u0026micro;m and 10.5 \u0026micro;m channels BT.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabd\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}=\\left(\\frac{{BT}_{3.8,t}}{{BT}_{10.5,t}}\\right)-\\left(\\frac{{BT}_{3.8,t-1}}{{BT}_{10.5,t-1}}\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e( 4 )\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Classification of wildfires and clouds\u003c/h2\u003e \u003cp\u003eThe enter conditions for the tests are set as multiples of the channel noise \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Cn\\)\u003c/span\u003e\u003c/span\u003e, which was calculated by accumulating \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}\\)\u003c/span\u003e\u003c/span\u003e over one day and determining its standard deviation. When \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Cn\\)\u003c/span\u003e\u003c/span\u003e was obtained separately by season and time, there was no change in the value. This suggests that the same value can be used regardless of the season, so it was fixed as 0.0011.\u003c/p\u003e \u003cp\u003eThe absolute fire test is designed to confirm pixels as wildfires that exhibit abnormally large signals. The enter condition is \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}\u0026gt;25Cn\\)\u003c/span\u003e\u003c/span\u003e, which is set to identify wildfire pixels with significantly stronger signals compared to other tests. In this test, if no clouds are present in the 8 surrounding pixels of the test pixel, the test pixel is immediately confirmed as a wildfire.\u003c/p\u003e \u003cp\u003eThe time correlation test identifies a pixel as a wildfire if the temperature at a fixed location increases consistently over time. Unlike wildfires, clouds move over time, so we use this feature to filter out pixels that move over time as cloud while fixed pixels are identified as wildfires. The starting conditions of this test are \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}\u0026gt;2Cn\\:and\\:\\:{S}_{t-1}\u0026gt;Cn\\)\u003c/span\u003e\u003c/span\u003e. Pixels meeting these conditions are designated as a test field, which includes the pixel itself and its 8 surrounding pixels (a total of 9 pixels). Additionally, 25 fields which have 3\u0026times;3 pixels around the test field from \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t-1}\\)\u003c/span\u003e\u003c/span\u003e are designated as compare fields. To detect movement, the correlation between the test field and the comparison field is calculated to determine whether the test field is a cloud or a forest fire. Among the calculated correlation values, if the compare field at the same position as the test field meets the following two conditions: 1) it has the highest value, and 2) the value is greater than 0.5, the pixel is determined to have not moved and is confirmed as a wildfire.\u003c/p\u003e \u003cp\u003eThe pop-up fire test is designed to identify wildfire signals that appear abruptly with high temperatures and then disappear without prior warning. The enter condition is \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}\u0026gt;2Cn\\)\u003c/span\u003e\u003c/span\u003e. For pixels meeting the enter condition, it is first checked whether the pixel was not a cloud on \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t-1}\\)\u003c/span\u003e\u003c/span\u003e. Then, on \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}\\)\u003c/span\u003e\u003c/span\u003e, the surrounding 24 pixels are checked to ensure there are no clouds present. This test adopts a method similar to the absolute fire test but is tailored to detect wildfires with slightly less temperature increase. The spatial resolution of GK2A is 2 km \u0026times; 2 km which is significantly larger than the size of most wildfires. As a result, the signal observed from a single pixel may appear smaller than the rapid temperature change caused by the wildfire. Therefore, this test was added as a separate method from the absolute fire test (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFinally, any pixel that passed at least one of the absolute fire test, time correlation test, or pop-up fire test was confirmed as a wildfire.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e4.3. Minor parts of the filtering algorithm\u003c/h2\u003e \u003cp\u003eIn the input data part, check all datasets exist on the correct directories. In the basic test part, checking test pixels are satisfied to enter the time correlation test, the pop-up fire test and the absolute fire test. If the test pixel is a clear sky, detected as fire in the GK2A product, located on land and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{BT}_{3.8,t}\\)\u003c/span\u003e\u003c/span\u003e is bigger than \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left[{BT}_{3.8,t}\\right]\\)\u003c/span\u003e\u003c/span\u003e, check \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{t}\\)\u003c/span\u003e\u003c/span\u003e to enter each test or not. The calculated \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Cn\\)\u003c/span\u003e\u003c/span\u003e was applied with different multipliers depending on the entry requirements for each test. The tests were conducted in three stages, in descending order of entry requirement magnitude.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e4.4. Methods for localization values\u003c/h2\u003e \u003cp\u003eThis filtering algorithm can be localized for specific regions, as each region has unique climatic characteristics and different latitudes and longitudes, making localization necessary.\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{L}_{3.8}^{sr}{(A}_{g}=1)\\)\u003c/span\u003e \u003c/span\u003e was calculated using a radiative transfer model to create a lookup table. This table was generated based on the latitude and longitude of Seoul (37.5\u0026deg; N, 127\u0026deg; E). Therefore, the model needs to be adjusted and recalculated to match the desired region specified by the user.\u003c/p\u003e \u003cp\u003eThe \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Cn\\:\\)\u003c/span\u003e\u003c/span\u003eused as a threshold in the algorithm was calculated using the standard deviation of the time difference temperature over one day. Although this value is fixed and used in the algorithm, it was derived based on data from South Korea. Therefore, it reflects regional characteristics and may require adjustment for other locations.\u003c/p\u003e \u003cp\u003eAll these adjustments were calibrated based on the Korean Peninsula, specifically South Korea. With a total area of 100,410 ㎢, all experiments were conducted within this geographical scope. Therefore, users should keep this in mind and adjust the thresholds accordingly when applying the algorithm to other regions.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Fire detection results and validation","content":"\u003cp\u003eThe filtered GK2A FF product successfully detected 28 out of 62 wildfire cases. For wildfires with an affected area of at least 1 ha, it detected 26 out of 38 cases, and for wildfires with an affected area of at least 10 ha, it detected 18 out of 19 cases. During the entire observation period, there were 25 false alarms despite no wildfires occurring. For all wildfires, the POD was 45.16%, and the FAR was 47.17%. For wildfires with an affected area of at least 1 ha, the POD was 71.05%, and the FAR was 49.02%. For wildfires with an affected area of at least 10 ha, the POD was 94.74%, and the FAR was 58.14%.\u003c/p\u003e \u003cp\u003eThe outputs used in the analysis are of two types: 1) the GK2A wildfire detection product, and 2) the filtered output, which applies the research algorithm to the GK2A results. Each output was evaluated based on the wildfire and fire incident data provided by the Korea Forest Service. The filter uses empirical thresholds and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Cn\\)\u003c/span\u003e\u003c/span\u003e. POD and FAR are compared to show the performance. as multiples of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Cn.\\)\u003c/span\u003e\u003c/span\u003e Additionally, this study examines how detection performance varies based on the area of the wildfire, as the wildfire cases used for validation differ in size. A total of 62 wildfire cases were used for validation, of which 38 had an affected area of at least 1 ha, and 19 had an affected area of at least 10 ha. And then present the performance of each test.\u003c/p\u003e \u003cp\u003eThe GK2A FF product successfully detected 33 out of 62 wildfire cases. For wildfires with an affected area of at least 1 ha, it detected 29 out of 38 cases, and for wildfires with an affected area of at least 10 ha, it detected all 19 cases. During the entire observation period, there were 869 false alarms despite no wildfires occurring (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). For all wildfires, the POD was 53.23%, and the FAR was 96.34%. For wildfires with an affected area of at least 1 ha, the POD was 76.32%, and the FAR was 96.77%. For wildfires with an affected area of at least 10 ha, the POD was 100%, and the FAR was 97.86%.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eContingency table of before and after filtering products.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabe\" border=\"1\"\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" morerows=\"1\" nameend=\"c2\" namest=\"c1\" rowspan=\"2\"\u003e \u003cp\u003eGK2A\u003c/p\u003e \u003cp\u003e(over 10 ha)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eReported\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFire\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-fire\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDetected\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFire\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e33 (19)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e869 (869)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNon-fire\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29 (0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabf\" border=\"1\"\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" morerows=\"1\" nameend=\"c2\" namest=\"c1\" rowspan=\"2\"\u003e \u003cp\u003eGK2A\u0026thinsp;+\u0026thinsp;the filter\u003c/p\u003e \u003cp\u003e(over 10 ha)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eReported\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFire\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-fire\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDetected\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFire\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28 (18)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25 (25)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNon-fire\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e34 (1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e844 (844)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows that the POD does not significantly decrease from the total cases to wildfires over 10 ha, even after applying the filter. However, the FAR decreases substantially compared to the slight drop in POD. The apparent increase in FAR for over 10 ha in the FAR graph occurs because the number of detected cases, which serves as the denominator, decreases significantly when the range shifts from over 1 ha to over 10 ha (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). The filtered wildfire detection product reduced 844 false alarms compared to the GK2A wildfire detection product (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). This corresponds to a reduction of approximately 97% of the total false alarms through filtering.\u003c/p\u003e \u003cp\u003eIn some instances, the alarm encompassed multiple tests on both sides. Consequently, the maximum number of true alarms for any individual test did not exceed 28. However, when the true alarms from all three tests were combined, the total exceeded 28. This was also observed for false alarms. The absolute fire test recorded 13 out of 28 true alarms and 3 out of 25 false alarms. The time correlation fire test reported 20 out of 28 true alarms and 8 out of 25 false alarms. Lastly, the pop-up fire test documented 21 out of 28 true alarms and 17 out of 25 false alarms. From a numerical perspective, the pop-up fire test appears to contribute significantly to false alarms, potentially rendering it seemingly less useful. However, the situation changes when examining the number of wildfires each test uniquely detected. The absolute fire test did not detect any wildfires independently, while the time correlation fire test detected 5, and the pop-up fire test independently detected 8. These results suggest that utilizing each test independently aligns with the goal of the algorithm in detecting wildfires as a critical disaster (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe detection results of each test on the filtering algorithm.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eAbsolute fire test\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eTime correlation test\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003ePop-up fire test\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrue alarm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFalse alarm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTrue alarm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFalse alarm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTrue alarm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eFalse alarm\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDetection result\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSolo detected result\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"6. Conclusion and Discussion","content":"\u003cp\u003eIn this study, the performance of the GK2A wildfire detection algorithm was improved by applying a filtering technique with three tests to solve the false alarms. After the filtering, the GK2A wildfire detection product had a POD of 45.16% and an FAR of 47.17% for the 62 wildfires, reducing the total false alarms to 25 \u0026mdash;the original GK2A wildfire detection product showed a POD of 53.23% and an FAR of 96.34% for a total of 62 wildfires, with 869 false alarms recorded.\u003c/p\u003e \u003cp\u003eGeostationary satellite-based wildfire detection still faces hardware limitations. Currently, the spatial resolution of GK2A is relatively large compared to the typical size of wildfires, making it less effective for detection. Most wildfires in South Korea affect areas smaller than 10 ha, yet the system requires wildfires larger than 10 ha to demonstrate sufficient detection capabilities. To address this issue, integrating low Earth orbit (LEO) satellites or linking detection methods with ground-based cameras may be a viable solution. Utilizing such satellites in combination with geostationary systems could enhance wildfire detection capabilities.\u003c/p\u003e \u003cp\u003eAdditionally, smoke is an important signal for wildfire detection; however, it has not been incorporated into this study. Traditionally, heat and smoke are the primary indicators used for wildfire detection, but smoke tends to lower channel temperatures, potentially delaying detection when using satellite data. Additionally, smoke from large wildfires is often misclassified as clouds, leading to the masking of wildfire-affected areas. Since smoke and wildfires are inherently linked, future research will consider this factor and incorporate it into the analysis.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2018R1A6A1A08025520).\u0026nbsp;This work was supported by the Specialized University Program for Confluence Analysis of Weather and Climate Data of the Korea Meteorological Institute (KMI), funded by the Korean government (KMA).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBarmpoutis P, Papaioannou P, Dimitropoulos K, Grammalidis N (2020) A Review on Early Forest Fire Detection Systems using Optical Remote Sensing. Sensors 20(22):6442\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCascio WE (2018) Wildland Fire Smoke Hum Health Sci Total Environ 624:586\u0026ndash;595\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi Y, Ho C, Ahn M, Kim Y (2007) An Exploratory Study of Cloud Remote Sensing Capabilities of the Communication, Ocean and Meteorological Satellite (COMS) Imagery. Int. J Remote Sens 28(21):4715\u0026ndash;4732\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDozier J (1981) A Method for Satellite Identification of Surface Temperature Fields of Subpixel Resolution. Remote Sens Environ 11:221\u0026ndash;229\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEmde C, Yu H, Kylling A, Van Roozendael M, Stebel K, Veihelmann B, Mayer B (2021) Impact of 3D Cloud Structures on the Atmospheric Trace Gas Products from UV-VIS Sounders\u0026ndash;Part I: Synthetic Dataset for Validation of Trace Gas Retrieval Algorithms. Atmospheric Measurement Techniques Discussions, 2021, 1\u0026ndash;28\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGiglio L, Csiszar I, Justice CO (2006) Global Distribution and Seasonality of Active Fires as Observed with the Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) Sensors. J Geophys Research: Biogeosciences, 111(G2).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoyce KE, Wright KC, Samsonov SV, Ambrosia VG (2009) Remote Sens Disaster Manage Cycle Adv Geoscience Remote Sens 48(7):317\u0026ndash;346\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeith H, Mackey BG, Lindenmayer DB (2009) Re-Evaluation of Forest Biomass Carbon Stocks and Lessons from the World's most Carbon-Dense Forests. Proceedings of the National Academy of Sciences, 106(28), 11635\u0026ndash;11640\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKorea Forest Service (2022) 2022 Annual Report on Wildfire Statistics. Publishing Korea Forest Service. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.index.go.kr/unity/potal/main/EachDtlPageDetail.do?idx_cd=1309\u003c/span\u003e\u003cspan address=\"https://www.index.go.kr/unity/potal/main/EachDtlPageDetail.do?idx_cd=1309\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, Accessed Nov. 18, 2024\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKorea Forest Service (2023) 2023 Annual Report on Wildfire Statistics. Publishing Korea Forest Service. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.index.go.kr/unity/potal/main/EachDtlPageDetail.do?idx_cd=1309\u003c/span\u003e\u003cspan address=\"https://www.index.go.kr/unity/potal/main/EachDtlPageDetail.do?idx_cd=1309\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, Accessed Nov. 18, 2024\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLitt AR, Steidl RJ (2011) Interactive Effects of Fire and Nonnative Plants on Small Mammals in Grasslands: Effets Interactifs Du Feu Et Des Plantes Non Indig\u0026egrave;nes Sur Les Petits Mammif\u0026egrave;res. Dans Les Prairies Wildl Monogr 176(1):1\u0026ndash;31\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMowat EJ, Webb JK, Crowther MS (2015) Fire-mediated Niche‐separation between Two Sympatric Small Mammal Species. Austral Ecol 40(1):50\u0026ndash;59\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Meteorological Satellite Center (2019) NMSC: GK2A AMI Algorithm Theoretical Basis Document Cloud Mask. Publishing NMSC. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://nmsc.kma.go.kr/homepage/html/base/cmm/selectPage.do?page=static.edu.atbdGk2a\u003c/span\u003e\u003cspan address=\"http://nmsc.kma.go.kr/homepage/html/base/cmm/selectPage.do?page=static.edu.atbdGk2a\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 25 No-vember 2024\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Meteorological Satellite Center (2019) NMSC: GK2A AMI Algorithm Theoretical Basis Document Forst Fire. Publishing NMSC. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://nmsc.kma.go.kr/homepage/html/base/cmm/selectPage.do?page=static.edu.atbdGk2a\u003c/span\u003e\u003cspan address=\"http://nmsc.kma.go.kr/homepage/html/base/cmm/selectPage.do?page=static.edu.atbdGk2a\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 25 No-vember 2024\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReid CE, Maestas MM (2019) Wildfire Smoke Exposure Under Climate Change: Impact on Respiratory Health of Affected Communities. Curr. Opin Pulm Med 25(2):179\u0026ndash;187\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShakesby RA (2011) Post-Wildfire Soil Erosion in the Mediterranean: Review and Future Research Directions. Earth-Sci Rev 105(3\u0026ndash;4):71\u0026ndash;100\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSullivan A, Baker E, Kurvits T, Popescu A, Paulson AK, Christianson C, Tulloch A, Bilbao A, Mathison B, Robinson C (2022) C. Spreading Like Wildfire: The Rising Threat of Extraordinary Landscape Fires.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVoigt S, Giulio-Tonolo F, Lyons J, Kučera J, Jones B, Schneiderhan T, Platzeck G, Kaku K, Hazarika MK, Czaran L (2016) Global Trends. Satellite-Based Emerg Mapp Sci 353(6296):247\u0026ndash;252\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYing L, Shen Z, Yang M, Piao S (2019) Wildfire Detection Probability of MODIS Fire Products Under the Constraint of Environmental Factors: A Study Based on Confirmed Ground Wildfire Records. Remote Sens 11(24):3031\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"natural-hazards","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nhaz","sideBox":"Learn more about [Natural Hazards](https://www.springer.com/journal/11069)","snPcode":"11069","submissionUrl":"https://submission.nature.com/new-submission/11069/3","title":"Natural Hazards","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Wildfire, Geostationary Satellite, False alarm, Filter","lastPublishedDoi":"10.21203/rs.3.rs-5749795/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5749795/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWildfires have caused significant damage to the economy, human health, and ecosystems over extended periods. Satellites offer advantages as wildfire detection tools because they provide near real-time, continuous observations in areas that are difficult to manage directly by humans. But in previous studies, satellite active wildfire detection products were prone to generating false alarms in areas with high reflectance, such as bare soils, urban regions, water bodies, and clouds. This issue arises because infrared channels were used for wildfire detection, but their observations include both solar and thermal radiances. In this study, we apply a calculation, based on a previous study, to wildfire detection that reduces solar reflectance radiance and propose a filtering method to eliminate false alarms occurring in cloud edge areas. South Korea, where detailed annual wildfire information is available, was selected as the study area to develop a region-specific false alarm filtering algorithm for wildfire detection product. The filtering algorithm was compared to the GEO-KOMPSAT-2A operational product using data from 62 wildfire cases that occurred in 2022. As a result, the number of false alarms was significantly reduced from 896 to 25, the False alarm ratio (FAR) decrease from 96.34\u0026ndash;47.17% and the Probability of Detection (POD) showed a slight decrease from 53.23\u0026ndash;45.16%. The results of this study are expected to contribute to more efficient disaster and risk management by enabling more accurate wildfire detection.\u003c/p\u003e","manuscriptTitle":"Development of the false alarm filtering method for GEO-KOMPSAT-2A wildfire detection product","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-01-21 11:03:45","doi":"10.21203/rs.3.rs-5749795/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2025-01-16T11:03:34+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-01-04T02:30:03+00:00","index":"","fulltext":""},{"type":"submitted","content":"Natural Hazards","date":"2025-01-02T02:45:38+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"natural-hazards","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nhaz","sideBox":"Learn more about [Natural Hazards](https://www.springer.com/journal/11069)","snPcode":"11069","submissionUrl":"https://submission.nature.com/new-submission/11069/3","title":"Natural Hazards","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"14ee9ee7-a319-42a2-a3b0-4ee4d65719ef","owner":[],"postedDate":"January 21st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-01-21T11:03:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-01-21 11:03:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5749795","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5749795","identity":"rs-5749795","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}