{"paper_id":"0776294c-c0f1-4de6-9080-3cf080080e1d","body_text":"Use of Fire Radiative Power as an Indicator for Forest Fire Prediction: A Study of Machine Learning Architectures | 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 Article Use of Fire Radiative Power as an Indicator for Forest Fire Prediction: A Study of Machine Learning Architectures Maria Santos This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8081211/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This project addresses contemporary climate, sociological, and environmental challenges by exploring the use of machine learning models to analyze fire radiative power (FRP) as an indicator for forest fire prediction. Climate change has intensified extreme weather events and biodiversity loss, demanding more accurate predictive tools. Machine learning enables the extraction of hidden patterns from large-scale environmental data, supporting more reliable forecasting and mitigation strategies. Using satellite data provided by the Brazilian National Institute for Space Research (INPE), this study evaluates fire occurrences across critical biomes, including the Amazon, Cerrado, and Pantanal. The dataset includes geospatial coordinates, time of detection, and complementary meteorological variables. By identifying patterns in wildfire behavior, this work aims to support the development of more effective prevention policies and environmental risk management approaches. Despite time constraints and the large volume of available data, the project highlights the potential of machine learning architectures to enhance climate resilience and guide data-driven decision-making in Brazil’s environmental context. Earth and environmental sciences/Climate sciences/Climate change/Climate-change mitigation Earth and environmental sciences/Climate sciences/Climate change/Projection and prediction Introduction Climate change represents one of the major global challenges of the 21st century, affecting not only the environment but also social, economic, and political dynamics (IPCC, 2021). Rising global temperatures, increasing extreme weather events, and biodiversity loss highlight the urgency of developing mitigation and adaptation strategies. In this context, machine learning models have emerged as promising tools capable of identifying complex patterns in large environmental datasets, enabling more accurate forecasting and supporting decision-making processes (Goodfellow, Bengio & Courville, 2016). This project focuses on analyzing Brazil’s position within this scenario, considering its vulnerability to wildfires and other climate-related events. The dataset used in this study is part of the “Queimadas” project developed by the Brazilian National Institute for Space Research (INPE), which monitors fire outbreaks in real time using satellite data. The platform provides valuable information including geolocation, intensity, and frequency of fire events, particularly in critical biomes such as the Amazon, Cerrado, and Pantanal. Using these data, this project aims to develop predictive models capable of identifying spatiotemporal patterns related to wildfire occurrence, contributing to more efficient mitigation strategies. Although the dataset is extensive, the limited execution time requires prioritization and careful selection of the most relevant variables. Methodology An exploratory data analysis was conducted to examine data distribution, identify patterns, and detect inconsistencies. Records from August to November 2024 were selected, corresponding to the period with historically higher wildfire activity. The DataHora column was converted to a datetime format, from which day, month, hour, and weekday features were extracted for a more granular temporal viewpoint. Invalid values flagged by INPE (e.g., −999 in FRP and DaysWithoutRain) were removed to ensure data integrity. Spatial visualizations were generated using Plotly, allowing the identification of municipalities with higher fire incidence, especially within the Amazon biome. Due to the large size of the dataset (approximately five million records), a stratified sampling approach was applied by geographic region to maintain representativeness while ensuring computational feasibility. This preserved the major spatial patterns observed in the complete dataset. Missing values in the Fire Radiative Power (FRP) time series were handled using temporal interpolation (interpolate(method='time')), followed by backfilling to correct initial gaps. Outliers in FRP and precipitation were mitigated by clipping values above the 99th percentile, reducing the influence of extreme events. A logarithmic transformation (np.log1p()) was applied to reduce distribution skewness and enhance model stability in regression tasks. Categorical variables such as biome and region were encoded using One-Hot Encoding with drop_first=True to prevent multicollinearity. Latitude and longitude coordinates were transformed into three-dimensional Cartesian space to avoid angular discontinuities and provide smoother spatial inputs for machine-learning algorithms. After preprocessing, multiple machine-learning architectures were implemented and evaluated, including Prophet for time-series forecasting, a Multi-Layer Perceptron (MLP) for nonlinear regression, XGBoost Regressor for gradient-boosted tree modeling, and TabNet for deep tabular feature representation. These models were compared to assess predictive performance and suitability for wildfire prediction in Brazilian biomes. Results The present study aimed to predict Fire Radiative Power (FRP), a metric directly associated with wildfire intensity, using a robust set of geospatial, meteorological, and temporal variables from the INPE database. The selection of FRP instead of traditional fire-risk indices reflects its ability to quantify the actual severity of fire events, as previously demonstrated by Wooster et al. (2003) and currently adopted by agencies such as NASA. Exploratory data analysis revealed that the most affected biomes are the Amazon and Cerrado, with particular relevance to the state of Pará. Within this region, the municipalities of São Félix do Xingu, Novo Progresso, and Altamira exhibit the highest concentrations of intense wildfire activity. These findings are consistent with results from MapBiomas (2023) and PRODES/INPE surveys, which indicate that Pará accounted for approximately 30% of the burned area in the Amazon between 2020 and 2022. These municipalities share climatic characteristics that favor ignition and propagation: São Félix do Xingu: Equatorial and humid, with an annual average of 1,423 mm of rainfall, high humidity, and mean temperatures of 26.7 °C. Novo Progresso: Hot and humid climate, with an average precipitation of 2,200 mm and a pronounced dry season between August and December. Altamira: Tropical climate with short dry periods that coincide with peaks in wildfire activity. In addition to environmental conditions, factors such as agricultural expansion, cattle ranching, and deforestation pressure contribute to an increased likelihood of high-intensity fire events. Accurate prediction of FRP in these areas can support specialists in devising targeted mitigation strategies and prioritizing surveillance efforts. Among the evaluated machine-learning architectures, the XGBoost Regressor combined with Grid Search optimization demonstrated superior performance in capturing nonlinear and complex patterns across the dataset. Its ability to handle heterogeneous features, irregular distributions, and interactions between climatic and spatial variables contributed to a higher predictive accuracy compared to the other tested models. This suggests its suitability for operational deployment in wildfire early-warning systems. Conclusion This study investigated the predictive modeling of Fire Radiative Power (FRP) using geospatial, meteorological, and temporal variables provided by INPE. By prioritizing FRP over traditional fire risk indices, the analysis incorporated an indicator directly associated with fire intensity and potential environmental impact, as previously supported in the literature. The exploratory results revealed that the Amazon and Cerrado biomes exhibit the highest concentration of high-intensity fire events, with particular incidence in the state of Pará. Municipalities such as São Félix do Xingu, Novo Progresso, and Altamira demonstrated recurrent hotspots, driven by favorable climatic conditions and anthropogenic pressures, including deforestation and the expansion of agricultural frontiers. Multiple machine learning architectures were evaluated, including Prophet, Multi-Layer Perceptron (MLP), TabNet, and XGBoost Regressor. Among these, the XGBoost Regressor optimized through Grid Search achieved superior performance, effectively capturing nonlinear interactions and complex variable dependencies. These results highlight the potential of ensemble-based approaches for environmental prediction tasks, particularly when dealing with heterogeneous and highly dynamic datasets. The predictive capability demonstrated by the proposed modeling approach may support environmental monitoring systems, enabling earlier identification of regions with elevated fire intensity. Such insights can contribute to mitigation planning, resource allocation, and the development of regional fire prevention strategies. Future research should consider incorporating satellite-derived vegetation indices, long-term climate trends, and explainable AI techniques to enhance model interpretability and operational applicability. Expanding the temporal resolution and integrating real-time data streams may further improve forecast reliability and support large-scale environmental policy decisions. References ARJONA-MÉNDEZ, G. et al. (2019). Wildfire prediction based on machine learning: Case study Mexico . arXiv preprint arXiv:1908.07442. Retrieved June 6, 2025, from https://arxiv.org/pdf/1908.07442 WORLD BANK. (2023). Climate and Development Report for Brazil 2023 . United Nations Brazil. Retrieved October 7, 2024, from https://brasil.un.org CLIMATE-DATA. (2025). Climate: Altamira . Retrieved April 20, 2025, from https://pt.climate-data.org/america-do-sul/brasil/para/altamira-43955/ CLIMATE TRACE. (2024). Climate Trace Data . Retrieved August 24, 2024, from https://climatetrace.org/data ECO DEBATE. (2024). Regional impacts of climate change . Retrieved October 7, 2024, from https://ecodebate.com.br FAYYAD, U., PIATETSKY-SHAPIRO, G., & SMYTH, P. (1996). From data mining to knowledge discovery in databases. AI Magazine , 17(3), 37–54. INPE – National Institute for Space Research. (2024). Burned Area Database – TerraBrasilis . São José dos Campos, Brazil. Retrieved May 30, 2025, from https://terrabrasilis.dpi.inpe.br/queimadas/bdqueimadas/ INPE – National Institute for Space Research. (2025). Frequently Asked Questions – Fire Monitoring Program . Retrieved June 6, 2025, from https://terrabrasilis.dpi.inpe.br/queimadas/portal/faq/index.html INPE – National Institute for Space Research. (n.d.). Fire risk: concise explanation . Retrieved June 6, 2025, from https://dataservercoids.inpe.br/queimadas/queimadas/PublicacoesImpacto/documentos/RiscoFogo_Sucinto.pdf Intergovernmental Panel on Climate Change (IPCC). (2021). Climate change 2021: The physical science basis . Retrieved May 23, 2025, from https://www.ipcc.ch/report/ar6/wg1/ MAPBIOMAS. (2025). Project MapBiomas – Collection 9 of the Annual Series of Land Cover and Use Maps of Brazil . Retrieved May 23, 2025, from https://brasil.mapbiomas.org/ NASA. (2024). Global Climate Change: Vital Signs of the Planet . Retrieved August 20, 2024, from https://climate.nasa.gov/ NOVO PROGRESSO CITY HALL. (2025). About the municipality . Retrieved April 20, 2025, from https://camaranovoprogresso.pa.gov.br/o-municipio/ WORLD METEOROLOGICAL ORGANIZATION. (2023). Report reveals that Brazil faced 12 extreme climate events in 2023 . ONU News. Retrieved October 7, 2024, from https://news.un.org PARÁ STATE GOVERNMENT. (2025). São Félix do Xingu: Municipal socioeconomic profile . Belém: Foundation for Support of Studies and Research of the Amazon (FAPESPA). Retrieved April 20, 2025, from https://www.fapespa.pa.gov.br/wp-content/uploads/2025/02/Sao-Felix-do-Xingu.pdf SANTOS, M. H. S. (2025). Using fire radiative power as an indicator for the prediction of forest fires: A study of machine learning architectures . Retrieved May 30, 2025, from https://github.com/mariahelenass/Wildfire-Brazil SILVA, I. N., SPATTI, D. H., & FLAUZINO, R. A. (2016). Artificial neural networks for engineering and applied sciences: Theoretical foundations and practical aspects (2nd ed., pp. 46–47). São Paulo: Artliber. WORLD METEOROLOGICAL ORGANIZATION (WMO). (2024). State of the Global Climate . Retrieved August 25, 2024, from https://public.wmo.int/en/resources/library/state-of-global-climate Additional Declarations There is NO Competing Interest. Cite Share Download PDF Status: Posted Version 1 posted 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-8081211\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Article\",\"associatedPublications\":[],\"authors\":[{\"id\":543663480,\"identity\":\"d844f09e-ed48-468b-b8cf-5e2951697099\",\"order_by\":0,\"name\":\"Maria Santos\",\"email\":\"data:image/png;base64,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\",\"orcid\":\"\",\"institution\":\"Itaú\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Maria\",\"middleName\":\"\",\"lastName\":\"Santos\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-11-11 01:25:20\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-8081211/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-8081211/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":96366634,\"identity\":\"4fdf607a-d067-48d4-ba2b-e892a70981d1\",\"added_by\":\"auto\",\"created_at\":\"2025-11-20 10:11:38\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":248351,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-8081211/v1/4c652c26-c256-45a5-86aa-516f57c513a8.pdf\"}],\"financialInterests\":\"There is \\u003cb\\u003eNO\\u003c/b\\u003e Competing Interest.\",\"formattedTitle\":\"\\u003cp\\u003eUse of Fire Radiative Power as an Indicator for Forest Fire Prediction: A Study of Machine Learning Architectures\\u003c/p\\u003e\",\"fulltext\":[{\"header\":\"Introduction\",\"content\":\"\\u003cp\\u003eClimate change represents one of the major global challenges of the 21st century, affecting not only the environment but also social, economic, and political dynamics (IPCC, 2021). Rising global temperatures, increasing extreme weather events, and biodiversity loss highlight the urgency of developing mitigation and adaptation strategies.\\u003c/p\\u003e\\n\\u003cp\\u003eIn this context, machine learning models have emerged as promising tools capable of identifying complex patterns in large environmental datasets, enabling more accurate forecasting and supporting decision-making processes (Goodfellow, Bengio \\u0026amp; Courville, 2016). This project focuses on analyzing Brazil\\u0026rsquo;s position within this scenario, considering its vulnerability to wildfires and other climate-related events.\\u003c/p\\u003e\\n\\u003cp\\u003eThe dataset used in this study is part of the \\u0026ldquo;Queimadas\\u0026rdquo; project developed by the Brazilian National Institute for Space Research (INPE), which monitors fire outbreaks in real time using satellite data. The platform provides valuable information including geolocation, intensity, and frequency of fire events, particularly in critical biomes such as the Amazon, Cerrado, and Pantanal.\\u003c/p\\u003e\\n\\u003cp\\u003eUsing these data, this project aims to develop predictive models capable of identifying spatiotemporal patterns related to wildfire occurrence, contributing to more efficient mitigation strategies. Although the dataset is extensive, the limited execution time requires prioritization and careful selection of the most relevant variables.\\u003c/p\\u003e\"},{\"header\":\"Methodology\",\"content\":\"\\u003cp\\u003eAn exploratory data analysis was conducted to examine data distribution, identify patterns, and detect inconsistencies. Records from August to November 2024 were selected, corresponding to the period with historically higher wildfire activity. The DataHora column was converted to a datetime format, from which day, month, hour, and weekday features were extracted for a more granular temporal viewpoint. Invalid values flagged by INPE (e.g., \\u0026minus;999 in FRP and DaysWithoutRain) were removed to ensure data integrity. Spatial visualizations were generated using Plotly, allowing the identification of municipalities with higher fire incidence, especially within the Amazon biome.\\u003c/p\\u003e\\n\\u003cp\\u003eDue to the large size of the dataset (approximately five million records), a stratified sampling approach was applied by geographic region to maintain representativeness while ensuring computational feasibility. This preserved the major spatial patterns observed in the complete dataset.\\u003c/p\\u003e\\n\\u003cp\\u003eMissing values in the Fire Radiative Power (FRP) time series were handled using temporal interpolation (interpolate(method=\\u0026apos;time\\u0026apos;)), followed by backfilling to correct initial gaps. Outliers in FRP and precipitation were mitigated by clipping values above the 99th percentile, reducing the influence of extreme events. A logarithmic transformation (np.log1p()) was applied to reduce distribution skewness and enhance model stability in regression tasks.\\u003c/p\\u003e\\n\\u003cp\\u003eCategorical variables such as biome and region were encoded using One-Hot Encoding with drop_first=True to prevent multicollinearity. Latitude and longitude coordinates were transformed into three-dimensional Cartesian space to avoid angular discontinuities and provide smoother spatial inputs for machine-learning algorithms.\\u003c/p\\u003e\\n\\u003cp\\u003eAfter preprocessing, multiple machine-learning architectures were implemented and evaluated, including Prophet for time-series forecasting, a Multi-Layer Perceptron (MLP) for nonlinear regression, XGBoost Regressor for gradient-boosted tree modeling, and TabNet for deep tabular feature representation. These models were compared to assess predictive performance and suitability for wildfire prediction in Brazilian biomes.\\u003c/p\\u003e\"},{\"header\":\"Results\",\"content\":\"\\u003cp\\u003eThe present study aimed to predict Fire Radiative Power (FRP), a metric directly associated with wildfire intensity, using a robust set of geospatial, meteorological, and temporal variables from the INPE database. The selection of FRP instead of traditional fire-risk indices reflects its ability to quantify the actual severity of fire events, as previously demonstrated by Wooster et al. (2003) and currently adopted by agencies such as NASA.\\u003c/p\\u003e\\n\\u003cp\\u003eExploratory data analysis revealed that the most affected biomes are the Amazon and Cerrado, with particular relevance to the state of Par\\u0026aacute;. Within this region, the municipalities of S\\u0026atilde;o F\\u0026eacute;lix do Xingu, Novo Progresso, and Altamira exhibit the highest concentrations of intense wildfire activity. These findings are consistent with results from MapBiomas (2023) and PRODES/INPE surveys, which indicate that Par\\u0026aacute; accounted for approximately 30% of the burned area in the Amazon between 2020 and 2022.\\u003c/p\\u003e\\n\\u003cp\\u003eThese municipalities share climatic characteristics that favor ignition and propagation:\\u003c/p\\u003e\\n\\u003cul type=\\\"disc\\\"\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eS\\u0026atilde;o F\\u0026eacute;lix do Xingu:\\u003c/strong\\u003e Equatorial and humid, with an annual average of 1,423 mm of rainfall, high humidity, and mean temperatures of 26.7 \\u0026deg;C.\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eNovo Progresso:\\u003c/strong\\u003e Hot and humid climate, with an average precipitation of 2,200 mm and a pronounced dry season between August and December.\\u003c/li\\u003e\\n \\u003cli\\u003e\\u003cstrong\\u003eAltamira:\\u003c/strong\\u003e Tropical climate with short dry periods that coincide with peaks in wildfire activity.\\u003c/li\\u003e\\n\\u003c/ul\\u003e\\n\\u003cp\\u003eIn addition to environmental conditions, factors such as agricultural expansion, cattle ranching, and deforestation pressure contribute to an increased likelihood of high-intensity fire events. Accurate prediction of FRP in these areas can support specialists in devising targeted mitigation strategies and prioritizing surveillance efforts.\\u003c/p\\u003e\\n\\u003cp\\u003eAmong the evaluated machine-learning architectures, the \\u003cstrong\\u003eXGBoost Regressor combined with Grid Search optimization\\u003c/strong\\u003e demonstrated superior performance in capturing nonlinear and complex patterns across the dataset. Its ability to handle heterogeneous features, irregular distributions, and interactions between climatic and spatial variables contributed to a higher predictive accuracy compared to the other tested models. This suggests its suitability for operational deployment in wildfire early-warning systems.\\u003c/p\\u003e\"},{\"header\":\"Conclusion\",\"content\":\"\\u003cp\\u003eThis study investigated the predictive modeling of Fire Radiative Power (FRP) using geospatial, meteorological, and temporal variables provided by INPE. By prioritizing FRP over traditional fire risk indices, the analysis incorporated an indicator directly associated with fire intensity and potential environmental impact, as previously supported in the literature. The exploratory results revealed that the Amazon and Cerrado biomes exhibit the highest concentration of high-intensity fire events, with particular incidence in the state of Par\\u0026aacute;. Municipalities such as S\\u0026atilde;o F\\u0026eacute;lix do Xingu, Novo Progresso, and Altamira demonstrated recurrent hotspots, driven by favorable climatic conditions and anthropogenic pressures, including deforestation and the expansion of agricultural frontiers.\\u003c/p\\u003e\\n\\u003cp\\u003eMultiple machine learning architectures were evaluated, including Prophet, Multi-Layer Perceptron (MLP), TabNet, and XGBoost Regressor. Among these, the XGBoost Regressor optimized through Grid Search achieved superior performance, effectively capturing nonlinear interactions and complex variable dependencies. These results highlight the potential of ensemble-based approaches for environmental prediction tasks, particularly when dealing with heterogeneous and highly dynamic datasets.\\u003c/p\\u003e\\n\\u003cp\\u003eThe predictive capability demonstrated by the proposed modeling approach may support environmental monitoring systems, enabling earlier identification of regions with elevated fire intensity. Such insights can contribute to mitigation planning, resource allocation, and the development of regional fire prevention strategies. Future research should consider incorporating satellite-derived vegetation indices, long-term climate trends, and explainable AI techniques to enhance model interpretability and operational applicability. Expanding the temporal resolution and integrating real-time data streams may further improve forecast reliability and support large-scale environmental policy decisions.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\n \\u003cli\\u003eARJONA-M\\u0026Eacute;NDEZ, G. et al. (2019). \\u003cem\\u003eWildfire prediction based on machine learning: Case study Mexico\\u003c/em\\u003e. arXiv preprint arXiv:1908.07442. Retrieved June 6, 2025, from https://arxiv.org/pdf/1908.07442\\u003c/li\\u003e\\n \\u003cli\\u003eWORLD BANK. (2023). \\u003cem\\u003eClimate and Development Report for Brazil 2023\\u003c/em\\u003e. United Nations Brazil. Retrieved October 7, 2024, from https://brasil.un.org\\u003c/li\\u003e\\n \\u003cli\\u003eCLIMATE-DATA. (2025). \\u003cem\\u003eClimate: Altamira\\u003c/em\\u003e. Retrieved April 20, 2025, from https://pt.climate-data.org/america-do-sul/brasil/para/altamira-43955/\\u003c/li\\u003e\\n \\u003cli\\u003eCLIMATE TRACE. (2024). \\u003cem\\u003eClimate Trace Data\\u003c/em\\u003e. Retrieved August 24, 2024, from https://climatetrace.org/data\\u003c/li\\u003e\\n \\u003cli\\u003eECO DEBATE. (2024). \\u003cem\\u003eRegional impacts of climate change\\u003c/em\\u003e. Retrieved October 7, 2024, from https://ecodebate.com.br\\u003c/li\\u003e\\n \\u003cli\\u003eFAYYAD, U., PIATETSKY-SHAPIRO, G., \\u0026amp; SMYTH, P. (1996). From data mining to knowledge discovery in databases.\\u0026nbsp;\\u003cem\\u003eAI Magazine\\u003c/em\\u003e, 17(3), 37\\u0026ndash;54.\\u003c/li\\u003e\\n \\u003cli\\u003eINPE \\u0026ndash; National Institute for Space Research. (2024). \\u003cem\\u003eBurned Area Database \\u0026ndash; TerraBrasilis\\u003c/em\\u003e. S\\u0026atilde;o Jos\\u0026eacute; dos Campos, Brazil. Retrieved May 30, 2025, from https://terrabrasilis.dpi.inpe.br/queimadas/bdqueimadas/\\u003c/li\\u003e\\n \\u003cli\\u003eINPE \\u0026ndash; National Institute for Space Research. (2025). \\u003cem\\u003eFrequently Asked Questions \\u0026ndash; Fire Monitoring Program\\u003c/em\\u003e. Retrieved June 6, 2025, from https://terrabrasilis.dpi.inpe.br/queimadas/portal/faq/index.html\\u003c/li\\u003e\\n \\u003cli\\u003eINPE \\u0026ndash; National Institute for Space Research. (n.d.). \\u003cem\\u003eFire risk: concise explanation\\u003c/em\\u003e. Retrieved June 6, 2025, from https://dataservercoids.inpe.br/queimadas/queimadas/PublicacoesImpacto/documentos/RiscoFogo_Sucinto.pdf\\u003c/li\\u003e\\n \\u003cli\\u003eIntergovernmental Panel on Climate Change (IPCC). (2021). \\u003cem\\u003eClimate change 2021: The physical science basis\\u003c/em\\u003e. Retrieved May 23, 2025, from https://www.ipcc.ch/report/ar6/wg1/\\u003c/li\\u003e\\n \\u003cli\\u003eMAPBIOMAS. (2025). \\u003cem\\u003eProject MapBiomas \\u0026ndash; Collection 9 of the Annual Series of Land Cover and Use Maps of Brazil\\u003c/em\\u003e. Retrieved May 23, 2025, from https://brasil.mapbiomas.org/\\u003c/li\\u003e\\n \\u003cli\\u003eNASA. (2024). \\u003cem\\u003eGlobal Climate Change: Vital Signs of the Planet\\u003c/em\\u003e. Retrieved August 20, 2024, from https://climate.nasa.gov/\\u003c/li\\u003e\\n \\u003cli\\u003eNOVO PROGRESSO CITY HALL. (2025). \\u003cem\\u003eAbout the municipality\\u003c/em\\u003e. Retrieved April 20, 2025, from https://camaranovoprogresso.pa.gov.br/o-municipio/\\u003c/li\\u003e\\n \\u003cli\\u003eWORLD METEOROLOGICAL ORGANIZATION. (2023). \\u003cem\\u003eReport reveals that Brazil faced 12 extreme climate events in 2023\\u003c/em\\u003e. ONU News. Retrieved October 7, 2024, from https://news.un.org\\u003c/li\\u003e\\n \\u003cli\\u003ePAR\\u0026Aacute; STATE GOVERNMENT. (2025). \\u003cem\\u003eS\\u0026atilde;o F\\u0026eacute;lix do Xingu: Municipal socioeconomic profile\\u003c/em\\u003e. Bel\\u0026eacute;m: Foundation for Support of Studies and Research of the Amazon (FAPESPA). Retrieved April 20, 2025, from https://www.fapespa.pa.gov.br/wp-content/uploads/2025/02/Sao-Felix-do-Xingu.pdf\\u003c/li\\u003e\\n \\u003cli\\u003eSANTOS, M. H. S. (2025). \\u003cem\\u003eUsing fire radiative power as an indicator for the prediction of forest fires: A study of machine learning architectures\\u003c/em\\u003e. Retrieved May 30, 2025, from https://github.com/mariahelenass/Wildfire-Brazil\\u003c/li\\u003e\\n \\u003cli\\u003eSILVA, I. N., SPATTI, D. H., \\u0026amp; FLAUZINO, R. A. (2016). \\u003cem\\u003eArtificial neural networks for engineering and applied sciences: Theoretical foundations and practical aspects\\u003c/em\\u003e (2nd ed., pp. 46\\u0026ndash;47). S\\u0026atilde;o Paulo: Artliber.\\u003c/li\\u003e\\n \\u003cli\\u003eWORLD METEOROLOGICAL ORGANIZATION (WMO). (2024). \\u003cem\\u003eState of the Global Climate\\u003c/em\\u003e. Retrieved August 25, 2024, from https://public.wmo.int/en/resources/library/state-of-global-climate\\u003c/li\\u003e\\n\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":true,\"hideJournal\":true,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true},\"keywords\":\"\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-8081211/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-8081211/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"This project addresses contemporary climate, sociological, and environmental challenges by exploring the use of machine learning models to analyze fire radiative power (FRP) as an indicator for forest fire prediction. Climate change has intensified extreme weather events and biodiversity loss, demanding more accurate predictive tools. Machine learning enables the extraction of hidden patterns from large-scale environmental data, supporting more reliable forecasting and mitigation strategies. Using satellite data provided by the Brazilian National Institute for Space Research (INPE), this study evaluates fire occurrences across critical biomes, including the Amazon, Cerrado, and Pantanal. The dataset includes geospatial coordinates, time of detection, and complementary meteorological variables. By identifying patterns in wildfire behavior, this work aims to support the development of more effective prevention policies and environmental risk management approaches. Despite time constraints and the large volume of available data, the project highlights the potential of machine learning architectures to enhance climate resilience and guide data-driven decision-making in Brazil’s environmental context.\",\"manuscriptTitle\":\"Use of Fire Radiative Power as an Indicator for Forest Fire Prediction: A Study of Machine Learning Architectures\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-11-18 10:32:45\",\"doi\":\"10.21203/rs.3.rs-8081211/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"researchsquare\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":true,\"externalIdentity\":\"\",\"sideBox\":\"\",\"snPcode\":\"\",\"submissionUrl\":\"/submission\",\"title\":\"Research Square\",\"twitterHandle\":\"researchsquare\",\"acdcEnabled\":true,\"dfaEnabled\":false,\"editorialSystem\":\"\",\"reportingPortfolio\":\"\",\"inReviewEnabled\":false,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"24d9ec20-7672-49b0-ae93-c45c2e871886\",\"owner\":[],\"postedDate\":\"November 18th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[{\"id\":58176170,\"name\":\"Earth and environmental sciences/Climate sciences/Climate change/Climate-change mitigation\"},{\"id\":58176171,\"name\":\"Earth and environmental sciences/Climate sciences/Climate change/Projection and prediction\"}],\"tags\":[],\"updatedAt\":\"2025-11-20T05:41:13+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-11-18 10:32:45\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-8081211\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-8081211\",\"identity\":\"rs-8081211\",\"version\":[\"v1\"]},\"buildId\":\"XKTyCvWXoU3ODBz1xrDgd\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}