Forecasting fire risk with machine learning and dynamic information derived from satellite vegetation index time-series

Spatial analysis and machine learning prediction of forest fire susceptibility: a comprehensive approach for effective management and mitigation

Forest fires (FF) in tropical seasonal forests impact ecosystem. Addressing FF in tropical ecosystems has become a priority to mitigate impacts on biodiversity loss and climate change. The escalating frequency and intensity of FF globally have become a mounting concern. Understanding their tendencies, patterns, and vulnerabilities is imperative for conserving ecosystems and facilitating the development of effective prevention and management strategies. This study investigates the trends, patterns, and spatiotemporal distribution of FF for the period of 2001-2022, and delineates the forest fire susceptibility zones in Odisha State, India. The study utilized: (a) MODIS imagery to examine active fire point data; (b) Kernel density tools; (c) FF risk prediction using two machine learning algorithms, namely Support Vector Machine (SVM) and Random Forest (RF); (d) Receiver Operating Characteristic and Area Under the Curve, along with various evaluation metrics; and (e) a total of 19 factors, including three topographical, seven climatic, four biophysical, and five anthropogenic, to create a map indicating areas vulnerable to FF. The validation results revealed that the RF model achieved a precision exceeding 94 % on the validation datasets, while the SVM model reached 89 %. The estimated forest fire susceptibility zones using RF and SVM techniques indicated that 20.14 % and 16.72 % of the area, respectively, fall under the "Very High Forest Fire" susceptibility class. Trend analysis reveals a general upward trend in forest fire occurrences (R2 = 0.59), with a notable increase after 2015, peaking in 2021. Notably, Angul district was identified as the most affected area, documenting the highest number of forest fire incidents over the past 22 years. Additionally, forest fire mitigation plans have been developed by drawing insights from forest fire management strategies implemented in various countries worldwide. Overall, this analysis provides valuable insights for policymakers and forest management authorities to develop effective strategies for forest fire prevention and mitigation.

Improving the prediction of wildfire susceptibility on Hawai'i Island, Hawai'i, using explainable hybrid machine learning models

This study presents a comparative analysis of four Machine Learning (ML) models used to map wildfire susceptibility on Hawai'i Island, Hawai'i. Extreme Gradient Boosting (XGBoost) combined with three meta-heuristic algorithms - Whale Optimization (WOA), Black Widow Optimization (BWO), and Butterfly Optimization (BOA) - were employed to map areas susceptible to wildfire. To generate a wildfire inventory, 1408 wildfire points were identified within the study area from 2004 to 2022. The four ML models (XGBoost, WOA-XGBoost, BWO-XGBoost, and BOA-XGBoost) were run using 14 wildfire-conditioning factors categorized into four main groups: topographical, meteorological, vegetation, and anthropogenic. Six performance metrics - sensitivity, specificity, positive predictive values, negative predictive values, the Area Under the receiver operating characteristic Curve (AUC), and the average precision (AP) of Precision-Recall Curves (PRCs) - were used to compare the predictive performance of the ML models. The SHapley Additive exPlanations (SHAP) framework was also used to interpret the importance values of the 14 influential variables for the modeling of wildfire on Hawai'i Island using the four models. The results of the wildfire modeling indicated that all four models performed well, with the BWO-XGBoost model exhibiting a slightly higher prediction performance (AUC = 0.9269), followed by WOA-XGBoost (AUC = 0.9253), BOA-XGBoost (AUC = 0.9232), and XGBoost (AUC = 0.9164). SHAP analysis revealed that the distance from a road, annual temperature, and elevation were the most influential factors. The wildfire susceptibility maps generated in this study can be used by local authorities for wildfire management and fire suppression activity.

Assessment of forest fire emissions in Uttarakhand State, India, using Open Geospatial data and Google Earth Engine

In the recent past, forest fires have increased due to the changing climate pattern. It is necessary to analyse and quantify various gaseous emissions so as to mitigate their harmful effects on air pollution. Satellite remote sensing data provides an opportunity to study the greenhouse gases in the atmosphere. The multispectral sensor of the Tropospheric Monitoring Instrument (Sentinel-5) is capable of recording the reflectance of wavelengths vital for measuring the atmospheric concentrations of methane, formaldehyde, aerosol, carbon monoxide, etc., at a spatial resolution of 0.01°. The present study utilized the Google Earth Engine (GEE) platform to study the emissions caused by forest fires in four districts of Uttarakhand State of India, which witnessed unprecedented fires in April-May 2021. All the datasets were ingested in GEE, which has the capability to analyse large datasets without the need to download them. The pre-fire period chosen was September 2020; the fire period was February-May 2021, and the post-fire period was June 2021. The variables chosen were aerosol absorbing index (AAI), carbon monoxide (CO) and nitrogen dioxide (NO2). The climate parameter temperature (Moderate Resolution Imaging Spectroradiometer Land Surface Temperature) and precipitation (from Climate Hazards Group InfraRed Precipitation (CHIRPS) Pentad) were also studied for the period mentioned. The results indicate a different trend for emissions in each district. For AAI, maximum emissions were noted in district Nainital followed by Almora, Tehri Garhwal and Garhwal. For CO emissions, the most affected district was Almora followed by Nainital, Garhwal and Tehri Garhwal. For NO2 emissions, the most affected district was Garhwal, followed by Nainital, Tehri Garhwal and Almora. Delta Normalized Burn Ratio was computed from Sentinel data (difference of pre-fire and post-fire images) to assess the burnt area severity. The Delta Normalized Burn Ratio values observed that the district with the most burnt area is Garhwal, followed by Nainital, Almora and Tehri Garhwal. The elevated temperatures and scanty rainfall patterns regulated the intensity and duration of forest fire. Monitoring the gaseous emissions as a consequence of forest fire in the GEE platform is much easier and more convenient at a regional level. Such data is much needed for mitigation measures to be implemented in time.

Future global concurrent droughts and their effects on maize yield

Droughts are one of the most devastating natural disasters. Droughts can co-exist in different forms (e.g. meteorological, hydrological, and agricultural) as concurrent droughts. Such concurrent droughts can have far reaching implications for crop yield and global food security. The present study aims to assess global concurrent drought traits and their effects on maize yield under climate change. The standardized indices of precipitation, runoff, and soil moisture incorporated as multivariate standardized drought index (MSDI) using copula functions are used to quantify the concurrent droughts. The ensemble data of several General Circulation Models (GCMs) considering the high emission scenario of Coupled Model Intercomparison Project phase 6 (CMIP6) are utilized. Applying run theory on a time series (1950-2100) of MSDI values, the duration, severity, areal coverage, and average areal intensity of concurrent droughts are computed. The temporal evolution of drought duration and severity are compared among historical (1950-2014), near future (2021-2060), and far future (2061-2100) timeframes. The results indicate that the most vulnerable regions in the late 21st century are Central America, the Mediterranean, Southern Africa, and the Amazon basin. The indices and spatial extent of the individual droughts are used as predictor variables to predict the country-level crop index of the top seven producers of maize. The historical dynamics between maize yield and different drought forms are projected using XGBoost (Extreme Gradient Boosting) algorithms. The future temporal changes in drought-crop yield dynamics are tracked using probabilities of various drought forms under yield-loss conditions. The conditional concurrent drought probabilities are as high as 84 %, 64 %, and 37 % in France, Mexico, and Brazil, revealing that concurrent drought affects the maize yield tremendously in the far future. This approach of applying statistical and soft-computing techniques could aid in drought mitigation under changing climatic conditions.

Design of a Forest Fire Early Alert System through a Deep 3D-CNN Structure and a WRF-CNN Bias Correction

Throughout the years, wildfires have negatively impacted ecological systems and urban areas. Hence, reinforcing territorial risk management strategies against wildfires is essential. In this study, we built an early alert system (EAS) with two different Machine Learning (ML) techniques to calculate the meteorological conditions of two Colombian areas: (i) A 3D convolutional neural net capable of learning from satellite data and (ii) a convolutional network to bias-correct the Weather Research and Forecasting (WRF) model output. The results were used to quantify the daily Fire Weather Index and were coupled with the outcomes from a land cover analysis conducted through a Naïve-Bayes classifier to estimate the probability of wildfire occurrence. These results, combined with an assessment of global vulnerability in both locations, allow the construction of daily risk maps in both areas. On the other hand, a set of short-term preventive and corrective measures were suggested to public authorities to implement, after an early alert prediction of a possible future wildfire. Finally, Soil Management Practices are proposed to tackle the medium- and long-term causes of wildfire development, with the aim of reducing vulnerability and promoting soil protection. In conclusion, this paper creates an EAS for wildfires, based on novel ML techniques and risk maps.

Fire regime of peatlands in the Angolan Highlands

The Angolan Highlands region includes the Angolan miombo woodland ecoregion which supports miombo woodland, grasslands, subsistence agricultural land, and peatland deposits. Extensive fires, slash and burn agriculture, peat fuel extraction, and peatland drainage are among the anthropogenic practices that threaten these peatland deposits. Peat fires cause peatland degradation, release significant amounts of greenhouse gases, deteriorate air quality, and contribute towards climate change and biodiversity loss. This study presents an analysis of the fire regimes over the period 2001 to 2020 in an under-studied area of the Angolan Highlands. Moderate Resolution Imaging Spectroradiometer (MODIS) fire and vegetation data were used in combination with a land use/land cover (LULC) classification map to calculate fire frequency, burn area, and fire regimes. The fire patterns within the study site are comparable to those found in African woodland savannas. Across the study site, 6976 km² (11.31%) of the land surface area burned at least nine times from 2001 to 2020, occurring largely within in the river valley environment. Considering the different LULC classes, peatlands were calculated to (a) burn more frequently (average fire frequency from 2001 to 2020 = 9.12), (b) have the smallest proportion (4.11%) of area which remained unburnt over the fire archive, and (c) have the largest average proportion (45.65% or 746 km²) of burnt area per year. Peatland burning occurred predominantly during drier months from May to September. The results of this study highlight the strong influence of LULC on the fire frequency and distribution in the study area, requiring unique fire management strategies. As has been documented for boreal and tropical peatlands across the globe, we stress the importance of peatland conservation and protection; continued unsustainable management practices may lead to the loss of these important peatland deposits.

Forest fire monitoring via uncrewed aerial vehicle image processing based on a modified machine learning algorithm

Forests are indispensable links in the ecological chain and important ecosystems in nature. The destruction of forests seriously influences the ecological environment of the Earth. Forest protection plays an important role in human sustainable development, and the most important aspect of forest protection is preventing forest fires. Fire affects the structure and dynamics of forests and also climate and geochemical cycles. Using various technologies to monitor the occurrence of forest fires, quickly finding the source of forest fires, and conducting early intervention are of great significance to reducing the damage caused by forest fires. An improved forest fire risk identification algorithm is established based on a deep learning algorithm to accurately identify forest fire risk in a complex natural environment. First, image enhancement and morphological preprocessing are performed on a forest fire risk image. Second, the suspected forest fire area is segmented. The color segmentation results are compared using the HAF and MCC methods, and the suspected forest fire area features are extracted. Finally, the forest fire risk image recognition processing is conducted. A forest fire risk dataset is constructed to compare different classification methods to predict the occurrence of forest fire risk to improve the backpropagation (BP) neural network forest fire identification algorithm. An improved machine learning algorithm is used to evaluate the classification accuracy. The results reveal that the algorithm changes the learning rate between 0.1 and 0.8, consistent with the cross-index verification of the 10x sampling algorithm. In the combined improved BP neural network and support vector machine (SVM) classifier, forest fire risk is recognized based on feature extraction and the BP network. In total, 1,450 images are used as the training set. The experimental results reveal that in image preprocessing, image enhancement technology using the frequency and spatial domain methods can enhance the useful information of the image and improve its clarity. In the image segmentation stage, MCC is used to evaluate the segmentationresults. The accuracy of this algorithm is high compared with other algorithms, up to 92.73%. Therefore, the improved forest fire risk identification algorithm can accurately identify forest fire risk in the natural environment and contribute to forest protection.

Electrical Responses of Pinus halepensis Mill. as an Indicator of Wildfire Risk in Mediterranean Forests by Complementing Live Fuel Moisture

Pinus halepensis forests, as Mediterranean-type ecosystems, are subject to high levels of wildfire risk in times of drought, with meteorological conditions of water stress and very high temperatures, mainly in summer. Considering the difficulty of knowing the phenological state of this species, the objective of this research was to evaluate the possibility of implementing the electrical responses (voltage and short-circuit current) as a variable in fire risk management models, compared to live fuel moisture. On the one hand, the obtained results demonstrate non-significant differences between the moisture content of the different fractions of the living branches (base and half of the branch and live fuel), even in times of drought with hydric stress and very high temperatures. Live fuel moisture of Pinus halepensis does not show significant seasonal variations under the influence of extreme fire risk factors. For this reason, it should be complemented with other variables for fire risk management models. On the other hand, the differences registered in the electrical signal show oscillations with significant variations, which are strongly correlated with the periods of extremely favourable meteorological conditions for wildfires. So, the voltages measured show ranges that correspond with great accuracy to the FWI. Voltage variation is dependent on the hydraulic dynamic plant behaviour and a result of the physiological response of pine trees to abiotic stress of drought. It is an easy-to-measure electrical parameter as well as a very reliable indicator with a high correlation with wildfire risk. Thus, electrical responses could add more knowledge about the phenological state of the trees in dependence on stress climatic conditions, allowing integration of these variables in the preventive wildfire modelling and management.

A Highly Accurate Forest Fire Prediction Model Based on an Improved Dynamic Convolutional Neural Network

In this work, an improved dynamic convolutional neural network (DCNN) model to accurately identify the risk of a forest fire was established based on the traditional DCNN model. First, the DCNN network model was trained in combination with transfer learning, and multiple pre-trained DCNN models were used to extract features from forest fire images. Second, principal component analysis (PCA) reconstruction technology was used in the appropriate subspace. The constructed 15-layer forest fire risk identification DCNN model named “DCN_Fire” could accurately identify core fire insurance areas. Moreover, the original and enhanced image data sets were used to evaluate the impact of data enhancement on the model’s accuracy. The traditional DCNN model was improved and the recognition speed and accuracy were compared and analyzed with the other three DCNN model algorithms with different architectures. The difficulty of using DCNN to monitor forest fire risk was solved, and the model’s detection accuracy was further improved. The true positive rate was 7.41% and the false positive rate was 4.8%. When verifying the impact of different batch sizes and loss rates on verification accuracy, the loss rate of the DCN_Fire model of 0.5 and the batch size of 50 provided the optimal value for verification accuracy (0.983). The analysis results showed that the improved DCNN model had excellent recognition speed and accuracy and could accurately recognize and classify the risk of a forest fire under natural light conditions, thereby providing a technical reference for preventing and tackling forest fires.

Improving WRF-Fire Wildfire Simulation Accuracy Using SAR and Time Series of Satellite-Based Vegetation Indices

Wildfire simulations depend on fuel representation. Present fuel models are mainly based on the density and properties of different vegetation types. This study aims to improve the accuracy of WRF-Fire wildfire simulations, by using synthetic-aperture radar (SAR) data to estimate the fuel load and the trend of vegetation index to estimate the dryness of woody vegetation. We updated the chaparral and timber standard woody fuel classes in the WRF-Fire fuel settings. We used the ESA global above-ground biomass (AGB) based on SAR data to estimate the fuel load, and the Landsat normalized difference vegetation index (NDVI) trends of woody vegetation to estimate the fuel moisture content. These fuel sub-parameters represent the dynamic changes and spatial variability of woody fuel. We simulated two wildfires in Israel while using three different fuel models: the original 13 Anderson Fire Behavior fuel model, and two modified fuel models introducing AGB alone, and AGB and dryness. The updated fuel model (the basic fuel model plus the AGB and dryness) improved the simulation results significantly, i.e., the Jaccard similarity coefficient increased by 283% on average. Our results demonstrate the potential of combining satellite SAR data and Landsat NDVI trends to improve WRF-Fire wildfire simulations.

Machine learning based prediction for China's municipal solid waste under the shared socioeconomic pathways

Reliable forecast of municipal solid waste (MSW) generation is crucial for sustainable and efficient waste management. Big data analysis is a novel method to forecast MSW more accurately. Thus, this study employs five kinds of supervised machine learning approaches including linear regression, polynomial regression, support vector machine, random forest, and extreme gradient boosting (XGBoost) to examine their forecast performances. China's MSW generation from 2020 to 2060 under five shared socioeconomic pathways (SSPs) is further predicted and the mechanisms between MSW generation and socioeconomic features are explored. Results show that population and GDP are two dominant indicators in MSW prediction, and XGBoost model is proved to be effective in MSW forecast. MSW generation of China in 2060 is estimated to be 464-688 megatons under different SSPs scenarios, about four to six times of that in 2000. SSP3 that has the most population, least GDP and the highest climate change challenges is the only scenario showing a potential of MSW peak during the study period. The key for MSW increase is mainly the increase of per capita MSW caused by GDP. Finally, several policy recommendations are raised to reduce the overall MSW generation.

Remote Sensing and Meteorological Data Fusion in Predicting Bushfire Severity: A Case Study from Victoria, Australia

The extent and severity of bushfires in a landscape are largely governed by meteorological conditions. An accurate understanding of the interactions of meteorological variables and fire behaviour in the landscape is very complex, yet possible. In exploring such understanding, we used 2693 high-confidence active fire points recorded by a Moderate Resolution Imaging Spectroradiometer (MODIS) sensor for nine different bushfires that occurred in Victoria between 1 January 2009 and 31 March 2009. These fires include the Black Saturday Bushfires of 7 February 2009, one of the worst bushfires in Australian history. For each fire point, 62 different meteorological parameters of bushfire time were extracted from Bureau of Meteorology Atmospheric high-resolution Regional Reanalysis for Australia (BARRA) data. These remote sensing and meteorological datasets were fused and further processed in assessing their relative importance using four different tree-based ensemble machine learning models, namely, Random Forest (RF), Fuzzy Forest (FF), Boosted Regression Tree (BRT), and Extreme Gradient Boosting (XGBoost). Google Earth Engine (GEE) and Landsat images were used in deriving the response variable–Relative Difference Normalised Burn Ratio (RdNBR), which was selected by comparing its performance against Difference Normalised Burn Ratio (dNBR). Our findings demonstrate that the FF algorithm utilising the Weighted Gene Coexpression Network Analysis (WGCNA) method has the best predictive performance of 96.50%, assessed against 10-fold cross-validation. The result shows that the relative influence of the variables on bushfire severity is in the following order: (1) soil moisture, (2) soil temperature, (3) air pressure, (4) air temperature, (5) vertical wind, and (6) relative humidity. This highlights the importance of soil meteorology in bushfire severity analysis, often excluded in bushfire severity research. Further, this study provides a scientific basis for choosing a subset of meteorological variables for bushfire severity prediction depending on their relative importance. The optimal subset of high-ranked variables is extremely useful in constructing simplified and computationally efficient surrogate models, which can be particularly useful for the rapid assessment of bushfire severity for operational bushfire management and effective mitigation efforts.

MDIR Monthly Ignition Risk Maps, an Integrated Open-Source Strategy for Wildfire Prevention

Countries unaccustomed to wildfires are currently experiencing wildfire as a new climate-change reality. Understanding how fire ignition and propagation are correlated with temperature, orography, humidity, wind, and the mixture and age of individual plants must be considered when designing prevention strategies. While wildfire prevention focuses on fire ignition avoidance, firefighting success depends on early ignition detection, meaning that, in either case, ignition plays a major role. The current case study considered three Portuguese municipalities that annually observe frequent fire ignitions (Tomar, Ourém, and Ferreira do Zêzere) as the testing ground for the Modernized Dynamic Ignition Risk (MDIR) strategy, thus evaluating the efficiency of MDIR and the efficacy of the variables used. This methodology uses geographic information systems technology sustained by open-source satellite imagery, along with the Habitat Risk Assessment model from the InVEST software package, as drivers for the MDIR application. The MDIR approach grants frequent update capabilities and fully open-sourced high ignition risk area identification, producing monthly ignition risk maps. The advantage of using this method is the ease of adaptation to any current monitoring strategy, awarding further efficiency and efficacy in reducing ignitions. The approach delivered adequate results in estimating ignitions for the three Portuguese municipalities, achieving, for several months, prediction accuracy percentages of over 70%. For the studied area, MDIR clearly identifies areas of high ignition risk and delivers an average of 62% success in predicting ignitions, thus showing potential for analyzing the impact of policy implementation and monitoring through the strategy design.

A wildfire growth prediction and evaluation approach using Landsat and MODIS data

The increasing wildfire occurrence due to global climate changes urged the improvement of present wildfire growth prediction and evaluation methods. This study aimed to propose novel solutions to their two primary limitations, including the lack of robust fuel classification method and the low spatial resolution of wildfire growth accuracy assessment while ensuring wide applicability using open data satellite missions and software. The first objective was to create a robust two-step fuel model classification method consisted of the supervised machine learning classification of generalized land cover classes in the 1st level and their individual unsupervised classification to vegetation subtypes in the 2nd level. The second objective was creating a wildfire prediction accuracy assessment method using MODIS 250 m images, which overcome the limitations of low spatial resolution while preserving sub-daily temporal resolution. The wildfire on the Korčula island in Croatia was analyzed in the study, being specific for its long duration from 18 to 24 July 2015. The wildfire ignition occurred in the isolated area, which prolonged the response time from emergency agencies. Random Forest (RF) with input Landsat 8 spectral bands and indices resulted in the highest classification accuracy in the 1st classification level with an overall agreement of 83.6%. The vegetation subclasses from the 2nd classification level were matched to the 13 standard fuel models for the input in FARSITE software. The predicted wildfire evaluation showed the highest mean accuracy of 0.906 for the first two days, which decreased to 0.722 in the latter stages of the active wildfire caused by overprediction. The proposed two-step fuel model classification presented a cost-efficient solution to the fuel map creation in any part of the world, with a disadvantage of no in-situ ground truth identification and accuracy assessment for 2nd classification level. The evaluation of wildfire growth prediction with 250 m images enabled high spatial and temporal resolution of the assessment, while its limitations of wildfire overprediction and the negative effects of wildfire smoke in MODIS images should be addressed in future research.

A Systematic Review of Applications of Machine Learning Techniques for Wildfire Management Decision Support

Wildfires threaten and kill people, destroy urban and rural property, degrade air quality, ravage forest ecosystems, and contribute to global warming. Wildfire management decision support models are thus important for avoiding or mitigating the effects of these events. In this context, this paper aims at providing a review of recent applications of machine learning methods for wildfire management decision support. The emphasis is on providing a summary of these applications with a classification according to the case study type, machine learning method, case study location, and performance metrics. The review considers documents published in the last four years, using a sample of 135 documents (review articles and research articles). It is concluded that the adoption of machine learning methods may contribute to enhancing support in different fire management phases.

Forest Land Cover Mapping at a Regional Scale Using Multi-Temporal Sentinel-2 Imagery and RF Models

Over the last several decades, thanks to improvements in and the diversification of open-access satellite imagery, land cover mapping techniques have evolved significantly. Notable changes in these techniques involve the automation of different steps, yielding promising results in terms of accuracy, class detection and efficiency. The most successful methodologies that have arisen rely on the use of multi-temporal data. Several different approaches have proven successful. In this study, one of the most recently developed methodologies is tested in the region of Galicia (in Northwestern Spain), with the aim of filling gaps in the mapping needs of the Galician forestry sector. The methodology mainly consists of performing a supervised classification of individual images from a selected time series and then combining them through aggregation using decision criteria. Several of the steps of the methodology can be addressed in multiple ways: pixel resolution selection, classification model building and aggregation methods. The effectiveness of these three tasks as well as some others are tested and evaluated and the most accurate and efficient parameters for the case study area are highlighted. The final land cover map that is obtained for Galicia has high accuracy metrics (an overall accuracy of 91.6%), which is in line with previous studies that have followed this methodology in other regions. This study has led to the development of an efficient open-access solution to support the mapping needs of the forestry sector.