Fire Prediction Research Articles

Knowledge Graph Representation Learning-Based Forest Fire Prediction

Forest fires destroy the ecological environment and cause large property loss. There is much research in the field of geographic information that revolves around forest fires. The traditional forest fire prediction methods hardly consider multi-source data fusion. Therefore, the forest fire predictions ignore the complex dependencies and correlations of the spatiotemporal kind that usually bring valuable information for the predictions. Although the knowledge graph methods have been used to model the forest fires data, they mainly rely on artificially defined inference rules to make predictions. There is currently a lack of a representation and reasoning methods for forest fire knowledge graphs. We propose a knowledge-graph- and representation-learning-based forest fire prediction method in this paper for addressing the issues. First, we designed a schema for the forest fire knowledge graph to fuse multi-source data, including time, space, and influencing factors. Then, we propose a method, RotateS2F, to learn vector-based knowledge graph representations of the forest fires. We finally leverage a link prediction algorithm to predict the forest fire burning area. We performed an experiment on the Montesinho Natural Park forest fire dataset, which contains 517 fires. The results show that our method reduces mean absolute deviation by 28.61% and root-mean-square error by 53.62% compared with the previous methods.

Forest Fire Prediction Using K-Mean Clustering and Random Forest Classifier

Prediction of fire forest will be needing several parameters that located on the same location and on the same time frame. This is important to have data on the same location and same time periods, since forest fire mostly triggered by weather or temperature condition on certain area on the certain time. Since there are many parameters involved, the preprocessing will need to made the data have standard structure. The clustering process needed to give a label to the data with five class label, very low risk, low risk, medium risk, high risk, very high risk. Based on the clustered data, the data training and data test given for random forest classifier for model development, the composition of the data training and data test is 70:30. The accuracies of both algorithm is very good 100%, precision, recall and f1-score also have very high score 100%. This meant that the forest fire prediction model will produce a good prediction.

Optimal Emergency Evacuation Route Planning Model Based on Fire Prediction Data

For the emergency evacuation of cruise ships in case of sudden fire, this research proposes a dynamic route optimization method based on the improved A∗ algorithm for real-time information, in order to obtain the real-time optimal evacuation route. Initially, a basic network topology diagram is established according to the internal structure of the cruise ship. Before the occurrence of the accident, the A∗ algorithm can be applied to obtain an a priori evacuation network consisting of all the optimal routes from each node to the exit. At the time of the accident, the dynamic diffusion of fire can be simulated using Fire Dynamics Simulator (FDS) based on the preliminary information of the fire, so as to estimate the impact of the fire domain on each node of the network. Then, according to the fire dynamic diffusion data, the evacuation route planning is carried out by the improved A∗ algorithm applying the breadth-first search strategy, so as to determine the optimal route from the current node to the safety exit and to reduce the possibility of casualties due to the uncertainty of the fire during the evacuation. This model allows for both people’s safety and evacuation time to dynamically avoid fire-affected nodes and helps people to reach the safe area as soon as possible. Finally, the evacuation model is established according to the open-source cruise ship structure, and the evacuation process of people under the dynamic spread of cruise ship fire is simulated. The results show that the route planning method proposed in this research works out well in evacuating mass people, which can effectively reduce the evacuation time and improve the safety of the evacuation process.

Predicting real-time fire heat release rate by flame images and deep learning

The heat release rate (HRR) is the most critical parameter in characterizing the fire behavior and thermal effects of a burning item. However, traditional fire calorimetry methods are not applicable due to the lack of equipment in most fire scenarios. This work explores the real-time fire heat release rate prediction by using fire scene images and deep learning algorithms. A big database of 112 fire tests from the NIST Fire Calorimetry Database is formed, and 69,662 fire scene images labeled by their transient heat release rate are adopted to train the deep learning model. The fire tests conducted in the lab environment and the real fire events are used to validate and demonstrate the reliability of the trained model. Results show that regardless of the fire sources, background, light conditions, and camera settings, the proposed AI-image fire calorimetry method can well identify the transient fire heat release rate using only fire scene images. This work demonstrates that the deep learning algorithms can provide an alternative method to measure the fire HRR when traditional calorimetric methods cannot be used, which shows great potential in smart firefighting applications.

Enhanced interpolation with semi-supervised algorithm and greedy forwarding technique for forest fire prediction over wireless sensor and actuator networks

Enhanced interpolation with semi-supervised algorithm and greedy forwarding technique for forest fire prediction over wireless sensor and actuator networks

Spatio-Temporal Knowledge Graph Based Forest Fire Prediction with Multi Source Heterogeneous Data

Forest fires have frequently occurred and caused great harm to people’s lives. Many researchers use machine learning techniques to predict forest fires by considering spatio-temporal data features. However, it is difficult to efficiently obtain the features from large-scale, multi-source, heterogeneous data. There is a lack of a method that can effectively extract features required by machine learning-based forest fire predictions from multi-source spatio-temporal data. This paper proposes a forest fire prediction method that integrates spatio-temporal knowledge graphs and machine learning models. This method can fuse multi-source heterogeneous spatio-temporal forest fire data by constructing a forest fire semantic ontology and a knowledge graph-based spatio-temporal framework. This paper defines the domain expertise of forest fire analysis as the semantic rules of the knowledge graph. This paper proposes a rule-based reasoning method to obtain the corresponding data for the specific machine learning-based forest fire prediction methods, which are dedicated to tackling the problem with real-time prediction scenarios. This paper performs experiments regarding forest fire predictions based on real-world data in the experimental areas Xichang and Yanyuan in Sichuan province. The results show that the proposed method is beneficial for the fusion of multi-source spatio-temporal data and highly improves the prediction performance in real forest fire prediction scenarios.

Automatic Data Generation Method for Precise Ceiling Temperature Prediction of Cables Fire in the Utility Tunnel and Full-Scale Experimental Verification

Automatic Data Generation Method for Precise Ceiling Temperature Prediction of Cables Fire in the Utility Tunnel and Full-Scale Experimental Verification

Forest Fire Prediction with Imbalanced Data Using a Deep Neural Network Method

Forests suffer from heavy losses due to the occurrence of fires. A prediction model based on environmental condition, such as meteorological and vegetation indexes, is considered a promising tool to control forest fires. The construction of prediction models can be challenging due to (i) the requirement of selection of features most relevant to the prediction task, and (ii) heavily imbalanced data distribution where the number of large-scale forest fires is much less than that of small-scale ones. In this paper, we propose a forest fire prediction method that employs a sparse autoencoder-based deep neural network and a novel data balancing procedure. The method was tested on a forest fire dataset collected from the Montesinho Natural Park of Portugal. Compared to the best prediction results of other state-of-the-art methods, the proposed method could predict large-scale forest fires more accurately, and reduces the mean absolute error by 3–19.3 and root mean squared error by 0.95–19.3. The proposed method can better benefit the management of wildland fires in advance and the prevention of serious fire accidents. It is expected that the prediction performance could be further improved if additional information and more data are available.

Arc Grounding Fault Monitoring and Fire Prediction Method Based on EEMD and Reconstruction

To solve the problem of the single-phase ground fault and occurrence of electrical fires due to the residual current in substation AC power systems, a residual current intelligent sensing technology is proposed based on ensemble empirical modal decomposition (EEMD), sample entropy (SE) reconstruction, and fire warning technology using a beetle antennae search algorithm. First, through the residual current monitoring device to collect residual current information, EEMD and SE reconstruction for arc-earth fault diagnosis and an analysis of the differences in the current characteristics of each line after reconstruction are used to determine the fault line. Second, residual current, temperature, and operating voltage as input parameters and fire probability are the output parameters. The input–output relationship is established by a back-propagation neural network (BPNN) and optimized by the beetle antennae search (BAS) algorithm to speed up the convergence and improve the prediction accuracy to establish a substation fire warning scheme. Through simulation experiments, this paper proposes the residual current as a monitoring object method can effectively diagnose ground faults and accurately predict the probability of fire occurrence to ensure the safe and stable operation of substations.

Parameter Flexible Wildfire Prediction Using Machine Learning Techniques: Forward and Inverse Modelling

Parameter identification for wildfire forecasting models often relies on case-by-case tuning or posterior diagnosis/analysis, which can be computationally expensive due to the complexity of the forward prediction model. In this paper, we introduce an efficient parameter flexible fire prediction algorithm based on machine learning and reduced order modelling techniques. Using a training dataset generated by physics-based fire simulations, the method forecasts burned area at different time steps with a low computational cost. We then address the bottleneck of efficient parameter estimation by developing a novel inverse approach relying on data assimilation techniques (latent assimilation) in the reduced order space. The forward and the inverse modellings are tested on two recent large wildfire events in California. Satellite observations are used to validate the forward prediction approach and identify the model parameters. By combining these forward and inverse approaches, the system manages to integrate real-time observations for parameter adjustment, leading to more accurate future predictions.

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.

Mapping Susceptibility of Forest in Taman Raya Raden Soerjo Using Machine Learning Random Forest

The increasing incidence of forest fires in Raden Soerjo’s Grand Park over the last 5 years has become a concern for conducting research on fire predictions. The availability of free remote sensing data makes it easier to analyze forest fires using Geographic Information Systems (GIS). and remote sensing. Varying publicly available spatial datasets were used to classify forest fires affected areas with an approach of Machine Learning (ML) using Random Forest (RF). Nine (9) variables suspected as potential cause of forest fires (NDVI. ET. LST. Aspect. Slope. Altitude. Distance to Road (ED_Road). Distance to Built-up area (ED_BUA). and Distance to river (ED_River) were used as determinants. About 80% of the data were used as Trained data and 20% as a validation. The results of the model produced an accuracy of 0.96. The model result was checked for its sensitivity with the Area Under the Curve (AUC) AUC results with a value of 0.89. Findings show that Random Forest could be applied to map forest fire severity classes with a good result.

Dynamic Fire Risk Classification Prediction of Stadiums: Multi-Dimensional Machine Learning Analysis Based on Intelligent Perception

Stadium fires can easily cause massive casualties and property damage. The early risk prediction of stadiums will be able to reduce the incidence of fires by making corresponding fire safety management and decision making in an early and targeted manner. In the field of building fires, some studies apply data mining techniques and machine learning algorithms to the collected risk hazard data for fire risk prediction. However, most of these studies use all attributes in the dataset, which may degrade the performance of predictive models due to data redundancy. Furthermore, machine learning algorithms are numerous and applied to fewer stadium fires, and it is crucial to explore models suitable for predicting stadium fire risk. The purpose of this study was to identify salient features to build a model for predicting stadium fire risk predictions. In this study, we designed an index attribute threshold interval to classify and quantify different fire risk data. We then used Gradient Boosting-Recursive Feature Elimination (GB-RFE) and Pearson correlation analysis to perform efficient feature selection on risk feature attributes to find the most informative salient feature subsets. Two cross-validation strategies were employed to address the dataset imbalance problem. Using the smart stadium fire risk data set provided by the Wuhan Emergency Rescue Detachment, the optimal prediction model was obtained based on the identified significant features and six machine learning methods of 12 combination forms, and full features were input as an experimental comparison study. Five performance evaluation metrics were used to evaluate and compare the combined models. Results show that the best performing model had an F1 score of 81.9% and an accuracy of 93.2%. Meanwhile, by introducing a precision-recall curve to explain the actual classification performance of each model, AdaBoost achieves the highest Auprc score (0.78), followed by SVM (0.77), which reveals more stable performance under such imbalanced data.

Fine fuel changes due to timber harvesting and frequent prescribed burning in eucalypt forests of southeastern Australia

Forests are managed for a range of ecological and economic values. Prescribed burning and timber harvesting are two widely used and often co-occurring forest management activities. Both may alter wildfire risk by changing fuel characteristics. Prescribed burning alters the amount and arrangement of fuel, which changes the probability of ignition and rate of fire spread. Timber harvesting may influence future fuel composition, amount and structure directly via the addition of harvesting residue and indirectly by changing stand structure and fine fuel accumulation. Whilst there is reasonable understanding of how fuel changes immediately after a single prescribed burn or harvest in temperate eucalypt forests, the long-term effects of frequent prescribed burning combined with timber harvesting are poorly understood. We used a long-term (∼25 year) prescribed burning experiment conducted in ∼1000 ha of temperate eucalypt forest in southeastern New South Wales, Australia to determine the effects of a single selective timber harvest and 0–6 subsequent prescribed burns on the accumulation and composition of fine fuel (<25 mm thickness). It was found that fine fuel loads differed between harvested and unharvested stands, with 30 % less leaves and fine twigs (<6 mm) in the litter bed but 24 % more green understorey fuel up to 10 years post-burn in harvested stands. Frequent prescribed burning did not change total fine fuel load in harvested stands but did change fine fuel composition, with 4 times more leaves in the litter bed and 6 times less green understorey fuel in plots burnt at high frequencies (>4 fires in 25 years) versus low frequencies (0 fires in 25 years). Similar trends were observed in unharvested stands following frequent burning, however, there was 45% more leaves in the litter bed of frequently burnt, unharvested stands compared to frequently burnt, harvested stands. Our results suggest that irrespective of harvesting, frequent burning (>4 fires over 25 years) in temperate eucalypt forest may be a suitable strategy for lowering wildfire risk as it can reduce the amount of green understorey fuel, which may reduce flame heights and the likelihood of canopy fire. However, the decrease in green understorey fuel may be offset by the increase in leaves in the litter bed, which support fire ignition and spread. Our results highlight how the presence of timber harvesting changes the accumulation and composition of fine fuel post-fire, yet timber harvesting is currently not well integrated into many landscape fuel models, which often focus solely on time since last fire. Overlooking key attributes of disturbance history (timber harvesting, number of past fires), may jeopardise the accuracy of fuel models, and therefore the reliability of fire behaviour predictions.

Identifying anthropogenic and natural causes of wildfires by maximum entropy method-based ignition susceptibility distribution models

Turkey has a high potential for wildfires along its Mediterranean coast because of its dense forest cover and mild climate. An average of 250 wildfires occurs every year with more than 10,000 hectares destroyed due to natural and human-related causes. The study area is sensitive to fires caused by lightning, stubble burning, discarded cigarette butts, electric arcing from power lines, deliberate fire setting, and traffic accidents. However, 52% of causes could not be identified due to intense wildfires occurring at the same time and insufficient equipment and personnel. Since wildfires destroy forest cover, ecosystems, biodiversity, and habitats, they should be spatially evaluated by separating them according to their causes, considering environmental, climatic, topographic and forest structure variables that trigger wildfires. In this study, wildfires caused by lightning, the burning of agriculture stubble, discarded cigarette butts and power lines were investigated in the provinces of Aydın, Muğla and Antalya, where 22% of Turkey’s wildfires occurred. The MaxEnt method was used to determine the spatial distribution of wildfires to identify risk zones for each cause. Wildfires were used as the species distribution and the probability of their occurrence estimated. Additionally, since the causes of many wildfires are unknown, determining the causes is important for fire prediction and prevention. The highest wildfire occurrence risks were 9.7% for stubble burning, 30.2% for lightning, 4.5% for power lines and 16.9% by discarded cigarette butts. In total, 1,266 of the 1,714 unknown wildfire causes were identified by the analysis of the cause-based risk zones and these were updated by including cause-assigned unknown wildfire locations for verification. As a result, the Area under the ROC Curve (AUC) values were increased for susceptibility maps.

The Alberta Wildland Fuels Inventory Program (AWFIP): data description and reference tables

Key messageThis document describes a dataset obtained from a field sampling program conducted in Alberta, Canada. Field data were used to describe the structure and composition of forest stands, including several fuel loads (e.g., surface, understory, canopy fuels). The dataset can be downloaded from https://doi.org/10.17605/OSF.IO/FZ8E4 and metadata is available at https://metadata-afs.nancy.inra.fr/geonetwork/srv/fre/catalog.search#/metadata/527efb49-43b4-43eb-88b2-70535ff99fc5We present a quality-checked and curated dataset obtained from a field sampling program conducted in the province of Alberta, Canada. Field data were used to describe the structure and composition of forest stands documented in 476 sampling events. Each sampling event record consists of 42 different variables, including several fuel loads (e.g., surface, understory, canopy fuels). The dataset has been created for operational and research applications including but not limited to fuel classification, estimation of fuel attributes from remote sensing technologies, fuel treatment planning, fire behavior prediction, and use in high resolution fire growth models.

The fuel-climate-fire conundrum: How will fire regimes change in temperate eucalypt forests under climate change?

Fire regimes are changing across the globe in response to complex interactions between climate, fuel, and fire across space and time. Despite these complex interactions, research into predicting fire regime change is often unidimensional, typically focusing on direct relationships between fire activity and climate, increasing the chances of erroneous fire predictions that have ignored feedbacks with, for example, fuel loads and availability. Here, we quantify the direct and indirect role of climate on fire regime change in eucalypt dominated landscapes using a novel simulation approach that uses a landscape fire modelling framework to simulate fire regimes over decades to centuries. We estimated the relative roles of climate‐mediated changes as both direct effects on fire weather and indirect effects on fuel load and structure in a full factorial simulation experiment (present and future weather, present and future fuel) that included six climate ensemble members. We applied this simulation framework to predict changes in fire regimes across six temperate forested landscapes in south‐eastern Australia that encompass a broad continuum from climate‐limited to fuel‐limited. Climate‐mediated change in weather and fuel was predicted to intensify fire regimes in all six landscapes by increasing wildfire extent and intensity and decreasing fire interval, potentially led by an earlier start to the fire season. Future weather was the dominant factor influencing changes in all the tested fire regime attributes: area burnt, area burnt at high intensity, fire interval, high‐intensity fire interval, and season midpoint. However, effects of future fuel acted synergistically or antagonistically with future weather depending on the landscape and the fire regime attribute. Our results suggest that fire regimes are likely to shift across temperate ecosystems in south‐eastern Australia in coming decades, particularly in climate‐limited systems where there is the potential for a greater availability of fuels to burn through increased aridity.

Research on Multi-Sensor Fusion Indoor Fire Perception Algorithm Based on Improved TCN.

Indoor fires cause huge casualties and economic losses worldwide. Thus, it is critical to quickly and accurately perceive the fire. In this work, an indoor fire perception algorithm based on multi-sensor fusion was proposed. Firstly, the sensor data features were fully extracted by improved temporal convolutional network (TCN). Then, the dimension of the extracted features was reduced by adaptive average pooling (AAP). Finally, the fire classification was realized by the support vector machine (SVM) classifier. Experimental results demonstrated that the proposed algorithm can improve accuracy of fire classification by more than 2.5% and detection speed by more than 15%, compared with TCN, back propagation (BP) neural network and long short-term memory (LSTM). In conclusion, the proposed algorithm can perceive the fire quickly and accurately, which is of great significance to improve the performance of the current fire prediction systems.

Fire probability mapping and prediction from environmental data: What a comprehensive savanna-forest transition can tell us

The Cerrado-Amazon transition (CAT) is located between the two largest biogeographic domains of the Neotropical region, coinciding with the “Arc of Deforestation”, an agricultural frontier that expands towards the Amazon. Fire plays an important role in the landscape changes that have been occurring in this transition. Thus, the objectives of this study were (i) evaluate the accuracy of different algorithms/techniques in predicting fires in the CAT; (ii) investigate the season of highest fire probabilities in the CAT; and (iii) identify, among anthropic, climatic, topographic and soil variables, the main drivers of fires spatial distribution in the CAT. Fire occurrence data (active fire pixels) from the MODIS sensor were used as input data in species distribution models. The SDM allowed the construction of monthly fire probability models for the study area. The model's quality was assessed by R-squared, RMSE, AUC and TSS metrics. In addition, we assessed the weight of variables in building the models. To obtain the fire seasonality, we used the bootstrap technique to calculate the 95% confidence interval (CI) about the mean. As result, we found that all algorithms obtained satisfactory performance, as well as the ensemble model technique used, which obtained an average AUC of 0.812 ± 0.001. The algorithm with the best individual performance was Random Forest, with a mean AUC 0.851 ± 0.002. Similar patterns were found when considering TSS metrics. The predicted and observed fire occurrences showed strong association (R2 = 0.81). The high fire probability area was higher during the dry season (15304.37 ± 41.65), showing a significant difference (p-value ≤ 0.05) from the rainy season. The variables that most explained the fire occurrences along the CAT were 'Silt Fraction', 'Distance from Agriculture' and 'Maximum Temperature'. In addition, ‘Distance from Roads’, ‘Distance from Protected Areas’ and Elevation’ were relevant for the monthly models. Anthropogenic variables were more important during the dry season. The results demonstrate that SDMs can be adequately used in fire prediction. Protected areas have shown an important role in acting as a fire barrier. However, many of these areas are vulnerable, surrounded by high fire probabilities during the dry season, predominantly anthropic, putting the natural dynamics of fire in the region and the local biodiversity at risk. Our approach can be useful in environmental protection and monitoring, in addition to facilitating the understanding about the CAT ecological complexity and the high anthropogenic pressure happening there.

Planning Satellite Swarm Measurements for Earth Science Models: Comparing Constraint Processing and MILP Methods

We compare two planner solutions for a challenging Earth science application to plan coordinated measurements (observations) for a constellation of satellites. This problem is combinatorially explosive, involving many degrees of freedom for planner choices. Each satellite carries two different sensors and is maneuverable to 61 pointing angle options. The sensors collect data to update the predictions made by a high-fidelity global soil moisture prediction model. Soil moisture is an important geophysical variable whose knowledge is used in applications such as crop health monitoring and predictions of floods, droughts, and fires. The global soil-moisture model produces soil-moisture predictions with associated prediction errors over the globe represented by a grid of 1.67 million Ground Positions (GPs). The prediction error varies over space and time and can change drastically with events like rain/fire. The planner's goal is to select measurements which reduce prediction errors to improve future predictions. This is done by targeting high-quality observations at locations of high prediction-error. Observations can be made in multiple ways, such as by using one or more instruments or different pointing angles; the planner seeks to select the way with the least measurement-error (higher observation quality). In this paper we compare two planning approaches to this problem: Dynamic Constraint Processing (DCP) and Mixed Integer Linear Programming (MILP). We match inputs and metrics for both DCP and MILP algorithms to enable a direct apples-to-apples comparison. DCP uses domain heuristics to find solutions within a reasonable time for our application but cannot be proven optimal, while the MILP produces provably optimal solutions. We demonstrate and discuss the trades between DCP flexibility and performance vs. MILP's promise of provable optimality.