Forest steppe fire risk assessment method and system

Abstract

The invention discloses a forest steppe fire risk assessment method and system, and belongs to the technical field of forest steppe fire prevention. The problem of high false alarm and missing report rate caused by lagging fire risk identification, weak multi-source data fusion and insufficient dynamic response in the prior art is solved. According to the technical scheme, the method comprises the following steps: deploying a ground sensing node array to obtain real-time environment data of a grassland region, fusing high-resolution satellite remote sensing and regional weather forecast data, and constructing a space-time aligned risk assessment data cube; establishing a grassland fire risk factor dynamic coupling model, and dynamically allocating factor weights in combination with the adaptive weight decision tree; generating a comprehensive risk index, carrying out nonlinear mapping to five risk levels, and linking a visual engine to generate a dynamic thermodynamic diagram and push the dynamic thermodynamic diagram to a command terminal; the system supports online updating of the model, and weight parameters are automatically optimized based on a new fire event. According to the invention, high-precision, real-time and spatialized evaluation is realized, the early warning capability and prevention and control decision efficiency are improved, and ecological and economic losses are reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of forest and grassland fire prevention, and particularly relates to a forest and grassland fire risk assessment method and system. BACKGROUND

[0002] With the intensification of global climate change and the expansion of human activities, forest and grassland fires have become an important natural disaster threatening ecological security, economic property and people's lives. The current mainstream risk assessment system mainly relies on the statistics of historical fire point data and the superposition analysis of static environmental factors, and its core logic is based on the macro mapping relationship between regional average meteorological conditions and vegetation coverage. However, the grassland ecosystem has high spatial heterogeneity and dynamic response characteristics: the load of surface combustible materials fluctuates sharply with the change of seasons, and the local microclimate shows nonlinear conduction under the influence of terrain shielding and wind field disturbance. However, the traditional method lacks the ability to model the spatio-temporal coupling of multi-source heterogeneous data, and it is difficult to capture the collaborative evolution law between fire risk factors, resulting in lagging identification of high-risk areas, rigid early warning thresholds, and high false alarm and missed alarm rates. Especially in the complex terrain interlaced zone or the area with frequent human disturbance, the existing model often ignores the instantaneous coupling effect of the thermodynamic conduction path and the continuity of fuel, which leads to the deviation of risk level assessment from the actual burning potential and misses the best prevention and control window period.

[0003] Among them, the dynamic monitoring architecture based on the fusion of remote sensing images and ground sensors has been initially established, but its risk quantification process still stays at the feature weighting and summing level, and it has not established a nonlinear response function driven by physical mechanism. A typical scheme generates a risk score by linearly combining normalized vegetation index NDVI and land surface temperature LST, and the formula is R = a·NDVI + b·LST + g. Such a model ignores the influence of water stress gradient on the critical ignition energy of combustible materials and the exponential amplification effect of wind speed turbulence on the flame spread rate, and it is prone to systematic underestimation under extremely dry or strong wind conditions. At the same time, the evaluation unit generally adopts fixed grid division, which cannot adaptively match the topological connectivity of fire behavior propagation, resulting in fuzzy risk boundary and ineffective prediction of hot spot diffusion path.

[0004] The prior art has three structural defects: firstly, the risk factor weight configuration relies on expert experience setting and lacks a parameter inversion mechanism for combustion kinetics equations; secondly, the evaluation model does not embed a spatiotemporal attention mechanism to capture cross-scale factor interactions, resulting in local sudden fire risk signals being covered up by global mean smoothing; and thirdly, the output result does not form a closed-loop feedback with the emergency resource scheduling module, and the risk priority cannot be dynamically corrected according to the real-time rescue force distribution. The above defects are particularly prominent during the key spring fire prevention period. When multiple high-risk patches concurrently catch fire within a short time, the traditional evaluation system often causes rescue force mismatching, isolation belt layout failure and other chain losses due to response delay and insufficient accuracy, and a new generation of fire risk assessment method with physical interpretability, spatiotemporal adaptability and decision linkage is urgently needed. SUMMARY

[0005] The purpose of the present application is to provide a forest and grassland fire risk assessment method and system, mainly solving the problems of low grassland fire hazard identification accuracy, poor early warning timeliness, high false alarm and missed alarm rates in the prior art, which are difficult to support accurate prevention and rapid response.

[0006] To achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A forest and grassland fire risk assessment method, comprising the following steps:

[0008] S1, deploying a distributed ground sensor node array to obtain real-time environmental data of the grassland area;

[0009] S2, synchronously accessing high-resolution satellite remote sensing image data to extract vegetation coverage feature information;

[0010] S3, accessing regional weather forecast data to obtain weather information in the future period;

[0011] S4, performing spatiotemporal alignment and normalization preprocessing on the real-time environmental data, vegetation coverage feature information and weather information in the future period, and constructing a grassland fire risk assessment data cube under a unified spatiotemporal reference;

[0012] S5, constructing a grassland fire risk factor dynamic coupling model, and obtaining a grassland fire comprehensive risk index according to the output value of the grassland fire risk factor dynamic coupling model;

[0013] S6, converting the grassland fire comprehensive risk index into a risk level identifier through a nonlinear mapping function, the risk level identifier being divided into five levels and corresponding to a preset prevention and control response plan;

[0014] S7, constructing a risk assessment result visualization engine, spatially rendering the grassland fire comprehensive risk index according to geographic grid units, and generating a dynamic risk heat map;

[0015] S8, setting a risk assessment model online updating mechanism, when receiving a new fire verification event data, automatically triggering a model parameter retraining process.

[0016] Further, in the present application, the real-time environmental data includes real-time ground surface temperature, soil moisture, vegetation water content, wind speed, wind direction and atmospheric pressure data; the vegetation coverage feature information includes vegetation coverage, surface albedo, thermal anomaly point distribution and historical fire point trajectory information; the future period weather information includes future 24-hour precipitation probability, air temperature change trend and relative humidity prediction value.

[0017] Further, in the present application, in the step S5, the grassland fire risk factor dynamic coupling model comprises:

[0018] a ground surface environmental factor sub-model, which is based on real-time ground surface temperature T s and soil moisture M s , calculates a ground surface dryness index I sd ; the expression is:

[0019]

[0020] wherein, k e is a preset environmental coefficient;

[0021] a meteorological driving factor sub-model, which is based on wind speed v, wind direction and atmospheric pressure data, calculates a fire spread potential index I sp ; the expression is:

[0022] I sp = v 2 × c w × f p ;

[0023] wherein, f p is an atmospheric pressure correction factor; c w is a wind direction stability coefficient, the value range is: 0-1;

[0024] a vegetation state factor sub-model, which is based on vegetation water content W v and vegetation coverage C v , calculates a combustible load index I cl ; the expression is:

[0025]

[0026] wherein, k t is a vegetation type correction coefficient;

[0027] A historical fire factor sub-model, which is based on historical fire point trajectory and thermal anomaly point distribution data, calculates a regional fire recurrence tendency index I fr ;

[0028] I fr =D h ×W t ;

[0029] wherein D h is the historical fire point density per unit area, and W t is the thermal anomaly point duration weighted value.

[0030] Further, in the present application, in the step S5, the process of obtaining the grassland fire comprehensive risk index is as follows:

[0031] S51, an adaptive weight allocation mechanism is introduced: a weight decision tree is constructed, the input nodes of the decision tree include the current month, the altitude interval, the dominant wind direction type and the precipitation probability level, and the output node is the normalized weight coefficient of the four risk factor sub-models;

[0032] The branch rules of the weight decision tree are obtained by training based on the historical fire case library, and the gradient boosting decision tree algorithm is used to optimize the split threshold and leaf node weight value in the training process;

[0033] S52, the weight proportion of each risk factor sub-model in the total risk assessment is dynamically adjusted according to the current season, regional geographical characteristics and real-time meteorological conditions;

[0034] S53, the output values of the four risk factor sub-models are multiplied by their corresponding normalized weight coefficients respectively, and then weighted summation is performed to obtain the grassland fire comprehensive risk index.

[0035] Further, in the present application, in the step S6, the nonlinear mapping function is realized in a piecewise linear interpolation manner, and the interpolation nodes are calibrated according to historical fire loss statistical data.

[0036] Further, in the present application, in the step S7, the specific process of constructing the risk assessment result visualization engine is as follows:

[0037] The grassland fire comprehensive risk index is spatially rendered according to the geographical grid unit to generate a dynamic risk heat map;

[0038] The heat map is superimposed to display the current wind direction arrow, the historical fire point mark and the high-risk area boundary contour;

[0039] Supporting the historical risk evolution process according to the time axis, and supporting the risk statistical analysis according to the administrative region, the vegetation type or the altitude gradient;

[0040] The risk thermal map is pushed to the terminal of the fire prevention command center at each level and the mobile terminal of the field patrol personnel in real time through a special communication link.

[0041] Further, in the present application, the risk assessment model is updated online, and the online updating mechanism comprises:

[0042] When receiving new fire verification event data, a model parameter retraining process is automatically triggered, wherein the retraining process comprises extracting input data of each risk factor sub-model within twenty-four hours before the occurrence of the verification event, calculating the contribution degree of each factor at the occurrence of the event, and adjusting the weight distribution coefficient of the corresponding branch in the weight decision tree.

[0043] The present application also provides a forest and grassland fire risk assessment system for forest and grassland fire risk assessment, comprising:

[0044] A distributed ground sensing node array module is configured to acquire real-time ground surface temperature, soil moisture, vegetation water content, wind speed, wind direction and atmospheric pressure data of the grassland area.

[0045] A satellite remote sensing data access module is configured to synchronously access high-resolution satellite remote sensing image data, and extract vegetation coverage, ground albedo, thermal anomaly point distribution and historical fire point trajectory information.

[0046] A meteorological forecast data access module is configured to access regional meteorological forecast data, and acquire precipitation probability, temperature change trend and relative humidity prediction value within the next twenty-four hours.

[0047] A data preprocessing module is configured to perform spatio-temporal alignment and normalization preprocessing on the multi-source heterogeneous data, and construct a grassland fire risk assessment data cube under a unified spatio-temporal reference.

[0048] A risk factor dynamic coupling modeling module is configured to construct a grassland fire risk factor dynamic coupling model comprising a ground environmental factor sub-model, a meteorological driving factor sub-model, a vegetation state factor sub-model and a historical fire factor sub-model.

[0049] An adaptive weight distribution module is configured to dynamically adjust the weight proportion of each risk factor sub-model in the total risk assessment according to the current season, regional geographical features and real-time meteorological conditions.

[0050] A comprehensive risk index calculation module is configured to multiply the output values of the four risk factor sub-models by their corresponding normalized weight coefficients, and then perform weighted summation to obtain a grassland fire comprehensive risk index.

[0051] A risk level mapping module is configured to convert the grassland fire comprehensive risk index into a risk level identifier through a nonlinear mapping function.

[0052] A visualization engine module is configured to spatially render the grassland fire comprehensive risk index by geographical grid unit to generate a dynamic risk heat map.

[0053] A model online updating module is configured to automatically trigger a model parameter retraining process when receiving new fire verification event data.

[0054] Further, in the present application, the distributed ground sensing node array is composed of a solar power supply module, a low-power microcontroller, a wireless communication module and a multi-parameter environmental sensor.

[0055] The multi-parameter environmental sensor includes an infrared ground surface temperature sensor, a capacitive soil moisture sensor, a near-infrared vegetation water content sensor, an ultrasonic wind speed and direction sensor, and a piezoresistive atmospheric pressure sensor.

[0056] Compared with the prior art, the present application has the following beneficial effects:

[0057] (1) The present application solves the problems of weak multi-source heterogeneous data fusion and poor spatio-temporal alignment in the traditional method by deploying a distributed ground sensing node array, synchronously accessing high-resolution satellite remote sensing data and regional weather forecast data, and constructing a "risk assessment data cube" under a unified spatio-temporal reference. Combined with a grassland fire risk factor dynamic coupling model and an adaptive weight decision tree, the present application realizes dynamic allocation of risk factor weights (such as adjusting factor contribution degree based on season, terrain and weather conditions), and breaks through the limitations of static weight relying on expert experience. Through nonlinear mapping, the comprehensive risk index is converted into a five-level risk grade, which can better capture the synergistic effect (such as the coupling amplification effect of high temperature, low humidity and strong wind) between fire risk factors than the traditional linear weighted model (such as R = a·NDVI + b·LST + g), so as to improve the risk assessment accuracy to more than 89% and reduce the false alarm rate to less than 7%, and realize the dynamic assessment goal of "high accuracy, real-time and spatialization".

[0058] (2) The present application solves the problems of "fuzzy boundary" and "ineffective prediction of hot spot diffusion path" of the traditional assessment result by using a risk assessment result visualization engine to render the comprehensive risk index into a dynamic heat map by geographical grid unit, superimpose key information such as wind direction arrow and historical fire point trajectory, and push it to the command terminal in real time. The five-level risk grade corresponds to a preset prevention and control response plan, realizing the rapid linkage of "risk grade-response measure", for example, automatically triggering the "isolation belt preset + unmanned aerial vehicle patrol" plan in the high risk grade. Pilot application shows that this mechanism makes the average early warning time more than 4 hours, improves the emergency response efficiency by 60%, effectively avoids the problem of "missing the best prevention and control window period", and reduces the ecological and economic losses.

[0059] (3) The application sets an online updating mechanism of the risk assessment model. When receiving new fire event data, the parameter retraining process (such as extracting factor data 24 hours before the fire, adjusting the weight decision tree branch coefficient) is automatically triggered, solving the defects of the traditional model, such as "parameter solidification and difficulty in adapting to environmental changes". Through the closed-loop feedback of the historical fire case library and real-time data, the model can dynamically optimize the coupling coefficient of the surface dryness index, the fire spread potential index and other sub-models, for example, automatically increasing the weight of the wind speed factor in the strong wind area. This mechanism reduces the evaluation deviation rate of the model in the complex terrain interlaced zone and the area with frequent human interference by 40%, maintains high prediction ability for a long time, and adapts to climate change and ecosystem evolution. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 The figure is a schematic diagram of the method of the application.

[0061] Figure 2 The figure is a principle block diagram of the system of the application. DETAILED DESCRIPTION

[0062] The application will be further described below in combination with the accompanying drawings and examples. The mode of the application includes but is not limited to the following examples.

[0063] Embodiment

[0064] As shown in Figure 1 , Figure 2 , the application discloses a forest and grassland fire risk assessment method. The core of the method is to build a risk assessment system that can dynamically perceive environmental changes, adaptively adjust the evaluation weight, and realize deep fusion of multi-source heterogeneous data. The method breaks through the limitations of traditional static threshold models and single factor evaluation, realizes fine, intelligent and forward-looking assessment of grassland fire risk by introducing an adaptive weight allocation mechanism and a dynamic coupling model of risk factors. In practical application, the occurrence of grassland fire often has the characteristics of strong suddenness, fast spreading speed, great difficulty in extinguishing, and serious economic loss.

[0065] As shown in Figure 1The shown. First, through the deployment of distributed sensing network in grassland area, real-time collection of multiple source environmental data directly related to fire risk. Environmental data at least includes surface temperature data, air humidity data, wind speed and direction data, vegetation water content data, historical fire point distribution data and human activity heat map data. Among them, the surface temperature data is obtained by infrared thermal imaging sensor array, the sampling frequency is not less than once per minute, and the spatial resolution reaches ten meters; air humidity data is collected by high-precision capacitive humidity sensor, the installation height is uniform for two meters from the ground, in order to eliminate the interference of surface evaporation; wind speed and direction data is obtained by ultrasonic anemometer, which has zero start-up wind speed and omnidirectional measurement capability; vegetation water content data is calculated by near infrared spectral reflectance inversion model, which is inputted into multispectral vegetation index and field sampling calibration value.

[0066] Historical fire point distribution data is derived from satellite remote sensing fire point product and local fire department historical record database; human activity heat map data is generated by mobile communication base station signaling data aggregation, reflecting the population density and activity track in a specific period. All raw data is immediately time stamped and spatial coordinate normalized after collection, to ensure the spatio-temporal consistency of subsequent fusion analysis. For example, in the application of Xinjiang Yili River Valley grassland, the spatial resolution of remote sensing image reaches 1 meter, which can clearly identify the change of local vegetation coverage, and mark the historical fire point trajectory. In addition, the distribution of thermal anomaly points is detected by the change of radiation intensity in short wave infrared band, which further improves the accuracy of fire risk identification.

[0067] Access to regional weather forecast data is completed by weather forecast data access module. Future weather information includes precipitation probability, temperature change trend and relative humidity prediction value in the next twenty-four hours. Meteorological data is obtained from regional meteorological bureau through standard communication interface, and after format conversion, it is aligned with real-time environmental data in unified space-time reference. For example, in the key period of spring fire prevention, the update frequency of meteorological data is set to once an hour, to ensure the accurate capture of future weather information. The fusion of meteorological data and ground sensing data improves the response ability of evaluation model to environmental change.

[0068] After the multi-source environmental data collection is completed, the risk factor feature extraction and standardization phase is entered. The core task of this phase is to convert the original observation values into risk factor indicators with physical meaning and comparability. For surface temperature data, its daily maximum, daily average and instantaneous gradient change rate are extracted, and are respectively mapped to the preset temperature risk interval; for air humidity data, the ratio of its saturation water vapor pressure is calculated to obtain the relative humidity risk coefficient; for wind speed and direction data, not only the instantaneous wind speed value is recorded, but also the standard deviation of the wind speed in the past 30 minutes is calculated to represent the wind field stability, and the wind direction and the spatial relationship of the vegetation distribution are combined to generate the wind-assisted fire spread potential index. For vegetation water content data, it is compared with the local vegetation type benchmark water content to calculate the water deficit percentage. For historical fire point distribution data, the kernel density estimation method is used to generate a historical fire occurrence probability heat map, and the current vegetation coverage type is superimposed for weighted correction. For human activity heat map data, the number of active people per unit area and its trend are calculated as a human-induced fire risk factor. All extracted risk factor indicators are normalized by the Z-score standardization method to eliminate dimensional differences and make the numerical distribution concentrated in the -3-3 interval, which is convenient for subsequent weight allocation and coupling calculation.

[0069] The risk factor dynamic coupling modeling module constructs a grassland fire risk factor dynamic coupling model to quantify the nonlinear interaction effects between factors. The model includes a surface environmental factor sub-model, a meteorological driving factor sub-model, a vegetation state factor sub-model, and a historical fire factor sub-model. Among them, the surface environmental factor sub-model calculates the land surface dryness index based on real-time land surface temperature and soil moisture data, and its expression is:

[0070]

[0071] where T is the real-time land surface temperature, M is the soil moisture, α and β are preset environmental coefficients, and k is a preset environmental coefficient. s s e

[0072] The meteorological driving factor sub-model calculates the fire spread potential index based on wind speed, wind direction and atmospheric pressure data, and its expression is:

[0073]

[0074] where v is the wind speed, f is the atmospheric pressure correction factor; c is the wind direction stability coefficient, with a value range of 0-1. p w

[0075] The vegetation state factor sub-model calculates the combustible load index based on vegetation water content and vegetation coverage, and its expression is:​​​​​

[0076]

[0077] wherein, W v is the vegetation water content, C v is the vegetation coverage, k t is the vegetation type correction coefficient.

[0078] The historical fire factor sub-model calculates the regional fire recurrence tendency index based on the historical fire point trajectory and hot anomaly point distribution data, and its expression is:

[0079] I fr = D h × W t ;

[0080] wherein, D h is the historical fire point density per unit area, and W t is the hot anomaly point duration weighted value.

[0081] After the risk factor standardization is completed, the adaptive weight allocation module enters the execution phase of the adaptive weight allocation mechanism. The design purpose of this mechanism is to automatically adjust the contribution weight of each risk factor in the final evaluation result according to the dynamic changes of the current environmental situation, avoiding the failure problem of fixed weight model in extreme weather or special scene. The weight allocation process is dynamically calculated based on the correlation between the factor fluctuation intensity and the historical fire event within the sliding time window. Specifically, the system maintains a rolling observation window with a length of 7 days, and updates the variance and skewness values of each risk factor daily as the measurement indicators of its fluctuation intensity. At the same time, the system has a built-in historical fire event library, which records the numerical sequence of each risk factor within 3 days before each fire occurs. By calculating the Pearson correlation coefficient between the current factor sequence and the historical fire factor sequence, the historical prediction ability score of the factor is obtained. Finally, the dynamic weight of each risk factor is determined by the weighted average of its fluctuation intensity score and historical prediction ability score, and the weight coefficient is automatically updated once a day at zero o'clock in the morning and remains constant throughout the day to ensure the stability of the evaluation result. For example, when the standard deviation of wind speed is observed to increase significantly for 3 consecutive days and the correlation coefficient with historical fire events exceeds zero point seven, the weight of wind speed factor will automatically increase to more than 30% of the total weight, thereby giving it higher decision-making influence in the evaluation model.

[0082] After obtaining the dynamic weights of each risk factor, the operation phase of the risk factor dynamic coupling model is entered. The core idea of this model is to break the traditional linear superposition evaluation mode and introduce nonlinear interaction terms to capture the synergistic amplification effect or inhibition effect between factors. For example, the combination of high temperature and low humidity has a much greater impact on fire risk than the sum of their individual effects. High wind speed can significantly accelerate the spread of fire when the vegetation water content is very low, but its impact is relatively limited when the vegetation water content is higher. To quantify such coupling effects, the model uses a tensor product kernel function to construct a factor interaction matrix. Specifically, let X be the standardized risk factor vector, with a dimension equal to the total number of factors. A symmetric interaction coefficient matrix W is constructed, with elements W ij representing the coupling strength between the i-th factor and the j-th factor. The initial value of this matrix is obtained through historical data training and is fine-tuned at each weight update. The final risk score R is calculated by the following formula:

[0083]

[0084] The first term is the linear contribution of each factor, and the second term is the nonlinear coupling contribution between factors. This formula ensures that the evaluation results can reflect the independent influence of single factors and capture the risk mutations under the synergistic action of multiple factors. The calculated risk score R is then mapped to the preset risk level interval: when R <-1.5, it is determined as low risk; when -1.5 < R < 0.5, it is determined as low-medium risk; when 0.5 < R < 1.8, it is determined as medium-high risk; when R > 1.8, it is determined as high risk. Each risk level corresponds to a set of pre-set warning response plans, including patrol frequency adjustment, fire prevention buffer zone inspection, emergency supplies prepositioning, public warning information release, and other specific measures.

[0085] To ensure the reliability and traceability of the evaluation results, the system automatically generates an evaluation log after each evaluation is completed. The log content includes the evaluation timestamp, the original data collection value, the standardized factor value, the dynamic weight allocation result, the coupling matrix parameters, the final risk score, the determined risk level, and the triggered response plan number. All log entries are stored in a tamper-proof database in chronological order and support multi-dimensional retrieval by region, time period, risk level, etc. At the same time, the system provides a visual interface to superimpose the evaluation results on the geographic information system base map in the form of a heat map, with different color blocks representing different risk levels, supporting zooming and clicking to view detailed evaluation parameters. Management personnel can intuitively grasp the global risk distribution situation through this interface and allocate resources and make command decisions based on the system recommended plans.

[0086] In terms of handling abnormal data, the system incorporates multiple verification and fault-tolerance mechanisms. When a sensor's data exceeds the physically reasonable range for three consecutive samplings, the system automatically marks the data as abnormal and initiates a data interpolation procedure. The interpolation strategy prioritizes using spatial interpolation results from adjacent sensors; if adjacent sensor data is also unavailable, the historical average value for the same period at that location is used as a substitute, and the interpolation source is noted in the log. For data loss due to communication interruptions, the system allows a maximum delay of two hours for retransmission. Upon arrival of the retransmitted data, a reassessment is immediately triggered, and the risk level for the affected time period is updated. If data loss exceeds two hours, the system enters a conservative assessment mode, automatically raising the risk level of the relevant area by one level until the data is recovered, ensuring safety redundancy.

[0087] Regarding model self-learning and optimization, the system performs parameter backtracking calibration quarterly. The calibration process uses actual fire events from the past three months as positive samples and high-risk assessment periods without fires as negative samples to retrain the interaction coefficient matrix W and the correlation threshold in the weight calculation model. Training employs a gradient descent optimization algorithm, with the objective function being the cross-entropy loss between the assessment results and real events. After calibration, the new parameters are automatically deployed to the production environment, while the old parameter version is retained for rollback. Furthermore, the system supports a manual intervention interface, allowing fire prevention experts to manually adjust the weight caps or coupling coefficients of specific factors based on their field survey experience. The adjustments are recorded and incorporated into the next round of automatic calibration, achieving human-machine collaborative knowledge evolution.

[0088] At the system deployment and operation level, this method can be implemented based on existing grassland fire monitoring infrastructure. Sensor network nodes can reuse existing weather stations, lookout towers, and drone patrol platforms, reducing hardware deployment costs. Data transmission employs dual-link backup via low-power wide-area networks and satellite communication to ensure data accessibility in remote, network-free areas. Edge computing nodes are deployed at county-level fire command centers, responsible for local data preprocessing and preliminary assessment; provincial data centers are responsible for full-domain data fusion, model training, and parameter distribution. Assessment results are pushed to fire command centers at all levels via the government intranet and synchronized to mobile emergency command apps, achieving second-level access to risk information. The entire system supports 24 / 7 uninterrupted operation, with an average mean time between failures (MTBF) exceeding 5000 hours, and assessment latency controlled within 5 minutes of data acquisition, meeting the timeliness requirements for forest and grassland fire prevention and control.

[0089] In practical application scenarios, the application is applied in typical regions such as the Inner Mongolia Hulunbuir grassland and the Xinjiang Yili river valley grassland. The pilot results show that the application improves the fire warning accuracy to 89%, reduces the false positive rate to less than 7%, and the average warning time is more than 4 hours, which is significantly better than the traditional evaluation method. At the same time, due to the strong correlation between the evaluation results and the response plan, the utilization rate of prevention and control resources is increased by 40%, the emergency response efficiency is improved by 60%, and the ecological and economic losses caused by grassland fires are effectively reduced. For example, in a certain pilot, the system issued a high-risk warning 4 hours in advance and suggested setting a fireproof isolation belt 3 kilometers in the southeast direction, which successfully controlled the fire in a local area and avoided larger-scale ecological damage.

[0090] In summary, the application provides a forest and grassland fire risk assessment method, which realizes real-time, accurate and intelligent assessment of grassland fire risk by constructing an adaptive weight distribution mechanism and a risk factor dynamic coupling model. The method not only has a complete technical scheme, rigorous logic and strong operability, but also shows significant performance advantages and social benefits in practical application, providing strong technical support for China's grassland ecological safety and fire prevention and disaster reduction work.

[0091] The above embodiment is only one of the preferred embodiments of the application and should not be used to limit the protection scope of the application. Any modification or polishing made within the main design idea and spirit of the application without substantial meaning, which still solves the technical problems consistent with the application, should be included in the protection scope of the application.

Claims

1. A method of forest steppe fire risk assessment, characterized in that, The method comprises the following steps: S1, deploying a distributed ground sensor node array to obtain real-time environmental data of the grassland area; S2, synchronously accessing high-resolution satellite remote sensing image data to extract vegetation coverage feature information; S3, accessing regional weather forecast data to obtain weather information in a future period; S4, performing spatio-temporal alignment and normalization preprocessing on the real-time environmental data, the vegetation coverage feature information, and the weather information in the future period to construct a grassland fire risk assessment data cube under a unified spatio-temporal reference; S5, constructing a grassland fire risk factor dynamic coupling model, and obtaining a grassland fire comprehensive risk index according to an output value of the grassland fire risk factor dynamic coupling model; S6, converting the grassland fire comprehensive risk index into a risk level identifier through a nonlinear mapping function, wherein the risk level identifier is divided into five levels and corresponds to a preset prevention and control response plan; S7, constructing a risk assessment result visualization engine, spatially rendering the grassland fire comprehensive risk index according to geographical grid units to generate a dynamic risk heat map; S8, setting an online updating mechanism of the risk assessment model, and automatically triggering a model parameter retraining process when receiving new fire verification event data.

2. The forest steppe fire risk assessment method according to claim 1, characterized in that, The real-time environmental data comprises real-time ground surface temperature, soil moisture, vegetation water content, wind speed, wind direction, and atmospheric pressure data; the vegetation coverage feature information comprises vegetation coverage, ground albedo, thermal anomaly point distribution, and historical fire point trajectory information; and the weather information in the future period comprises precipitation probability, temperature change trend, and relative humidity prediction value in the next 24 hours.

3. The method of claim 2, wherein, In the step S5, the grassland fire risk factor dynamic coupling model comprises: a surface environmental factor sub-model, which is based on real-time surface temperature T s and soil moisture M s to calculate a surface dryness index I sd , whose expression is: wherein k e is a predetermined environmental coefficient; a weather driving factor sub-model, which calculates a fire spread potential index I based on wind speed v, wind direction and atmospheric pressure data sp ; the expression of which is: I sp = v 2 x c w x f p ; Wherein, f p is the atmospheric pressure correction factor; c w is the wind direction stability coefficient, the value range: 0~1; a vegetation state factor sub-model, which is based on the water content of the vegetation W v and the vegetation cover C v , to calculate the fuel load index I cl ; the expression of which is: wherein k t is a vegetation type correction factor; a historical fire factor sub-model, which is based on historical fire point trajectory and hot anomaly point distribution data to calculate a regional fire recurrence tendency index I fr ; I fr = D h x W t ; Wherein, D h is the historical fire point density per unit area, W t is the thermal anomaly point duration weighting value.

4. The method of claim 3, wherein, In the step S5, the process of obtaining the grassland fire comprehensive risk index is as follows: S51, introducing an adaptive weight allocation mechanism: constructing a weight decision tree, wherein input nodes of the decision tree comprise a current month, an altitude interval, a dominant wind direction type, and a precipitation probability level, and an output node is a normalized weight coefficient of four risk factor sub-models; The branch rule of the weight decision tree is obtained based on training of a historical fire case library, and a gradient boosting decision tree algorithm is used to optimize a split threshold and a leaf node weight value in the training process; S52, dynamically adjusting a weight proportion of each risk factor sub-model in the total risk assessment according to a current season, regional geographical features, and real-time weather conditions; S53, multiplying output values of the four risk factor sub-models by corresponding normalized weight coefficients respectively, and then performing weighted summation to obtain the grassland fire comprehensive risk index.

5. The method of claim 4, wherein, In the step S6, the nonlinear mapping function is realized in a piecewise linear interpolation manner, and interpolation nodes are calibrated according to historical fire loss statistical data.

6. The method of claim 5, wherein, In the step S7, the specific process of constructing the risk assessment result visualization engine is as follows: spatially rendering the grassland fire comprehensive risk index according to geographical grid units to generate a dynamic risk heat map; the heat map is superimposed with a current wind direction arrow, historical fire point markers, and a high-risk area boundary contour; supporting backtracking of a historical risk evolution process according to a time axis, and supporting risk statistical analysis according to an administrative region, a vegetation type, or an altitude gradient; The risk heat map is pushed to the terminal of the fire command center at all levels and the mobile terminal of the field patrol personnel in real time through a special communication link.

7. The method of claim 6, wherein, The risk assessment model online updating mechanism is: When receiving new fire verification event data, a model parameter retraining process is automatically triggered; the retraining process includes extracting input data of each risk factor submodel within 24 hours before the occurrence of the verification event, calculating the contribution degree of each factor at the occurrence of the event, and adjusting the weight distribution coefficient of the corresponding branch in the weight decision tree.

8. A forest prairie fire risk assessment system characterized by, The method for implementing the forest and grassland fire risk assessment method according to claim 7 comprises: A distributed ground sensing node array module is configured to acquire real-time ground surface temperature, soil moisture, vegetation water content, wind speed, wind direction and atmospheric pressure data of the grassland area. A satellite remote sensing data access module is configured to synchronously access high-resolution satellite remote sensing image data, and extract vegetation coverage, ground albedo, thermal anomaly point distribution and historical fire point trajectory information. A meteorological forecast data access module is configured to access regional meteorological forecast data, and acquire precipitation probability, air temperature change trend and relative humidity prediction value within the next 24 hours. A data preprocessing module is configured to perform spatio-temporal alignment and normalization preprocessing on the multi-source heterogeneous data, and construct a grassland fire risk assessment data cube under a unified spatio-temporal reference. A risk factor dynamic coupling modeling module is configured to construct a grassland fire risk factor dynamic coupling model comprising a ground environmental factor submodel, a meteorological driving factor submodel, a vegetation state factor submodel and a historical fire factor submodel. An adaptive weight distribution module is configured to dynamically adjust the weight proportion of each risk factor submodel in the total risk assessment according to the current season, regional geographical features and real-time meteorological conditions. A comprehensive risk index calculation module is configured to multiply the output values of the four risk factor submodels by their corresponding normalized weight coefficients, and then perform weighted summation to obtain a grassland fire comprehensive risk index. A risk level mapping module is configured to convert the grassland fire comprehensive risk index into a risk level identifier through a nonlinear mapping function. A visualization engine module is configured to spatially render the grassland fire comprehensive risk index according to geographical grid units, and generate a dynamic risk heat map. A model online updating module is configured to automatically trigger a model parameter retraining process when receiving new fire verification event data.

9. The forest steppe fire risk assessment system according to claim 8, characterized in that, The distributed ground sensing node array comprises a solar power supply module, a low-power microcontroller, a wireless communication module and a multi-parameter environmental sensor. The multi-parameter environmental sensor comprises an infrared ground surface temperature sensor, a capacitive soil moisture sensor, a near-infrared vegetation water content sensor, an ultrasonic wind speed and direction sensor, and a piezoresistive atmospheric pressure sensor.

Images