Forest ground surface fine combustible moisture content prediction method and system

Abstract

The invention discloses a forest surface fine combustible moisture content prediction method and system, and the method comprises the steps: collecting basic information data of a to-be-predicted forest region, including combustible moisture content and meteorological factor, vegetation and terrain factor information; based on the basic information data, constructing a combustible moisture content prediction model; and completing the prediction of the moisture content of the forest surface fine combustibles by using the combustible moisture content prediction model. According to the method, the accurate combustible matter moisture content prediction model is constructed by comprehensively collecting the combustible matter moisture content, the meteorological factors and the vegetation and terrain factor information of the forest area, the model can effectively predict the moisture content of the fine combustible matter on the forest ground surface, the accuracy of forest fire early warning and risk assessment is remarkably improved, and the forest fire early warning and risk assessment efficiency is improved. The method has great value and practical significance for improving the forest fire prevention and control effect.

Description

Technical Field

[0001] The invention relates to the field of moisture content prediction of combustibles, and in particular to a method and system for predicting moisture content of fine combustibles on a forest surface. Background Art

[0002] Forest fire prevention and control is a challenge and practical need to ensure the national forest ecological security and the safety of people's lives and property. The moisture content of fine surface combustibles directly affects the possibility of forest fire ignition and potential forest fire behavior. It is considered to be the core and direct reference indicator for regional forest fire warning and risk assessment, and is crucial to improving the effectiveness of forest fire prevention and control. Therefore, revealing the spatiotemporal heterogeneity of the moisture content of fine surface combustibles in forests and constructing a spatiotemporally clear and effective combustible moisture content prediction model are of great value and practical significance for achieving accurate and efficient forest fire prevention and control. The current bottlenecks in the estimation of fuel moisture content include: (1) The data on fuel moisture content cannot keep up with the demand, and there is a lack of long-term fixed monitoring plot data, which affects the reliability and accuracy of the assessment of the spatiotemporal heterogeneity of fuel moisture content; (2) More consideration is given to the relationship between fuel moisture content and meteorological factors, and the joint mechanism and interactive influence path of meteorological, vegetation, topographic and other factors are unclear; (3) There is little research on remote sensing inversion of fuel moisture content, which makes it difficult to reveal the spatial heterogeneity of fuel moisture content, affecting the practicality of forest fire warning systems at the regional scale. Summary of the invention

[0003] This invention aims to directly address the problem of how to improve the accuracy of the combustible moisture content prediction model in forest fire prevention and control. It deeply explores the core scientific issues behind the construction of the forest combustible moisture content prediction model technology, and then proposes the optimal combustible moisture content prediction model to serve forest fire early warning and prevention.

[0004] To achieve the above object, the present invention provides a method for predicting the moisture content of fine combustibles on the forest floor, the steps comprising:

[0005] Collect basic information data of the forest area to be predicted, including: moisture content of fuels and meteorological factors, vegetation and terrain factors;

[0006] Based on the basic information data, a combustible moisture content prediction model is constructed;

[0007] The moisture content of fine combustibles on the forest surface is predicted using the combustible moisture content prediction model.

[0008] Preferably, the basic information data includes: information acquisition of moisture content of combustibles and meteorological factors as well as vegetation and terrain factors;

[0009] By setting up monitoring instruments to measure the moisture content of combustibles in real time 24 hours a day, and obtaining information on meteorological factors such as temperature, precipitation, wind speed, relative humidity, and light intensity at the monitoring point;

[0010] Land cover type data, forest phase map data and topographic map data were obtained through open source platforms.

[0011] Preferably, the moisture content of the combustibles is determined by a weighing method, a 50 cm × 50 cm sample plot is set up with the monitor as the center, all the fine surface combustibles in the sample plot are collected, and they are dried continuously at 105°C for 24 hours until the combustibles are absolutely dry, and their dry mass is weighed with an electronic scale.

[0012] Preferably, the method for constructing the combustible moisture content prediction model includes:

[0013] The wavelet analysis method is used to analyze the periodic variation law and mutation characteristics of the moisture content of combustibles on different time scales; the non-parametric variance analysis and gradient analysis method are used to analyze the influence of the basic information data on the moisture content of combustibles; finally, the structural equation model is used to construct the moisture content prediction model of combustibles.

[0014] The present invention also provides a system for predicting moisture content of fine combustibles on forest floor, the system being used to implement the above method, comprising: a collection module, a construction module and a prediction module;

[0015] The acquisition module is used to collect basic information data of the forest area to be predicted, including: moisture content of combustibles and meteorological factors, vegetation and terrain factors;

[0016] The construction module is used to construct a combustible moisture content prediction model based on the basic information data;

[0017] The prediction module uses the combustible moisture content prediction model to complete the prediction of the moisture content of fine combustibles on the forest surface.

[0018] Preferably, the workflow of the construction module includes: using wavelet analysis method to analyze the periodic variation law and mutation characteristics of the moisture content of combustibles on different time scales; using non-parametric variance analysis and gradient analysis method to analyze the influence of the basic information data on the moisture content of combustibles; and finally using structural equation model to construct the moisture content prediction model of combustibles.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The present invention constructs an accurate forecasting model for moisture content of combustible materials by comprehensively collecting information on moisture content of combustible materials and meteorological factors, vegetation and terrain factors in forest areas. This model can effectively predict the moisture content of fine combustible materials on the forest surface, significantly improving the accuracy of forest fire warning and risk assessment, and has great value and practical significance for improving the effectiveness of forest fire prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0022] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0023] Figure 2 is a schematic diagram of a detection device according to an embodiment of the present invention;

[0024] Figure 3 This is a conceptual diagram of the path of factors affecting the moisture content of combustibles in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Embodiment 1

[0028] like Figure 1 FIG. 1 is a schematic diagram of the method flow of this embodiment, and the steps include:

[0029] S1. Collect basic data information of the forest area to be predicted.

[0030] In this embodiment, the basic information data includes: moisture content of combustibles and meteorological factors as well as vegetation and terrain factor information.

[0031] (1) Obtaining moisture content of combustibles and meteorological factors

[0032] This example selects the forest area in southern Jiangxi as the prediction object for illustration, which includes Yinkeng Ecological Forest Farm in Yudu County, Nanshan Forest Park in Nankang District, and Jinpenshan Forest Farm in Xinfeng County. 60 combustible moisture content monitoring instruments (such as Figure 2 The monitored vegetation types are Masson pine forest, Chinese fir forest, Elliptic pine forest, coniferous mixed forest, coniferous and broadleaved mixed forest, broadleaved forest, and bamboo forest. These monitoring points can measure the moisture content of combustibles in real time 24 hours a day, and can also obtain information on meteorological factors such as temperature, precipitation, wind speed, relative humidity, and light intensity at the monitoring points at the same time.

[0033] The traditional weighing method is used to determine the moisture content of combustibles. A 50cm×50cm sample plot is set with the monitor as the center. All the fine surface combustibles in the sample plot (composed of dead surface herbaceous plants and branches with a diameter of less than 0.64cm, including dead leaves, thin dead branches, dead grass, needles, etc.) are collected and brought back to the laboratory and placed in an oven. Dry continuously at 105°C for 24 hours until the combustibles reach absolute dryness (the weight remains unchanged after continuous drying for 2 hours), and weigh their dry mass with an electronic scale. The combustibles dried indoors are placed in the net bag of the monitoring instrument and laid flat on the ground in a manner similar to the undamaged style in the field. The automatic weighing rod of the monitoring instrument weighs the weight of the combustibles continuously within the set unit time (this embodiment is set to weigh once per hour). The real-time moisture content of the combustibles can be calculated by the weight of the combustibles every hour and the initial dry weight, and recorded in the storage device of the instrument host. The combustible samples are replaced at intervals of approximately one month (appropriately adjusted according to actual conditions).

[0034] The moisture content of combustibles is calculated using the following formula:

[0035]

[0036] (2) Obtaining vegetation and terrain factor information

[0037] Obtain land cover type data, forest phase map data and topographic map data on an open source platform. Vegetation (including forest phase map data) and topographic data are mainly used for:

[0038] ① Extract forest stand factor information such as forest stand density;

[0039] ②Construct the relationship between factors such as vegetation and topography and the moisture content of combustible materials.

[0040] S2. Based on basic information data, a prediction model for moisture content in combustibles is constructed.

[0041] The evolution cycle of moisture content in combustibles will change with different observation time scales. Changes in small time scales are often nested in the change cycle of large time scales, reflecting the multi-level structure and local change characteristics in the time domain. Wavelet analysis method is proposed to explore the periodic change law and mutation characteristics of moisture content in combustibles on different time scales.

[0042] Wavelet analysis using the Wavelet package of R language includes the following three main steps:

[0043] Step 1: Model parameter selection and data calculation

[0044] This step has two purposes: a) Select a reasonable wavelet function. Selecting a suitable wavelet function is a prerequisite when using wavelet analysis. There are many wavelet functions to choose from, such as Mexican hat wavelet, Meyer wavelet, Morlet wavelet, etc. In view of the continuous time series characteristics of the moisture content of combustibles, it is appropriate to use the Morlet continuous complex wavelet function to analyze its multi-time scale variation characteristics; b) Determine the corresponding time scale (or wavelet decomposition level), that is, use the wavelet function to decompose the time series data in order to express the variation characteristics of the moisture content on the time scale; c) Analyze the moisture content sequence of combustibles by the continuous wavelet transform method to obtain the variance, modulus square and real part of the continuous wavelet transform coefficients at each scale.

[0045] Step 2: Identification of time series complexity

[0046] By analyzing the changing rules of wavelet coefficients, we can understand the complex characteristics of the sequence: including: a) directly analyzing the changing rules of wavelet coefficients to identify the composition and multi-time scale changing characteristics of the sequence. For example, when the wavelet coefficient is 0, it corresponds to the mutation point of the moisture content of the combustible material. The larger the absolute value of the wavelet change coefficient, the more significant the change of the moisture content of the combustible material on this time scale; b) taking corresponding methods to process the wavelet coefficients can achieve other research purposes, such as wavelet threshold denoising, separation of different determined components, etc.; c) using wavelet coefficients to solve some quantitative indicators, such as wavelet entropy, wavelet mutual entropy, fractal dimension, etc., to quantitatively describe the complexity of the sequence; d) by analyzing the relationship between the wavelet coefficients of different sequences, we can reveal the correlation between the moisture content sequences of combustible materials on different time scales.

[0047] Step 3: Time Series Structure Identification

[0048] Subsequences on different time scales are obtained through wavelet reconstruction method, and the composition and change characteristics of the sequence are analyzed by using the subsequence decomposition results. For example, the real part of the continuous wavelet transform coefficient of the moisture content sequence of combustibles at various scales is plotted over time, and the high and low change trends of the moisture content of combustibles at the main scales are determined from the transformation of the real part of the continuous wavelet transform from positive to negative.

[0049] In addition, the moisture content of fuels will change with different vegetation (type, canopy density, density), topography (altitude, slope, aspect), meteorology (temperature, precipitation, wind speed, light intensity) and other conditions. The response of fuel moisture content to environmental factors includes the comparison of differences between different levels and the analysis of environmental gradient characteristics.

[0050] ① Comparison of intra-group and inter-group differences in environmental factors

[0051] Non-parametric analysis of variance (such as Kruskal-Wallis H test) was used to analyze whether there were significant differences in the moisture content of fuels between different levels of vegetation, terrain and meteorological conditions (P<0.05), as well as the differences in the moisture content of fuels between different levels of environmental factors (P<0.05).

[0052] ② Environmental gradient ranking analysis

[0053] Gradient Analysis is used to reveal the relationship between vegetation and environment, and give corresponding explanations, so as to better express the environmental gradient and structural gradient of the community and reflect the spatial variation characteristics of the environment and the community. It is proposed to use Detrended Canonical Correspondence Analysis (DCCA) to explore the gradient structural relationship between the moisture content of combustibles and environmental factors. DCCA sorting classification combines vegetation factors and environmental factors at the same time, which can better express the environmental gradient of the community and reflect the spatial variation characteristics. DCCA sorting requires two data matrices, one is the vegetation data matrix and the other is the environmental factor data matrix. The vegetation data matrix of this embodiment is a P×N-dimensional combustible moisture content matrix, where P is the number of vegetation types and N is the number of sample plots. Environmental factors include forest stands (canopy density, density), topography (altitude, slope, slope aspect), and meteorology (precipitation, temperature, relative humidity, wind speed, light intensity). DCCA analysis uses CANOCO software for sorting. The "Monte Carlo permutation test" is used to test the significance of the DCCA sorting axis eigenvalue.

[0054] Based on the above steps, a combustible moisture content prediction model is constructed.

[0055] (1) Screening of key factors affecting moisture content of combustibles and analysis of their influencing paths

[0056] Structural equation modeling (SEM) is used to reveal the influence pattern of vegetation, topographic and meteorological factors on the moisture content of fuels. Structural equation modeling integrates factor analysis and path analysis, and simultaneously examines the relationship between explicit variables, latent variables, interference or error terms contained in the model, thereby obtaining the direct, indirect or total effect of the independent variable on the dependent variable. The SEM package of R software is proposed to conduct structural equation model analysis, which includes the following main steps:

[0057] Step 1: Screening of impact factors

[0058] There are many factors that affect the moisture content of combustibles, and many of them have high collinearity or correlation. The high correlation between explanatory variables may distort model estimation or interfere with the accuracy of simulation estimation, and it is difficult to find the key factors that affect the moisture content of combustibles. This embodiment explores the correlation between factors through correlation analysis, and uses the method of stepwise multiple linear regression to eliminate redundant factors. Factor elimination is based on the AIC (Akaike Information Criterion) information criterion. AIC is a standard for measuring the goodness of fit of a statistical model. It can calculate the AIC values ​​of multiple models at a time, and find the model factor corresponding to the minimum AIC value as the selection object.

[0059] Step 2: Build a conceptual model

[0060] The core of constructing a conceptual model is to determine the relationship (influence path) between latent variables and observed variables. In this embodiment, vegetation, terrain conditions and meteorological conditions are used as factors to describe the variability of moisture content in combustibles, that is, latent variables; forest stand type (composition), forest stand canopy density, forest stand density, altitude, slope, slope aspect, precipitation, temperature, relative humidity, wind speed, light intensity, etc. are usually used to measure vegetation, terrain conditions, and meteorological conditions information, that is, defined as observed variables. The conceptual diagram of the influencing factor path of the moisture content of combustibles is shown in Figure 1. Figure 3 shown.

[0061] Step 3: Model Fitting

[0062] That is, run the SEM package process. The goal of model fitting is to make the model's implicit covariance matrix, that is, the model's "path coefficient matrix (characterizing the relationship between latent variables)" as close as possible to the sample covariance matrix. This project intends to use the two-stage least squares method (Two-Stage Least Squares: TSLS), because the TSLS method does not rely on iteration and can quickly calculate the values ​​of each parameter.

[0063] Step 4: Model Evaluation

[0064] The model was evaluated using the Comparative Fit Index (CFI), the root mean square error (RMSEA), the standardized root mean square residual (SEMR), and the chi-square value (P-value). Generally, the model is considered to be excellent when CFI>0.9, RMSEA<0.05, SEMR<0.05, and P-value<0.05.

[0065] (2) Modeling and accuracy evaluation of moisture content in combustible materials under environmental heterogeneity

[0066] Compare the differences in fuel moisture content estimation with and without considering environmental heterogeneity (vegetation and terrain conditions). Through comparative analysis, identify the impact of environmental heterogeneity on the accuracy of fuel moisture content estimation and determine the prediction model.

[0067] ① Estimation of moisture content of combustible materials considering environmental heterogeneity

[0068] Step 1: Cluster analysis of environmental factors

[0069] Hierarchical cluster analysis was used to cluster the field observation data into several heterogeneous clusters. Hierarchical cluster analysis used Euclidean distances as a measure of similarity and clustered according to Ward's method. Before clustering, all variable values ​​were standardized to Z scores. The differences between different variables were tested using the nonparametric Kruskal-Wallis test. Hierarchical cluster analysis was implemented using the Hclust() function in R language.

[0070] Step 2: Combustible moisture content model simulation

[0071] Based on the screened key indicators that affect the moisture content of combustibles, 70% of the data were used to build the model.

[0072] This embodiment adopts the Boosted Regression Tree model (BRT) fitting model. The BRT model is a computer algorithm that can continuously learn by itself and is developed based on the traditional Classification and Regression Tree algorithm. The core idea of ​​the BRT model is to generate multiple regression trees through self-learning and random selection, and then combine these regression trees to form a more stable tree, thereby improving the stability and resolution of the model. The BRT model effectively combines the advantages of boosting trees and regression trees, and can better handle the influence of factors on the result variables. Compared with the requirements of other models for variable types, the BRT model can handle different types of result variables. In the boosted regression tree model, it is assumed that y=g(x), that is, g(x) is the influence function of the independent variable x to the dependent variable y. The loss function of the evaluation model is established as follows:

[0073] g(x)=Σ k g k (x) = Σ k β k b(x;γ k )

[0074] In the formula, b(x;γk) represents the countless subtree nodes constructed, γ k Represents each variable and the observed and predicted values ​​at each child node. k Represents the total number of hits for each tree node, and can determine the optimal prediction value obtained from different subtree combinations.

[0075] Step 3: Model Accuracy Evaluation

[0076] The 30% data that is not involved in modeling is used as validation data. Model accuracy evaluation index: relative average error Relative mean absolute error (MAE%), relative root mean square error (RMSE%) and Nash-Sutcliffe coefficient of certainty (ME). The mean deviation and the accuracy of model simulation were estimated. MAE% and RMSE% were used to evaluate the accuracy of model simulation values ​​using absolute error. ME was used to estimate the error between model simulation values ​​and field observations.

[0077]

[0078] In the formula, O i is the field observation value, P i is the model simulation value, and n is the number of verification samples.

[0079] ② Estimation of moisture content of combustible materials without considering environmental heterogeneity

[0080] The simulation model and model evaluation method used without considering environmental heterogeneity are consistent with the method considering environmental heterogeneity. The difference is that all observation data are combined to establish the estimation model.

[0081] ③Comparison of the effects of fuel moisture content estimation models

[0082] Comparison of the two estimation methods: 1) Comparative analysis of the moisture content of fuels under different time scales and vegetation and terrain conditions to determine the extent to which the estimation accuracy can be corrected by considering environmental heterogeneity; 2) 30% of the data were randomly selected and the sample non-parametric U test was used to identify the significance of the difference in the moisture content of fuels estimated by the two methods.

[0083] S3. Use the combustible moisture content prediction model to predict the moisture content of fine combustibles on the forest surface.

[0084] Finally, the constructed combustible moisture content prediction model is used to predict the moisture content of fine combustibles on the forest surface.

[0085] Embodiment 2

[0086] This embodiment also provides a forest fuel moisture content model based on remote sensing technology as a comparison item of the present invention. The specific steps are as follows:

[0087] First, obtain remote sensing image data.

[0088] MODIS image data was downloaded from the official website of NASA data in the United States, Landsat images and Sentinel-2 images were downloaded from the website of the United States Geological Survey; GF satellite images were downloaded from the website of the Land Observation Satellite Data Service Platform of the China Resources Satellite Application Center. When selecting remote sensing images, images with no clouds or cloud cover <10% were selected. Due to the long time span of the study, in order to ensure the integrity and uniformity of the data, this embodiment only uses GF-2 / 6 images, and selects MODIS, Landsat8, and Sentinel-2 remote sensing images with strong accessibility to invert the moisture content of combustibles.

[0089] Then the remote sensing image data is preprocessed.

[0090] The preprocessing of MODIS images and Landsat images is similar, mainly for atmospheric radiation correction. For example, Landsat8 data has undergone geometric correction and terrain correction, and only multispectral data needs to be calibrated for radiation and atmosphere. The FLAASH module provided by ENVI5.3.1 is used to calibrate the panchromatic band. The spatial resolution of Landsat8 multispectral data is increased to 15m by automatic registration and image fusion of multispectral data and panchromatic data. Finally, Landsat8 data with 30m and 15m resolutions are obtained.

[0091] The L1C product of Sentinel-2A / B is the atmospheric apparent reflectance product after orthorectification and sub-pixel geometric precision correction. It is planned to realize the atmospheric correction of Sentinel data through the Sen2cor processing module of the SNAP software platform (sentinels application platform, SNAP). The 10m image with more bands and higher spatial resolution after atmospheric correction is selected as the resampling data source. Then, the Sentinel image is mosaicked, cropped, masked and processed by ENVI5.3.1 software to generate L2A data. The multispectral images of GF-2 / 6 are radiometrically corrected, atmospherically corrected and orthorectified by ENVI5.3.1 remote sensing software. At the same time, the panchromatic bands of GF-2 and GF-6 are radiometrically calibrated and orthorectified. The spatial resolution of GF-6 multispectral data is increased to 2m and the spatial resolution of GF-2 multispectral data is increased to 1m through automatic registration and image fusion.

[0092] Calculate spectral reflectance, remote sensing indices, and texture features.

[0093] Spectral information refers to the reflectance of each band, which can be extracted from each band of each remote sensing data source. There are many remote sensing vegetation indices, and each vegetation index has its adaptability and limitations in application. Among the indices for evaluating the moisture content of combustibles internationally, the indices selected in this embodiment mainly include: Temperature Vegetation Dryness Index (TVDI), Perpendicular Drought Index (PDI), Temperature Condition Index (TCI), Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Normalized Difference Moisture Index (NDMI), Global Vegetation Moisture Index (GVMI), Normalized Soil Moisture Index (NSMI), Normalized Difference Infrared Index (NDII), Visible Atmospherically Resistant Index (VRAI), etc. Texture is a very complex spatial attribute that changes significantly with the sun, observation angle, terrain, target of interest and its environment. The uniqueness of different vegetation types and their structures makes their texture features different. The differences in different texture features are used as a supplement to the image spectrum-based evaluation, which can improve the accuracy to a certain extent. This embodiment calculates 8 texture features of each band of the image based on the gray level co-occurrence matrix (GLCM), which are mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment and correlation. The texture features are calculated using the Co-occurrence of ENVI5.3.1 software. Taking into account the scale problem of spatial heterogeneity, this embodiment takes the combustible moisture content monitor as the center and sets buffer zones with radii of 30m, 90m, 150m, 250m, 500m, 100m, 2500m, and 5000m to calculate the corresponding texture features.

[0094] Finally, a model for estimating moisture content in combustibles is constructed.

[0095] The remote sensing spectral reflectance, vegetation index and texture features of each field monitoring point in each time period were extracted using R language software. 70% of the data were used for model construction, and the boosted regression tree model (BRT) was used to fit the prediction model. The explanatory power of spectral reflectance, vegetation index and texture features under different remote sensing image data sources and different spatial resolutions was obtained through model fitting (as shown in Table 1).

[0096] Table 1

[0097]

[0098] Finally, the model accuracy is evaluated as follows:

[0099] The 30% sample plot data that are not involved in the modeling are used as verification data to evaluate the accuracy of remote sensing assessment of moisture content in fuels. The accuracy evaluation is planned to use indicators such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Nash-Sutcliffe Efficiency coefficient (NSE).

[0100] This model is used to perform analysis and comparison with the model in Example 1.

[0101] The above-mentioned differences in the results of estimating the moisture content of combustibles based on traditional meteorological factors and remote sensing are compared at different time scales (hours, days, months, seasons) and between different types of environmental heterogeneity. The Normalized Mean Bias (NMB) and the Normalized Mean Error (NME) are used to evaluate the degree of agreement between the estimation of meteorological factors and remote sensing estimation. NMB reflects the average deviation of the results of the two estimation methods, and NME reflects the mean absolute error. These are two dimensionless statistics. The closer they are to 0, the better the simulation effect. Generally, if both are less than 50%, the results are considered to be similar. The calculation formula is as follows:

[0102]

[0103] In the formula, C m represents the remote sensing estimated value; C o Represents the estimated value of meteorological factors.

[0104] Through the above comparative analysis, it can be seen that the model accuracy of embodiment 1 (ie, the present invention) is better.

[0105] Embodiment 3

[0106] The present embodiment also provides a system for predicting the moisture content of fine combustibles on the forest surface, including: a collection module, a construction module and a prediction module; the collection module is used to collect basic information data of the forest area to be predicted, including: moisture content of combustibles and meteorological factors, vegetation and terrain factor information; the construction module is used to construct a combustible moisture content prediction model based on the basic information data; the prediction module uses the combustible moisture content prediction model to complete the prediction of the moisture content of fine combustibles on the forest surface.

[0107] The following will describe in detail how the present invention solves the technical problems in practical work in conjunction with this embodiment.

[0108] First, the acquisition module is used to collect basic data information of the forest area to be predicted.

[0109] In this embodiment, the basic information data includes: moisture content of combustibles, meteorological factors, and vegetation and terrain factor information, and the data is collected by a combustible moisture content monitoring instrument.

[0110] (1) Obtaining moisture content of combustibles and meteorological factors

[0111] This embodiment selects the forest area in southern Jiangxi as the prediction object for illustration, which includes 60 combustible moisture content monitoring instruments built in Yinkeng Ecological Forest Farm in Yudu County, Nanshan Forest Park in Nankang District, and Jinpenshan Forest Farm in Xinfeng County according to different vegetation and terrain conditions. The monitored vegetation types are Masson pine forest, Chinese fir forest, Elliptic pine forest, coniferous mixed forest, coniferous and broad-leaved mixed forest, broad-leaved forest, and bamboo forest. These monitoring points can measure the moisture content of combustibles in real time 24 hours a day, and can also obtain meteorological factor information such as temperature, precipitation, wind speed, relative humidity, and light intensity at the monitoring points at the same time.

[0112] The traditional weighing method is used to determine the moisture content of combustibles. A 50cm×50cm sample plot is set with the monitor as the center. All the fine surface combustibles in the sample plot (composed of dead surface herbaceous plants and branches with a diameter of less than 0.64cm, including dead leaves, thin dead branches, dead grass, needles, etc.) are collected and brought back to the laboratory and placed in an oven. Dry continuously at 105°C for 24 hours until the combustibles reach absolute dryness (the weight remains unchanged after continuous drying for 2 hours), and weigh their dry mass with an electronic scale. The combustibles dried indoors are placed in the net bag of the monitoring instrument and laid flat on the ground in a manner similar to the undamaged style in the field. The automatic weighing rod of the monitoring instrument weighs the weight of the combustibles continuously within the set unit time (this embodiment is set to weigh once per hour). The real-time moisture content of the combustibles can be calculated by the weight of the combustibles every hour and the initial dry weight, and recorded in the storage device of the instrument host. The combustible samples are replaced at intervals of approximately one month (appropriately adjusted according to actual conditions).

[0113] The moisture content of combustibles is calculated using the following formula:

[0114]

[0115] (2) Obtaining vegetation and terrain factor information

[0116] Obtain land cover type data, forest phase map data and topographic map data on an open source platform. Vegetation (including forest phase map data) and topographic data are mainly used for:

[0117] ① Extract forest stand factor information such as forest stand density;

[0118] ②Construct the relationship between factors such as vegetation and topography and the moisture content of combustible materials.

[0119] Then, the construction module is used to build a combustible moisture content prediction model based on the basic information data.

[0120] The evolution cycle of moisture content in combustibles will change with different observation time scales. Changes in small time scales are often nested in the change cycle of large time scales, reflecting the multi-level structure and local change characteristics in the time domain. Wavelet analysis method is proposed to explore the periodic change law and mutation characteristics of moisture content in combustibles on different time scales.

[0121] Wavelet analysis using the Wavelet package of R language includes the following three main steps:

[0122] Step 1: Model parameter selection and data calculation

[0123] This step has two purposes: a) Select a reasonable wavelet function. Selecting a suitable wavelet function is a prerequisite when using wavelet analysis. There are many wavelet functions to choose from, such as Mexican hat wavelet, Meyer wavelet, Morlet wavelet, etc. In view of the continuous time series characteristics of the moisture content of combustibles, it is appropriate to use the Morlet continuous complex wavelet function to analyze its multi-time scale variation characteristics; b) Determine the corresponding time scale (or wavelet decomposition level), that is, use the wavelet function to decompose the time series data in order to express the variation characteristics of the moisture content on the time scale; c) Analyze the moisture content sequence of combustibles by the continuous wavelet transform method to obtain the variance, modulus square and real part of the continuous wavelet transform coefficients at each scale.

[0124] Step 2: Identification of time series complexity

[0125] By analyzing the changing rules of wavelet coefficients, we can understand the complex characteristics of the sequence: including: a) directly analyzing the changing rules of wavelet coefficients to identify the composition and multi-time scale changing characteristics of the sequence. For example, when the wavelet coefficient is 0, it corresponds to the mutation point of the moisture content of the combustible material. The larger the absolute value of the wavelet change coefficient, the more significant the change of the moisture content of the combustible material on this time scale; b) taking corresponding methods to process the wavelet coefficients can achieve other research purposes, such as wavelet threshold denoising, separation of different determined components, etc.; c) using wavelet coefficients to solve some quantitative indicators, such as wavelet entropy, wavelet mutual entropy, fractal dimension, etc., to quantitatively describe the complexity of the sequence; d) by analyzing the relationship between the wavelet coefficients of different sequences, we can reveal the correlation between the moisture content sequences of combustible materials on different time scales.

[0126] Step 3: Time Series Structure Identification

[0127] Subsequences on different time scales are obtained through wavelet reconstruction method, and the composition and change characteristics of the sequence are analyzed by using the subsequence decomposition results. For example, the real part of the continuous wavelet transform coefficient of the moisture content sequence of combustibles at various scales is plotted over time, and the high and low change trends of the moisture content of combustibles at the main scales are determined from the transformation of the real part of the continuous wavelet transform from positive to negative.

[0128] In addition, the moisture content of fuels will change with different vegetation (type, canopy density, density), topography (altitude, slope, aspect), meteorology (temperature, precipitation, wind speed, light intensity) and other conditions. The response of fuel moisture content to environmental factors includes the comparison of differences between different levels and the analysis of environmental gradient characteristics.

[0129] ① Comparison of intra-group and inter-group differences in environmental factors

[0130] Non-parametric analysis of variance (such as Kruskal-Wallis H test) was used to analyze whether there were significant differences in the moisture content of fuels between different levels of vegetation, terrain and meteorological conditions (P<0.05), as well as the differences in the moisture content of fuels between different levels of environmental factors (P<0.05).

[0131] ② Environmental gradient ranking analysis

[0132] Gradient Analysis is used to reveal the relationship between vegetation and environment, and give corresponding explanations, so as to better express the environmental gradient and structural gradient of the community and reflect the spatial variation characteristics of the environment and the community. It is proposed to use Detrended Canonical Correspondence Analysis (DCCA) to explore the gradient structural relationship between the moisture content of combustibles and environmental factors. DCCA sorting classification combines vegetation factors and environmental factors at the same time, which can better express the environmental gradient of the community and reflect the spatial variation characteristics. DCCA sorting requires two data matrices, one is the vegetation data matrix and the other is the environmental factor data matrix. The vegetation data matrix of this embodiment is a P×N-dimensional combustible moisture content matrix, where P is the number of vegetation types and N is the number of sample plots. Environmental factors include forest stands (canopy density, density), topography (altitude, slope, slope aspect), and meteorology (precipitation, temperature, relative humidity, wind speed, light intensity). DCCA analysis uses CANOCO software for sorting. The "Monte Carlo permutation test" is used to test the significance of the DCCA sorting axis eigenvalue.

[0133] Based on the above steps, a combustible moisture content prediction model is constructed.

[0134] (1) Screening of key factors affecting moisture content of combustibles and analysis of their influencing paths

[0135] Structural equation modeling (SEM) is used to reveal the influence pattern of vegetation, topographic and meteorological factors on the moisture content of fuels. Structural equation modeling integrates factor analysis and path analysis, and simultaneously examines the relationship between explicit variables, latent variables, interference or error terms contained in the model, thereby obtaining the direct, indirect or total effect of the independent variable on the dependent variable. The SEM package of R software is proposed to conduct structural equation model analysis, which includes the following main steps:

[0136] Step 1: Screening of impact factors

[0137] There are many factors that affect the moisture content of combustibles, and many of them have high collinearity or correlation. The high correlation between explanatory variables may distort model estimation or interfere with the accuracy of simulation estimation, and it is difficult to find the key factors that affect the moisture content of combustibles. This embodiment explores the correlation between factors through correlation analysis, and uses the method of stepwise multiple linear regression to eliminate redundant factors. Factor elimination is based on the AIC (Akaike Information Criterion) information criterion. AIC is a standard for measuring the goodness of fit of a statistical model. It can calculate the AIC values ​​of multiple models at a time, and find the model factor corresponding to the minimum AIC value as the selection object.

[0138] Step 2: Build a conceptual model

[0139] The core of constructing a conceptual model is to determine the relationship (influence path) between latent variables and observed variables. In this embodiment, vegetation, terrain conditions and meteorological conditions are used as factors to describe the variability of moisture content in combustibles, that is, latent variables; forest stand type (composition), forest stand canopy density, forest stand density, altitude, slope, slope aspect, precipitation, temperature, relative humidity, wind speed, light intensity, etc. are usually used to measure vegetation, terrain conditions, and meteorological conditions information, that is, defined as observed variables. The conceptual diagram of the influencing factor path of the moisture content of combustibles is shown in Figure 1. Figure 3 shown.

[0140] Step 3: Model Fitting

[0141] That is, run the SEM package process. The goal of model fitting is to make the model's implicit covariance matrix, that is, the model's "path coefficient matrix (characterizing the relationship between latent variables)" as close as possible to the sample covariance matrix. This project intends to use the two-stage least squares method (Two-Stage Least Squares: TSLS), because the TSLS method does not rely on iteration and can quickly calculate the values ​​of each parameter.

[0142] Step 4: Model Evaluation

[0143] The model was evaluated using the Comparative Fit Index (CFI), the root mean square error (RMSEA), the standardized root mean square residual (SEMR), and the chi-square value (P-value). Generally, the model is considered to be excellent when CFI>0.9, RMSEA<0.05, SEMR<0.05, and P-value<0.05.

[0144] (2) Modeling and accuracy evaluation of moisture content in combustible materials under environmental heterogeneity

[0145] Compare the differences in fuel moisture content estimation with and without considering environmental heterogeneity (vegetation and terrain conditions). Through comparative analysis, identify the impact of environmental heterogeneity on the accuracy of fuel moisture content estimation and determine the prediction model.

[0146] ① Estimation of moisture content of combustible materials considering environmental heterogeneity

[0147] Step 1: Cluster analysis of environmental factors

[0148] Hierarchical cluster analysis was used to cluster the field observation data into several heterogeneous clusters. Hierarchical cluster analysis used Euclidean distances as a measure of similarity and clustered according to Ward's method. Before clustering, all variable values ​​were standardized to Z scores. The differences between different variables were tested using the nonparametric Kruskal-Wallis test. Hierarchical cluster analysis was implemented using the Hclust() function in R language.

[0149] Step 2: Combustible moisture content model simulation

[0150] Based on the screened key indicators that affect the moisture content of combustibles, 70% of the data were used to build the model.

[0151] This embodiment adopts the Boosted Regression Tree model (BRT) fitting model. The BRT model is a computer algorithm that can continuously learn by itself and is developed based on the traditional Classification and Regression Tree algorithm. The core idea of ​​the BRT model is to generate multiple regression trees through self-learning and random selection, and then combine these regression trees to form a more stable tree, thereby improving the stability and resolution of the model. The BRT model effectively combines the advantages of boosting trees and regression trees, and can better handle the influence of factors on the result variables. Compared with the requirements of other models for variable types, the BRT model can handle different types of result variables. In the boosted regression tree model, it is assumed that y=g(x), that is, g(x) is the influence function of the independent variable x to the dependent variable y. The loss function of the evaluation model is established as follows:

[0152] g(x)=Σ k g k (x) = Σ k β k b(x;γ k )

[0153] In the formula, b(x;γk) represents the countless subtree nodes constructed, γ k Represents each variable and the observed and predicted values ​​at each child node. k Represents the total number of hits for each tree node, and can determine the optimal prediction value obtained from different subtree combinations.

[0154] Step 3: Model Accuracy Evaluation

[0155] The 30% data that is not involved in modeling is used as validation data. Model accuracy evaluation index: relative average error Relative mean absolute error (MAE%), relative root mean square error (RMSE%) and Nash-Sutcliffe coefficient of certainty (ME). The mean deviation and the accuracy of model simulation were estimated. MAE% and RMSE% were used to evaluate the accuracy of model simulation values ​​using absolute error. ME was used to estimate the error between model simulation values ​​and field observations.

[0156]

[0157] In the formula, O i is the field observation value, P i is the model simulation value, and n is the number of verification samples.

[0158] ② Estimation of moisture content of combustible materials without considering environmental heterogeneity

[0159] The simulation model and model evaluation method used without considering environmental heterogeneity are consistent with the method considering environmental heterogeneity. The difference is that all observation data are combined to establish the estimation model.

[0160] ③Comparison of the effects of fuel moisture content estimation models

[0161] Comparison of the two estimation methods: 1) Comparative analysis of the moisture content of fuels under different time scales and vegetation and terrain conditions to determine the extent to which the estimation accuracy can be corrected by considering environmental heterogeneity; 2) 30% of the data were randomly selected and the sample non-parametric U test was used to identify the significance of the difference in the moisture content of fuels estimated by the two methods.

[0162] Finally, the prediction module uses the constructed combustible moisture content prediction model for prediction.

[0163] The embodiments are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting the moisture content of fine combustibles on the forest floor, characterized in that the steps include: Collect basic information data of the forest area to be predicted, including: moisture content of fuels and meteorological factors, vegetation and terrain factors; Based on the basic information data, a combustible moisture content prediction model is constructed; The moisture content of fine combustibles on the forest surface is predicted using the combustible moisture content prediction model.

2. The method for predicting moisture content of fine combustibles on forest floor according to claim 1, characterized in that: The basic information data include: the moisture content of combustibles and meteorological factors as well as the acquisition of vegetation and terrain factors; By setting up monitoring instruments to measure the moisture content of combustibles in real time 24 hours a day, and obtaining information on meteorological factors such as temperature, precipitation, wind speed, relative humidity, and light intensity at the monitoring point; Land cover type data, forest phase map data and topographic map data were obtained through open source platforms.

3. The method for predicting moisture content of fine combustibles on forest floor according to claim 2, characterized in that: The moisture content of combustibles is determined by weighing method. A 50cm×50cm sample plot is set up with the monitor as the center. All the fine surface combustibles in the sample plot are collected and dried continuously at 105℃ for 24h until the combustibles are absolutely dry. The dry mass is weighed with an electronic scale.

4. The method for predicting moisture content of fine combustibles on forest floor according to claim 1, characterized in that: The method for constructing the combustible moisture content prediction model includes: The wavelet analysis method is used to analyze the periodic variation law and mutation characteristics of the moisture content of combustibles on different time scales; the non-parametric variance analysis and gradient analysis method are used to analyze the influence of the basic information data on the moisture content of combustibles; finally, the structural equation model is used to construct the moisture content prediction model of combustibles.

5. A system for predicting moisture content of fine combustibles on forest floor, the system being used to implement the method according to any one of claims 1 to 4, characterized in that: include: Acquisition module, construction module and prediction module; The acquisition module is used to collect basic information data of the forest area to be predicted, including: moisture content of combustibles and meteorological factors, vegetation and terrain factors; The construction module is used to construct a combustible moisture content prediction model based on the basic information data; The prediction module uses the combustible moisture content prediction model to complete the prediction of the moisture content of fine combustibles on the forest surface.

6. The system for predicting moisture content of fine combustibles on forest floor according to claim 5, characterized in that: The workflow of the construction module includes: using wavelet analysis method to analyze the periodic variation law and mutation characteristics of the moisture content of combustibles on different time scales; using non-parametric variance analysis and gradient analysis method to analyze the influence of the basic information data on the moisture content of combustibles; and finally using structural equation model to construct the moisture content prediction model of combustibles.

Images