A method and system for predicting future debris flow activity

Abstract

The present application provides a future debris flow activity prediction method and system, and relates to the technical field of geological disaster activity prediction. The present application is based on a debris flow historical activity catalog database, combines a multi-source debris flow disaster-pregnant background and an excitation factor, respectively constructs a whole-basin debris flow activity quantity and a small-basin debris flow outbreak probability prediction model, and realizes the prediction of the number and the location of the future target period debris flow through the cooperative constraint of the regional scale activity quantity on the single-gully outbreak probability.

Description

Technical Field

[0001] This invention relates to the field of geological disaster activity prediction technology, and in particular to a method and system for predicting future debris flow activity. Background Technology

[0002] Debris flows are a common and highly destructive geological hazard in mountainous areas. Their formation is typically controlled by a combination of factors, including loose solid material, water supply, and steep terrain. In recent years, with the intensification of global climate change, the frequency of extreme weather events has increased significantly, especially in high-altitude or high-latitude mountainous areas. The superposition of climate patterns such as high summer temperatures, heavy rainfall, and rapid snowmelt has made the gestation conditions and triggering mechanisms of debris flows more complex. Debris flows often exhibit characteristics such as short activity cycles, sudden onset, and significant recurrence, and their activity patterns are sensitive to climate change.

[0003] In debris flow disaster research, debris flow activity is used to characterize the historical activity features and future development trends of debris flows. It is influenced by multiple factors, including sediment source, water source, and topography. For example, the watershed needs to have a certain amount of loose solid material, and rainfall in the area needs to reach or exceed the debris flow initiation threshold. Furthermore, human engineering activities can alter sediment source conditions. Therefore, the activity of different debris flow gullies, or even the same debris flow gully at different times, exhibits significant differences. Debris flow activity is consistent with debris flow frequency; high-frequency debris flows are more active, while low-frequency debris flows are less active. However, debris flow activity is time-limited, meaning its activity needs to be studied within a specific period.

[0004] Modeling debris flow activity prediction faces the following challenges: using the number of debris flows as the modeling objective only yields the quantity of debris flows occurring in a region, but not the location of the debris flows; using debris flow susceptibility as the modeling objective results in a skewed distribution of the target variable due to the presence of records of events that did not occur in the debris flow sample, increasing the difficulty of model training; and the complex nonlinear relationship between multi-source feature variables and debris flow activity makes it difficult for a single modeling strategy to objectively determine the debris flow activity prediction threshold.

[0005] Against the backdrop of a complex coupling between significant climate change and intensified human activities, the conditions for triggering and causing debris flows are changing. There is an urgent need for a method that, based on an understanding of the historical activity of debris flows, can extrapolate and predict the probability and activity changes of debris flows under different future scenarios. This method should be able to predict both the number of debris flows that may occur in the future and the possible locations of future debris flows, so as to achieve cross-scale (regional and single-gully scale) prediction of future debris flow activity. This will support the improvement of the scientificity and reliability of debris flow risk assessment, monitoring and early warning, and disaster prevention and mitigation engineering planning. Summary of the Invention

[0006] To address the aforementioned shortcomings in existing technologies, this invention provides a method and system for predicting future debris flow activity, used to predict future debris flow activity trends at regional and single-gully scales. This overcomes the problem that existing technologies have a single scale for debris flow activity prediction, making it difficult to take into account both the overall trend of the entire watershed and the differences between single gullies in small watersheds.

[0007] To achieve the above objectives, the technical solution adopted by this invention is: a method for predicting future debris flow activity, comprising the following steps: S1. Construction of a historical debris flow activity cataloging database: obtaining a sequence of debris flow occurrences over many years; S2. Prediction of the number of debris flow activities in the whole basin: Based on the annual debris flow occurrence sequence over many years, a debris flow activity quantity model for the whole basin is constructed, and the total number of debris flow occurrences in the whole basin within the target period is predicted using the debris flow activity quantity model for the whole basin, so as to obtain the prediction result of the number of debris flow activities in the whole basin. S3. Small watershed debris flow outbreak probability prediction: Collect multi-source data characterizing debris flow activity and construct a small watershed debris flow outbreak probability prediction model by combining the annual debris flow occurrence sequence over many years. Using the small watershed debris flow outbreak probability prediction model, obtain the debris flow outbreak probability prediction results at the single gully scale within the future target period. S4. Prediction of future debris flow locations: Based on the prediction results of the number of debris flow activities in the entire basin and the prediction results of debris flow outbreak probability, the debris flow outbreak probability threshold is determined, and the spatial distribution prediction results of debris flow locations within the future target period are obtained.

[0008] The beneficial effects of this invention are as follows: This invention constructs a catalog database of historical debris flow activities and, based on this, establishes a basin-wide debris flow activity quantity prediction model and a small-basin debris flow outbreak probability prediction model, achieving collaborative modeling of debris flow activity prediction at both the regional and single-gully scales. By introducing a small-basin debris flow outbreak probability prediction model based on multi-source factors at the single-gully scale, the prediction results can quantify the relative probability of debris flows occurring in different debris flow gullies within the future target period in probabilistic form. Simultaneously, the predicted quantity of future debris flow activities across the entire basin is used as a constraint at the regional scale to determine the threshold for the small-basin debris flow outbreak probability. This ensures that the spatial distribution of predicted debris flow locations is consistent with the predicted activity quantity at the regional scale, thereby achieving prediction of debris flow locations within the future target period and avoiding the problem of mismatch between single-gully outbreak probability prediction results and the overall regional activity level. This technical solution can determine the location of debris flows in the future target period while ensuring a reasonable number of debris flow activities at the regional scale. This provides a scientific basis for the analysis of the spatiotemporal distribution of debris flow activities, as well as for future debris flow risk assessment, monitoring and early warning, and disaster prevention and mitigation decision-making.

[0009] Further, S1 includes the following steps: Based on the interpretation results of interannual multi-temporal remote sensing images, field survey data, and records of historical debris flow disaster events, debris flow gullies are identified and labeled. Based on the identification and annotation results, debris flow events that occurred year by year in each debris flow gully during the historical record period were statistically analyzed to form a multi-year sequence of debris flow occurrences and to construct a catalog database of historical debris flow activities.

[0010] The beneficial effect of the above-mentioned further scheme is that the database provides high-quality historical data support for the training of debris flow occurrence quantity prediction models and debris flow outbreak probability prediction models, thereby ensuring the reliability of the models.

[0011] Furthermore, S2 includes the following steps: Based on the annual sequence of debris flow occurrences over many years, the number of interannual debris flow activities within the historical record period is calculated. We collected precipitation and temperature data related to debris flow activities in historical and future scenarios, performed time-scale uniform processing on the collected data, and extracted meteorological characteristic variables. Using the interannual activity of debris flows within the historical record period as the modeling target variable and the meteorological characteristic variables corresponding to the historical and future periods as input characteristic variables, a machine learning algorithm is used to construct a model of debris flow activity in the entire watershed. A model for the quantity of debris flow activities across the entire basin was trained, and the trained model was used to predict the number of debris flows in the entire basin within a future target period, thus obtaining the predicted results of debris flow activity quantity across the entire basin.

[0012] The beneficial effect of the above-mentioned further scheme is that by collecting and processing historical and future precipitation and temperature data, combined with the interannual debris flow activity, and using machine learning algorithms to construct a basin-wide debris flow activity quantity model, the quantity of debris flow activity in the future target period can be accurately predicted. Furthermore, S3 includes the following steps: Multi-source data characterizing debris flow activity were collected, and the multi-source data were processed at a unified spatiotemporal scale. Multi-source characteristic variables corresponding to each debris flow gully were extracted. The multi-source data included: static variables, which characterize the debris flow gully and its corresponding watershed characteristics, topography and sediment source conditions; and dynamic characteristic variables, which characterize the interannual variation characteristics of meteorological conditions related to the debris flow triggering process. Multi-source feature variables are normalized, and Pearson correlation coefficients are calculated based on the normalization results. Highly collinear and redundant variables are removed to form a feature set. Based on the annual debris flow occurrence sequence over many years, the normalized outbreak frequency of each debris flow gully was calculated. By merging and normalizing the outbreak frequency and feature set, a debris flow sample dataset containing historical debris flow outbreak information and multi-source disaster-causing background feature variables is constructed. Construct binary classification labels based on normalized outbreak frequency. Furthermore, machine learning algorithms were used to construct a prediction model for the probability of debris flow outbreaks in small watersheds. Using the multi-source feature variables corresponding to each debris flow gully in the debris flow sample dataset, a small watershed debris flow outbreak probability prediction model is trained to learn the mapping relationship between the multi-source feature variables and the probability of debris flow outbreak. Using a trained small watershed debris flow outbreak probability prediction model, we obtained the predicted probability of debris flow outbreaks at the single-gully scale within the target period.

[0013] The beneficial effects of the above-mentioned further solutions are: by processing multi-source data at a unified spatiotemporal scale, normalizing and optimizing features, data redundancy and high collinearity are effectively reduced, while improving the accuracy of debris flow outbreak probability prediction models.

[0014] Furthermore, the calculation of the normalized outbreak frequency of each debris flow gully includes the following steps: Based on the annual occurrence frequency of debris flows over many years, let the length of the historical record period be... Year, the The debris flow gully is in The number of mudslides occurring annually is ; Based on the Number of mudslides per year The cumulative number of debris flow gullies during the historical record period was calculated to be: ; Cumulative number of outbreaks The cumulative number of outbreaks was obtained by performing a logarithmic transformation to compress the scale. ; Cumulative number of outbreaks Normalization is performed, mapping the frequency to [0,1], to obtain the normalized debris flow outbreak frequency. :

[0015] .

[0016] The beneficial effect of the above-mentioned further scheme is that by performing logarithmic transformation and normalization on the cumulative number of debris flow outbreaks, the scale difference of debris flow outbreaks is resolved, so that the outbreak frequency of each debris flow gully has a consistent standard.

[0017] Furthermore, during the training process of the small watershed debris flow outbreak probability prediction model, based on binary classification labels... The debris flow sample dataset is divided into a positive sample set and a negative sample set. When the... The normalized outbreak frequency of each debris flow gully is greater than [missing information]. When the normalized outbreak frequency is [value], the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], ... the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], the corresponding sample is When this happens, the corresponding sample is marked as a negative sample, indicating that no debris flow event occurred during the historical record period; Based on the positive and negative sample sets, all debris flow samples are divided into K non-overlapping subsets according to the same ratio of positive to negative samples, so that the ratio of positive to negative samples in each subset is consistent with the original sample dataset. In each round of cross-validation, one subset is selected as the validation subsample of the current fold, and the K-1 subsamples other than the selected subset are used as training subsamples. A machine learning model is trained using the current training subsample to learn the mapping relationship between debris flow feature variables and debris flow outbreak probability. Multiple machine learning models are trained on the same training subsample, and the prediction results of each sub-machine learning model are weighted and averaged to obtain the debris flow outbreak probability prediction value of the current validation subsample. If the cross-validation reaches the threshold, the predicted debris flow probability obtained from multiple rounds of cross-validation is integrated to obtain the final predicted debris flow probability, thus completing the training of the small watershed debris flow probability prediction model.

[0018] The beneficial effects of the above-mentioned further scheme are: by cross-validation and weighted integration of multiple sub-models, the training and validation of the small watershed debris flow outbreak probability prediction model can be ensured on different subsets, effectively solving the sample imbalance problem.

[0019] Furthermore, the expression for the mapping relationship between the learned debris flow characteristic variables and the probability of debris flow outbreaks is as follows:

[0020]

[0021] in, This represents the normalized frequency of debris flow outbreaks. Indicates the first Predicted probability of mudslide outbreaks in each debris flow gully. Indicates the first The feature vector corresponding to each debris flow gully.

[0022] The beneficial effect of the above-mentioned further scheme is that by establishing a mapping relationship between debris flow characteristic variables and outbreak probability, the debris flow outbreak probability can be accurately calculated for each sample.

[0023] Furthermore, step S4 includes the following steps: Based on the predicted number of debris flow events in the entire basin, the total predicted number of debris flow events in the entire basin within the future target period is determined. Based on the predicted probability of debris flow outbreaks at the single-gully scale during the future target period, all predicted probability of debris flow outbreaks are sorted from largest to smallest and from highest to lowest to form a debris flow gully outbreak probability sorting sequence. Based on the predicted total number of occurrences, the debris flow outbreak probability value at the corresponding position in the debris flow gully outbreak probability ranking sequence is determined by mapping, and the debris flow outbreak probability value is used as the debris flow outbreak probability threshold. Considering the uncertainty of the prediction results, a tolerance adjustment is made to the debris flow outbreak probability threshold, and a range is set. p, making the probability of debris flow outbreaks within [ Screening was conducted on debris flow gullies within the specified range, among which, This represents the threshold for the probability of a debris flow outbreak. Screening for debris flow outbreak probability prediction values ​​greater than or equal to [ Within the target period, debris flow gullies are identified and determined as gullies where debris flows may occur in the future, in order to determine the spatial distribution prediction results of debris flow locations in the future target period.

[0024] The beneficial effects of the above-mentioned further scheme are: by comprehensively considering the number of debris flow activities in the entire basin and the probability of debris flow outbreaks in each gully, the location of debris flow occurrence can be accurately predicted, and by adjusting the tolerance of the uncertainty of the prediction results, it can better cope with changes under different climate scenarios.

[0025] The present invention also provides a system for predicting future debris flow activity, comprising: The first processing module is used to obtain the sequence of annual debris flow occurrences over many years; The second processing module is used to construct a debris flow activity quantity model for the entire basin based on the annual debris flow occurrence sequence over many years, and to use the debris flow activity quantity model for the entire basin to predict the total number of debris flow occurrences in the entire basin within the future target period, thereby obtaining the debris flow activity quantity prediction results for the entire basin. The third processing module is used to collect multi-source data characterizing debris flow activity and construct a small watershed debris flow outbreak probability prediction model by combining the annual debris flow occurrence sequence over many years. Using the small watershed debris flow outbreak probability prediction model, the probability prediction results of debris flow outbreaks at the single gully scale within the future target period are obtained. The fourth processing module is used to determine the debris flow outbreak probability threshold based on the predicted number of debris flow activities in the entire watershed and the predicted probability of debris flow outbreaks in small watersheds, and to obtain the spatial distribution prediction results of debris flow occurrence locations in the future target period.

[0026] The beneficial effects of this invention are as follows: Based on the historical debris flow activity catalog database, this invention combines the disaster-causing background and triggering factors of multi-source debris flows to construct prediction models for the number of debris flow activities in the whole watershed and the probability of debris flow outbreaks in small watersheds. By synergistically constraining the probability of single-gully outbreaks with the number of regional-scale activities, this invention can predict the number and location of debris flows in the future target period. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention.

[0028] Figure 2 This is a schematic diagram of the method framework of the present invention.

[0029] Figure 3 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0030] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0031] Example 1 like Figure 1 and Figure 2 As shown, this invention provides a method for predicting future debris flow activity, the implementation of which is as follows: S1. Construction of a historical debris flow activity cataloging database: Obtaining the annual sequence of debris flow occurrences over many years, the implementation method is as follows: Based on the interpretation results of interannual multi-temporal remote sensing images, field survey data, and records of historical debris flow disaster events, debris flow gullies are identified and labeled. Based on the identification and annotation results, debris flow events that occurred year by year in each debris flow gully during the historical record period were statistically analyzed to form a multi-year sequence of debris flow occurrences and to construct a catalog database of historical debris flow activities.

[0032] In this embodiment, hydro-topographic analysis is performed on digital elevation model (DEM) data to extract confluence paths and divide watershed units. The watershed units are verified and supplemented by combining multi-temporal remote sensing images, field survey data, and existing national debris flow databases to identify debris flow gullies. Based on this, debris flow events occurring year by year in each debris flow gully are statistically analyzed during the historical record period to form a multi-year debris flow occurrence sequence and construct a debris flow historical activity catalog database.

[0033] In this embodiment, the interannual multi-temporal remote sensing images include, but are not limited to, satellite remote sensing images (such as Landsat, Sentinel, etc.) and UAV aerial images. At the same time, the interpretation results are verified and supplemented by field survey data and historical debris flow disaster records.

[0034] In the specific implementation process, hydrological and topographical analysis is first conducted based on the Digital Elevation Model (DEM). Through operations such as depression filling, flow direction extraction, and flow accumulation, debris flow watershed units are delineated to determine the potential debris flow gully range. Next, combined with multi-temporal remote sensing imagery, the debris flow is visually interpreted based on the differences in hue, texture, and morphology of the debris flow depositional fan relative to the surrounding surface. Dynamic information such as the debris flow source area, flow area, and deposition area within the watershed in the imagery is used to assist in identifying signs of debris flow activity. Finally, the interpretation results are compared, verified, and corrected with field survey data and existing debris flow databases to complete the identification and labeling of debris flow gullies. On this basis, combined with the interpretation results of interannual remote sensing imagery and historical debris flow disaster records, debris flow events occurring year by year in each debris flow gully during the historical record period are statistically analyzed to form a multi-year annual debris flow occurrence sequence, thereby constructing a catalog database of historical debris flow activity.

[0035] Specifically, let the length of the historical record period be... T In the year, for the first A debris flow gully, in its first t The number of mudslide events per year is Then, the sequence of annual debris flow occurrences in each debris flow gully over the historical record period can be represented as { }

[0036] S2. Prediction of the Number of Debris Flow Activities in the Entire Basin: Based on the annual debris flow occurrence sequence over many years, a model for the number of debris flow activities in the entire basin is constructed. This model is then used to predict the total number of debris flow occurrences in the entire basin within the target period, yielding the predicted number of debris flow activities in the entire basin. The implementation method is as follows: Based on the annual sequence of debris flow occurrences over many years, the number of interannual debris flow activities within the historical record period is calculated. We collected precipitation and temperature data related to debris flow activities in historical and future scenarios, performed time-scale uniform processing on the collected data, and extracted meteorological characteristic variables. Using the interannual activity of debris flows within the historical record period as the modeling target variable and the meteorological characteristic variables corresponding to the historical and future periods as input characteristic variables, a machine learning algorithm is used to construct a model of debris flow activity in the entire watershed. A model for the quantity of debris flow activities across the entire basin was trained, and the trained model was used to predict the number of debris flows in the entire basin within a future target period, thus obtaining the predicted results of debris flow activity quantity across the entire basin.

[0037] In this embodiment, based on the annual debris flow occurrence sequence over many years, the interannual number of debris flows during the historical record period is calculated; using the debris flow activity quantity index as the modeling target variable, and combining historical and future precipitation and temperature data, a debris flow activity quantity model for the entire basin is constructed to predict the total number of debris flows in the future target period of the entire basin, thus obtaining the regional scale debris flow activity quantity prediction results.

[0038] In this embodiment, the quantitative model of debris flow activity across the entire basin adopts the Generalized Additive Model (GAM), whose mathematical expression is:

[0039] in, Indicates the first Predicted annual number of debris flow events across the entire watershed. Represents the model constant term. Indicates the first Year One climate characteristic variable, This represents a nonlinear smoothing function corresponding to the meteorological characteristic variables. This indicates the number of meteorological characteristic variables.

[0040] Based on the annual debris flow occurrence sequence over many years, the interannual number of debris flows within the historical record period is calculated. Specifically, based on the annual debris flow occurrence sequence over many years { }, calculate the total number of debris flows occurring in the entire watershed each year, i.e. = , =1,2,……,T; where, Indicates the first Total number of debris flows in the entire basin in a year Indicates the first Debris flow gully The number of times it occurs, This represents the total number of debris flow gullies in the entire watershed. Then, the total number of debris flow occurrences for each year is sorted by year to obtain the interannual debris flow activity sequence for the entire watershed. }

[0041] We collected historical and future precipitation and temperature data related to debris flow activity, unified the time scales of the data, and extracted meteorological characteristic variables for modeling. These meteorological characteristic variables include, but are not limited to, annual average precipitation, maximum precipitation over three consecutive days (Rx3day), maximum precipitation over five consecutive days (Rx5day), R95p (95th percentile precipitation), R99p (99th percentile precipitation), annual average temperature, daily maximum temperature, and diurnal temperature range.

[0042] Using debris flow activity quantity as the modeling target variable and historical and future meteorological characteristic variables as input characteristic variables, a machine learning algorithm is used to construct a basin-wide debris flow activity quantity prediction model. During the training process of the basin-wide debris flow activity quantity prediction model, the following indicators, including but not limited to, are used to evaluate the model performance: coefficient of determination (COP). ), root mean square error (RMSE) and mean absolute error (MAE).

[0043] The coefficient of determination is used to measure the goodness of fit of a model, and its calculation formula is as follows: ; in, Indicates the first The number of historical debris flow events in the entire basin in a given year The model predicts the first... The number of debris flow activities in the entire basin this year This represents the average number of historical events.

[0044] The root mean square error (RMSE) measures the average deviation between the predicted results of a basin-wide debris flow activity prediction model and the actual values. A smaller RMSE value indicates higher prediction accuracy. The formula for calculating RMSE is: ; in, This represents the total number of samples.

[0045] This represents the average absolute value of the error between the predicted and actual values ​​of debris flow activity prediction models for the entire watershed. A smaller value indicates a smaller prediction deviation. The calculation formula is: .

[0046] After training the model for predicting the number of debris flows in the entire basin, the model is used to predict the number of debris flows in the entire basin during the target period, thus obtaining the prediction results for the number of debris flows in the entire basin.

[0047] S3. Small watershed debris flow outbreak probability prediction: Collect multi-source data characterizing debris flow activity and construct a small watershed debris flow outbreak probability prediction model by combining the annual debris flow occurrence sequence over many years. Using the small watershed debris flow outbreak probability prediction model, obtain the predicted debris flow outbreak probability results at the single-gully scale within the future target period. The implementation method is as follows: Multi-source data characterizing debris flow activity were collected, and the multi-source data were processed at a unified spatiotemporal scale. Multi-source characteristic variables corresponding to each debris flow gully were extracted. The multi-source data included: static variables, which characterize the debris flow gully and its corresponding watershed characteristics, topography and sediment source conditions; and dynamic characteristic variables, which characterize the interannual variation characteristics of meteorological conditions related to the debris flow triggering process. Multi-source feature variables are normalized, and Pearson correlation coefficients are calculated based on the normalization results. Highly collinear and redundant variables are removed to form a feature set. Based on the annual debris flow occurrence sequence over many years, the normalized outbreak frequency of each debris flow gully was calculated. By merging and normalizing the outbreak frequency and feature set, a debris flow sample dataset containing historical debris flow outbreak information and multi-source disaster-causing background feature variables is constructed. Construct binary classification labels based on normalized outbreak frequency. Furthermore, machine learning algorithms were used to construct a prediction model for the probability of debris flow outbreaks in small watersheds. Using the multi-source feature variables corresponding to each debris flow gully in the debris flow sample dataset, a small watershed debris flow outbreak probability prediction model is trained to learn the mapping relationship between the multi-source feature variables and the probability of debris flow outbreak. Using a trained small watershed debris flow outbreak probability prediction model, we obtained the predicted probability of debris flow outbreaks at the single-gully scale within the target period.

[0048] In this embodiment, the TabNet algorithm is used to predict the probability of debris flow outbreaks, and its expression is as follows:

[0049] in, Indicates the first The predicted probability of a sample represents the probability of a debris flow outbreak; This represents the standardized feature vector; The TabNet model, representing the probability prediction model for small watershed debris flow outbreaks after training, includes the network structure, feature selection, and decision-making process, with parameters of... .

[0050] The TabNet model, which predicts the probability of debris flow outbreaks in small watersheds, is trained using a cross-entropy loss function. The goal is to minimize this loss function to optimize the model's parameters. Its cross-entropy loss function can be expressed as:

[0051] in, Indicates the first The predicted probability of each sample; Indicates the first The true label of each sample; This represents the parameters of the TabNet model, including the weights and biases of all neural network layers.

[0052] The parameters of TabNet, a small watershed debris flow outbreak probability prediction model, were optimized using the gradient descent method. That is, minimizing the loss function :

[0053] in, Indicates the model parameters for the current iteration; Indicates the learning rate; This indicates the loss function relative to the parameters. The gradient.

[0054] In this embodiment, during the training process of the small watershed debris flow outbreak probability prediction model, based on binary classification labels... The debris flow sample dataset is divided into a positive sample set and a negative sample set. When the... The normalized outbreak frequency of each debris flow gully is greater than [missing information]. When the normalized outbreak frequency is [value], the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], ... the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], the corresponding sample is When a debris flow event occurs, the corresponding sample is labeled as a negative sample, indicating that no debris flow event has occurred during the historical record period. The corresponding sample is a combination of the feature vector and label corresponding to each debris flow gully during the historical observation period. Based on the positive and negative sample sets, all debris flow samples are divided into K non-overlapping subsets according to the same positive-to-negative sample ratio, ensuring that the ratio of positive to negative samples in each subset is consistent with the original sample dataset (the original sample dataset refers to the complete dataset containing all debris flow gully samples, where each sample includes the multi-source feature vector of the debris flow gully and its binary label corresponding to its historical normalized outbreak frequency). In each round of cross-validation, one subset is selected. The set of K-1 subsamples other than the selected subset is used as the training subsample. The machine learning model is trained using the current training subsample to learn the mapping relationship between debris flow characteristic variables and debris flow outbreak probability. Multiple machine learning models are trained on the same training subsample, and the prediction results of each sub-machine learning model are weighted and averaged to obtain the debris flow outbreak probability prediction value of the current validation subsample. If the cross-validation reaches the threshold, the debris flow outbreak probability prediction values ​​obtained in multiple rounds of cross-validation are integrated to obtain the final debris flow outbreak probability prediction value, thus completing the training of the small watershed debris flow outbreak probability prediction model.

[0055] In this embodiment, a logarithmic transformation normalization process is performed on the annual debris flow occurrence sequence over many years to calculate the normalized debris flow outbreak frequency. This normalized frequency is then merged with the final modeling feature set to construct a debris flow sample dataset. A binary classification label is constructed based on the normalized debris flow outbreak frequency, where, when the... The normalized outbreak frequency of each debris flow gully is greater than [missing information]. When the normalized outbreak frequency is [value], the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], ... the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], the corresponding sample is At that time, the corresponding samples were marked as negative samples, indicating that no debris flow events had occurred during the historical record period; multi-source disaster-causing background data affecting debris flow activity were collected, and the multi-source data underwent unified spatiotemporal scale transformation and preprocessing to extract debris flow activity-related feature variables; the feature variables were normalized to obtain an initial set of feature variables for modeling; the Pearson correlation coefficients among the feature variables in the initial set of feature variables were calculated, and highly collinear redundant variables were removed to form the final modeling feature set; the multi-source feature variable vectors corresponding to each debris flow gully in the final modeling feature set were used as the basis for modeling. As input to the small watershed debris flow outbreak probability prediction model, a machine learning algorithm is used for training to obtain the future debris flow outbreak probability of each debris flow gully, with the value ranging from 0 to 1.

[0056] In this embodiment, considering that debris flow activity is influenced by both relatively stable topographic and source conditions and time-varying meteorological conditions, multi-source characteristic variables affecting debris flow activity are collected and extracted from two aspects: relatively stable static variables and time-varying dynamic characteristic variables. Static characteristic variables are mainly used to characterize debris flow gullies and their corresponding watershed features, topography, and source conditions, including but not limited to watershed area, channel gradient, relative elevation difference, Melthon ratio, elevation variation coefficient, topographic relief, surface roughness, slope, aspect, average elevation, loose material volume, shape factor, planar curvature, profile curvature, topographic humidity index (TWI), and topographic power index (SPI). Dynamic characteristic variables are mainly used to characterize the interannual variation of meteorological conditions related to debris flow triggering processes, including but not limited to annual average precipitation, precipitation variation coefficient, maximum precipitation over three consecutive days (Rx3day), maximum precipitation over five consecutive days (Rx5day), R95p (95th percentile precipitation), R99p (99th percentile precipitation), annual average temperature, daily maximum temperature, diurnal temperature range, and annual positive cumulative temperature. Static characteristic variables are mainly extracted from watershed analysis results and digital elevation model (DEM) data, while dynamic characteristic variables are derived from climate model datasets (such as CMIP6). Due to differences in the time scale, spatial resolution, and data sources of the above data, a unified spatiotemporal scale transformation is first required. Subsequently, characteristic variables corresponding to each debris flow gully are obtained through feature extraction, and all characteristic variables are normalized to obtain the set of characteristic variables used for modeling.

[0057] Because strong linear correlations may exist among multi-source feature variables, directly inputting all features into a small-basin debris flow outbreak probability prediction model for training can easily lead to the model struggling to distinguish the independent contributions of each feature during training, potentially resulting in overfitting. Therefore, Pearson correlation analysis is performed on the modeling feature variables to calculate the correlation coefficient between any two features. The formula for the Pearson correlation coefficient is: r = ; in, r This represents the Pearson correlation coefficient. , They represent the characteristic variables respectively. , The mean, and All represent data points in the sample. n Indicates the number of samples.

[0058] Typically, the correlation coefficient r The value of is between -1 and +1. rIf >0, it indicates a positive correlation between the two feature variables; if r If <0, it indicates a negative correlation between the two feature variables; if r =0 indicates that there is no linear correlation between the two feature variables; correlation coefficient r The larger the absolute value, the stronger the linear correlation between the two feature variables.

[0059] To avoid multicollinearity issues caused by highly correlated feature variables during the training of debris flow outbreak probability prediction models in small watersheds, a correlation coefficient threshold is usually set (e.g., When the absolute value of the correlation coefficient between two feature variables exceeds 0.8, one of the feature variables is selectively removed, and the final modeling feature set used for model training is obtained.

[0060] Because different debris flow gullies vary significantly in terms of topographic conditions, water supply, and debris source scale, directly using the cumulative occurrence count as the regression target can easily overestimate or underestimate the probability of debris flow outbreaks. Furthermore, many debris flow gullies did not experience any outbreaks during the historical record period. In one implementation, based on the multi-year debris flow occurrence count sequence, a logarithmic transformation is used to normalize the data, and the normalized outbreak frequency of each debris flow gully is calculated as a subsequent modeling variable.

[0061] Specifically, let the length of the historical record period be... T Year, the The debris flow gully is in The number of mudslides occurring annually is The cumulative number of mudslides in this gully during the historical record period is: = ; Based on this, the cumulative number of outbreaks Scale compression is performed using a logarithmic transformation to obtain = ,right By performing maximum value normalization and mapping it to [0,1], the normalized debris flow outbreak frequency is obtained: ; Next, the normalized debris flow outbreak frequency was merged with the final modeling feature set to construct a debris flow sample dataset containing historical debris flow outbreak information and multi-source disaster-causing background characteristic variables. This debris flow sample dataset serves as the input data basis for subsequent small watershed debris flow outbreak probability prediction models.

[0062] Based on the normalized debris flow outbreak frequency, a binary classification label is constructed. Using machine learning algorithms, a small watershed debris flow outbreak probability prediction model is built to obtain the predicted debris flow outbreak probability at the single-gully scale within the future target period.

[0063] Specifically, let the first The normalized outbreak frequency of each debris flow gully is Then binary category labels Defined as ; in, This indicates that at least one debris flow event occurred within the historical record period, and is considered a positive sample. This indicates that no mudslide events occurred during the historical record period, and is considered a negative sample.

[0064] The multi-source feature variable vectors corresponding to each debris flow gully in the debris flow sample dataset As input to the small watershed debris flow outbreak probability prediction model, a machine learning model is used for training to learn the mapping relationship between feature variables and the probability of debris flow outbreaks: ; in, This represents the normalized frequency of debris flow outbreaks. Indicates the first Predicted probability of mudslide outbreaks in each debris flow gully. Indicates the first The feature vector corresponding to each debris flow gully.

[0065] In order to reduce the impact of sample imbalance and randomness in data partitioning on the prediction results of the small watershed debris flow outbreak probability prediction model during the training process, this embodiment adopts the Stratified K-Fold Cross Validation strategy based on the debris flow sample dataset.

[0066] Specifically, the debris flow sample dataset is first divided into a positive sample set and a negative sample set based on the binary classification labels. Then, all samples are divided into K non-overlapping subsets according to the same ratio of positive to negative samples, so that the ratio of positive to negative samples in each subset is consistent with the original sample dataset. The value of K is set according to the sample size and computing resources, for example, K=5 or K=10.

[0067] In each round of cross-validation, one subset is selected as the validation subsample for the current fold, and the remaining K... A subset of samples is used as the training subset. A machine learning model is trained based on this training subset to learn the mapping relationship between debris flow characteristic variables and debris flow outbreak probabilities. Simultaneously, a bagging ensemble strategy is introduced in each fold cross-validation process. This involves training multiple machine learning models on the same training subset and then taking a weighted average of the predictions from each model to obtain the debris flow outbreak probability prediction for the validation subset at that fold. This process is repeated K times for cross-validation, ensuring that each subset serves as a validation subset in the model evaluation. Furthermore, the predicted probability results obtained during the K times cross-validation are ensembled, meaning the arithmetic mean of the predicted probabilities obtained for the same sample at different folds is taken as the final debris flow outbreak probability prediction for that sample.

[0068] In the performance evaluation phase of the small watershed debris flow outbreak probability prediction model, the evaluation metrics of the model are calculated during each fold cross-validation process, and the average value of the evaluation results of each fold is taken as the overall performance evaluation result of the machine learning model. The evaluation metrics include accuracy, precision, recall, F1 score, and area undercurve (AUC) of the receiver operating curve (ROC).

[0069] Accuracy measures the proportion of correct predictions made by the model overall, and its calculation formula is as follows: ; Wherein, TP (True Positive) represents the number of samples where a debris flow actually occurred and the model predicted that a debris flow would occur; TN (True Negative) represents the number of samples where a debris flow did not actually occur and the model predicted that a debris flow would not occur; FP (False Positive) represents the number of samples where a debris flow did not actually occur but the model predicted that a debris flow would occur; and FN (False Negative) represents the number of samples where a debris flow actually occurred but the model predicted that a debris flow would not occur.

[0070] Accuracy reflects the proportion of debris flows that actually occur in the sample predicted by the model. The calculation formula is as follows: .

[0071] Recall rate represents the proportion of samples from all actual debris flows that were successfully identified as debris flow outbreaks by the model. Its calculation formula is: .

[0072] The F1 score is the harmonic mean of precision and recall, and its calculation formula is as follows: ; Precision and Recall are the two metrics, respectively.

[0073] Based on the output probability and historical records of a small watershed debris flow outbreak probability prediction model, a receiver operating characteristic (ROC) curve was plotted, and the area under the curve (AUC) was calculated as an evaluation index for the model to distinguish between debris flow outbreaks and non-outbreaks. The AUC value ranges from 0 to 1, with a larger value indicating a stronger discriminative ability of the model; when the AUC is close to 0.5, it indicates that the model's discriminative performance is close to random guessing.

[0074] S4. Prediction of Future Debris Flow Locations: Based on the predicted number of debris flow activities across the entire basin and the predicted probability of debris flow outbreaks, a debris flow outbreak probability threshold is determined, yielding a spatial distribution prediction of debris flow locations within the target future period. The implementation method is as follows: Based on the predicted number of debris flow events in the entire basin, the total predicted number of debris flow events in the entire basin within the future target period is determined. Based on the predicted probability of debris flow outbreaks at the single-gully scale during the future target period, all predicted probability of debris flow outbreaks are sorted from largest to smallest and from highest to lowest to form a debris flow gully outbreak probability sorting sequence. Based on the predicted total number of occurrences, the debris flow outbreak probability value at the corresponding position in the debris flow gully outbreak probability ranking sequence is determined by mapping, and the debris flow outbreak probability value is used as the debris flow outbreak probability threshold. Considering the uncertainty of the prediction results, a tolerance adjustment is made to the debris flow outbreak probability threshold, and a range is set. p, which makes the probability of a debris flow outbreak within [ Screening was conducted on debris flow gullies within the specified range, among which, This represents the threshold for the probability of a debris flow outbreak. Screening for debris flow outbreak probability prediction values ​​greater than or equal to [ Within the target period, debris flow gullies are identified and determined as gullies where debris flows may occur in the future, in order to determine the spatial distribution prediction results of debris flow locations in the future target period.

[0075] In this embodiment, based on the predicted probability of debris flow outbreaks in a small watershed, all debris flow gullies are sorted according to their predicted outbreak probabilities. Combining the predicted number of future debris flow activities in the entire watershed, the debris flow outbreak probability value at the corresponding position in the sorted sequence is determined based on the predicted number mapping, and used as an initial threshold. Then, debris flow gullies with a debris flow outbreak probability greater than or equal to the threshold are selected, and the threshold is allowed to be adjusted within a certain tolerance range, thereby obtaining the spatial distribution prediction result of debris flow occurrence locations in the future target period.

[0076] In this embodiment, based on the predicted number of debris flow activities across the entire basin, a debris flow outbreak probability threshold is determined, and the spatial distribution prediction of debris flow locations within the future target period is obtained. Specifically, based on the output of the basin-wide debris flow activity prediction model, the total predicted number of debris flows across the entire basin within the future target period is determined. Based on the small basin debris flow outbreak probability prediction model, the predicted outbreak probability value for each debris flow gully within the future target period is obtained. The predicted outbreak probability values ​​of all debris flow gullies are sorted from largest to smallest and from highest to lowest to form a debris flow gully outbreak probability ranking sequence. The predicted outbreak probability value corresponding to the Nth debris flow gully in the ranking sequence is used as the initial debris flow outbreak probability threshold p. Considering the uncertainty of the prediction results, a tolerance adjustment is made to the outbreak probability threshold p. In this embodiment, a certain range is set above and below the threshold based on the model confidence interval. p, such that the probability of outbreak is in [ The debris flow gullies within the specified range are screened, and those with a predicted outbreak probability greater than or equal to the threshold value are identified as gullies that may experience debris flows in the future target period, in order to determine the spatial distribution prediction results of debris flow occurrence locations in the future target period.

[0077] In summary, this invention can determine the location of debris flows in the future target period while ensuring a reasonable number of debris flow activities at the regional scale. This provides a scientific basis for the analysis of the spatiotemporal distribution of debris flow activities, as well as for future debris flow risk assessment, monitoring and early warning, and disaster prevention and mitigation decision-making.

[0078] Example 2 like Figure 3 As shown, the present invention provides a future debris flow activity prediction system for executing the future debris flow activity prediction method described in Example 1, comprising: The first processing module is used to obtain the sequence of annual debris flow occurrences over many years; The second processing module is used to construct a debris flow activity quantity model for the entire basin based on the annual debris flow occurrence sequence over many years, and to use the debris flow activity quantity model for the entire basin to predict the total number of debris flow occurrences in the entire basin within the future target period, thereby obtaining the debris flow activity quantity prediction results for the entire basin. The third processing module is used to collect multi-source data characterizing debris flow activity and construct a small watershed debris flow outbreak probability prediction model by combining the annual debris flow occurrence sequence over many years. Using the small watershed debris flow outbreak probability prediction model, the probability prediction results of debris flow outbreaks at the single gully scale within the future target period are obtained. The fourth processing module is used to determine the debris flow outbreak probability threshold based on the predicted number of debris flow activities in the entire watershed and the predicted probability of debris flow outbreaks in small watersheds, and to obtain the spatial distribution prediction results of debris flow occurrence locations in the future target period.

[0079] In this embodiment, the functional units can be divided according to the future debris flow activity prediction method. For example, each function can be divided into its own functional units, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this invention is illustrative and only represents a logical division; other division methods may be used in actual implementation.

[0080] In this embodiment, the future debris flow activity prediction system, in order to realize the principle and beneficial effects of the future debris flow activity prediction method, includes hardware structures and / or software modules corresponding to the execution of various functions. Those skilled in the art should readily recognize that, in conjunction with the illustrative units and algorithm steps described in the embodiments disclosed herein, the present invention can be implemented in hardware and / or a combination of hardware and computer software. Whether a function is executed by hardware or computer software depends on the specific application and design constraints of the technical solution. Different methods can be used to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A method for predicting future debris flow activity, characterized in that, Includes the following steps: S1. Construction of a historical debris flow activity cataloging database: obtaining a sequence of debris flow occurrences over many years; S2. Prediction of the number of debris flow activities in the whole basin: Based on the annual debris flow occurrence sequence over many years, a debris flow activity quantity model for the whole basin is constructed, and the total number of debris flow occurrences in the whole basin within the target period is predicted using the debris flow activity quantity model for the whole basin, so as to obtain the prediction result of the number of debris flow activities in the whole basin. S3. Small watershed debris flow outbreak probability prediction: Collect multi-source data characterizing debris flow activity and construct a small watershed debris flow outbreak probability prediction model by combining the annual debris flow occurrence sequence over many years. Using the small watershed debris flow outbreak probability prediction model, obtain the debris flow outbreak probability prediction results at the single gully scale within the future target period. S4. Prediction of future debris flow locations: Based on the prediction results of the number of debris flow activities in the entire basin and the prediction results of debris flow outbreak probability, the debris flow outbreak probability threshold is determined, and the spatial distribution prediction results of debris flow locations within the future target period are obtained.

2. The method for predicting future debris flow activity according to claim 1, characterized in that, S1 includes the following steps: Based on the interpretation results of interannual multi-temporal remote sensing images, field survey data, and records of historical debris flow disaster events, debris flow gullies are identified and labeled. Based on the identification and annotation results, debris flow events that occurred year by year in each debris flow gully during the historical record period were statistically analyzed to form a multi-year sequence of debris flow occurrences and to construct a catalog database of historical debris flow activities.

3. The method for predicting future debris flow activity according to claim 1, characterized in that, S2 includes the following steps: Based on the annual sequence of debris flow occurrences over many years, the number of interannual debris flow activities within the historical record period is calculated. We collected precipitation and temperature data related to debris flow activities in historical and future scenarios, performed time-scale uniform processing on the collected data, and extracted meteorological characteristic variables. Using the interannual activity of debris flows within the historical record period as the modeling target variable and the meteorological characteristic variables corresponding to the historical and future periods as input characteristic variables, a machine learning algorithm is used to construct a model of debris flow activity in the entire watershed. A model for the quantity of debris flow activities across the entire basin was trained, and the trained model was used to predict the number of debris flows in the entire basin within a future target period, thus obtaining the predicted results of debris flow activity quantity across the entire basin.

4. The method for predicting future debris flow activity according to claim 1, characterized in that, S3 includes the following steps: Multi-source data characterizing debris flow activity were collected, and the multi-source data were processed at a unified spatiotemporal scale. Multi-source characteristic variables corresponding to each debris flow gully were extracted. The multi-source data included: static variables, which characterize the debris flow gully and its corresponding watershed characteristics, topography and sediment source conditions; and dynamic characteristic variables, which characterize the interannual variation characteristics of meteorological conditions related to the debris flow triggering process. Multi-source feature variables are normalized, and Pearson correlation coefficients are calculated based on the normalization results. Highly collinear and redundant variables are removed to form a feature set. Based on the annual debris flow occurrence sequence over many years, the normalized outbreak frequency of each debris flow gully was calculated. By merging and normalizing the outbreak frequency and feature set, a debris flow sample dataset containing historical debris flow outbreak information and multi-source disaster-causing background feature variables is constructed. Construct binary classification labels based on normalized outbreak frequency. Furthermore, machine learning algorithms were used to construct a prediction model for the probability of debris flow outbreaks in small watersheds. Using the multi-source feature variables corresponding to each debris flow gully in the debris flow sample dataset, a small watershed debris flow outbreak probability prediction model is trained to learn the mapping relationship between the multi-source feature variables and the probability of debris flow outbreak. Using a trained small watershed debris flow outbreak probability prediction model, we obtained the predicted probability of debris flow outbreaks at the single-gully scale within the target period.

5. The method for predicting future debris flow activity according to claim 4, characterized in that, The calculation of the normalized outbreak frequency of each debris flow gully includes the following steps: Based on the annual occurrence frequency of debris flows over many years, let the length of the historical record period be... Year, the The debris flow gully is in The number of mudslides occurring annually is ; Based on the Number of mudslides per year The cumulative number of debris flow gullies during the historical record period was calculated to be: ; Cumulative number of outbreaks The cumulative number of outbreaks was obtained by performing a logarithmic transformation to compress the scale. ; Cumulative number of outbreaks Normalization is performed, mapping the frequency to [0,1], to obtain the normalized debris flow outbreak frequency. : 。 6. The method for predicting future debris flow activity according to claim 4, characterized in that, During the training process of the debris flow outbreak probability prediction model in a small watershed, based on binary classification labels The debris flow sample dataset is divided into a positive sample set and a negative sample set. When the... The normalized outbreak frequency of each debris flow gully is greater than [missing information]. When the normalized outbreak frequency is [value], the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], ... the corresponding sample is marked as a positive sample, indicating that at least one debris flow event occurred within the historical record period; when the normalized outbreak frequency is [value], the corresponding sample is When a debris flow event occurs, the corresponding sample is marked as a negative sample, indicating that no debris flow event has occurred during the historical record period; where the corresponding sample is a combination of the feature vector and label corresponding to each debris flow gully during the historical observation period; Based on the positive and negative sample sets, all debris flow samples are divided into K non-overlapping subsets according to the same ratio of positive to negative samples, so that the ratio of positive to negative samples in each subset is consistent with the original sample dataset. In each round of cross-validation, one subset is selected as the validation subsample of the current fold, and the K-1 subsamples other than the selected subset are used as training subsamples. A machine learning model is trained using the current training subsample to learn the mapping relationship between debris flow feature variables and debris flow outbreak probability. Multiple machine learning models are trained on the same training subsample, and the prediction results of each sub-machine learning model are weighted and averaged to obtain the debris flow outbreak probability prediction value of the current validation subsample. If the cross-validation reaches the threshold, the predicted debris flow probability obtained from multiple rounds of cross-validation is integrated to obtain the final predicted debris flow probability, thus completing the training of the small watershed debris flow probability prediction model.

7. The method for predicting future debris flow activity according to claim 6, characterized in that, The expression for the mapping relationship between the learned debris flow characteristic variables and the probability of debris flow outbreaks is as follows: in, This represents the normalized frequency of debris flow outbreaks. Indicates the first Predicted probability of mudslide outbreaks in each debris flow gully. Indicates the first The feature vector corresponding to each debris flow gully.

8. The method for predicting future debris flow activity according to claim 1, characterized in that, S4 includes the following steps: Based on the predicted number of debris flow events in the entire basin, the total predicted number of debris flow events in the entire basin within the future target period is determined. Based on the predicted probability of debris flow outbreaks at the single-gully scale during the future target period, all predicted probability of debris flow outbreaks are sorted from largest to smallest and from highest to lowest to form a debris flow gully outbreak probability sorting sequence. Based on the predicted total number of occurrences, the debris flow outbreak probability value at the corresponding position in the debris flow gully outbreak probability ranking sequence is determined by mapping, and the debris flow outbreak probability value is used as the debris flow outbreak probability threshold. Considering the uncertainty of the prediction results, a tolerance adjustment is made to the debris flow outbreak probability threshold, and a range is set. p, making the probability of debris flow outbreaks within [ Screening was conducted on debris flow gullies within the specified range, among which, This represents the threshold for the probability of a debris flow outbreak. Screening for debris flow outbreak probability prediction values ​​greater than or equal to [ Within the specified range, debris flow gullies are identified and determined as gullies where debris flows may occur in the future target period, in order to determine the spatial distribution prediction results of debris flow occurrence locations in the future target period.

9. A future debris flow activity prediction system, used to execute the future debris flow activity prediction method according to any one of claims 1-8, characterized in that, include: The first processing module is used to obtain the sequence of annual debris flow occurrences over many years; The second processing module is used to construct a debris flow activity quantity model for the entire basin based on the annual debris flow occurrence sequence over many years, and to use the debris flow activity quantity model for the entire basin to predict the total number of debris flow occurrences in the entire basin within the future target period, thereby obtaining the debris flow activity quantity prediction results for the entire basin. The third processing module is used to collect multi-source data characterizing debris flow activity and construct a small watershed debris flow outbreak probability prediction model by combining the annual debris flow occurrence sequence over many years. Using the small watershed debris flow outbreak probability prediction model, the probability prediction results of debris flow outbreaks at the single gully scale within the future target period are obtained. The fourth processing module is used to determine the debris flow outbreak probability threshold based on the predicted number of debris flow activities in the entire watershed and the predicted probability of debris flow outbreaks in small watersheds, and to obtain the spatial distribution prediction results of debris flow occurrence locations in the future target period.

Images