Method and system for identifying rainfall-driven snow melting disaster in high-cold and data-lacking mountainous area, and electronic equipment

Abstract

The invention relates to the technical field of snow melting disasters, and provides an identification method and system for rainfall-driven snow melting disasters in an alpine and cold data-deficient mountainous area and electronic equipment, and the method comprises the steps: collecting meteorological observation sequence data in a set time period of the alpine and cold mountainous area; performing multi-dimensional feature extraction on the meteorological observation sequence data to obtain multi-dimensional meteorological feature sequence data; inputting the multi-dimensional meteorological sequence feature data into a pre-trained snow surface rain machine learning discrimination model to obtain a discrimination result of a single-point snow surface rain snow-melting disaster in the high-cold region; and based on the judgment result of the single-point snow surface rain snow melting disaster in the high and cold area and the simulation result of the regional climate mode weather, obtaining the judgment result of the regional snow surface rain event in the set time period in the high and cold mountainous area.

Description

Technical Field

[0001] The present invention relates to the technical field of snowmelt disasters, and in particular to a method, system and electronic equipment for identifying snowmelt disasters driven by rainfall in high-altitude and cold mountainous areas with insufficient data. Background Art

[0002] The phenomenon of rainfall falling on snow is called snow surface rain event. Snow surface rain has the superposition effect of rainfall and snowmelt. The rapid snowmelt process caused by it can easily cause disasters such as large-scale floods, avalanches and landslides, posing a huge threat to regional water resources, ecological environment and socio-economic development, and bringing great instability to regional socio-economic construction and sustainable ecological environment development. Studying the spatiotemporal changes of snowmelt disasters driven by rainfall under climate change has theoretical and practical significance for regional snowmelt flood prediction and water resources management.

[0003] The process of snow surface rain disaster is complex. Rainfall will increase the water content of the snow layer, reduce the albedo, increase the energy of the snow layer, and enhance the metamorphic effect; it will cause the snow particle size to become coarser, the internal structure to change, the snow to sink and compact, the thermal conductivity to increase, and the snow to melt faster. At the same time, rainwater penetrates into the snow layer or freezes on the exposed surface, thus affecting vegetation growth, animal survival and surface heat exchange. The snow melting rate is closely related to the amount of snow, and is extremely sensitive to changes in temperature and precipitation form (rain or snow). For example, warming has caused climate volatility to increase, resulting in intensified water cycles in mountainous areas, water and heat redistribution processes tending to be extreme, and snowfall rates, snow characteristics and melting patterns have changed significantly. Using land surface models / hydrological models to investigate the changing process of large-scale snow surface rain and its snowmelt effect has become a hot topic in recent years.

[0004] Snow surface rain is a complex process in which multiple hydrological and meteorological variables work together. Affected by the combined effects of precipitation form, temperature and snow accumulation, the combination of threshold conditions based on rainfall and snow accumulation has become the main method for determining the occurrence of snow surface rain. At present, the critical temperature threshold method is mainly used to estimate the precipitation type, separate rain and snow, and identify whether snow surface rain events occur. At present, there is a lack of unified standards for the identification of snow surface rain events with disaster potential and the combination of threshold conditions for rainfall and snow accumulation. In addition, the critical temperature and precipitation process for rain and snow separation are highly spatiotemporally heterogeneous in high-altitude areas, which further increases the difficulty of identifying snow surface rain and snowmelt disasters. In addition, due to the sparse ground observation network and uneven elements in high-altitude areas, the accuracy of snow surface rain disaster identification is reduced. Regional climate models have gradually become the main method for studying snow surface rain disaster identification in areas with insufficient data.

[0005] The current method for identifying rainfall and snowmelt in alpine mountainous areas is mainly based on the combination of threshold conditions of rainfall and snow accumulation. However, there is a lack of unified standards for the combination conditions, and the reconstruction of rainfall and snowmelt events with disaster potential still faces great challenges. The critical temperature threshold method is used to estimate the precipitation type, separate rain and snow, and identify whether snow-surface rain events occur. The scarcity of ground monitoring data is the main factor limiting the identification of snow-surface rain in alpine mountainous areas. Generally speaking, there are the following problems: Problem 1: The traditional critical temperature threshold method based on meteorological station interpolation for rain-snow separation has large uncertainty. At the same time, the accuracy of large-scale meteorological monitoring in mountainous areas is limited by the amount of station data and its representativeness, which reduces the accuracy of rainfall and snowmelt disaster identification.

[0006] Question 2: Combining 19GHz and 37GHz passive microwave brightness temperature and optical remote sensing technology, the inversion algorithm is not accurate enough and the spatial resolution is low. The existing remote sensing means are not capable of continuous monitoring of precipitation patterns and wet snow in complex terrain areas.

[0007] Question 3: Regional climate models can achieve detailed rainfall / snowfall characteristics and continuous snow accumulation processes, and have gradually become the main means of studying snow surface rain in data-deficient areas. However, their discrimination methods still have large deviations due to the lack of reliable measured data calibration. Summary of the invention

[0008] In view of the above problems, the present invention is proposed to provide a method, system and electronic equipment for identifying rainfall-driven snowmelt disasters in high-altitude and cold mountainous areas with insufficient data to overcome the above problems.

[0009] One aspect of the present invention provides a method for identifying rainfall-driven snowmelt disasters in high-altitude and cold mountainous areas with insufficient data, comprising: Collect meteorological observation sequence data within a set time period in high-cold mountainous areas; By performing multidimensional feature extraction on the meteorological observation sequence data, multidimensional meteorological feature sequence data is obtained; By inputting the multi-dimensional meteorological sequence feature data into a pre-trained snow surface rain machine learning discrimination model, a discrimination result of a single-point snow surface rain snowmelt disaster in a high-cold area is obtained; Based on the judgment results of the single-point snow surface rain and snowmelt disasters in the alpine area and the simulation results of the regional climate model meteorology, the judgment results of the regional snow surface rain events within the set time period in the alpine mountainous area are obtained.

[0010] Preferably, the meteorological observation sequence data includes at least atmospheric pressure, solar radiation, rainfall, snowfall, sleet, snow water equivalent, air temperature, humidity, surface temperature, and snow coverage.

[0011] Preferably, the multidimensional meteorological characteristic sequence data at least includes: snow surface rain intensity characteristic sequence data, rainfall-related characteristic sequence data, snowmelt process-related characteristic sequence data, snow water equivalent-related characteristic sequence data and snow cover-related characteristic sequence data.

[0012] Preferably, the step of extracting multidimensional features from the meteorological observation sequence data to obtain multidimensional meteorological feature sequence data comprises: Determine whether the rainfall in the meteorological observation sequence data is 0; If it is determined that the rainfall in the meteorological data is 0, the daily precipitation type is determined based on the wet-bulb temperature method; When it is determined that the daily precipitation type is a rainfall type, precipitation sequence data is obtained, and deviation correction processing is performed on the precipitation sequence data based on a precipitation correction model to obtain deviation-corrected precipitation sequence data; When it is determined that the daily precipitation type is a snowfall type, snowfall amount sequence data is obtained, and the snowfall amount sequence data is subjected to deviation correction processing based on a snowfall correction model to obtain the deviation-corrected snowfall amount sequence data; The precipitation amount sequence data after the deviation correction or the snowfall amount sequence data after the deviation correction is processed to obtain rainfall amount sequence data, and the rainfall amount sequence data is used as meteorological characteristic sequence data.

[0013] Preferably, the step of extracting multidimensional features from the meteorological observation sequence data to obtain multidimensional meteorological feature sequence data comprises: Snowmelt sequence data are estimated based on the snow water equivalent and snowfall in the meteorological observation sequence data and the bias-corrected snowfall sequence data, and the snowmelt sequence data are used as meteorological characteristic sequence data.

[0014] Preferably, the trained snow surface rain machine learning discriminant model includes: Collect meteorological data within a set time period in multiple high-cold mountainous areas and the actual snow surface rain event type corresponding to the set time period in each high-cold mountainous area, and obtain multidimensional meteorological feature data by performing multidimensional feature extraction on each of the meteorological data; A snow surface rain machine learning discriminant model is constructed, and the multidimensional meteorological characteristic data within a set time period in each high-cold mountainous area is used as the model input. At the same time, the actual snow surface rain event type corresponding to the set time period in each high-cold mountainous area is used as the model output. The snow surface rain machine learning discriminant model is trained to obtain a trained snow surface rain machine learning discriminant model.

[0015] Preferably, the snow-surface rain events include: snowmelt flood events, rain-snow combined flood events and no flood events.

[0016] Another aspect of the present invention provides a system for identifying rainfall-driven snowmelt disasters in high-altitude and cold mountainous areas with insufficient data, comprising: The collection module is used to collect meteorological observation sequence data within a set time period in the high-cold mountainous area; A multidimensional feature extraction module, used for obtaining multidimensional meteorological feature sequence data by performing multidimensional feature extraction on the meteorological observation sequence data; The identification module is used to obtain the discrimination result of single-point snow surface rain and snowmelt disaster in the high-cold area by inputting the multi-dimensional meteorological sequence feature data into a pre-trained snow surface rain machine learning discrimination model; based on the discrimination result of single-point snow surface rain and snowmelt disaster in the high-cold area and the simulation result of the regional climate model meteorology, obtain the discrimination result of regional snow surface rain events within a set time period in the high-cold mountainous area.

[0017] Preferably, the snow-surface rain events include: snowmelt flood events, rain-snow combined flood events and no flood events.

[0018] Another aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the method for preventing rainfall-driven snowmelt disasters in high-altitude and data-deficient mountainous areas as described above are implemented.

[0019] The method, system and electronic device for rain-driven snowmelt disasters in high-cold and data-deficient mountainous areas provided by the embodiments of the present invention, by coupling the convective allowable scale simulation of the regional climate model of the land snow accumulation process, finely depicts the rain and snow phase and energy exchange process under the conditions of complex mountain terrain; and then uses machine learning to deeply mine multi-source earth system data (meteorological observations, remote sensing, historical disaster cases, etc.) to perform high-precision identification of "snow surface rain events". Compared with traditional regional models, the present invention effectively reduces the uncertainty of precipitation simulation at the convective allowable scale, and combines the measured and historical disaster data for correction, overcoming the problem of lack of reliability of regional climate models in high-cold mountainous areas, and further improving the accuracy of distinguishing rain and snow events with the potential for floods or snow disasters.

[0020] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Various other advantages and benefits will become apparent to those skilled in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only used for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. In the accompanying drawings: Figure 1 It is a flow chart of a method for identifying rainfall-driven snowmelt disasters in high-altitude and cold mountainous areas with insufficient data according to an embodiment of the present invention; Figure 2 This is a general flow chart of a method for identifying rainfall-driven snowmelt disasters in high-altitude and cold mountainous areas with insufficient data proposed in an embodiment of the present invention; Figure 3 The present invention is a schematic diagram of the structure of a system for identifying rainfall-driven snowmelt disasters in high-altitude and data-deficient mountainous areas proposed by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0023] Those skilled in the art will understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined.

[0024] Figure 1 Flow chart of a method for identifying rainfall-driven snowmelt disasters in high-altitude and data-deficient mountainous areas according to an embodiment of the present invention. Figure 1 As shown, including: Step S101: Collecting meteorological observation sequence data within a set time period in a high-cold mountain area; In one embodiment of the present invention, the meteorological observation sequence data at least includes atmospheric pressure, solar radiation, rainfall, snowfall, sleet, snow water equivalent, air temperature, humidity, surface temperature, and snow coverage.

[0025] Step S102: extracting multidimensional features from the meteorological observation sequence data to obtain multidimensional meteorological feature sequence data; In one embodiment of the present invention, the multidimensional meteorological characteristic sequence data at least includes: snow surface rain intensity characteristic sequence data, rainfall related characteristic sequence data, snowmelt process related characteristic sequence data, snow water equivalent related characteristic sequence data and snow cover related characteristic sequence data.

[0026] In one embodiment of the present invention, the step of extracting multidimensional features from the meteorological observation sequence data to obtain multidimensional meteorological feature sequence data includes: Determine whether the rainfall in the meteorological observation sequence data is 0; If it is determined that the rainfall in the meteorological data is 0, the daily precipitation type is determined based on the wet-bulb temperature method; When it is determined that the daily precipitation type is a rainfall type, precipitation sequence data is obtained, and deviation correction processing is performed on the precipitation sequence data based on a precipitation correction model to obtain deviation-corrected precipitation sequence data; When it is determined that the daily precipitation type is a snowfall type, snowfall amount sequence data is obtained, and the snowfall amount sequence data is subjected to deviation correction processing based on a snowfall correction model to obtain the deviation-corrected snowfall amount sequence data; The precipitation amount sequence data after the deviation correction or the snowfall amount sequence data after the deviation correction is processed to obtain rainfall amount sequence data, and the rainfall amount sequence data is used as meteorological characteristic sequence data.

[0027] In one embodiment of the present invention, the step of extracting multidimensional features from the meteorological observation sequence data to obtain multidimensional meteorological feature sequence data includes: Snowmelt sequence data are estimated based on the snow water equivalent and snowfall in the meteorological observation sequence data and the bias-corrected snowfall sequence data, and the snowmelt sequence data are used as meteorological characteristic sequence data.

[0028] Step S103: obtaining a discrimination result of single-point snow surface rain and snowmelt disaster in high-cold areas by inputting the multi-dimensional meteorological sequence feature data into a pre-trained snow surface rain machine learning discrimination model; In one embodiment of the present invention, the trained snow surface rain machine learning discriminant model includes: Collect meteorological data within a set time period in multiple high-cold mountainous areas and the actual snow surface rain event type corresponding to the set time period in each high-cold mountainous area, and obtain multidimensional meteorological feature data by performing multidimensional feature extraction on each of the meteorological data; A snow surface rain machine learning discriminant model is constructed, and the multidimensional meteorological characteristic data within a set time period in each high-cold mountainous area is used as the model input. At the same time, the actual snow surface rain event type corresponding to the set time period in each high-cold mountainous area is used as the model output. The snow surface rain machine learning discriminant model is trained to obtain a trained snow surface rain machine learning discriminant model.

[0029] Step S104: Based on the judgment result of the single-point snow surface rain and snowmelt disaster in the alpine area and the simulation result of the regional climate model meteorology, the judgment result of the regional snow surface rain event within the set time period in the alpine mountain area is obtained.

[0030] In one embodiment of the present invention, the snow surface rain event includes: snowmelt flood event, rain and snow flood event and no flood event.

[0031] The present invention provides a method for driving snowmelt disasters caused by rainfall in high-altitude and data-deficient mountainous areas. By simulating the convection-allowed scale of the regional climate model coupled with the land snow accumulation process, the rain-snow phase and energy exchange process under the complex terrain conditions of the mountain are finely portrayed; and then the multi-source earth system data (meteorological observations, remote sensing, historical disaster cases, etc.) are deeply mined with the help of machine learning methods to perform high-precision identification of "snow surface rain events". Compared with traditional regional models, the present invention effectively reduces the uncertainty of precipitation simulation at the convection-allowed scale, and combines the measured and historical disaster data for correction, overcoming the problem of the lack of reliability of regional climate models in high-altitude and cold mountainous areas, and further improving the accuracy of distinguishing rain and snow events with the potential for floods or snow disasters.

[0032] For the method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0033] The embodiment of the present invention proposes a method for identifying rainfall-driven snowmelt disasters in high-altitude and data-deficient mountainous areas. Figure 2 As shown, the specific steps include: Step 1: Collection of site rainfall and snowmelt data Collect meteorological data from monitoring sites, including temperature, rainfall, relative humidity, air pressure, wind speed, snow depth, snow water equivalent, surface temperature, etc.

[0034] Rain gauge observation of daily rainfall data: mainly used to obtain elements such as cumulative rainfall and rainfall intensity. If there is no direct rainfall observation, the rain and snow phases are divided using the wet-bulb temperature method based on variables such as precipitation, air pressure, relative humidity and temperature (maximum temperature, minimum temperature and average temperature of the day) to obtain the daily precipitation type.

[0035] The method for dividing the rain and snow phase at a station based on the wet-bulb temperature method is as follows: ; in P type The precipitation type of the day (snow, sleet or rain), T w is the daily wet-bulb temperature, Tmin represents the temperature at which the probability of snowfall and sleet is equal. T max Indicates the temperature at which the probability of rain and sleet is equal.

[0036] It is proposed to use a precipitation correction model based on the Chinese standard rain gauge to correct the systematic errors of station observations caused by meteorological factors and precipitation types. It can be expressed as: ; in P c is the corrected daily precipitation; P g ,△ P w ,△ P e and P t They represent daily observed precipitation, wet loss, evaporation loss and trace precipitation respectively. K (≥1) represents the correction factor for wind-induced errors.

[0037] Based on the above method, the daily corrected snowfall data is estimated to be: ; in P is the corrected daily precipitation (mm) (same as above P c ).

[0038] Furthermore, rainfall is the amount of precipitation (i.e. the corrected daily precipitation P c ) minus the snowfall (i.e. the corrected daily precipitation Snowfall).

[0039] Sonar observation of snow water equivalent data Based on the observations of the sonar instrument, the daily snow water equivalent data is obtained, and combined with the snowfall estimated in step 1, the current snowmelt amount is obtained.

[0040] Site snowmelt estimates are as follows: ; where snowmelt(t) is the amount of snowmelt at the current moment, SWE(t) and SWE (t-1) are the snow water equivalent at the current moment and the previous moment, respectively, and snowfall(t) is the amount of snowfall at the current moment.

[0041] Collect daily MODIS snow cover data Select a snow cover fraction (SCF) product that can provide daily and cloud-free snow cover. Through cloud detection and cloud removal algorithms from multi-source satellite data, obtain the spatial distribution of snow cover during the "cloud-free" period. If there is occasional cloud cover or data is missing, interpolation or synthesis can be performed by combining adjacent daily data to ensure time series integrity.

[0042] Step 2: Calculation of regional meteorological-snow cover variable parameters Through the simulation results of the regional climate model (Weather Research and Forecasting model, WRF) and its coupled land surface model convection allowed scale, we obtain gridded precipitation (rainfall / snowfall), surface air temperature, humidity, snow water equivalent, snowmelt, snow cover, snow depth and other variables with a daily resolution of 1-4 km.

[0043] Step 3: Machine learning optimizes the method of identifying single rain points on snow surface 301. Data Preparation and Feature Engineering Sample set construction The site rainfall and snowmelt data collection and snow cover remote sensing observation are based on the loose standards of 1) daily rainfall ≥1 mm, 2) falling on snow with a snow water equivalent of ≥1 mm, and 3) snow cover ≥ 1% mm. All observed snow surface rain events are screened to obtain all "snow surface rain event" samples, where rainfall + snowmelt is the intensity of snow surface rain. Each sample usually corresponds to a set of (event, time period), such as: event start time, event end time; snow surface rain intensity, rainfall, snow water equivalent (SWE), snowmelt, snow cover and other key elements in this period. If there is historical flood disaster information measured or recorded in literature, it can be marked whether "snowmelt flood" or "flood caused by simultaneous rain and snow" has occurred; events without measured disasters are marked as "no flood occurred".

[0044] Feature selection and construction Characteristics of snow surface rain intensity: liquid rainfall + snowmelt, that is, intensity change characteristics.

[0045] Precipitation-related characteristics: cumulative rainfall on the day / event period, average rainfall intensity, maximum rainfall intensity, etc.

[0046] Snow water equivalent (SWE) related characteristics: initial SWE, final SWE, maximum / minimum SWE during the event period; average SWE change rate during the event period.

[0047] Related characteristics of the snowmelt process: total snowmelt volume during the event period, snowmelt rate (such as mm / day), average daily temperature, daily maximum / minimum temperature, etc.

[0048] Snow cover fraction (SCF) related characteristics: change in snow cover, difference in coverage ratio before / after the event.

[0049] Characteristics of meteorological elements: wet-bulb temperature, wind speed, relative humidity, etc., can also be included in the model if they play an important role in the snowmelt process.

[0050] 302. Model Algorithm Selection Two commonly used supervised learning classification algorithms are mainly considered: (input data include: rainfall, air temperature, relative humidity, snowmelt, snow cover, snow water equivalent, surface temperature, rainfall + snowmelt when snow rain events occur; output is: meteorological snow variable thresholds that are closely related to snow rain events, for example: daily-scale snow rain is defined as: 1) daily rainfall ≥5 mm in the model grid, 2) falling on snow with snow water equivalent ≥3 mm, 3) daily snow cover ≥ 25% (used to mask relatively small snow rain events)).

[0051] Support Vector Machine (SVM) Suitable for small samples or medium-sized data sets; has strong expression ability for high-dimensional data, nonlinear decision boundaries, etc. (Gaussian kernel or polynomial kernel can be selected). Parameters mainly include penalty coefficient C , kernel function type and its parameters (such as the γ value of the RBF kernel), etc.

[0052] Random Forest (RF) By integrating multiple decision trees, the features are interpretable to a certain extent; they are more robust to missing values ​​or noise; the parameters mainly include the number of trees, maximum depth, minimum number of samples per node, etc. According to the required goal (identifying snow-surface rain events that "may cause flood disasters"), the problem is modeled as a binary classification task: positive class: events with the potential to cause flood disasters (or have indeed caused flood disasters); negative class: events that do not have the potential to cause disasters.

[0053] 303. Training and Verification Process Data partitioning On the constructed event sample data set (including feature vectors and "whether there is a flood" label), the training set / test set is divided into two parts according to a ratio of 8:2 or 7:3. If the data scale allows, K-fold cross-validation can also be further performed on the training set.

[0054] Model Training For SVM: adjust parameters by grid search or Bayesian optimization on the training set (find the optimal C and γ, etc.). Select the optimal hyperparameters based on the metrics on the validation set / cross-validation set (accuracy, recall, F1 score, etc.).

[0055] For random forests: train with different combinations of parameters such as the number of trees and maximum depth; evaluate and determine the model configuration with the best performance. If necessary, perform a weighted ensemble of the two algorithms or compare with other classifiers.

[0056] Test and Evaluation Evaluate model performance on an independent test set. Common indicators: accuracy, precision, recall (for disaster prediction problems, recall is often of particular concern), F1 score, area under the ROC curve (AUC), etc. If the number of test sets is limited or the distribution is uneven, additional methods such as leave-one-out / multiple resampling can be used to ensure robust results.

[0057] Further verification and revision of identification results The "high-risk snow-surface rain events" identified by the model are compared with the measured hydrological and meteorological data and historical flood disaster events: if the model fails to identify some flood events (missing reports), it is necessary to find deficiencies in the feature structure or model principle; if there are a large number of "false alarms" / "false positives", it may be necessary to adjust the classification threshold, feature selection, or balance the weights of positive and negative classes. After a certain number of iterations, a machine learning discriminant model suitable for snow-surface rain with disaster potential in high-altitude and cold mountainous areas is finally formed.

[0058] 304、Model Application and Interpretation Variable Importance Analysis (Taking Random Forest as an Example) Random forest can output feature importance, such as the contribution of rainfall to model decision-making, the contribution of snowmelt, the contribution of SWE, etc. By quantifying "which variable is most critical to flood potential identification", it can further provide reference for early warning business.

[0059] Decision boundary and visualization (taking SVM as an example) Visualize the decision boundary of SVM in 2D or 3D feature subspace; help understand which combination of conditions (e.g., rainfall + SWE exceeding a certain threshold at the same time) is most likely to lead to flood risk.

[0060] Ultimately, once the machine learning model identifies a "high-risk" combination of key indicators such as rainfall, snowmelt rate, and snow cover when high snow surface rain intensity (snowmelt + rainfall) occurs, it can be judged as a snow surface rain event with disaster potential, thereby obtaining a method for identifying snow surface rain with disaster potential based on observational data.

[0061] Step 4: Method for determining regional snow surface rain Based on the single-point snow surface rain event identification method with disaster-causing potential obtained in the previous step, combined with the reanalysis data-driven (such as ERA5, MERRA2, etc.) regional climate model (such as WRF, etc.) convection-allowed scale simulation (1-4 km spatial resolution), the output is the daily / hourly regional large-scale rainfall, snowmelt, snow cover, snow depth, snow water equivalent, temperature, humidity and other related variables, forming a regional meteorological-snow variable threshold judgment combination, and finally forming a method for identifying snow surface rain events with disaster-causing potential in large-scale ranges in high-altitude and data-deficient areas.

[0062] In summary, the embodiments of the present invention have the following advantages: (1) Identification system of rain-snow coupling in high-altitude and data-deficient mountainous areas Based on the regional climate model and land snow process simulation at the convection-allowed scale, the complex terrain and rain-snow phase state of the high-cold mountainous area are finely portrayed; in order to solve the problem of insufficient observation data, multi-source remote sensing and site observation fusion are used to accurately obtain the dynamic information of regional precipitation (rain / snow) and snow accumulation process. Effectively improve the recognition accuracy of snow-surface rain events with disaster potential in high-cold mountainous areas with insufficient data: through the simulation of the convection-allowed scale of the regional climate model coupled with the land snow process, the rain-snow phase state and energy exchange process under the conditions of complex mountain terrain are finely portrayed; and then with the help of machine learning methods, multi-source earth system data (meteorological observations, remote sensing, historical disaster cases, etc.) are deeply mined to identify "snow-surface rain events" with high precision. Compared with the traditional regional model, the present invention effectively reduces the uncertainty of precipitation simulation at the convection-allowed scale, and at the same time combines the measured and historical disaster data to calibrate the snow-surface rain event discrimination method, overcoming the problem of the lack of measured data in the discrimination method based on the output of the regional climate model in the high-cold mountainous area. The discrimination accuracy of rain-snow events with flood or snow disaster potential is further improved.

[0063] (2) Machine learning-driven method for identifying the disaster potential of “snow-surface rain events” Multidimensional factors such as precipitation patterns, snow accumulation and snowmelt are input into machine learning models (such as support vector machines and random forests) to identify "snow surface rain events" with the potential to cause disasters such as floods and avalanches; historical snowmelt flood cases, disaster records, etc. are used for calibration and verification to improve recognition accuracy and practicality. In other words, the combination of statistical analysis and machine learning significantly improves the recognition speed and data utilization efficiency: the present invention uses the rapid iterative training and prediction capabilities of statistical analysis and machine learning (such as support vector machines, random forests, etc.) to quickly process multi-source, heterogeneous observation and remote sensing data while ensuring data reliability, and automatically identify potential "snow surface rain" risk events. In the high-cold mountain environment with limited manpower and spatiotemporal observation resources, this automated method greatly reduces manual intervention and data processing time, and has higher efficiency and accuracy than traditional pure manual judgment or single model analysis methods.

[0064] (3) Rapid identification and application Providing a scientific basis for flood and avalanche risk assessment and assisting regional disaster prevention: The "snow surface rain event identification method with flood disaster potential" proposed in this invention can not only identify snowmelt floods caused by the superposition of rain and snow, but also provide key process indicators for avalanche risk assessment (such as the impact of high-intensity rainfall superimposed on rapid snowmelt on snow stability). By combining this identification method with a regional emergency management system or early warning platform, disaster signs can be captured in real time or quasi-real time, providing quantitative and scientific decision-making support for the prevention of rain and snow superimposed disasters (including snowmelt floods, avalanches, etc.) in high-altitude mountainous areas. Under the premise of ensuring data reliability, rapid identification of rain and snow superimposed disaster events can be achieved, improving the efficiency of emergency response and disaster prevention and mitigation; providing scientific support and operational technical means for the monitoring, early warning and management decision-making of rain and snow disasters in high-altitude mountainous areas.

[0065] Figure 3 The structure diagram of a system for identifying rainfall-driven snowmelt disasters in high-altitude and data-deficient mountainous areas proposed by an embodiment of the present invention is as follows: Figure 3 As shown, it includes: a collection module, which is used to collect meteorological observation sequence data within a set time period in the high-cold mountainous area; a multidimensional feature extraction module, which is used to obtain multidimensional meteorological feature sequence data by performing multidimensional feature extraction on the meteorological observation sequence data; an identification module, which is used to obtain a discrimination result of a single-point snow surface rain and snowmelt disaster in the high-cold area by inputting the multidimensional meteorological sequence feature data into a pre-trained snow surface rain machine learning discrimination model; based on the discrimination result of the single-point snow surface rain and snowmelt disaster in the high-cold area and the simulation result of the regional climate model meteorology, the discrimination result of the regional snow surface rain event within the set time period in the high-cold mountainous area is obtained.

[0066] In the specific implementation process of this system embodiment, reference may be made to the above method embodiment, which has corresponding technical effects.

[0067] Another aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the above-mentioned method for identifying rainfall-driven snowmelt disasters in high-cold and data-deficient mountainous areas are implemented. Figure 1 Alternatively, when the processor executes the computer program, the functions of each module in the above-mentioned vehicle fuel cell performance prediction device embodiments are realized, for example Figure 3 The collection module, multi-dimensional feature extraction module and recognition module are shown.

[0068] During the specific implementation of the device embodiment, the above method embodiment can be referred to, and the corresponding technical effects are achieved.

[0069] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, any one of the claimed embodiments may be used in any combination.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be used to modify the technical solutions or to perform equivalent replacements on some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying rainfall-driven snowmelt disasters in high-altitude and data-deficient mountainous areas, characterized in that: include: Collect meteorological observation sequence data within a set time period in high-cold mountainous areas; By performing multidimensional feature extraction on the meteorological observation sequence data, multidimensional meteorological feature sequence data is obtained; By inputting the multi-dimensional meteorological sequence feature data into a pre-trained snow surface rain machine learning discrimination model, a discrimination result of a single-point snow surface rain snowmelt disaster in a high-cold area is obtained; Based on the judgment results of the single-point snow surface rain and snowmelt disasters in the alpine area and the simulation results of the regional climate model meteorology, the judgment results of the regional snow surface rain events within the set time period in the alpine mountainous area are obtained.

2. The method according to claim 1, characterized in that The meteorological observation sequence data at least includes atmospheric pressure, solar radiation, rainfall, snowfall, sleet, snow water equivalent, air temperature, humidity, surface temperature, and snow coverage.

3. The method according to claim 2, characterized in that The multi-dimensional meteorological characteristic sequence data at least include: snow surface rain intensity characteristic sequence data, rainfall related characteristic sequence data, snowmelt process related characteristic sequence data, snow water equivalent related characteristic sequence data and snow cover related characteristic sequence data.

4. The method according to claim 3, characterized in that The multidimensional meteorological feature sequence data obtained by performing multidimensional feature extraction on the meteorological observation sequence data includes: Determine whether the rainfall in the meteorological observation sequence data is 0; If it is determined that the rainfall in the meteorological observation sequence data is 0, determining the daily precipitation type based on the wet-bulb temperature method; When it is determined that the daily precipitation type is a rainfall type, precipitation sequence data is obtained, and deviation correction processing is performed on the precipitation sequence data based on a precipitation correction model to obtain deviation-corrected precipitation sequence data; When it is determined that the daily precipitation type is a snowfall type, snowfall amount sequence data is obtained, and the snowfall amount sequence data is subjected to deviation correction processing based on a snowfall correction model to obtain the deviation-corrected snowfall amount sequence data; The precipitation amount sequence data after the deviation correction or the snowfall amount sequence data after the deviation correction is processed to obtain rainfall amount sequence data, and the rainfall amount sequence data is used as meteorological characteristic sequence data.

5. The method according to claim 4, characterized in that The multidimensional meteorological feature sequence data obtained by performing multidimensional feature extraction on the meteorological observation sequence data includes: Snowmelt sequence data are estimated based on the snow water equivalent and snowfall in the meteorological observation sequence data and the bias-corrected snowfall sequence data, and the snowmelt sequence data are used as meteorological characteristic sequence data.

6. The method according to claim 1, characterized in that The trained snow surface rain machine learning discriminant model includes: Collect meteorological data within a set time period in multiple high-cold mountainous areas and the actual snow surface rain event type corresponding to the set time period in each high-cold mountainous area, and obtain multidimensional meteorological feature data by performing multidimensional feature extraction on each of the meteorological data; A snow surface rain machine learning discrimination model is constructed, and the multidimensional meteorological characteristic data within a set time period in each high-cold mountainous area is used as the model input. At the same time, the actual snow surface rain event type corresponding to the set time period in each high-cold mountainous area is used as the model output. The snow surface rain machine learning discrimination model is trained to obtain a trained snow surface rain discrimination model for the target area.

7. The method according to claim 6, characterized in that The snow-surface rain events include: snowmelt flood events, rain-snow combined flood events and no flood events.

8. A system for identifying rainfall-driven snowmelt disasters in high-altitude and data-deficient mountainous areas, characterized in that: include: The collection module is used to collect meteorological observation sequence data within a set time period in the high-cold mountainous area; A multidimensional feature extraction module, used for obtaining multidimensional meteorological feature sequence data by performing multidimensional feature extraction on the meteorological observation sequence data; An identification module is used to obtain a discrimination result of a single-point snow surface rain and snowmelt disaster in a high-cold area by inputting the multi-dimensional meteorological sequence feature data into a pre-trained snow surface rain machine learning discrimination model; Based on the judgment results of the single-point snow surface rain and snowmelt disasters in the alpine area and the simulation results of the regional climate model meteorology, the judgment results of the regional snow surface rain events within the set time period in the alpine mountainous area are obtained.

9. The system according to claim 8, characterized in that The snow-surface rain events include: snowmelt flood events, rain-snow combined flood events and no flood events.

10. An electronic device, characterized in that: It comprises a memory, a processor and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the method for identifying rainfall-driven snowmelt disasters in high-altitude and cold mountainous areas with insufficient data as described in any one of claims 1 to 7 are implemented.

Images