Meteorological and hydrological data analysis method and system based on multi-source data fusion

Abstract

The invention provides a meteorological and hydrological data analysis method and system based on multi-source data fusion, and relates to the technical field of data analys.The method comprises the steps that a to-be-analyzed area is divided into grids, and radar grid point rainfall intensity is calculated in combination with meteorological radar reflectivity factors and rainfall station data; correcting the soil water content by using the gradient, and calculating a grid disaster potential index in combination with historical disaster situation data; predicting a future direct runoff depth based on an LSTM model, and calculating a disaster risk index in combination with the disaster potential index; and finally, fusing the elevation change rate and the people flow density, constructing a multi-layer early warning trigger mechanism, and dynamically issuing disaster early warnings of different levels. Prediction precision is optimized through multi-source data fusion and machine learning, early recognition and graded early warning of disaster risks are achieved, scientificity and timeliness of disaster prevention and reduction decisions are improved, and the method is suitable for refined monitoring and early warning of meteorological and hydrological disasters such as urban waterlogging and mountain torrent disasters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a meteorological and hydrological data analysis method and system based on multi-source data fusion. BACKGROUND

[0002] Under the background of global climate change intensification, extreme precipitation events occur frequently, and disasters such as mountain torrents and urban waterlogging pose a serious threat to people's life and property safety. Traditional meteorological and hydrological monitoring methods mainly rely on rainfall station data, but due to factors such as sparse site distribution and terrain influence, it is difficult to achieve high spatial and temporal precision precipitation estimation and runoff prediction. In addition, existing hydrological models use static parameters, which cannot dynamically respond to changes in soil moisture, rainfall intensity and historical disaster conditions, resulting in insufficient accuracy and timeliness of flood warning.

[0003] Traditional meteorological and hydrological monitoring mainly relies on ground rainfall station data, but due to uneven site distribution, especially in complex terrain areas (such as mountainous and hilly areas), there are monitoring blind spots, resulting in low spatial interpolation precision of precipitation. In addition, although radar precipitation estimation can provide high spatial and temporal resolution rainfall information, it is affected by factors such as radar beam shielding and attenuation, and the estimated rainfall intensity often has systematic bias. Traditional meteorological and hydrological disaster warning methods mainly rely on rainfall stations and single hydrological models, but due to sparse site distribution and low data fusion, it is difficult to achieve fine risk assessment. Especially in complex terrain areas, factors such as soil moisture content and slope have not been fully quantified to affect runoff formation, resulting in insufficient prediction accuracy and poor warning timeliness. In addition, existing warning mechanisms are mostly based on a single indicator, lack comprehensive consideration of historical disasters, terrain changes and population density, and are difficult to dynamically adjust warning levels, which cannot meet the needs of precise disaster prevention.

[0004] The above information disclosed in the background section is only intended to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute the prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a meteorological and hydrological data analysis method and system based on multi-source data fusion to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] A meteorological and hydrological data analysis method based on multi-source data fusion, the specific steps comprising:

[0008] Step 1: Divide the area to be analyzed into several grids of the same size, collect the radar reflectivity factor of the grid through the meteorological radar, and convert it into radar estimated rainfall intensity;

[0009] Step 2: Collect the rain intensity of the rain gauge station in the area to be analyzed and the historical disaster data, calculate the radar rain intensity deviation based on the rain intensity of the rain gauge station, determine the radar grid point rain intensity of the grid, obtain the slope of the grid to correct the soil moisture content of the grid, determine the historical disaster intensity of the grid through expert scoring method, and calculate the disaster potential index of the grid according to the historical disaster intensity and the distance from the grid to the disaster location;

[0010] Step 3: Obtain the direct runoff depth, radar grid point rain intensity, slope and corrected soil moisture content of each grid in the current monitoring time period in the area to be analyzed, and train the LSTM model based on the above data, and output the direct runoff depth prediction value of each grid at the future time based on the LSTM model;

[0011] Step 4: Combine the direct runoff depth prediction value and the disaster potential index to determine the disaster risk index of each grid, and set a multi-layer early warning trigger mechanism combining the elevation change rate and the human flow density of each grid in the current monitoring time period to issue different degrees of disaster warning for each grid.

[0012] Further, the radar estimated rain intensity specifically includes:

[0013] The area to be analyzed is divided into a plurality of 1km×1km grids, the radar reflectivity factor of the grid is converted into the radar estimated rain intensity through the Z-R relationship, and the calculation formula is:

[0014] P radar (m)=a -1 / b ·10 0.1 / b·Z

[0015] Wherein, P radar (m) is the radar estimated rain intensity of the mth grid, Z is the radar reflectivity factor, a represents the proportionality constant, b represents the nonlinearity degree, m is the grid index, M is the total number of grids, and m∈[1,M].

[0016] Further, the radar grid point rain intensity specifically includes:

[0017] Based on the rain intensity data of the rain gauge station, the radar rain intensity deviation is calculated, and the formula is as follows:

[0018]

[0019] Wherein, ΔP represents the radar rain intensity deviation, k is the index of the rain gauge station, n represents the total number of rain gauge stations, and k∈[1,n], P station (k) represents the rain intensity measured by the kth rain gauge station, and P radar (k) represents the radar estimated rain intensity of the grid where the kth rain gauge station is located.

[0020] The radar estimated rain intensity of each grid is corrected, and the calculation formula is as follows:

[0021]

[0022] wherein, represents the radar grid rain intensity of the mth grid.

[0023] Further, the soil moisture of the grid is specifically corrected, and the correction of the soil moisture of the grid specifically includes:

[0024] The soil moisture sensors are arranged around and in the center of the region to be analyzed to collect the soil moisture data of the positions, and the average value is taken as the soil moisture of the region to be analyzed.

[0025] The slope of each grid is obtained, and the soil moisture of each grid is corrected according to the slope, and the calculation formula is as follows:

[0026]

[0027] wherein, θ adjusted (m) represents the corrected soil moisture of the mth grid, θ raw is the soil moisture of the region to be analyzed, β φ is a slope influence coefficient, and φ(m) represents the slope of the mth grid.

[0028] Further, the calculation of the disaster potential index specifically includes:

[0029] The historical disaster data includes historical disaster points, disaster duration, disaster type and damage degree, the historical disaster point refers to the center point coordinate of the disaster area, and the historical disaster intensity of each grid is determined based on the historical disaster data through the expert scoring method, and the calculation formula of the disaster potential index of each grid is as follows:

[0030]

[0031] wherein, H risk (m) represents the disaster potential index of the mth grid, I(m,i) represents the historical disaster intensity of the mth grid in the ith historical disaster, D(m,i) represents the distance from the mth grid to the historical disaster point in the ith historical disaster, m is the grid index, M is the total number of grids, and m∈[1,M], i is the historical disaster index, U is the total number of historical disasters in the historical period, and i∈[1,U].

[0032] Further, the training of the LSTM model specifically includes:

[0033] A water level gauge is arranged at the center of each grid to obtain the direct runoff depth of each grid. The direct runoff depth, radar grid rain intensity, slope and corrected soil moisture data of each grid in the current monitoring period are divided into a training set, a validation set and a test set in a ratio of 7:1.5:1.5. The model is trained by using the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture data of the grid in the first ten time points in the training set as input, and the direct runoff depth of the grid in the next time point as output label. If the loss of the validation set changes by less than a preset threshold in ten iterations, the training is automatically terminated. The data in the test set is input for testing, and the mean square error and mean absolute error are used as evaluation indexes. If the testing is unqualified, the model is retrained. If the testing is qualified, the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture data of the grid in the current time point and the previous nine time points are input into the model to obtain the direct runoff depth prediction value of the grid in the next time point.

[0034] Further, determining the disaster risk index specifically includes:

[0035] The disaster potential index and the direct runoff depth prediction value of each grid are normalized, and the normalized data is used for calculation, and the formula is as follows:

[0036]

[0037] wherein Risk(m) is the disaster risk index of the mth grid, Q(m) represents the direct runoff depth prediction value of the mth grid, Q c represents the direct runoff depth threshold, H(m) represents the normalized disaster potential index of the mth grid, and a and λ are weight coefficients of the corresponding items.

[0038] Further, setting the multi-layer early warning triggering mechanism specifically includes:

[0039] An early warning triggering mechanism with three levels of determination is set, and the levels are upgraded one by one. Each level of early warning triggering mechanism includes level, judgment condition and warning level, and specifically: the first level, if the disaster risk index of the grid is higher than the preset risk threshold, a yellow warning is issued; the second level, based on the judgment condition of the first level, the elevation change rate is calculated, if the elevation change rate is greater than 15%, an orange warning is issued; the third level, based on the judgment condition of the second level, the residential density of the grid is determined, if the residential density is higher than 50 people / km 2 , a red warning is issued;

[0040] The formula for calculating the elevation change rate is as follows:

[0041]

[0042] Wherein, ΔZ(m) represents the elevation change rate of the mth grid, Z(m, t) represents the elevation value of the mth table at the current monitoring moment, Z(m, t-Δt) represents the elevation value of the mth table at the previous monitoring moment, and Δt is the interval duration.

[0043] The application further provides a meteorological and hydrological data analysis system based on multi-source data fusion, which is used for realizing the meteorological and hydrological data analysis method based on multi-source data fusion.

[0044] The grid radar data acquisition module is used for dividing the region to be analyzed into a plurality of grids of the same size, collecting the radar reflectivity factor of the grid through the meteorological radar, and converting the radar reflectivity factor into the radar estimated rain intensity.

[0045] The rain intensity deviation and disaster potential evaluation module is used for collecting the rain intensity of the rain gauge station and the historical disaster data of the region to be analyzed, calculating the radar rain intensity deviation based on the rain intensity of the rain gauge station, determining the radar grid point rain intensity of the grid, obtaining the slope of the grid to correct the soil moisture content of the grid, determining the historical disaster intensity of the grid through the expert scoring method, and calculating the disaster potential index of the grid according to the historical disaster intensity and the distance from the grid to the disaster location.

[0046] The runoff depth prediction module is used for obtaining the direct runoff depth, the radar grid point rain intensity, the slope and the corrected soil moisture content of each grid in the region to be analyzed in the current monitoring period, training the LSTM model based on the direct runoff depth, the radar grid point rain intensity, the slope and the corrected soil moisture content, and outputting the direct runoff depth prediction value of each grid at the future moment based on the LSTM model.

[0047] The risk assessment and early warning module is used for combining the direct runoff depth prediction value and the disaster potential index to determine the disaster risk index of each grid, combining the elevation change rate and the human flow density of each grid in the current monitoring period, and setting a multi-layer early warning triggering mechanism to issue different degrees of disaster warning for each grid.

[0048] In the above technical solution, the application provides the following technical effects and advantages:

[0049] The scheme effectively solves the problems of sparse data, insufficient prediction accuracy and single early warning mechanism in traditional meteorological and hydrological warning by multi-source data fusion and machine learning optimization. Firstly, the spatial resolution and accuracy of rainfall monitoring are significantly improved by combining radar remote sensing and ground rain station data; secondly, the calculation of soil moisture content with slope correction and the assessment of historical disaster intensity make the runoff prediction more in line with the actual topography and disaster distribution characteristics; finally, the dynamic runoff prediction based on the LSTM model combined with the multi-layer early warning trigger mechanism realizes the adaptive response to the change of elevation and the density of people flow, thereby providing more accurate and graded disaster warning. The method can significantly improve the disaster risk assessment ability under extreme weather conditions and provide more timely and scientific decision-making basis for urban waterlogging and mountain flood prevention. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 It is a whole method flowchart of the present application;

[0051] Figure 2 It is a direct runoff depth prediction value and disaster risk index relationship diagram of the present application;

[0052] Figure 3 It is a disaster potential index and disaster risk index relationship diagram of the present application;

[0053] Figure 4 It is a system structure schematic diagram of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with specific embodiments.

[0055] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like only represent relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0056] Embodiment:

[0057] Please refer to Figure 1 The present application provides a technical scheme:

[0058] A meteorological and hydrological data analysis method based on multi-source data fusion, the specific steps comprising:

[0059] Step 1: Divide the area to be analyzed into several grids of the same size, collect the radar reflectivity factor of the grid by meteorological radar, and convert it into radar estimated rainfall intensity;

[0060] In this embodiment, obtaining the radar estimated rainfall intensity specifically comprises:

[0061] Divide the area to be analyzed into a plurality of 1km×1km grids. If the resolution of the radar is greater than 1km, for example 2km×2km, multiple grids may appear, for example each 1km×1km grid contains 4 radar scanning units, at this time the average value of the radar data in the grid is taken to adapt to the 1km×1km grid. The radar reflectivity factor of the grid is converted into radar estimated rainfall intensity by Z-R relationship, and the calculation formula is:

[0062] P radar (m)=a -1 / b ·10 0.1 / b·Z

[0063] Wherein, P radar (m) is the radar estimated rainfall intensity of the mth grid, Z is the radar reflectivity factor, a represents the proportionality constant, and b represents the nonlinearity. The values of a and b are determined by M-P relationship, in which b=1.6 and a=200. m is the grid index, M is the total number of grids, and m∈[1,M].

[0064] By dividing the area to be analyzed into high-resolution (1km×1km) regular grids, combining the reflectivity factor data of the meteorological radar, and using the Z-R conversion formula optimized by the M-P relationship to accurately retrieve the radar estimated rainfall intensity, this step significantly improves the spatial resolution and quantitative accuracy of precipitation monitoring, overcomes the defect of insufficient spatial coverage of traditional rain gauges, and provides a unified spatial reference for subsequent multi-source data fusion through standardized gridding processing, so that the precipitation data can accurately reflect the details of local heavy rain and maintain the overall consistency of large-scale precipitation field, laying a reliable data foundation for fine meteorological and hydrological analysis.

[0065] Step 2: Collect the rainfall station rainfall intensity and historical disaster data of the area to be analyzed, calculate the radar rainfall intensity deviation based on the rainfall station rainfall intensity, determine the radar grid rainfall intensity of the grid, obtain the slope of the grid to correct the soil moisture content of the grid, determine the historical disaster intensity of the grid by expert scoring method, and calculate the disaster potential index of the grid according to the historical disaster intensity and the distance from the grid to the disaster location;

[0066] In this embodiment, the radar grid rain intensity specifically includes:

[0067] Based on the rain intensity data of the rain gauge station, the radar rain intensity deviation is calculated, and the formula is as follows:

[0068]

[0069] Wherein, ΔP represents the radar rain intensity deviation, k is the rain gauge station index, n represents the total number of rain gauge stations, and k ∈ [1, n], P station (k) represents the rain intensity measured by the kth rain gauge station, P1 radar (k) represents the radar estimated rain intensity of the grid where the kth rain gauge station is located;

[0070] The core purpose of calculating the radar rain intensity deviation is to quantify the systematic difference between the radar estimated rain intensity and the ground rain gauge station measured rain intensity, and to establish a regional precipitation error correction model. This step uses the rain gauge station as the ground truth to calculate the spatial average deviation, thereby eliminating the systematic deviation caused by radar calibration error, beam shielding, Z-R relationship applicability and other factors. The radar grid rain intensity obtained by subsequent linear correction can not only retain the advantage of high spatial resolution of radar data, but also significantly improve the absolute accuracy of precipitation estimation, and finally realize the complementary advantages of radar remote sensing and ground observation data, and provide more reliable precipitation input field for subsequent soil moisture correction and disaster potential analysis.

[0071] The radar estimated rain intensity of each grid is corrected, and the calculation formula is as follows:

[0072]

[0073] Wherein, represents the radar grid rain intensity of the mth grid.

[0074] In this embodiment, the soil water content of the corrected grid specifically includes:

[0075] Soil moisture sensors are arranged around and in the center of the area to be analyzed to collect soil water content data at the location, and the average value is taken as the soil water content of the area to be analyzed;

[0076] Get DEM data of the area to be analyzed, read DEM data matrix, and then calculate the slope of each grid, correct the soil water content of each grid according to the slope, and the calculation formula is as follows:

[0077]

[0078] Wherein, θ adjusted (m) represents the corrected soil water content of the mth grid, θ raw is the soil water content of the area to be analyzed, and β φis the slope influence coefficient, and φ(m) represents the slope of the mth grid.

[0079] The slope influence coefficient can be obtained by inversion of measured data. For example, a plurality of small sample areas with different slopes are selected in the area to be analyzed, the area of each small sample area is consistent, and the non-slope factors such as soil type and vegetation coverage are consistent. Soil moisture sensors are arranged at the center of each small sample area, and the soil water content is recorded for a continuous period of time (such as 24 hours after rainfall), and then the average slope of each sample area is calculated using a total station or DEM data, and the soil water content in the flat area is measured as a reference value. For each small sample area, record the data pair [tan average slope, ln(soil water content / reference value)], and fit by linear regression to obtain the slope influence coefficient. The value of the slope influence coefficient of sandy soil is in the range of [0.1, 0.3], and the value of the slope influence coefficient of clay soil is in the range of [0.3, 0.5]. The soil water content here refers to the mass water content, which represents the ratio of water mass to soil dry mass. Here, the slope data is used to correct the soil water content, which reflects the consistency of the physical mechanism and the experience of hydrological observation: the slope directly affects the drainage efficiency of surface runoff, and a steeper slope will accelerate the loss of rainwater, reduce the amount of soil infiltration, and result in a decrease in surface water content. Flat areas are more likely to store water, so by introducing a slope influence coefficient to dynamically adjust the original water content, the redistribution of topography on soil moisture can be more realistically reflected. The correction formula is more consistent with the nonlinear hydrological process, thereby improving the accuracy of the subsequent warning model.

[0080] In this embodiment, calculating the disaster potential index specifically includes:

[0081] The historical disaster data includes historical disaster points, disaster duration, disaster type, and damage degree; the historical disaster point refers to the center point coordinate of the disaster area;

[0082] The damage degree is classified into one to four levels, representing slight, moderate, severe, and extremely severe, respectively, and is determined according to the damaged area of the grid. Level one is for a damaged area less than 10% of the grid area, level two is for a damaged area between 10% and 30% of the grid area, level three is for a damaged area between 30% and 60% of the grid area, and level four is for a damaged area higher than 60% of the grid area, and the scores for levels one to four are 1, 3, 5, and 7, respectively.

[0083] The disaster types are divided into flood disaster, geological disaster, and waterlogging disaster. The flood disaster includes river overflow, urban waterlogging, and mountain flood, with corresponding weights of 1.2, 1, and 1.5. The geological disaster is subdivided into landslide, debris flow, and ground subsidence, with corresponding weights of 1.3, 1.6, and 1.1. The waterlogging disaster is divided into farmland waterlogging and road waterlogging, with corresponding weights of 0.8 and 0.5.

[0084] The disaster duration is divided into short-term, medium-term, and long-term, with short-term being 1-3 days, medium-term being 4-7 days, and long-term being more than 8 days. The short-term weight is 1, the medium-term weight is 1.2, and the long-term weight is 1.5. The damage degree score is assigned as the base score. The disaster type weighted base score is obtained by multiplying the corresponding disaster type weight and the base score. The disaster duration weight is combined with the disaster type weighted base score to obtain the score of the historical disaster intensity of the grid. The score is used as the historical disaster intensity of the corresponding grid.

[0085] The calculation formula of the historical disaster intensity is as follows:

[0086] I(m, i) = S d (m, i) x β d (m, i) x β T (m, i)

[0087] where I(m, i) represents the historical disaster intensity of the mth grid in the ith historical disaster, β d (m, i) represents the disaster type weight of the mth grid in the ith historical disaster, β T (m, i) represents the disaster duration weight of the mth grid in the ith historical disaster. In the calculation of the disaster potential index, the scoring and weight setting of damage degree, disaster type, and duration are not randomly determined, but are based on the results of comprehensive weighing of disaster impact mechanism, historical disaster statistics, and expert experience. This weighting method aims to objectively reflect the contribution of different disaster elements to risk, while taking into account the availability of actual data and the operability of calculation.

[0088] The scoring of damage degree uses simple equal interval assignment. The destructive power of disasters often shows exponential growth. For example, when the grid damage area exceeds 60% (level 4), the resulting chain reaction (such as infrastructure paralysis, personnel evacuation, etc.) far exceeds the linear superposition of slight damage. This scoring design can more sensitively capture the impact of high damage level disasters and avoid underestimating the potential risk of extreme events. At the same time, the interval setting of odd values also leaves a clearer distinction for intermediate levels, which facilitates the reduction of boundary ambiguity when classifying actual disaster data.

[0089] The weights of various disasters are mainly based on the destructive power, influence range, suddenness, and social harm degree. In flood disasters, the weight of mountain torrents is the highest because of their strong suddenness and impact, often accompanied by mudslides and causing heavy casualties; the weight of river overflow is second, because of its wide influence range and the possibility of causing basin disasters, but the warning time is relatively long; the weight of urban waterlogging is the lowest, because the damage is usually limited to infrastructure and the recovery is fast. In geological disasters, the destructive power of mudslides is the largest because of their high-speed movement and large-scale solid material characteristics; the weight of landslides is second, which is large in size but slow in movement; the influence range of ground subsidence is usually small and develops slowly. The overall weight of waterlogging disasters is low, among which the weight of farmland waterlogging is higher than that of road waterlogging, because it will lead to long-term economic losses such as crop reduction, while road waterlogging mainly causes short-term traffic interruption. This weight setting not only considers the physical characteristics of various disasters, but also takes into account their social and economic impact, ensuring that the disaster potential assessment results can objectively reflect the actual risk level.

[0090] The weight of duration reflects the cumulative effect and recovery cost of disasters. Short-term disasters may only cause instantaneous damage, while long-term disasters will cause irreversible losses such as soil saturation and crop failure, and even trigger secondary disasters (such as landslides). The setting of this linear weighting method, on the one hand, avoids overestimating the impact of duration, and on the other hand, can effectively distinguish the difference between instantaneous impact and continuous pressure disasters.

[0091] The disaster potential index of each grid is calculated as follows:

[0092]

[0093] where H risk (m) represents the disaster potential index of the mth grid, D(m, i) represents the distance from the mth grid to the historical disaster point at the ith historical disaster, m is the grid index, M is the total number of grids, and m ∈ [1, M], i is the historical disaster index, U is the total number of historical disasters in the historical period, and i ∈ [1, U], usually the disaster data of the latest complete hydrological year, usually 12 months.

[0094] The multi-dimensional weighting method is used to comprehensively evaluate the disaster potential, avoiding the one-sidedness of a single indicator. The division and weighting of the three core elements of damage degree, disaster type, and duration are consistent with the general rules of disaster impact. For example, the damage degree is classified based on the damaged area of the grid, which can intuitively reflect the direct impact range of the disaster; the weight setting of disaster type considers the damage difference of different disasters, which is consistent with the cognition that geological disasters are usually more destructive than waterlogging disasters in actual disaster situations; and the weighting of duration reflects the cumulative effect of disasters, and long-term disasters often lead to more serious secondary risks. This hierarchical weighting design not only retains the judgment of expert experience, but also reduces subjective bias through quantitative rules. The calculation of historical disaster intensity is weighted twice by the type and duration of the disaster, making the results not only distinguish the instantaneous destructive power of the disaster, but also reflect the potential impact of long-term disasters.

[0095] In the denominator of the disaster potential index calculation formula, the distance D(m, i) is divided by 500, with the unit being meters, which is actually a standardization process for distance. 500 can be regarded as a threshold that adjusts the impact of distance on the disaster potential index. Specifically, when the distance D(m, i) is less than 500 meters, the value of the denominator is small, and the disaster potential index is greatly affected by the historical disaster intensity. When the distance exceeds 500, the value of the denominator increases rapidly, making the value of the disaster potential index decrease significantly, indicating that the farther the distance, the weaker the potential impact of the grid.

[0096] As the dependent variable, the disaster potential index reflects the potential risk level of the mth grid in the future disaster, and its value range is determined by the historical disaster intensity and the spatial decay law. This index quantifies the spatial correlation between the grid and the historical disaster point and the severity of the disaster, converting discrete historical disaster data into continuous spatial risk distribution. The technical effect is to realize the dynamic spatial visualization of disaster risk, making it possible to accurately identify high-risk areas. The historical disaster intensity directly reflects the destructive power of the disaster, and its value is derived from the expert's comprehensive evaluation of disaster type, damage degree, and other factors, determining the basic magnitude of the risk. The distance is simulated through the decay function, which models the physical law of the decrease of disaster impact with distance. The synergistic effect of the two shows that the closer the distance to the disaster point and the more severe the historical disaster, the higher the potential risk of the grid. The independent variable historical disaster intensity is positively correlated with the disaster potential index, and the index value increases linearly with the increase of 1 unit of disaster intensity. The distance and the disaster potential index are nonlinearly negatively correlated, and the index rapidly decays according to the inverse square law when the distance increases. This relationship is consistent with the transmission characteristics of natural disasters (such as the rapid contraction of flood impact range with distance), ensuring that the model can highlight the extreme risk of the core disaster area and reasonably reflect the risk gradient change in the marginal area.

[0097] The measured data of the rain stations is fused to dynamically correct the radar estimated rain intensity, effectively eliminating the system bias and significantly improving the absolute accuracy of the precipitation data; the slope factor calculated based on the DEM data is used to spatially differentiate the soil moisture, so that the hydrological characteristics of the underlying surface are more in line with the actual topographic distribution; the historical disaster intensity is quantified based on the expert scoring method and the distance attenuation function is introduced to calculate the disaster potential index, which not only objectively reflects the spatial aggregation and historical recurrence of disasters, but also realizes the transformation of disaster risk from point disaster records to face grid evaluation, thereby building a multi-dimensional risk assessment system integrating real-time precipitation, soil moisture and historical disaster memory, and providing a comprehensive research basis with physical mechanism and statistical law for fine meteorological and hydrological warning.

[0098] Step 3: Obtain the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture of each grid in the current monitoring time period in the area to be analyzed, and train the LSTM model based on the direct runoff depth prediction value of each grid at the future time output by the LSTM model;

[0099] In this embodiment, training the LSTM model specifically includes:

[0100] The length of the current monitoring time period can be set to 24 hours, and the time interval between adjacent collection time points can be set to a sampling frequency of 15 minutes. 24 hours can cover a complete small rainfall process, such as short-term heavy rain or continuous moderate rain, and this time length can provide sufficient time series data without causing LSTM training efficiency to decrease due to excessive data volume. For the time interval between adjacent collection time points, a sampling frequency of 15 minutes is recommended, which not only meets the hydrological monitoring specification, but also effectively captures the rapid change of runoff depth during heavy rain, while avoiding data redundancy and storage pressure caused by high-frequency sampling. A high-precision water level meter, such as a pressure-type water level sensor or a radar water level meter, is arranged at the center of each grid to obtain the direct runoff depth of each grid. The direct runoff depth, radar grid rain intensity, slope and corrected soil moisture data of each grid in the current monitoring time period are divided into a training set, a validation set and a test set in a ratio of 7:1.5:1.5. The model is trained by the training set, specifically by using the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture data of the grid at the first ten time points in the training set as input, and using the direct runoff depth of the grid at the next time point as output label to train the model;

[0101] The LSTM layer can be set to 2 to 3 layers, the number of neurons in each layer is set to 64 to 256, the activation function is tanh, which is used to capture the long and short term dependence relationship of runoff formation; the fully connected layer is connected with 1 neuron at the end (linear activation) to output the direct runoff depth prediction value, and Dropout (ratio 0.2 to 0.5) is added between layers to prevent overfitting. If the loss of the validation set changes less than a preset threshold in ten iteration processes, the training is automatically terminated, and the threshold can be set to 1e-5. Then the data in the test set is input for testing, and the mean square error and the mean absolute error are used as evaluation indexes, the excellent threshold of the mean square error is set to less than 0.5, and the excellent threshold of the mean absolute error is set to less than 0.5, and if the evaluation indexes are lower than the excellent thresholds, the model performance is excellent, if both evaluation indexes are lower than the excellent thresholds, the model is retrained, and if the model performance is excellent, the grid direct runoff depth, radar grid rain intensity, slope and corrected soil moisture data at the current time and the previous nine times are input into the model to obtain the direct runoff depth prediction value of the grid at the next time.

[0102] Step 4: combining the direct runoff depth prediction value and the disaster potential index to determine the disaster risk index of each grid, and combining the elevation change rate and the human flow density of each grid in the current monitoring period, a multi-layer early warning triggering mechanism is set to issue different degrees of disaster warning for each grid.

[0103] In this embodiment, the disaster risk index is determined specifically as follows:

[0104] The disaster potential index and the direct runoff depth prediction value of each grid are normalized, and the normalized data is used for calculation, and the formula is as follows:

[0105]

[0106] Wherein, Risk(m) is the disaster risk index of the mth grid, Q(m) represents the direct runoff depth prediction value of the mth grid, Q c represents the direct runoff depth threshold, which can be analyzed from the direct runoff depth when the historical disaster data occurs, and the minimum runoff depth when the disaster occurs is taken as the threshold, for example, the historical disaster occurs when the direct runoff depth is not less than 80mm, then Q c = 80mm, H(m) represents the normalized disaster potential index of the mth grid, and a and l are weight coefficients of the corresponding items, and l > a > 1.

[0107] In the formula of disaster risk index, the dependent variable specifically reflects the disaster risk level of the mth grid. Its meaning is to quantify the potential disaster risk of the grid under certain conditions by considering the influence of direct runoff depth and disaster potential index. This design can provide scientific basis for disaster management and emergency response, help relevant departments to develop more accurate early warning and response measures, and thus reduce the potential threat of disasters to the population and assets.

[0108] The independent variables include the predicted value of direct runoff depth and the normalized disaster potential index, which have a close correlation with the dependent variable. The predicted value of direct runoff depth reflects the possible water flow depth of a grid under the influence of precipitation or other factors, while the disaster potential index assesses the vulnerability and bearing capacity of the region under various natural disasters. These two independent variables together determine the disaster risk of the grid, and their changes directly affect the calculation results of the disaster risk index, indicating the risk status of the grid under different precipitation conditions and disaster potential.

[0109] In the formula, the dependent variable Risk(m) has a positive correlation with the independent variables Q(m) and H(m). That is, as the predicted value of direct runoff depth Q(m) increases, or the normalized disaster potential index H(m) increases, the disaster risk index Risk(m) will also increase accordingly. This positive correlation shows that when the water flow depth of the grid increases or the disaster potential increases, the risk of disaster also increases, prompting relevant departments to take more urgent early warning and prevention measures.

[0110] The size relationship between the weight coefficients a and l generally depends on the relative importance of direct runoff depth and disaster potential index to disaster risk in a specific situation in the disaster risk index calculation formula. Generally speaking, if the hydrological characteristics of an area make the runoff depth particularly significant to disaster impact, the value of a can be appropriately increased to emphasize the contribution of runoff depth to risk; conversely, if the disaster potential index of the region plays a more critical role in decision-making, the value of l should be increased. By reasonably setting the two weight coefficients, the influence of different factors on disaster risk can be more accurately reflected, so that the calculated risk index is more in line with the actual situation, helping relevant departments to make more effective strategy adjustments when responding to disasters. Here, the influence of the disaster potential index is emphasized, so l > a > 1 is set. The influence of different factors on disaster risk can be quantified by statistical analysis methods such as regression analysis or principal component analysis to determine the value of the weight coefficient, and the disaster risk index, direct runoff depth prediction value and direct runoff depth threshold value of each grid are obtained by collecting historical disaster data, and the least squares method is used to fit to obtain the weight coefficients a and l; the importance of direct runoff depth and disaster potential in a specific situation can also be determined by expert evaluation, such as the Delphi method, in which hydrological and geological experts independently score the contribution of direct runoff depth prediction value and disaster potential index, and the weight coefficients a and l are obtained by converting the weight proportion.

[0111] In this embodiment, 30 groups of direct runoff depth prediction value, disaster potential index and disaster risk index data are collected, the direct runoff depth threshold is set to 80mm, a = 1.2, and l = 1.5, and the specific data are shown in the following table:

[0112] Table 1: Sample data table for analyzing disaster risk index

[0113] Sample No. Direct runoff depth prediction value Disaster potential index Disaster risk index 1 10 0.1 0.022 2 12 0.12 0.032 3 15 0.15 0.049 4 18 0.18 0.071 5 20 0.2 0.089 6 22 0.22 0.11 7 25 0.25 0.147 8 28 0.28 0.19 9 30 0.3 0.226 10 32 0.32 0.266 11 35 0.35 0.334 12 38 0.38 0.413 13 40 0.4 0.476 14 42 0.42 0.545 15 45 0.45 0.666 16 48 0.48 0.806 17 50 0.5 0.909 18 52 0.52 1.022 19 55 0.55 1.219 20 58 0.58 1.44 21 60 0.6 1.618 22 62 0.62 1.81 23 65 0.65 2.126 24 68 0.68 2.475 25 70 0.7 2.755 26 72 0.72 3.055 27 75 0.75 3.543 28 78 0.78 4.074 29 80 0.8 4.5 30 82 0.82 4.952

[0114] Referring to Figures 2-3 It can be seen from the above table that the disaster risk index and the independent variables direct runoff depth prediction value and disaster potential index exhibit a positive correlation relationship. That is, as the direct runoff depth prediction value increases, or the normalized disaster potential index increases, the disaster risk index also increases accordingly.

[0115] In this embodiment, the multi-level early warning triggering mechanism specifically includes:

[0116] A three-level grading warning trigger mechanism is set, and each level is upgraded, each level of the warning trigger mechanism includes level, judgment condition and warning level, specifically: the first level, if the disaster risk index of the grid is higher than the preset risk threshold, a yellow warning is issued; the second level, on the basis of the judgment condition of the first level, the elevation change rate of the grid is calculated, if the elevation change rate is greater than 15%, an orange warning is issued; the third level, on the basis of the judgment condition of the second level, the residential density of the grid is judged, if the residential density is higher than 50 people / km 2 , a red warning is issued; here, the residential density judgment basis can be adjusted according to the characteristics of different regions, such as the density standards of cities and villages may be different. In addition, residential density includes daily resident population, night resident population and mobile flow, for different disaster types, different types of residential density can be referred to, such as urban waterlogging or daytime geological disasters, daily resident population can be referred to, night landslide or earthquake can be referred to night resident population, holiday scenic spot mountain flood, mobile flow can be referred to.

[0117] By analyzing the DEM data matrix, the elevation value of each grid is obtained, and then the elevation change rate calculation formula is as follows:

[0118]

[0119] Where, ΔZ(m) represents the elevation change rate of the mth grid, Z(m,t) represents the elevation value of the mth table at the current monitoring time, Z(m,t-Δt) represents the elevation value of the mth table at the previous monitoring time, and Δt is the interval time, which can be set to 6 hours.

[0120] Here, the interval time can be set to 6 hours, and the 6-hour elevation change calculation can cover multiple rainfall intensity peaks, avoiding misjudgment caused by instantaneous rainfall, such as short-time heavy rainfall may cause temporary subsidence of the ground, but not necessarily constitute a sustained disaster risk.

[0121] Please refer to Figure 2 , the application further provides a meteorological and hydrological data analysis system based on multi-source data fusion, which is used to realize the meteorological and hydrological data analysis method based on multi-source data fusion, comprising:

[0122] The grid radar data acquisition module is used for dividing the area to be analyzed into a plurality of grids of the same size, collecting the radar reflectivity factor of the grid through the weather radar, and converting it into the radar estimated rain intensity;

[0123] The rain intensity deviation and disaster potential assessment module is used for collecting rain intensity of a rain gauge station and historical disaster data of a region to be analyzed, calculating radar rain intensity deviation based on the rain intensity of the rain gauge station, determining radar grid point rain intensity of a grid, obtaining slope of the grid to correct soil moisture content of the grid, determining historical disaster intensity of the grid through expert scoring method, and calculating disaster potential index of the grid according to the historical disaster intensity and distance from the grid to a disaster location;

[0124] The runoff depth prediction module is used for obtaining direct runoff depth, radar grid point rain intensity, slope and corrected soil moisture content of each grid in a current monitoring time period in the region to be analyzed, training an LSTM model based on the above, and outputting direct runoff depth prediction value of each grid at a future time based on the LSTM model.

[0125] The risk assessment and early warning module is used for combining the direct runoff depth prediction value and the disaster potential index to determine disaster risk index of each grid, and combining elevation change rate and people flow density of each grid in the current monitoring time period to set a multi-layer early warning triggering mechanism to issue disaster early warning of different degrees for each grid.

[0126] The above formulas are all dimensionless values, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the nearest real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0127] The above embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0128] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiments according to actual needs.

[0129] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A meteorological hydrological data analysis method based on multi-source data fusion, characterized in that, The specific steps include: Step 1: Divide the area to be analyzed into a plurality of grids of the same size, collect the radar reflectivity factor of the grid by the weather radar, and convert it into the radar estimated rainfall intensity; Step 2: Collect the rainfall station rainfall intensity and historical disaster data of the area to be analyzed, calculate the radar rainfall intensity deviation based on the rainfall station rainfall intensity, determine the radar grid rainfall intensity of the grid, obtain the slope of the grid to correct the soil moisture content of the grid, determine the historical disaster intensity of the grid by expert scoring method, and calculate the disaster potential index of the grid according to the historical disaster intensity and the distance from the grid to the disaster location; Step 3: Obtain the direct runoff depth, radar grid rainfall intensity, slope and corrected soil moisture content of each grid in the area to be analyzed in the current monitoring period, and train the LSTM model based on the direct runoff depth prediction value of each grid at the future time output by the LSTM model; Step 4: Combine the direct runoff depth prediction value and the disaster potential index to determine the disaster risk index of each grid, and set a multi-layer early warning trigger mechanism according to the elevation change rate and the human flow density of each grid in the current monitoring period to issue different degrees of disaster warning for each grid. 2.The weather and hydrological data analysis method based on multi-source data fusion according to claim 1, characterized in that, The radar estimated rainfall intensity is obtained specifically as follows: Divide the area to be analyzed into a plurality of 1km×1km grids, convert the radar reflectivity factor of the grid into the radar estimated rainfall intensity by the Z-R relationship, and the calculation formula is as follows: P radar (m) = a -1 / b ·10 0.1 / b·z where P radar (m) is the radar-estimated rain rate for the mth grid cell, Z is the radar reflectivity factor, a represents a proportionality constant, b represents a degree of nonlinearity, m is the grid index, M is the total number of grid cells, and m ∈ [1, M]. 3.The weather and hydrological data analysis method based on multi-source data fusion according to claim 2, characterized in that, The radar grid rainfall intensity is determined specifically as follows: Based on the rainfall station rainfall intensity data, calculate the radar rainfall intensity deviation, and the formula is as follows: where ΔP represents the radar rain rate bias, k is the index of the rain gauge, n represents the total number of rain gauges, and k ∈ [1, n], P station (k) represents the rain rate measured by the kth rain gauge, P radar (k) represents the radar estimated rain rate of the grid where the kth rain gauge is located. The radar estimated rainfall intensity of each grid is corrected, and the calculation formula is as follows: wherein, Rm represents the radar grid rain intensity of the mth grid. 4.The weather and hydrological data analysis method based on multi-source data fusion according to claim 3, characterized in that, The soil moisture content of the grid is corrected specifically as follows: Soil moisture sensors are arranged around and in the center of the area to be analyzed to collect the soil moisture content data at the location, and the average value is taken as the soil moisture content of the area to be analyzed; The slope of each grid is obtained, and the soil moisture content of each grid is corrected according to the slope, and the calculation formula is as follows: where θ adjusted (m) represents the soil moisture content of the mth grid after correction, θ raw is the soil moisture content of the region to be analyzed, β φ is the slope influence coefficient, and φ(m) represents the slope of the mth grid. 5.The weather and hydrological data analysis method based on multi-source data fusion according to claim 1, characterized in that, The disaster potential index is calculated specifically as follows: The historical disaster data includes historical disaster points, disaster duration, disaster types and damage degree, and the historical disaster point refers to the center point coordinate of the disaster area, the historical disaster intensity of each grid is determined by expert scoring method based on the historical disaster data, and the calculation formula of the disaster potential index of each grid is as follows: where H risk (m) represents the disaster potential index of the mth grid, I(m, i) represents the historical disaster intensity of the mth grid at the ith historical disaster, D(m, i) represents the distance from the mth grid to the historical disaster point at the ith historical disaster, m is the grid index, M is the total number of grids, and m ∈ [1, M], i is the historical disaster index, U is the total number of historical disasters in the historical period, and i ∈ [1, U]. 6.The weather and hydrological data analysis method based on multi-source data fusion according to claim 1, characterized in that, The LSTM model is trained specifically as follows: A water level gauge is arranged at the center of each grid to obtain the direct runoff depth of each grid. The direct runoff depth, radar grid rain intensity, slope and corrected soil moisture of each grid in the current monitoring period are divided into a training set, a validation set and a test set in a ratio of 7:1.5:1.

5. The model is trained by using the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture of the grid in the first ten time points in the training set as input, the direct runoff depth of the grid in the next time point as output label, and the direct runoff depth prediction value of the grid in the next time point as output. If the loss of the validation set changes by less than a preset threshold in ten iterations, the training is automatically terminated. The data in the test set is input for testing, and the mean square error and mean absolute error are used as evaluation indexes. If the testing is not qualified, the model is retrained. If the testing is qualified, the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture of the grid in the current time point and the previous nine time points are input into the model to obtain the direct runoff depth prediction value of the grid in the next time point. 7.The weather and hydrological data analysis method based on multi-source data fusion according to claim 1, characterized in that, The determination of the disaster risk index specifically includes: The disaster potential index and the direct runoff depth prediction value of each grid are normalized, and the normalized data are used for calculation according to the following formula: wherein Risk(m) is the disaster risk index of the mth grid, Q(m) represents the direct runoff depth prediction value of the mth grid, Q c represents the direct runoff depth threshold value, H(m) represents the normalized disaster potential index of the mth grid, and a and l are weight coefficients of the corresponding items. 8.The weather and hydrological data analysis method based on multi-source data fusion according to claim 1, characterized in that, The multi-layer early warning triggering mechanism is set specifically as follows: The early warning trigger mechanism of three-level grade judgment is set, and is upgraded layer by layer, each layer of early warning trigger mechanism includes level, judgment condition and early warning level, specifically: the first level, if the disaster risk index of the grid is higher than the preset risk threshold, a yellow early warning is issued; the second level, on the basis of the judgment condition of the first level, the elevation change rate is calculated, if the elevation change rate is greater than 15%, an orange early warning is issued; the third level, on the basis of the judgment condition of the second level, the residential density of the grid is judged, if the residential density is higher than 50 people / km 2 , a red early warning is issued; The change rate of elevation is calculated according to the following formula: wherein, ΔZ(m) represents the change rate of elevation of the mth grid, Z(m,t) represents the elevation value of the mth table at the current monitoring time, Z(m,t-Δt) represents the elevation value of the mth table at the previous monitoring time, and Δt is the interval length. 9.A meteorological and hydrological data analysis system based on multi-source data fusion, characterized in that, The meteorological and hydrological data analysis system based on multi-source data fusion is used to realize the meteorological and hydrological data analysis method based on multi-source data fusion according to any one of claims 1-8, comprising: A gridded radar data acquisition module is configured to divide the area to be analyzed into a plurality of grids of the same size, acquire the radar reflectivity factor of each grid by a weather radar, and convert the radar reflectivity factor into a radar estimated rain intensity. A rain intensity deviation and disaster potential evaluation module is configured to acquire the rain intensity of a rain gauge in the area to be analyzed and historical disaster data, calculate the radar rain intensity deviation based on the rain intensity of the rain gauge, determine the radar grid rain intensity of each grid, acquire the slope of each grid to correct the soil moisture of each grid, determine the historical disaster intensity of each grid by an expert scoring method, and calculate the disaster potential index of each grid according to the historical disaster intensity and the distance from each grid to the location where the disaster occurred. A runoff depth prediction module is configured to acquire the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture of each grid in the area to be analyzed in the current monitoring period, train an LSTM model based on the direct runoff depth, radar grid rain intensity, slope and corrected soil moisture, and output the direct runoff depth prediction value of each grid in the future time based on the LSTM model. The risk assessment and early warning module is used for combining the direct runoff depth prediction value and the disaster potential index to determine the disaster risk index of each grid, and combining the elevation change rate and the people flow density of each grid in the current monitoring time period to set a multi-layer early warning trigger mechanism to issue disaster early warnings of different degrees for each grid.

Images