Precipitation estimation method and system based on land rainfall multi-data fusion and application

Abstract

The invention discloses a precipitation estimation method and system based on territorial rainfall multi-data fusion and application, and relates to the technical field of weather forecast, by fusing radar, satellite and ground observation data, the accuracy and reliability of precipitation prediction are remarkably improved, the fusion of multi-source data can make up for the deficiency of a single data source, and the prediction accuracy and reliability are improved. Through space-time registration and a weighted average fusion algorithm, a macroscopic rainfall background is provided by introducing territorial rainfall data, and radar, satellite and ground observation rainfall data are combined, so that rainfall estimation is more comprehensive, ground observation stations are supplemented and fused, effective data of the stations are increased, the real time of rainfall can be reflected more accurately, and the rainfall estimation accuracy is improved. In addition, the fusion of multi-source data can make up for the deficiency of a single data source, and the advantages of different data sources can be effectively integrated through the space-time registration and weighted average fusion algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecasting, and in particular to a precipitation estimation method, system and application based on national land rainfall multi-data fusion. Background Art

[0002] Precipitation is an important part of the water cycle and is of great significance to agricultural production, water resources management, urban planning, disaster prevention and mitigation. Accurate precipitation forecasts can not only help farmers arrange planting plans reasonably, but also provide a scientific basis for the planning and construction of urban drainage systems. Traditional precipitation forecasting methods mainly rely on data from ground observation stations. Although these data are accurate, their coverage is limited, especially in remote areas or when observation stations are sparsely distributed, the accuracy and reliability of the forecast are greatly reduced. In recent years, with the development of remote sensing technology, radar and satellite observation data have gradually become an important data source for precipitation forecasting. These data can provide large-scale, high-resolution precipitation information, making up for the lack of data from ground observation stations.

[0003] Then, radar observation data is easily affected by terrain, buildings and other obstacles, resulting in obstruction and reflection of radar beams, thus generating errors. Secondly, satellite observation data is greatly affected by clouds and atmospheric conditions, especially in heavy rain or thunderstorm weather, the quality of satellite data will be significantly reduced. In addition, the inconsistent spatiotemporal resolution between different data sources makes data fusion a complex task. Most existing data fusion methods rely on simple linear combinations or fixed weight allocations, which cannot effectively cope with data changes under different weather conditions, resulting in low prediction accuracy. Therefore, it is particularly important to develop a precipitation estimation method that can effectively fuse multiple data sources. Summary of the invention

[0004] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide a precipitation estimation method, system and application based on the fusion of national land rainfall multi-data. It can integrate radar, satellite and ground observation data. The fusion of multi-source data can make up for the shortcomings of a single data source. Through spatiotemporal alignment and weighted average fusion algorithm, it can effectively integrate the advantages of different data sources to generate high-resolution and high-precision precipitation prediction results.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: On the one hand, a precipitation estimation method based on land rainfall multi-data fusion, the specific steps of the method are:

[0006] S1. Rainfall data collection and integration: Collect national rainfall data, radar observation precipitation data, satellite observation precipitation data and ground observation precipitation data, conduct preliminary integration of various types of collected data to ensure unified data format and time stamp alignment, and conduct preliminary checks on the integrity and accuracy of the data, including:

[0007] For national rainfall data, we obtain the rainfall accumulation values ​​of different terrains and climate regions that have been observed and recorded for a long time from the meteorological data storage center, clean the data, and check for missing and incorrect data. If there are data problems, we repair and supplement them by comparing them with the data of surrounding areas and referring to the historical data of the same period.

[0008] For radar observations, precipitation data are collected from a meteorological radar system, which includes multiple radar stations distributed across the country. The collected data include information on radar reflectivity and radial velocity at different angles and distances;

[0009] The satellite-observed precipitation data is obtained by receiving data from meteorological satellites, which use different sensors to observe the earth's surface and convert the satellite data into a uniform format;

[0010] For ground-based precipitation data, which comes from a distributed network of ground-based rainfall observation stations, during the data collection process, ensure that the equipment at each observation station is operating normally and data transmission is stable, and perform quality control on the collected data;

[0011] S2, data preprocessing: preprocessing the radar observed precipitation data, satellite observed precipitation data and ground observed precipitation data, wherein;

[0012] For radar observed precipitation data, the calibration algorithm is used for preprocessing. The noise signal caused by the radar equipment itself and the external environment interference is removed by the filtering algorithm. Based on the reflectivity data and calibration equipment, the radar's transmission frequency, gain, and beam width parameters are calibrated to make the radar observation data accurately reflect the actual precipitation situation.

[0013] For the precipitation data observed by satellite, format conversion and geolocation processing are performed. The format conversion of the claim form is based on the format specifications of different satellite data, and the conversion algorithm is used to unify them into a standard format. The geolocation processing adjusts the inconsistency of spatial resolution and temporal resolution between different satellite data due to sensor differences and orbital characteristics through resampling and time interpolation methods. The satellite data is geolocated using the positioning information and ground calibration points of the satellite itself to ensure that the precipitation data observed by the satellite can accurately correspond to the actual position on the earth;

[0014] Quality inspection is performed on the ground-based precipitation data. The quality inspection is performed by comparing the data with the data of multiple surrounding observation stations. If the data of an observation station deviates from the average value of the surrounding stations, the data is marked and further inspected. For abnormal data caused by equipment failure, the maintenance personnel are notified in time to repair it, and the abnormal value is eliminated in data processing. The further data inspection replaces the abnormal value by estimating the standard value using the data of surrounding stations through the spatial interpolation method;

[0015] S3, multi-source data fusion processing: perform spatiotemporal registration of the pre-processed radar, satellite, and ground-based precipitation data with the national rainfall data, and fuse the registered data using the weighted average method;

[0016] The spatiotemporal registration converts the coordinate system of radar and satellite data into a geographic coordinate system consistent with ground observation data in space. For radar data, the polar coordinate data detected by the radar is converted into two-dimensional plane data under geographic coordinates through the known position of the radar station and the beam scanning geometry. For satellite data, it is projected into the geographic coordinate system of the earth's surface using satellite orbit parameters and geographic positioning algorithms.

[0017] In terms of time, data from different data sources are aligned to a unified time scale based on the data collection timestamp;

[0018] The data fusion adopts a weighted average fusion algorithm, which assigns an initial weight to each data source under different circumstances through comparative analysis and statistics of historical data. In the real-time fusion process, the weight is dynamically adjusted according to the current weather conditions and data quality indicators, and the processed radar, satellite and ground observation precipitation data are fused into a comprehensive precipitation data set;

[0019] S4. Construction and training of precipitation estimation model: A precipitation estimation model is established based on the fused multi-source precipitation data. The precipitation estimation model includes: an estimation model based on physical principles based on the physical process of precipitation formation, and an estimation model based on a data-driven multi-layer perceptron;

[0020] The model based on physical principles simulates precipitation by establishing physical equations of atmospheric movement and precipitation formation;

[0021] The multi-layer perceptron estimation model uses the fused precipitation data as the input feature vector and the historical precipitation data as the target output. The data set is divided into different subsets according to 70%-80% as the training set, 10%-15% as the validation set, and the rest as the test set. The model is trained using the training set. During the training process, the hyperparameters of the model are adjusted through the validation set, and the performance of the model is evaluated using the test set.

[0022] Furthermore, the S1 is for land rainfall data, and its land rainfall data set is G = {g1, g2, ..., g n}, where g i represents the rainfall data of the ith region. For missing data, the missing data location is defined as j. There is a rainfall data set S = {s1, s2, ..., s m}, then fix it to That is, the average rainfall data of the surrounding area is used to repair the missing values; for the erroneous data, the historical data set H = {h1, h2, ..., h l}Compared with the rainfall data in the same area, when g i If the difference between the corresponding position data in H exceeds the threshold T1, then g i Considered as wrong data, repaired to g i =hi, that is, directly replace the wrong data with historical data of the same period;

[0023] For radar observed precipitation data, the radar observed precipitation data set is R = {r1, r2, ..., r p}, where r i Includes radar reflectivity r i1 and radial velocity r i2 For data collected at different angles θ and different distances d, the data collection is represented by r i =f(θ, d), where f is the radar data acquisition function, for radar reflectivity r i1 and radial velocity r i2 The data collection function f(θ, d) is: and Where θ is the elevation angle of the radar beam, d is the distance between the target and the radar, P(θ, d) is the distribution function of the radar transmission power in the (θ, d) direction, λ is the wavelength of the electromagnetic wave emitted by the radar, and |K| 2 is the coefficient of precipitation particles, σ k is the radar cross section of the kth precipitation particle at a distance d from the radar, n is the total number of precipitation particles within the effective illumination volume of the radar beam, and f d is the Doppler shift;

[0024] For satellite observed precipitation data, the satellite observed precipitation data set is S = (s1, s2, ..., s q}, the satellite sensor type set is For each sensor t j Collected data i j, the format is converted to in For sensor t j Format conversion function;

[0025] For ground-based precipitation data, the ground-based precipitation data set is L = {l1, l2, ..., l r}, the observation station data is l i , the data set of the surrounding n observation stations is N = {n1, n2, ..., n n}, and its average value is Quality control is: Then mark l i is abnormal data, where T2 is the data abnormality threshold. For abnormal data, when it is judged to be caused by equipment failure, it is repaired to l through the spatial interpolation function I. i =I(N), that is, the replacement value is estimated by spatial interpolation method using the surrounding station data.

[0026] Furthermore, the radar observed precipitation data is obtained by S2 according to the collected radar observed precipitation data set R = {r1, r2, ..., r p}, where r i =(r i1 , r i2 ), r i1 is the radar reflectivity, r i2 is the radial velocity, and the filtered radar observation precipitation data set is defined as R f = {r f1 , r f2 ,…,r fp}, for each r i , using a threshold-based filtering algorithm, the noise threshold is T n , then the filtering is: That is, when the radar reflectivity and radial velocity are both greater than the noise threshold, the original data are retained, otherwise they are regarded as noise data and set to 0, and the calibrated radar observation precipitation data set is defined as R c = {r c1 , r c2 ,…,r cp}, for each r fi , whose standard reflectivity is r s1 , the standard gain is G s , the standard beam width is θ s , the current radar reflectivity is r fi1 ,, the current gain is G i , the current beam width is θ i , then the calibration is

[0027] For the calibration of radar observation precipitation data parameters, the standard reflectivity is r s , the standard gain is G s, the standard beam width is B s , the current radar reflectivity is The current gain is G i , the current beam width is B i , the calibrated radar reflectivity is calculated as:

[0028] Furthermore, when marking outliers in the ground observation precipitation data quality inspection in S2, using P i as the precipitation observation value of the i-th ground observation station, as the average precipitation of multiple surrounding observation stations, σ P as the standard deviation of precipitation of multiple surrounding observation stations, when satisfying n is a constant, then P i is marked as an outlier.

[0029] Furthermore, the time interpolation method for aligning data in time in the spatio-temporal registration in S3 is as follows: P1(t1) and P2(t2) are precipitation data at two adjacent time points, and the precipitation data P(t) at time t (t1 < t < t2) is obtained by interpolation, then:

[0030] Furthermore, the integrated precipitation data after fusion in S3 is P fusion , the radar observation precipitation data is P r , the satellite observation precipitation data is P s , the ground observation precipitation data is P g , then the weighted average fusion formula is: P fusion = W r P r + W s P s + W g P g , where W r , W s , W g are the weights of radar, satellite, and ground observation data respectively, and satisfy W r + W s + W g = 1, and W r , W s , W g in the weighted average fusion algorithm, using A r , A s , A g are the comprehensive evaluation scores of radar, satellite, and ground observation data under the current weather conditions and data quality indicators respectively, then: Among them, the comprehensive evaluation score of radar data is obtained through Ar =a1Q r +a2C r +a3E r Similarly, we get the comprehensive evaluation score of satellite and ground data, Q r is the radar data quality, C r is the matching degree between radar data and current weather conditions, E r is the historical accuracy of radar data in the current geographical area, and a1, a2, and a3 are the corresponding weight coefficients.

[0031] On the other hand, a precipitation estimation system based on multi-data fusion of land rainfall is provided, the system comprising: a data acquisition module, a data processing module, a data fusion module, and a precipitation estimation module;

[0032] The data acquisition module accurately collects national rainfall data, radar observation precipitation data, satellite observation precipitation data and ground observation precipitation data through the device end, and transmits it to the data center;

[0033] The data preprocessing module preprocesses the radar observed precipitation data, satellite observed precipitation data and ground observed precipitation data;

[0034] The data fusion module performs spatiotemporal registration of the pre-processed radar, satellite, and ground-observed precipitation data with the land rainfall data, and fuses the registered data by a weighted average method;

[0035] The precipitation estimation module establishes a precipitation estimation model based on the fused multi-source precipitation data.

[0036] Furthermore, the application of the system in meteorological disaster warning and infrastructure construction specifically includes:

[0037] Early warning system: When the estimated precipitation exceeds the warning threshold, the system automatically generates early warning information, including the type of disaster, expected time of occurrence, and scope of impact. Through cooperation with communication operators and radio and television stations, the early warning information is sent to the public, emergency management departments, and water conservancy departments in a timely manner;

[0038] Infrastructure construction system: Combine precipitation estimation data with geographic information system (GIS) to analyze precipitation risks in different regions;

[0039] Emergency preparedness system: When emergency disastrous weather occurs, the availability of ground observation station data is monitored in real time. For stations that cannot provide data normally, the interpolation algorithm is used to calculate the estimated data of the missing stations based on the spatial distribution of surrounding station data and fused precipitation data. The data is then transmitted to the emergency command center to provide data support for emergency decision-making.

[0040] Compared with the existing technology, the precipitation estimation method, system and application based on land rainfall multi-data fusion have the following beneficial effects:

[0041] 1. The present invention provides a macro precipitation background by introducing national land rainfall data, and combines radar, satellite, and ground observation precipitation data to make precipitation estimation more comprehensive. The supplementation of ground observation sites and data fusion increase the effective data of the sites, which can more accurately reflect the actual precipitation time and strive for more advance time for disaster warning. In addition, the fusion of multi-source data can make up for the shortcomings of a single data source. Through spatiotemporal registration and weighted average fusion algorithm, it can effectively integrate the advantages of different data sources to generate high-resolution and high-precision precipitation prediction results, further improving the stability of the prediction. This high-precision precipitation prediction is of great significance for agricultural irrigation, urban drainage system planning and water resources management.

[0042] 2. The present invention uses a spatial interpolation method based on the geographic information system (GIS) to quickly calculate missing values ​​based on the spatial distribution characteristics of surrounding normal station data and fused precipitation data. This ensures that the early warning system can still accurately determine whether the precipitation exceeds the disaster threshold in an emergency and issue disaster warnings in a timely manner. In terms of infrastructure construction, the continuous supply of data in emergency situations ensures the dynamic adjustment of construction plans, effectively reduces disaster losses, and improves the ability of the entire society to respond to emergency disastrous weather.

[0043] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 It is an operation flow chart of a precipitation estimation method based on multi-data fusion of land rainfall;

[0046] Figure 2 This is a module composition diagram of a precipitation estimation system based on the fusion of national land rainfall multi-data. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Embodiment 1

[0049] This embodiment elaborates on the application of the precipitation estimation method based on the fusion of national land rainfall multi-data in urban rainstorm and flood warning. By integrating national land rainfall data, radar observation precipitation data, satellite observation precipitation data and ground observation precipitation data, and using technical means such as data preprocessing, fusion and estimation model construction, accurate estimation of urban precipitation is achieved, and rainstorm and flood warnings are issued in time, providing strong support for urban disaster prevention and mitigation.

[0050] In the specific implementation, firstly, the rainfall data of the land industry in Guizhou is obtained from the Guizhou Meteorological Data Center, and the land rainfall data of different terrains (mountains, plains, and hills) and climate zones (monsoon areas, arid areas, and humid areas) in the country within a specific time period are obtained. These data are in the form of a rainfall data set G = {g1, g2, ..., g n}, where g i represents the rainfall data of the ith area, and then the collected data is cleaned. In the data cleaning process, for the case of missing data, that is, the rainfall data of area j is missing, by finding the rainfall data set S = {s1, s2, ..., s m},according to Repair is performed by using the average rainfall data of the surrounding areas to fill in the missing values. For erroneous data, the historical data set H = {h1, h2, ..., h l}Compare with the rainfall data in the same area, set the threshold T1, when |g i -h i |>T1, g i Consider it as wrong data and directly use the historical data of the same period i Replace. Collect data from a weather radar system consisting of multiple radar stations distributed across the country. The collected data set is R = {r1, r2, ..., r p}, where each r i Includes radar reflectivity r i1 and radial velocity r i2 information, and data collection is represented by r i =f(θ, d), where θ is the elevation angle of the radar beam, d is the distance between the target and the radar, and f is the radar data acquisition function. For the radar reflectivity r i1and radial velocity r i2 The data collection functions are: Among them, P(β, d) is the distribution function of the radar transmission power in the (θ, d) direction, λ is the wavelength of the electromagnetic wave emitted by the radar, |K| 2 is the coefficient of precipitation particles, σ k is the radar cross section of the kth precipitation particle at a distance d from the radar, n is the total number of precipitation particles within the effective illumination volume of the radar beam, and f d is the Doppler frequency shift. The satellite observed precipitation data is obtained by receiving data from meteorological satellites. The data set is S = {s1, s2, ..., s q}, the satellite sensor type set is For each sensor t j Collected data ij , using the format conversion function for this sensor After format conversion, the converted data is The ground-based precipitation data are collected from a widely distributed network of ground-based rainfall observation stations. The data set is L = {l1, l2, ..., l r}, where the data of each observation station is l i , calculate the data set of n surrounding observation stations N = {n1, n2, ..., n n The average value of Set the data abnormal threshold T2, when When i Marked as abnormal data, for abnormal data judged to be caused by equipment failure, the spatial interpolation function I is used according to l i =I(N) uses the data of surrounding stations for estimation and substitution, adopts the inverse distance weighted interpolation method, and performs weighted average calculation based on the power of the inverse of the distance between the surrounding stations and the abnormal station as the weight to obtain the estimated value of the abnormal station.

[0051] Then, for the collected radar observation precipitation data set R = {r1, r2, ..., r p}, where r i =(r i1 , r i2 ), a threshold-based filtering algorithm is used to remove noise signals, and the noise threshold T is set n , the filtering formula is: That is, when the radar reflectivity and radial velocity are both greater than the noise threshold, the original data is retained, otherwise it is regarded as noise data and set to 0, and the filtered radar observed precipitation data is calibrated. The standard reflectivity is known to be r s1 , the standard gain is G s , the standard beam width is θs , the current radar reflectivity is r fi1 , the current gain is G i , the current beam width is θ i , the calibration formula is: By calibration, the radar observation data can more accurately reflect the actual precipitation situation. After format conversion of the satellite observation precipitation data, geolocation processing is carried out. It is known that the original coordinates of the satellite observation point in the satellite's own coordinate system are (x s , y s , z s ). The Earth model is a sphere with a radius of R, and the satellite orbit parameters are (θ, φ, h) (representing the orbital inclination, the right ascension of the ascending node, and the orbital altitude respectively). The coordinates of the ground calibration point in the geographic coordinate system are (x g , y g , z g ). Through the coordinate transformation matrix M, the corrected geographic coordinates of the satellite observation point The calculation formula is: Determine the coordinate transformation matrix M according to the satellite orbit mechanics and coordinate transformation principle, substitute it into the formula to calculate the corrected geographic coordinates, so as to ensure that the satellite observation data can accurately correspond to the actual position on the Earth. During the data acquisition process, quality inspection and outlier marking have been carried out on the ground observation precipitation data. For the data marked as abnormal, if it is judged to be caused by equipment failure, the surrounding station data is used to estimate the substitution value through the spatial interpolation function to ensure the reliability and accuracy of the data for subsequent data fusion.

[0052] Subsequently, the preprocessed radar, satellite, ground observation precipitation data and national land rainfall data are spatiotemporally registered, and the registered data are fused by the weighted average method. Spatially, the coordinate systems of the radar and satellite data are converted into the same geographic coordinate system as the ground observation data. For the radar data, through the known position of the radar station and the beam scanning geometric relationship, the polar coordinate data detected by the radar is converted into two-dimensional plane data in the geographic coordinate system. For the satellite data, using the satellite orbit parameters and geolocation algorithm, it is projected into the geographic coordinate system on the Earth's surface to ensure that the data from different data sources can accurately correspond in space. Temporally, the time interpolation method is used to align the data from different data sources to a unified time scale. Let P1(t1) and P2(t2) be the precipitation data at two adjacent time points. To interpolate the precipitation data P(t) at time t (t1 < t < t2), the linear interpolation formula is used: In this way, the radar, satellite, ground observation precipitation data and national rainfall data are accurately registered in time and space, laying the foundation for subsequent fusion processing. The weighted average fusion algorithm is used to fuse the registered data. The fused comprehensive precipitation data is set as P fusion , radar observation precipitation data is P r , satellite observation precipitation data is P s , ground observation precipitation data P g ,, the weighted average fusion formula is P fusion =W r P r +W s P s +W g P g , where W r , W s , W g are the weights of radar, satellite, and ground observation data, respectively, and satisfy W r +W s +W g =1, the weight determination method in the weighted average fusion algorithm is as follows: using A r , A s , A g are the comprehensive evaluation scores of radar, satellite and ground observation data under current weather conditions and data quality indicators, then: Among them, the comprehensive evaluation score of radar data is A r =a1Q r +a2C r +a3E r Similarly, the comprehensive evaluation scores of satellite and ground data can be obtained. Here, Q r is the radar data quality, C r is the matching degree between radar data and current weather conditions, E r is the historical accuracy of radar data in the current geographic area (obtained by comparing and analyzing historical precipitation events), a1, a2, and a3 are the corresponding weight coefficients (the sum of which is 1). The comprehensive evaluation scores of satellite and ground observation data are calculated in the same way, and then their respective weights are determined. Finally, the processed radar, satellite, and ground observation precipitation data are integrated into a comprehensive precipitation dataset.

[0053] Next, a precipitation estimation model is established based on the fused multi-source precipitation data, including: establishing an estimation model based on physical principles based on the physical process of precipitation formation, and an estimation model based on data-driven multi-layer perceptron. Taking water vapor transport and precipitation formation as an example, the process of building an estimation model based on physical principles is as follows: Calculate the horizontal water vapor flux Q x and Q y , and the vertical water vapor flux Qz Assume that the wind speed components in the x, y, and z directions are u, v, and w respectively, and the water vapor density is ρ v , then the horizontal water vapor flux formula is: Q x =ρ v u,Q y =ρ v v, the vertical water vapor flux formula is: Q z =ρ v w+EC, where E is the evaporation rate of water vapor and C is the condensation rate of water vapor. Considering the influence of terrain, the terrain height function is h(x, y), and the terrain slope in the x and y directions is When airflow encounters terrain, the wind speed and water vapor flux are adjusted to Through the water vapor continuity equation (in, ) In actual calculations, it is necessary to combine meteorological observation data (temperature, humidity, air pressure) as well as terrain and geographic information, solve the equation through numerical calculation methods, and obtain an estimated value of precipitation. The construction process of the data-driven multi-layer perceptron estimation model is as follows: the fused precipitation data and other relevant meteorological parameters (temperature, humidity, air pressure, wind speed, wind direction) are used as input feature vectors, and these input variables are normalized to improve the efficiency and stability of neural network training. At the same time, feature engineering is performed on the data. For historical precipitation data, as the output target of the neural network, according to the characteristics of the time series, a multi-layer perceptron (MIP) is selected as the estimation model, and the number of neurons in the input layer is determined to be the same as the number of input features. For the input of the jth neuron in the lth layer, Suppose the output of the i-th neuron in the previous layer (the l-1 layer) is The connection weight is Bias is but: The output of this neuron It is obtained through the activation function. Here we choose the commonly used ReLU function (y=max(0,x)), that is: During the training process, the data set is divided into different subsets according to 70%-80% as the training set, 10%-15% as the validation set, and the rest as the test set. The model is trained using the training set, and the loss function is set to mean square error (MSE). For the true precipitation value yi of the i-th sample and the precipitation value predicted by the model, The MSE calculation formula is: Where N is the total number of training samples. The weights of the neural network are adjusted through the optimization algorithm to minimize the loss function. In each iteration, according to the gradient descent algorithm, the weight update formula is (the learning rate is η). At the same time, the validation set is used to monitor the performance of the model during the training process, and the hyperparameters of the model are adjusted according to the loss value of the validation set.

[0054] Finally, based on historical precipitation data and records of urban flood disaster events, precipitation thresholds and corresponding flood disaster levels are set for different areas of the city. When the precipitation predicted by the precipitation estimation model reaches or exceeds the set threshold, the early warning system automatically generates early warning information. The early warning information includes the type of disaster, expected time of occurrence, scope of impact, and recommended measures. The early warning information is released through multiple channels, and in cooperation with communication operators, the SMS group sending function is used to send early warning text messages to users in potentially affected areas to ensure that the public can obtain early warning information in a timely manner and take corresponding preventive measures.

[0055] In summary, this embodiment demonstrates the complete application process of the precipitation estimation method based on national land rainfall multi-data fusion in urban rainstorm and flood warning. Through accurate data collection, preprocessing, fusion and estimation model construction, combined with reasonable warning threshold setting and multi-channel information release, it can effectively improve the city's ability to respond to rainstorm and flood disasters, provide citizens with timely and accurate warning information, help them make preventive preparations, minimize disaster losses, and ensure the safety and stable operation of the city.

[0056] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A precipitation estimation method based on land rainfall multi-data fusion, characterized in that: The specific steps of this method are: S1. Rainfall data collection and integration: Collect national rainfall data, radar observation precipitation data, satellite observation precipitation data and ground observation precipitation data, conduct preliminary integration of various types of collected data to ensure unified data format and time stamp alignment, and conduct preliminary checks on the integrity and accuracy of the data, including: For national rainfall data, we obtain the observed and recorded latitude and longitude, time and rainfall accumulation values ​​from the meteorological data storage center, clean the data, and check for missing and incorrect data. If there are data problems, we repair and supplement them by comparing them with the data of surrounding areas and referring to the historical data of the same period. For radar observations, precipitation data are collected from a meteorological radar system, which includes data collected by multiple radar stations and contains information on radar reflectivity and radial velocity at different angles and distances; The satellite-observed precipitation data is obtained by receiving data from meteorological satellites, which use different sensors to observe the earth's surface and convert the satellite data into a uniform format; For ground-based precipitation data from distributed meteorological observation rain gauges, during the data collection process, ensure that the equipment at each observation station is operating normally and data transmission is stable, and perform quality control on the collected data; S2, data preprocessing: preprocessing the radar observed precipitation data, satellite observed precipitation data and ground observed precipitation data, wherein; For radar observed precipitation data, the calibration algorithm is used for preprocessing. The noise signal caused by the radar equipment itself and the external environment interference is removed by the filtering algorithm. Based on the reflectivity data and calibration equipment, the radar's transmission frequency, gain, and beam width parameters are calibrated to make the radar observation data accurately reflect the actual precipitation situation. For the precipitation data observed by satellite, format conversion and geolocation processing are performed. According to the format specifications of different satellite data, a conversion algorithm is used to unify them into a standard format. The geolocation processing adjusts the inconsistency of spatial resolution and temporal resolution between different satellite data due to sensor differences and orbital characteristics through resampling and time interpolation methods. The satellite data is geolocated using the positioning information and ground calibration points of the satellite itself to ensure that the precipitation data observed by the satellite can accurately correspond to the actual position on the earth. Two quality checks are implemented for ground-based precipitation data. The first quality check compares the data with the data of multiple surrounding observation stations. If the data of an observation station deviates from the average value of the surrounding stations, the data will be marked and further checked. For abnormal data caused by equipment failure, the maintenance personnel will be notified in time to repair it and the abnormal value will be removed in data processing. The second data check replaces the abnormal value by estimating the standard value using the data of surrounding stations through spatial interpolation method. S3, multi-source data fusion processing: perform spatiotemporal registration of the pre-processed radar, satellite, and ground-based precipitation data with the national rainfall data, and fuse the registered data using the weighted average method; The spatiotemporal registration converts the coordinate system of radar and satellite data into a geographic coordinate system consistent with ground observation data in space. For radar data, the polar coordinate data detected by the radar is converted into two-dimensional plane data under geographic coordinates through the known position of the radar station and the beam scanning geometry. For satellite data, it is projected into the geographic coordinate system of the earth's surface using satellite orbit parameters and geographic positioning algorithms. In terms of time, data from different data sources are aligned to a unified time scale based on the data collection timestamp; The data fusion adopts a weighted average fusion algorithm, which assigns an initial weight to each data source under different circumstances through comparative analysis and statistics of historical data. In the real-time fusion process, the weight is dynamically adjusted according to the current weather conditions and data quality indicators, and the processed radar, satellite and ground observation precipitation data are fused into a multi-source fusion precipitation data set; S4. Based on the fused multi-source fusion precipitation dataset, the early warning system, infrastructure construction system and emergency preparedness system are used to provide site data for emergency disastrous weather.

2. The precipitation estimation method based on land rainfall multi-data fusion according to claim 1 is characterized in that: The S1 is for national rainfall data, and its national rainfall data set is G = {g1, g2, ..., g n }, where g i represents the rainfall data of the ith region. For missing data, the missing data location is defined as j. There is a rainfall data set S = {s1, s2, …, s m }, then fix it to That is, the missing values ​​are repaired by using the average rainfall data of the surrounding area; for erroneous data, the historical data set H = {h1, h2, ..., h l }Compared with the rainfall data in the same area, when g i If the difference between the corresponding position data in H exceeds the threshold T1, then g i Considered as wrong data, repaired to g i =h i , that is, directly replace the wrong data with historical data of the same period; For radar observed precipitation data, the radar observed precipitation data set is R = {r1, r2, …, r p }, where r i Includes radar reflectivity r i1 and radial velocity r i2 For data collected at different angles θ and different distances d, the data collection is represented by r i =f(θ,d), where f is the radar data acquisition function, for radar reflectivity r i1 and radial velocity r i2 The data collection function f(θ,d) is: and Where θ is the elevation angle of the radar beam, d is the distance between the target and the radar, P(θ, d) is the distribution function of the radar transmission power in the (θ, d) direction, and λ is the wavelength of the electromagnetic wave emitted by the radar. | K | 2 is the coefficient of precipitation particles, σ k is the radar cross section of the kth precipitation particle at a distance d from the radar, n is the total number of precipitation particles within the effective illumination volume of the radar beam, and f d is the Doppler shift; For satellite-observed precipitation data, format conversion is performed to unify different satellite data formats and perform geolocation, and accurately map satellite observation data to the geographic coordinate system; For ground-based precipitation data, the ground-based precipitation data set is L = {l1,l2,…,l r }, the observation station data is l i , the data set of the surrounding n observation stations is N = {n1,n2,…,n n }, and its average value is Quality control is: Then mark l i is abnormal data, where T2 is the data abnormality threshold. For abnormal data, when it is judged to be caused by equipment failure, it is repaired to l through the spatial interpolation function I. i =I(N), that is, the replacement value is estimated by spatial interpolation method using the surrounding station data.

3. The precipitation estimation method based on land rainfall multi-data fusion according to claim 1 is characterized in that: The S2 is for radar observed precipitation data, according to the collected radar observed precipitation data set R = {r1, r2, ..., r p }, where r i =(r i1 ,r i2 ), r i1 is the radar reflectivity, r i2 is the radial velocity, and the filtered radar observation precipitation data set is defined as R f = {r f1 ,r f2 ,…,r fp }, for each r i , using a threshold-based filtering algorithm, the noise threshold is T n , then the filtering is: That is, when the radar reflectivity and radial velocity are both greater than the noise threshold, the original data are retained, otherwise they are regarded as noise data and set to 0, and the calibrated radar observation precipitation data set is defined as R c = {r c1 ,r c2 ,…,r cp }, for each r fi , whose standard reflectivity is r s1 , the standard gain is G s , the standard beam width is θ s , the current radar reflectivity is r fi1, , the current gain is G i , the current beam width is θ i , then the calibration is For the calibration of radar observation precipitation data parameters, the standard reflectivity is r s , the standard gain is G s , the standard beam width is B s , the current radar reflectivity is The current gain is G i , the current beam width is B i , calibrated radar reflectivity Calculated as:

4. The precipitation estimation method based on land rainfall multi-data fusion according to claim 1 is characterized in that: When marking abnormal values ​​in the quality check of ground-based precipitation data in S2, P i is the precipitation observation value of the i-th ground observation station, is the average precipitation of multiple observation stations in the surrounding area, σ P is the standard deviation of precipitation at multiple surrounding observation stations. If n is a constant, then P i Mark as an outlier.

5. The precipitation estimation method based on land rainfall multi-data fusion according to claim 1 is characterized in that: The time interpolation method for spatio-temporal registration in S3 to align data in time is as follows: P1(t1) and P2(t2) are precipitation data at two adjacent time points. The precipitation data P(t) at time t (t1 < t < t2) is obtained by interpolation. Then:

6. The precipitation estimation method based on land rainfall multi-data fusion according to claim 1 is characterized in that: The integrated precipitation data after fusion in S3 is P fusion , radar observation precipitation data is P r , satellite observation precipitation data is P s , the ground observation precipitation data is P g , then the weighted average fusion formula is: P fusion =W r P r +W s P s +W g P g , where W r , W s , W g are the weights of radar, satellite, and ground observation data, respectively, and satisfy W r +W s +W g =1, W in the weighted average fusion algorithm r , W s , W g , using A r , A s , A g are the comprehensive evaluation scores of radar, satellite and ground observation data under current weather conditions and data quality indicators, then: Among them, the comprehensive evaluation score of radar data is A r =a1Q r +a2C r +a3E r Similarly, we get the comprehensive evaluation score of satellite and ground data, Q r is the radar data quality, C r is the matching degree between radar data and current weather conditions, E r is the historical accuracy of radar data in the current geographical area, and a1, a2, and a3 are the corresponding weight coefficients.

7. A precipitation estimation system based on land rainfall multi-data fusion, characterized in that: The system consists of: data acquisition module, data processing module, data fusion module, precipitation estimation module; The data acquisition module accurately collects national rainfall data, radar observation precipitation data, satellite observation precipitation data and ground observation precipitation data through the device end, and transmits it to the data center; The data preprocessing module preprocesses the radar observed precipitation data, satellite observed precipitation data and ground observed precipitation data; The data fusion module performs spatiotemporal registration of the pre-processed radar, satellite, and ground-observed precipitation data with the land rainfall data, and fuses the registered data by a weighted average method; The precipitation estimation module establishes a precipitation estimation model based on the fused multi-source precipitation data.

8. The application of the precipitation estimation system based on land rainfall multi-data fusion according to claim 7 is characterized in that: The application of the system in meteorological disaster early warning and infrastructure construction specifically includes: Early warning system: When the estimated precipitation exceeds the warning threshold, the system automatically generates early warning information, including the type of disaster, expected time of occurrence, and scope of impact. Through cooperation with communication operators and radio and television stations, the early warning information is sent to the public, emergency management departments, and water conservancy departments in a timely manner; Infrastructure construction system: Combine precipitation estimation data with geographic information system (GIS) to analyze precipitation risks in different regions; Emergency preparedness system: When emergency disastrous weather occurs, the availability of ground observation station data is monitored in real time. For stations that cannot provide data normally, the interpolation algorithm is used to calculate the estimated data of the missing stations based on the spatial distribution of surrounding station data and fused precipitation data. The data is then transmitted to the emergency command center to provide data support for emergency decision-making.

Images