Mountain area rainfall station network layout optimization method based on radar rainfall measurement data

Abstract

The invention discloses a mountainous area rainfall station network layout optimization method based on radar rain measurement data. The method comprises the following steps: step 1, rain measurement data collection and target area grid division; step 2, identifying abnormal stations of rainfall stations; step 3, determining a candidate position of the compensation station of the rainfall station; step 4, evaluating precision effects of different rainfall station network optimization schemes; and step 5, determining an optimal rainfall station network layout scheme. According to the method, radar rain measurement data is combined, effective planning and optimization of the rainfall station network under the complex terrain condition of the mountainous area are achieved, then the accuracy and reliability of rainfall observation of the mountainous area drainage basin are improved, and more accurate data support is provided for management, prevention and control of flood disasters in the mountainous area.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydrometeorological monitoring, and particularly relates to a method for optimizing the layout of mountain rainfall stations based on radar rainfall data. Background Art

[0002] With the increasingly significant impact of global climate change, extreme weather events occur frequently. Especially, the flood disasters in mountainous basins pose a serious threat to people's lives and property. Accurate and reliable rainfall monitoring is of great significance for flood forecasting and early warning. The existing layout of rainfall stations mostly relies on empirical judgment and limited meteorological observation data, without fully considering the influence of complex mountain terrain on rainfall distribution. In the face of the strong spatial heterogeneity and spatio-temporal variation characteristics of mountain rainfall, it is easy to result in blind spots in rainstorm observation, unable to accurately capture local heavy precipitation and rainstorm events, thus affecting the accuracy of flood forecasting and disaster early warning.

[0003] Currently, the existing methods for optimizing rainfall station networks are mostly based on certain optimization algorithms, such as genetic algorithms, particle swarm optimization (PSO), etc., to improve the station network coverage rate and monitoring accuracy by selecting the optimal locations. However, most of the existing optimization methods rely on ground rainfall station data and ignore the potential of remote sensing technology (such as radar rainfall measurement). Radar data has high spatio-temporal resolution and can provide more comprehensive precipitation monitoring information, but the traditional methods fail to make full use of this information. Therefore, although the existing optimization methods can improve the performance of the station network under certain conditions, they still face challenges in terms of accuracy and efficiency in complex mountainous environments and cannot achieve the ideal monitoring effect. Therefore, how to optimize the traditional rainfall station network under complex mountain terrain by combining high-precision remote sensing data is a technical problem that urgently needs to be solved at present. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for optimizing the layout of mountain rainfall stations based on radar rainfall data to solve the above technical problems.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] The present invention discloses a method for optimizing the layout of mountain rainfall stations based on radar rainfall data, and the method includes the following steps:

[0007] Step 1: Collection of rainfall data and grid division of the target area: Collect the actual rainfall data P of all M rainfall stations in the target area 雨量站 and the radar rainfall data P corresponding to the time series of P 雨量站 , and perform the same grid division on the target area according to the grid distribution of the radar rainfall data. There are a total of J, and a unique identifier is assigned to each grid; 雷达

[0008] ​Step 2, Identification of abnormal rain gauge stations: Use the graph convolutional network model, i.e., the GCN model, to identify abnormal rain gauge stations. First, construct the node features of the rain gauge stations. If the actual rainfall measurement data of each rain gauge station is N hours, then use the node features of M rain gauge stations and the corresponding actual rainfall measurement data for N hours as the training dataset to train the GCN model. Take the node features of the rain gauge stations as the model input and the corresponding actual rainfall measurement data as the model output. Then, use the trained GCN model to identify abnormal rain gauge stations, eliminate the identified abnormal stations, and record the number of rain gauge stations operating normally as K;

[0009] Step 3, Determination of candidate positions for supplementary rain gauge stations: Use Kriging interpolation method to perform spatial interpolation on the actual rainfall measurement data of K rain gauge stations hour by hour to generate rainfall field data P of the target basin range 站点插值 , then for the J grids divided in Step 1, there are two sets of rainfall data at the center point of each grid, namely radar rainfall measurement data P 雷达 and interpolated rainfall field data P 站点插值 . Calculate the relative error RE 雷达 between P 站点插值 and P j at the center of each grid hour by hour, j = 1, 2, …, J. Count the number of grids N j where RE 2 > D D2,t . Set that the time periods when N D2,t > D 3 belong to the time periods with significantly different rainfall distributions. Denote the total number of such time periods as T, where the thresholds D 2 and D 3 are both determined according to the natural geographical conditions and data situation of the basin. Statistically calculate the average value of the relative error RE j of each grid within the T time periods and arrange them from large to small. Take the first γ grid center points as candidate positions for supplementary stations, numbered 1, 2, …, γ;

[0010] Step 4, Evaluation of the accuracy effect of different rain gauge network optimization schemes: Set to supplement 1 to γ rain gauge stations to generate a total of γ supplementary station schemes. The γ supplementary station schemes are respectively to supplement a station at candidate position No. 1, to supplement stations at candidate positions No. 1 and 2, and so on, to supplement stations at candidate positions No. 1, 2, …, γ. Then the total number of rain gauge stations corresponding to the γ supplementary station schemes is K + 1, K + 2, …, K + γ respectively;

[0011] Use a machine learning model to predict the rainfall conditions of different rainfall station supplementation schemes. For each rainfall station location s, construct a feature vector X(s), including the geographical location, terrain features, meteorological features of the station, and the radar rainfall measurement data corresponding to the grid where it is located. Then, use the feature vectors and actual rainfall measurement data of all K rainfall stations as the training dataset to train the machine learning model f, with the feature vector X of the rainfall station as the model input and the corresponding actual rainfall measurement data as the model output; then use the trained model to, for each candidate location s* for supplementation, based on its feature vector X(s * ), predict the rainfall conditions at its location It is:

[0012]

[0013] For all T time periods, predict the rainfall data of each candidate location for supplementation in γ rainfall station supplementation schemes period by period, and use the Kriging interpolation method to perform spatial interpolation on the prediction results under each rainfall station supplementation scheme to generate the corresponding rainfall field data;

[0014] For the T time periods, calculate RE j > D 2 The number of grids is denoted as That is, the number of grids where RE j > D 2 in the t-th time period, t = 1, 2,..., T, and then calculate the average percentage reduction per of the number of grids in all time periods:

[0015]

[0016] The larger per is, the closer the rainfall field data of the basin after supplementation is to the radar rainfall measurement data;

[0017] Step 5, determination of the optimal rainfall station network layout scheme: Considering the construction cost, maintenance cost of rainfall stations, and potential social and economic benefits, construct a penalty function ω to represent the adverse impact of field conditions on the construction of rainfall stations. Let C represent the construction cost, W represent the maintenance cost, and B represent the social and economic benefits. C, W, and B are all functions of the supplementation scheme and change with the scheme. Then, the penalty function ω is defined as:

[0018]

[0019] In the formula: α, β, are weight coefficients;

[0020] Combining per and ω, construct an optimization objective function φ, making it positively correlated with per and negatively correlated with ω, that is:

[0021] φ = μ 1 × per - μ 2 × ω (6)

[0022] where: μ 1 and μ 2 are weight coefficients;

[0023] For each supplementary station scheme, calculate its corresponding φ value, and select the scheme with the highest φ value as the optimal rainfall station layout scheme.

[0024] Furthermore, the specific operation of the same grid division of the target area according to the grid distribution of radar rainfall data in step 1 is as follows: Analyze the radar rainfall data file, extract its spatial resolution to determine the basic unit of the grid; then, based on the spatial position of the radar data, use the row-column coordinate method to systematically number the grid, with the row number increasing from north to south and the column number increasing from west to east.

[0025] Furthermore, the specific operation of constructing the node features of the rainfall station in step 2 is as follows: Convert the geographical location relationship between rainfall stations into a graph structure, where each rainfall station is regarded as a node in the graph structure, and the edges between nodes represent the spatial correlation between rainfall stations, and the weights of the edges are set through distance metrics or similarity metrics based on geographical distances; the node features of the rainfall station include rainfall time series features and geographical information features; the rainfall time series features include rainfall statistical features and time feature information in historical periods; the geographical information features include the longitude, latitude, altitude, and terrain features of each rainfall station.

[0026] Furthermore, the specific operation of using the trained GCN model to identify abnormal rainfall stations in step 2 is as follows: For each rainfall station, for all N periods, calculate its reconstruction error RE after prediction by the trained GCN model:

[0027]

[0028] where: and are respectively the measured value of the rainfall station at the nth time and the predicted value based on the GCN model;

[0029] For rainfall stations with RE > D 1 mark them as abnormal stations, where the threshold D 1 is determined by combining the local weather conditions and natural geographical features.

[0030] Furthermore, the terrain features in step 4 include altitude, slope, and aspect; the meteorological features include temperature, humidity, and wind speed.

[0031] The beneficial effect of the present invention is that the method of the present invention combines radar rainfall measurement data to achieve effective planning and optimization of rainfall station networks under complex terrain conditions in mountainous areas, thereby improving the accuracy and reliability of precipitation observations in mountainous river basins, and providing more accurate data support for the management and prevention of flood disasters in mountainous areas.

[0032] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0034] The present invention discloses a method for optimizing the layout of a rainfall station network in a mountainous area based on radar rainfall measurement data. Figure 1 As shown, the method comprises the following steps:

[0035] Step 1: Collect rainfall data and divide the target area into grids.

[0036] For the target area where the layout of the rainfall station network is to be optimized, the actual rainfall data P of all M rainfall stations in the area are first collected. 雨量站 and P 雨量站 Radar rainfall data corresponding to the time series P 雷达 , giving priority to rainfall during the flood season. Secondly, the target area is divided into the same grids (a total of J) according to the grid distribution of radar rainfall data, and a unique identifier is assigned to each grid. Specifically, the radar rainfall data file is parsed and its spatial resolution is extracted to determine the basic unit of the grid; then based on the spatial position of the radar data, the grids are systematically numbered using the row and column coordinate method, with row numbers increasing from north to south and column numbers increasing from west to east, to ensure that each grid has a unique identifier for subsequent data processing and analysis.

[0037] Step 2: Identify abnormal rainfall stations.

[0038] Considering that the existing rain gauges may have abnormal operation, the GCN (graph convolutional network) model is used to identify abnormal sites to ensure that the existing station network is composed of normally operating sites. First, the node features of the rain gauges are constructed. If the rainfall data of each rain gauge is N hours, the node features of M rain gauges and the corresponding N hours of actual rainfall data are used as training data sets to train the GCN model. The node features of the rain gauges are used as model inputs, and the corresponding actual rainfall data are used as model outputs. Then, the trained GCN model is used to identify abnormal sites of rain gauges, and the identified abnormal sites are eliminated. The number of rain gauges that are operating normally is recorded as K. The subsequent steps are only based on K rain gauges, and abnormal sites are no longer considered.

[0039] The specific process is as follows:

[0040] 1) Construct the node features of the rain gauges. Transform the geographical location relationship between rain gauges into a graph structure. Each rain gauge is regarded as a node in the graph structure, and the edges between nodes represent the spatial correlation between rain gauges. The weights of the edges are set by distance metrics (such as Euclidean distance) or similarity metrics based on geographical distance. The node features of the rain gauges include rainfall time series features and geographical information features; the rainfall time series features include statistical features of rainfall amounts in historical periods (such as the mean and variance of precipitation in the past few hours or days), time features (the year, month, season, moment, etc. of the current period), etc.; the geographical information features include the longitude, latitude, altitude, terrain features, etc. of each rain gauge.

[0041] 2) Initialize the model parameters and train the GCN model using the training dataset. Set the number of layers L of the model, the number of neurons d in each layer 1 , and the activation function σ. The l-th (l = 1, 2, …, L) layer of the GCN model can be expressed as:

[0042]

[0043] In the formula: H (l) is the node feature matrix of the l-th layer; is the adjacency matrix after adding self-loops; is the degree matrix; W (l) is the weight matrix of the l-th layer; σ is the activation function, such as ReLU.

[0044] 3) Identify abnormal stations. For each rain gauge, for all N periods, calculate the reconstruction error RE after prediction by the trained GCN model:

[0045]

[0046] In the formula: and are the measured value and the predicted value based on the GCN model of the rain gauge at time n, respectively.

[0047] For rain gauges with RE > D 1 , mark them as abnormal stations, where the threshold D 1 is determined by combining the regional weather conditions and natural geographical features, etc.

[0048] Step 3: Determine the candidate positions for supplementing rain gauges.

[0049] Interpolate the rainfall data of K rain gauges spatially hour by hour using the Kriging interpolation method to generate the rainfall field data P of the target basin range 站点插值So far, for the J grids marked in step 1, there are two sets of rainfall data at the center point of each grid, namely the radar rainfall measurement data P 雷达 and the interpolated rainfall field data P 站点插值 . Calculate the relative error RE 雷达 between P 站点插值 and P j at the center of each grid hourly (j = 1, 2, …, J), and count the number of grids N j where RE 2 > D D2,t . Set the periods when N D2,t > D 3 as the periods with obvious differences in the two rainfall distributions, and record the total number of such periods as T. The thresholds D 2 and D 3 are both determined according to the natural geographical conditions of the basin and data conditions, etc. Then, within the T periods, count the average value of the relative error RE j of each grid, arrange them from large to small, and select the first γ grid center points as the candidate positions for supplementary stations, numbered 1, 2, …, γ respectively.

[0050] Step 4: Evaluate the accuracy effect of different rain gauge network optimization schemes.

[0051] Set up 1 to γ supplementary rain gauges to generate a total of γ supplementary station schemes. The γ supplementary station schemes are respectively adding a station at the candidate position numbered 1, adding stations at the candidate positions numbered 1 and 2, and so on, adding stations at the candidate positions numbered 1, 2, …, γ. Then the total number of rain gauges corresponding to the γ supplementary station schemes are K + 1, K + 2, …, K + γ respectively;

[0052] Use machine learning models (such as random forest, gradient boosting tree, etc.) to predict the rainfall conditions of different supplementary station schemes. For each rain gauge position s, construct a feature vector X(s), including the geographical location of the station, topographic features (including altitude, slope, aspect, etc.), meteorological features (including temperature, humidity, wind speed, etc.), and the radar rainfall measurement data corresponding to the grid where it is located. Then use the feature vectors and actual rainfall measurement data of all K rain gauges as the training dataset to train the machine learning model f, with the feature vector X of the rain gauge as the model input and the corresponding actual rainfall measurement data as the model output. Then, use the trained model to predict the rainfall condition at each supplementary station candidate position s* based on its feature vector X(s * ) (that is, the rainfall measured assuming there is a ground rain gauge here) as:

[0053]

[0054] For all T time periods, predict the rainfall data of each candidate rain gauge location in γ rain gauge addition schemes period by period, and use Kriging interpolation method to perform spatial interpolation on the prediction results under each rain gauge addition scheme to generate corresponding rainfall field data;

[0055] For the T time periods, calculate RE period by period again j > D 2 The number of grids, denoted as (the number of grids where RE j > D 2 at the t-th time period, t = 1, 2, …, T), and then calculate the average percentage reduction per of the number of grids for all time periods:

[0056]

[0057] The larger per is, the closer the rainfall field data of the basin after rain gauge addition is to the radar rainfall measurement data; the more significant the improvement in the ability of the ground rain gauge network to capture the rainfall situation in the basin.

[0058] Step 5: Determine the optimal rain gauge network layout scheme.

[0059] Combined with the actual economy, construct an objective function for network layout optimization to determine the optimal rain gauge network layout scheme. Considering the construction cost, maintenance cost of rain gauges and potential social and economic benefits, construct a penalty function ω to represent the adverse impact of field conditions on the construction of rain gauges. Let C represent the construction cost, W represent the maintenance cost, and B represent the social and economic benefits. C, W, and B are all functions of the rain gauge addition scheme and change with the scheme. Then the penalty function ω can be defined as:

[0060]

[0061] In the formula: α, β, are weight coefficients.

[0062] Combined with per and ω, construct an optimization objective function φ, making it positively correlated with per and negatively correlated with ω, that is:

[0063] φ = μ 1 × per - μ 2 × ω (6)

[0064] In the formula: μ 1 、μ 2 are weight coefficients, allocated according to importance.

[0065] For each rain gauge addition scheme, calculate its corresponding φ value, and select the scheme with the highest φ value as the optimal rain gauge layout scheme. This scheme not only maximally improves the accuracy of rainfall observation, but also ensures the economic rationality of construction and maintenance, achieving the comprehensive optimum of technology and economy.

[0066] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred arrangement, those of ordinary skill in the art should understand that the technical solution of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for optimizing the layout of rainfall station networks in mountainous areas based on radar rainfall measurement data, characterized in that: The method comprises the following steps: Step 1: Collect rainfall data and divide the target area into grids: Collect the actual rainfall data P of all M rainfall stations in the target area. 雨量站 and P 雨量站 Radar rainfall data corresponding to the time series P 雷达 , divide the target area into the same grids according to the grid distribution of radar rainfall data, record a total of J, and assign a unique identifier to each grid; Step 2, Identification of abnormal rain gauge stations: Use the graph convolutional network model, i.e., the GCN model, to identify abnormal rain gauge stations. First, construct the node features of the rain gauge stations. If the actual rainfall data of each rain gauge station is N hours, then use the node features of M rain gauges and the corresponding N hours of actual rainfall data as training data sets to train the GCN model. Use the node features of the rain gauge stations as model input, and use the corresponding actual rainfall data as model output. Then use the trained GCN model to identify abnormal rain gauge stations, remove the identified abnormal stations, and record the number of rain gauge stations that are operating normally as K. Step 3: Determine candidate locations for additional rain gauge stations: Use Kriging interpolation method to spatially interpolate the actual rainfall data of K rain gauge stations hourly to generate rainfall field data P for the target basin. 站点插值 , then for the J grids divided in step 1, each grid center point has two sets of rainfall data, namely radar rainfall data P 雷达 and interpolated rainfall field data P 站点插值 , calculate P at the center of each grid hourly 雷达 and P 站点插值 Relative error RE j , j=1,2,…,J, count RE in each period j >Number of grids in D2 D2,t , set N D2,t The time period of >D3 belongs to the period of time when the two rainfall distributions are obviously different. The total number of such time periods is T. The thresholds D2 and D3 are determined according to the natural geographical conditions and data conditions of the basin. The relative error RE of each grid is calculated in T time periods. j The average value is arranged from large to small, and the first γ grid center points are taken as candidate locations for supplementary stations, numbered 1, 2, …, γ respectively; Step 4, accuracy evaluation of different rainfall station network optimization schemes: set 1 to γ ​​rainfall stations to be supplemented, and generate a total of γ supplementary station schemes, wherein the γ supplementary station schemes are to supplement a station at candidate position number 1, to supplement a station at candidate positions numbered 1 and 2, and so on, to supplement a station at candidate positions numbered 1 and 2, …, γ. Then, the total number of rainfall stations corresponding to the γ supplementary station schemes is K+1, K+2, …, K+γ respectively; The machine learning model is used to predict rainfall conditions for different rain gauge replacement schemes. For each rain gauge location s, a feature vector X(s) is constructed, including the geographical location, terrain characteristics, meteorological characteristics of the station, and the radar rainfall data corresponding to the grid. The feature vectors and actual rainfall data of all K rain gauges are then used as training data sets to train the machine learning model f, with the feature vector X of the rain gauge as the model input and the corresponding actual rainfall data as the model output. The trained model is then used to predict rainfall conditions for each candidate location s* based on its feature vector X(s * ), predicting rainfall conditions at its location for: For all T time periods, the rainfall data of each candidate location of the station in γ station replacement schemes is predicted time period by time period, and the prediction results under each station replacement scheme are spatially interpolated using the Kriging interpolation method to generate the corresponding rainfall field data; For T periods, RE is calculated again period by period. j >The number of grids in D2 is expressed as That is, RE in the tth period j > The number of grids of D2, t = 1, 2, ..., T, and then calculate the average percentage of reduction in the number of grids in all time periods per: The larger the per is, the closer the basin rainfall data after the station is added is to the radar rainfall data; Step 5: Determine the optimal layout of the rain gauge network: Considering the construction cost, maintenance cost and potential social and economic benefits of the rain gauge station, a penalty function ω is constructed to represent the adverse effects of field conditions on the construction of rain gauge stations. Let C represent the construction cost, W represent the maintenance cost, and B represent the social and economic benefits. C, W, and B are all functions of the supplementary station plan and change with the plan. The penalty function ω is defined as: Where: α, β, is the weight coefficient; Combining per and ω, we construct an optimization objective function φ so that it is positively correlated with per and negatively correlated with ω, that is: φ=μ1×per-μ2×ω (6) Where: μ1, μ2 are weight coefficients; For each rainfall station replenishment plan, the corresponding φ value is calculated, and the plan with the highest φ value is selected as the optimal rainfall station layout plan.

2. The method for optimizing the layout of rainfall stations in mountainous areas based on radar rainfall data according to claim 1 is characterized in that: The specific steps of dividing the target area into the same grid according to the grid distribution of radar rainfall data in step 1 are as follows: parsing the radar rainfall data file, extracting its spatial resolution, and determining the basic unit of the grid; then based on the spatial position of the radar data, using the row and column coordinate method to systematically number the grids, with the row numbers increasing from north to south and the column numbers increasing from west to east.

3. The method for optimizing the layout of rainfall stations in mountainous areas based on radar rainfall measurement data according to claim 1 is characterized in that: The node features of the rain gauges constructed in step 2 are as follows: the geographical location relationship between the rain gauges is converted into a graph structure, each rain gauge is regarded as a node in the graph structure, the edges between the nodes represent the spatial correlation between the rain gauges, and the weights of the edges are set by distance measurement or similarity measurement based on geographical distance; the node features of the rain gauges include rainfall time series features and geographic information features; the rainfall time series features include rainfall statistical features and time feature information in historical periods; the geographic information features include the latitude and longitude, altitude, and terrain features of each rain gauge.

4. The method for optimizing the layout of rainfall stations in mountainous areas based on radar rainfall data according to claim 1 is characterized in that: The specific method of using the trained GCN model to identify abnormal rainfall stations in step 2 is as follows: for each rainfall station, for all N time periods, calculate the reconstruction error RE predicted by the trained GCN model: Where: and are the measured value at the rainfall station at time n and the predicted value based on the GCN model; Rainfall stations with RE>D1 are marked as abnormal stations, where the threshold D1 is determined in combination with regional weather conditions and natural geographical features.

5. The method for optimizing the layout of rainfall stations in mountainous areas based on radar rainfall data according to claim 1, characterized in that: The terrain features in step 4 include altitude, slope, and slope direction; the meteorological features include temperature, humidity, and wind speed.

Images