Optimization method for mountainous rainfall observation network layout based on radar rainfall data

Abstract

This invention discloses a method for optimizing the layout of a rain gauge network in mountainous areas based on radar rainfall data, which includes Step 1: collecting rainfall data and dividing the target area into meshes, Step 2: identifying abnormal rain gauge station locations, Step 3: determining candidate locations for additional rain gauge stations, Step 4: evaluating the accuracy and effectiveness of different rain gauge network optimization methods, and Step 5: determining the optimal rain gauge network layout method. [Effects] The method of the present invention, combined with radar rainfall data, realizes effective planning and optimization of rainfall observation networks in complex mountainous terrain conditions, further improves the accuracy and reliability of precipitation observation in mountainous watersheds, and provides more accurate data support for the management and prevention of mountainous flood disasters.

Description

[Technical Field]

[0001] The present invention relates to the field of hydrological meteorological observation technology, and in particular to a method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data. [Background technology]

[0002] As the impact of global climate change becomes increasingly pronounced, extreme weather events are becoming more frequent, and flood disasters, especially in mountainous basins, pose a serious threat to the safety of people's lives and property. Accurate and reliable rainfall observation is crucial for flood forecasting and early warning. Traditional rainfall observation station construction relies on empirical judgment and limited meteorological observation data, and does not fully consider the impact of the complex terrain of mountainous areas on rainfall distribution. Faced with the strong spatial heterogeneity and spatiotemporal fluctuation characteristics of rainfall in mountainous areas, blind spots are likely to occur in storm observation, making it difficult to accurately capture localized heavy precipitation and storm events, thereby affecting the accuracy of flood forecasting and disaster early warning.

[0003] Currently, existing rainfall observation network optimization methods often use certain optimization algorithms, such as genetic algorithms and particle swarm optimization (PSO), to select optimal locations to improve network coverage and observation accuracy. However, traditional optimization methods often rely on data from ground-based rainfall observation stations and ignore the potential of remote sensing technologies (e.g., radar rainfall). Radar data has relatively high spatiotemporal resolution and can provide more comprehensive precipitation observation information, but traditional methods are unable to fully utilize this information. Therefore, while traditional optimization methods can improve observation network performance in certain conditions, they still face accuracy and efficiency challenges in complex mountainous environments, preventing them from achieving ideal observation results. Therefore, how to combine high-precision remote sensing data to optimize traditional rainfall observation networks in complex mountainous terrain is a technical problem that urgently needs to be solved. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] (None in particular) Summary of the Invention [Problem to be solved by the invention]

[0005] SUMMARY OF THE INVENTION The objective of the present invention is to provide a method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data in order to solve the above technical problems. [Means for solving the problem]

[0006] To achieve the above objectives, the present invention provides the following technical solutions: The present invention discloses a method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data, the method including the following steps 1 to 5: Step 1: Collecting rainfall data and dividing the target area into meshes: The actual rainfall data P 雨量観測所 and P 雨量観測所 Time series of radar rainfall data P レーダー Step 1: Collecting the data, dividing the target area into the same meshes based on the mesh distribution of the radar rainfall data, and dividing the total into J meshes, and assigning a unique identifier to each mesh; Step 2: Identifying abnormal rain gauge stations: Using a graph convolutional network (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 the node features of M rain gauge stations and the corresponding actual rainfall data of N hours are used as the training data set to train the GCN model. The node features of the rain gauge stations are used as the model input and the corresponding actual rainfall data are used as the model output. Then, the trained GCN model is used to identify abnormal rain gauge stations, and the identified abnormal rain gauge stations are removed, and the number of normally operating rain gauge stations is recorded as K. Step 3: Determining candidate locations for additional rainfall stations: Using the Kriging interpolation method, spatial interpolation is performed on the actual rainfall data from K rainfall stations every hour to obtain the rainfall field data P 観測点補間 Then, for the J meshes divided in step 1, there are two sets of rainfall data at the center of each mesh, and each set is the radar rainfall data P レーダー and the interpolated rainfall field data P 観測点補間 and P at each mesh center every hour. レーダー and P 観測点補間 Relative error RE j Calculate the RE for j=1, 2, ..., J for each time period. j > Number of meshes in D2, N D2,t Statistical analysis of N D2,t The time period D3 is set as belonging to two types of time periods where the difference in rainfall distribution is significant, and the total number of such time periods is denoted as T. Here, the thresholds D2 and D3 are both determined based on the natural geographical conditions and data conditions of the basin, and the relative error RE of each mesh within T time periods is calculated. j Step 3: Count the average values ​​of the γ mesh centers and sort them in descending order. The first γ mesh centers are selected as candidate locations for addition, and are numbered 1, 2, ..., γ, respectively. Step 4: Evaluate the accuracy and effectiveness of different rain gauge network optimization methods: Set 1 to γ ​​rain gauge stations to be added, generating a total of γ types of additional methods, where the γ types of additional methods are added at candidate location number 1, candidate locations number 1 and 2, ..., γ candidate locations, respectively. The total number of rain gauge stations corresponding to the γ types of additional methods is K+1, K+2, ..., K+γ, respectively. A machine learning model is used to predict the rainfall conditions of different additional means. For each rain gauge location s, a feature vector X(s) is constructed, which includes the geographic location of the observation point, topographical features, meteorological features, and radar rainfall data corresponding to the mesh where it is located. Then, the feature vectors and actual rainfall data of all K rain gauge stations are used as the training data set to train the machine learning model f. The rain gauge feature vector X is used as the model input, and the corresponding actual rainfall data is used as the model output. Then, using the trained model, for each additional candidate location s*, the feature vector X(s * ) based on the rainfall situation at that location

number

number

number

number

[0007] Furthermore, in step 1, the same mesh division is performed on the target area based on the mesh distribution of the radar rainfall data, specifically by analyzing the radar rainfall data file, extracting its spatial resolution, and determining the basic unit of the mesh. Then, based on the spatial position of the radar data, the mesh is systematically numbered using the matrix coordinate method, with the row number increasing from north to south and the column number increasing from west to east.

[0008] Furthermore, constructing the node features of rain gauge stations in step 2 specifically involves converting the geographical location relationships between rain gauge stations into a graph structure, with each rain gauge station being regarded as a node in the graph structure, the edges between the nodes indicating the spatial correlation between the rain gauge stations, and the edge weights being set by a distance metric or a similarity metric based on geographical distance, the node features of the rain gauge stations including rainfall time series features and geographic information features, the rainfall time series features including rainfall statistical features and time feature information for historical time periods, and the geographic information features including the latitude and longitude, altitude, and topographical features of each rain gauge station.

[0009] Furthermore, in step 2, identifying abnormal observation points of rain gauge stations using the trained GCN model specifically includes calculating the reconstruction error RE after prediction by the trained GCN model for each rain gauge station for all N time periods;

number

number

[0010] Furthermore, in step 4, the terrain features include elevation, slope, and slope direction, and the meteorological features include temperature, humidity, and wind speed. [Effects of the Invention]

[0011] The beneficial effects of the present invention are as follows: the method of the present invention, combined with radar rainfall data, realizes effective planning and optimization of rainfall observation networks in complex mountainous terrain, further improves the accuracy and reliability of precipitation observation in mountainous watersheds, and provides more accurate data support for the management and prevention of mountainous flood disasters.

[0012] The invention will now be described in more detail with reference to the drawings and specific embodiments. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a flow chart of a method according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0014] The present invention discloses a method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data, and as shown in FIG. 1, the method includes the following steps:

[0015] Step 1: Collect rainfall data and mesh the target area.

[0016] For the target area where the layout of the rain gauge network is to be optimized, first, the actual rainfall data P 雨量観測所 and P 雨量観測所 Time series of radar rainfall data P レーダー The radar rainfall data is collected, with priority given to rainfall during the flood season. Next, based on the mesh distribution of the radar rainfall data, the same mesh division is performed on the target area (a total of J meshes), and each mesh is assigned a unique identifier. Specifically, the radar rainfall data file is analyzed to extract its spatial resolution and determine the basic unit of the mesh. Then, based on the spatial location of the radar data, the matrix coordinate method is used to systematically number the meshes, with row numbers increasing from north to south and column numbers increasing from west to east, ensuring that each mesh has a unique identifier and facilitating subsequent data processing and analysis.

[0017] Step 2, Identification of abnormal rain gauge observation points.

[0018] Considering the possibility of operational anomalies in existing rain gauge stations, we use a graph convolutional network (GCN) model to identify anomalous rain gauge stations and ensure that the existing observation network is composed of normally operating stations. First, we construct node features for each rain gauge station. If each station's rainfall data spans N hours, we train a GCN model using the node features of M rain gauge stations and the corresponding actual rainfall data for N hours as a training data set. The rain gauge station's node features serve as model input and the corresponding actual rainfall data serves as model output. Next, we use the trained GCN model to identify anomalous rain gauge stations. The identified anomalous stations are removed, and the number of normally operating rain gauge stations is recorded as K. Subsequent steps are based only on the K rain gauge stations, ignoring the anomalous stations.

[0019] The specific process is as follows: 1) Construct node features for rain gauge stations. The geographical location relationships of rain gauge stations are converted into a graph structure, with each rain gauge station considered as a node in the graph structure. Edges between nodes indicate the spatial correlation between the rain gauge stations, and edge weights are set by a distance metric (e.g., Euclidean distance) or a similarity metric based on geographic distance. The node features for rain gauge stations include rainfall time series features and geographic information features. The rainfall time series features include information such as historical time-series rainfall statistical features (e.g., average and variance of precipitation over the past few hours or days) and time features (e.g., the year, month, quarter, and time of day in which the current time period is located), while the geographic information features include the longitude and latitude, altitude, and topographical features of each rain gauge station. 2) Initialize the model parameters and train the GCN model using the training dataset. Set the model layer L, the number of neurons in each layer d1, and the activation function σ. The lth (l=1, 2, ..., L) layer of the GCN model can be expressed as follows:

number

number

number

number

number

[0020] Step 3: Identifying candidate locations for additional rain gauge stations.

[0021] The actual rainfall data from K rainfall stations are spatially interpolated every hour using the Kriging interpolation method, and the rainfall field data P 観測点補間 In the above steps, for the J meshes divided in step 1, there are two sets of rainfall data at the center of each mesh, and each set of radar rainfall data P レーダー and the interpolated rainfall field data P 観測点補間 P at each mesh center every hour レーダー and P 観測点補間 The relative error RE of the two types of rainfall j (j=1, 2, ..., J) and calculate the RE for each time period. j > Number of meshes in D2, ND2,t Statistical analysis of N D2,t The time period D3 is set as belonging to two types of time periods where the difference in rainfall distribution is significant, and the total number of such time periods is denoted as T. Here, the thresholds D2 and D3 are both determined based on the natural geographical conditions of the basin and the data situation. Next, the relative error RE of each mesh within T time periods is calculated. j The average values ​​of are calculated and arranged in descending order. The first γ mesh center points are selected as additional candidate positions, and are numbered 1, 2, ..., γ, respectively.

[0022] Step 4: Accuracy effect evaluation of different rain gauge network optimization measures.

[0023] Set 1 to γ ​​rain gauge stations to be supplemented, generating a total of γ types of additional means, and the γ types of additional means are respectively added at candidate position number 1, candidate positions number 1 and 2, and similarly added at candidate positions number 1 and 2, ..., γ, so that the total number of rain gauge stations corresponding to the γ types of additional means is K+1, K+2, ..., K+γ, respectively; A machine learning model (such as random forest or gradient lift tree) is used to predict the rainfall conditions for different additional methods. For each rain gauge location s, a feature vector X(s) is constructed, which includes the geographic location of the station, terrain features (including elevation, slope, slope direction, etc.), meteorological features (including temperature, humidity, wind speed, etc.), and radar rainfall data corresponding to the mesh where it is located. Next, the feature vectors and actual rainfall data of all K rain gauge stations are used as a training dataset to train a machine learning model f, with the rain gauge feature vector X as the model input and the corresponding actual rainfall data as the model output. Then, using the trained model, for each additional candidate location s*, the feature vector X(s * ) based on the rainfall situation at that location

number

number

number

number

[0024] Step 5: Determining the optimal rainfall observation network deployment method.

[0025] Combined with actual economics, an optimization objective function for the placement of the observation network is constructed to determine the optimal rainfall observation network placement method. Taking into account the construction cost, maintenance cost and potential socio-economic effects of the rainfall observation station, a penalty function ω is constructed to indicate the adverse impact of the actual conditions on the construction of the rainfall observation station, where C is the construction cost, W is the maintenance cost and B is the socio-economic effect, and C, W and B are all functions of additional means and change with the change of means. Then, the penalty function ω is defined as follows: ω=αC+βW-φB (5) where: α, β, φ are weighting coefficients. Combined with per and ω, we construct an optimization objective function φ, which is positively correlated with per and negatively correlated with ω, i.e., φ=μ1×per-μ2×ω (6) where μ1, μ2 are weighting coefficients, assigned according to importance. For each type of additional method, the corresponding φ value is calculated, and the method with the highest φ value is selected as the optimal rain gauge station placement method, which not only maximizes the accuracy of rainfall observation, but also ensures the economic rationality of construction and maintenance, thereby achieving comprehensive optimization of technology and economy.

[0026] Finally, it should be mentioned that the above is used to explain but not limit the technical solutions of the present invention, and the present invention has been described in detail with reference to preferred arrangements, but it is understood by those skilled in the art that the technical solutions of the present invention can be modified or equivalently substituted without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data, comprising: Step 1, Collecting rainfall data and meshing the target area: Actual rainfall data P of all M rainfall stations within the target area 雨量観測所 and P 雨量観測所 Time series corresponding radar rainfall data P レーダー Step 1: Collecting the data, dividing the target area into the same meshes based on the mesh distribution of the radar rainfall data, and dividing the total into J meshes, and assigning a unique identifier to each mesh; Step 2, Identifying abnormal rainfall observation points: Step 2: Identify abnormal rain gauge stations using a graph convolutional network (GCN) model, by first constructing node features for the rain gauge stations; if the actual rainfall data from each rain gauge station is N hours, train the GCN model using the node features of M rain gauge stations and the corresponding N hours of actual rainfall data as a training data set; use the node features of the rain gauge stations as model input and the corresponding actual rainfall data as model output; then identify abnormal rain gauge stations using the trained GCN model; remove the identified abnormal rain gauge stations; and record K as the number of normally operating rain gauge stations; Step 3: Identifying candidate locations for additional rain gauge stations: The actual rainfall data from K rainfall observation stations is spatially interpolated every hour using the Kriging interpolation method, and the rainfall field data P 観測点補間 Then, for the J meshes divided in step 1, there are two sets of rainfall data at the center of each mesh, and each set of radar rainfall data P レーダー and the interpolated rainfall field data P 観測点補間 and P at each mesh center every hour. レーダー and P 観測点補間 Relative error RE ｊ , where j = 1, 2, ..., J, and the RE ｊ >D ２ Number of meshes N Ｄ２，ｔ Statistical analysis of N Ｄ２，ｔ >D ３ The time periods are set to belong to two types of time periods where the difference in rainfall distribution is significant, and the total number of such time periods is denoted as T. Here, the threshold D ２ and D ３ are determined based on the natural geographical conditions and data conditions of the basin, and the relative error RE of each mesh within T time periods ｊ Step 3: Count the average values ​​of the above and sort them in descending order, and select the first γ mesh center points as additional candidate positions, numbered 1, 2, ..., γ, respectively; Step 4: Evaluate the accuracy and effectiveness of different rain gauge network optimization measures: Set 1 to γ ​​rain gauge stations to be supplemented, generating a total of γ types of additional means, and the γ types of additional means are respectively added at candidate position number 1, at candidate positions number 1 and 2, and by analogy, at candidate positions number 1 and 2, ..., γ, so that the total number of rain gauge stations corresponding to the γ types of additional means is K+1, K+2, ..., K+γ, respectively; A machine learning model is used to predict the rainfall conditions of different additional means. For each rain gauge location s, a feature vector X(s) is constructed, which includes the geographic location of the observation point, topographical features, meteorological features, and radar rainfall data corresponding to the mesh where it is located. Then, the feature vectors and actual rainfall data of all K rain gauge locations are used as a training data set to train a machine learning model f. The feature vector X of the rain gauge location is used as the model input, and the corresponding actual rainfall data is used as the model output. Then, using the trained model, for each additional candidate location s*, its feature vector X(s ＊ ) based on the rainfall situation at that location [Equation 1] is predicted as follows: [Equation 2] For all T time periods, predict rainfall data for each additional candidate location in γ types of additional means for each time period, and perform spatial interpolation on the prediction results of each additional means using Kriging interpolation to generate corresponding rainfall field data; For T time periods, RE is again performed for each time period. ｊ >D ２ Calculate the number of meshes in [Equation 3] That is, the RE in the t-th time period ｊ >D ２ The number of meshes is set to t=1, 2, ..., T, and then the average percentage reduction in the number of meshes per for all time periods is calculated. [Equation 4] Step 4: The larger the per, the closer the basin's rainfall field data and radar rainfall data are after adding; Step 5: Determine the optimal rainfall network deployment method: Considering the construction cost, maintenance cost and potential socio-economic effects of the rain gauge station, construct a penalty function ω to indicate the adverse impact of field conditions on the construction of the rain gauge station. Let C be the construction cost, W be the maintenance cost, and B be the socio-economic effect. C, W, and B are all functions of additional means and change with the changes in means. Then, the penalty function ω is defined as follows: ω=αC+βW−φB (5) where: α, β, φ are weighting coefficients; Combined with per and ω, we construct an optimization objective function φ, which is positively correlated with per and negatively correlated with ω, i.e., f=m １ ×per-m ２ ×ω (6) In the formula: μ １ , μ ２ is a weighting coefficient, Step 5: for each type of additional means, calculate the corresponding φ value, and select the means with the highest φ value as the optimal rain gauge station placement means; A method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data, comprising:

2. The method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data, as described in claim 1, characterized in that in step 1, the same mesh division is performed on the target area based on the mesh distribution of radar rainfall data, specifically by analyzing the radar rainfall data file, extracting its spatial resolution, determining the basic unit of the mesh, and then systematically numbering the meshes using the matrix coordinate method based on the spatial position of the radar data, with the row numbers increasing from north to south and the column numbers increasing from west to east.

3. 2. The method for optimizing the layout of a mountainous rainfall observation network based on radar rainfall data according to claim 1, wherein constructing node features of rainfall observation stations in step 2 specifically involves converting the geographical location relationships between the rainfall observation stations into a graph structure, with each rainfall observation station being regarded as a node in the graph structure, the edges between the nodes indicating the spatial correlation between the rainfall observation stations, and the weights of the edges being set by a distance metric or a similarity metric based on geographical distance, the node features of the rainfall observation stations including rainfall time series features and geographic information features, the rainfall time series features including rainfall statistical features and time feature information for historical time periods, and the geographic information features including the latitude and longitude coordinates, altitude, and topographical features of each rainfall observation station.

4. In step 2, identifying abnormal observation points of rain gauge stations using the trained GCN model specifically includes calculating, for each rain gauge station, a reconstruction error RE predicted by the trained GCN model for all N time periods; [Equation 5] During the ceremony: [Equation 6] are the actual measured value at the rain gauge station at time n and the predicted value based on the GCN model, respectively. RE>D １ The rainfall observation station is marked as an anomaly observation point, and the threshold D １ 2. The method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data according to claim 1, wherein the is determined in combination with the local weather conditions and natural geographical features.

5. 2. The method for optimizing the layout of a mountain rainfall observation network based on radar rainfall data according to claim 1, wherein in step 4, the topographical features include elevation, gradient, and gradient direction, and the meteorological features include temperature, humidity, and wind speed.

Images