River domain upstream snow depth data set construction method based on deep learning

Abstract

The invention discloses a deep learning-based river region upstream snow depth data set construction method. The method comprises the steps of collecting remote sensing snow depth data and ground station snow depth data; the remote sensing snow depth data and the ground station snow depth data with the unified resolution are matched through geographic coordinates, and snow depth matrix sample data and time sequence sample data are generated; learning the snow depth matrix sample data by using a multi-layer perceptron and learning time sequence sample data by using a one-dimensional convolutional neural network model; inputting the feature vectors into a model for data fusion to obtain a unified snow depth data set; and comparing the snow depth data set with observation data of the ground station, and evaluating the snow depth data set after data fusion. According to the method, multiple remote sensing data can be effectively utilized, accurate fusion of snow depth data is realized through a deep learning method, the method better serves the fields of water resource management, ecological monitoring, climate change research and the like, and a reliable scientific basis is provided for decision making.

Description

Technical Field

[0001] The present invention relates to a method for constructing snow depth data in the upper reaches of a river basin, specifically a method for constructing snow depth data in the upper reaches of a river basin based on deep learning, and belongs to the technical field of hydrometeorological data applications. Background Art

[0002] Snow depth, as an important physical parameter of snow accumulation, is an important indicator for measuring the thickness and changes of the snow layer, and is of great significance for hydrometeorological research, ice and snow disaster early warning, water resource management, etc. Although certain progress has been made in snow depth monitoring technology, in some remote areas, especially those with complex terrain and high altitude, the observation of snow depth still faces many challenges.

[0003] In the prior art, the method for measuring snow depth mainly relies on manual observation, that is, measuring the snow depth through on-site investigation. Although this method of measuring snow depth is accurate, it is limited by manual operation and time cost, and it is difficult to cover a wide area. Especially for large-scale, high-altitude or complex terrain areas, the monitoring work is very difficult. In addition, the manual measurement process is often affected by weather changes and environmental conditions, resulting in the inability to guarantee the timeliness and accuracy of the data.

[0004] With the progress of remote sensing technology, satellite observation data has gradually become an important tool for snow depth monitoring. Satellite remote sensing can obtain large-scale and continuous surface information through sensors in different bands. Especially in winter, the acquisition of snow depth data is relatively effective. By using microwave remote sensing technology, the snow cover fraction and snow depth information can be extracted more accurately. However, although satellite remote sensing data can provide snow depth monitoring results for large-scale areas, its resolution is limited by the accuracy of satellite sensors and observation conditions. In addition, factors such as weather changes, cloud cover, and precipitation may affect the transmission of remote sensing signals, resulting in data loss or reduced accuracy in some areas.

[0005] Generally speaking, although the existing snow depth monitoring technologies have their own advantages, they still face problems such as limited spatial coverage and unstable accuracy. Summary of the Invention

[0006] The technical solution of the present invention aims to solve the technical problems existing in the background art and provides a solution significantly different from the prior art. Specifically, it is a method for constructing a snow accumulation data set in the upper reaches of a river basin based on deep learning. This method uses deep learning technology to extract effective features from various types of data and combines multi-source remote sensing data, so as to more comprehensively and accurately reflect the change process of snow depth, improve the accuracy and timeliness of snow depth data, and further provide more reliable data support for hydrological forecast analysis, etc.

[0007] To achieve the above object, the present invention adopts the following technical solutions: A method for constructing a snow depth data set for the upper reaches of the river basin based on deep learning, and the construction method includes the following steps:

[0008] S1. Collect remote sensing snow depth data and ground station snow depth data;

[0009] S2. Match the remote sensing snow depth data with the same resolution and the ground station snow depth data through geographic coordinates to generate snow depth matrix sample data and time series sample data;

[0010] S3. Use a multi-layer perceptron to learn the snow depth matrix sample data to obtain a snow depth matrix feature vector, and use a one-dimensional convolutional neural network model to learn the time series sample data to obtain a time series feature vector;

[0011] S4. Input the snow depth matrix feature vector and the time series feature vector into the one-dimensional convolutional neural network model for data fusion to obtain a unified snow depth data set;

[0012] S5. Compare the snow depth data set with the observation data of the ground station, evaluate the snow depth data set after data fusion, and ensure the accuracy and reliability of the snow depth data set.

[0013] As a further technical solution of the present invention: In S1, the remote sensing snow depth data includes AMSR-E microwave snow cover snow depth products and SSMI / S remote sensing snow depth data, and the ground station snow depth data is the measured snow depth data on a daily scale.

[0014] As a further technical solution of the present invention: In S2, the same resolution means that the AMSR-E microwave snow cover snow depth products and SSMI / S remote sensing snow depth data are unified in spatial resolution by using the bilinear interpolation method.

[0015] As a further technical solution of the present invention: In S2, the geographic coordinate matching means that the remote sensing snow depth data after unified resolution is based on the ground station, and the nearest grid point is selected as the snow depth data.

[0016] As a further technical solution of the present invention: In S3, the specific formula for using a multi-layer perceptron to learn the snow depth matrix sample data is:

[0017]

[0018] In the formula: FV m is the feature vector in the matrix mode sample data of the m-th month; is the connection weight between the k-th neuron in the l-1 layer and the z-th neuron in the l layer; is the bias of the z-th neuron in the l layer; x Zis the z - input variable, and L is the number of layers of the neural network.

[0019] As a further technical solution of the present invention: In S3, the specific process of using the one - dimensional convolutional neural network model to learn the time - series sample data is to input the time - series sample data into the one - dimensional convolutional neural network model. The input time - series sample data is a one - dimensional time - series vector, denoted as X = [X1, X2, …, X n , where X i is the snow depth value at the i - th moment. The local time features of the time - series sample data are extracted through convolution operations, and then the time - series feature values are output;

[0020] Among them, the formula of the one - dimensional convolutional neural network model is:

[0021]

[0022] In the formula: Z t is the output of the convolution operation, w k is the convolution kernel weight, b is the bias term, K is the size of the convolution kernel, t represents the time step; f is the activation function ReLU, A t is the output of the convolutional layer; P t is the output of the pooling layer, and S is the pooling window size, indicating that the maximum value is selected from consecutive S time steps as the output;

[0023] The time - series feature values extracted by the convolutional layer and the pooling layer are concatenated or weighted and fused to form the final time - series feature vector. The specific formula is:

[0024] F final = [P1, P2,..., P m

[0025] In the formula: F final is the finally output feature vector, and m is the number of time steps of the output of the pooling layer.

[0026] As a further technical solution of the present invention: In S4, the data fusion is specifically to use the Concat layer to concatenate the snow depth matrix feature vector and the time - series feature vector, and then input them into the fully - connected layer for fusion

[0027] As a further technical solution of the present invention: In S5, the accuracy and reliability of the fused snow depth data set are evaluated by using the root - mean - square error RMSE and the mean absolute error MAE indicators;

[0028] Among them, the formula of the root - mean - square error RMSE is:

[0029]

[0030] ​The formula for the mean absolute error (MAE) is as follows:

[0031]

[0032] In the formula: n is the number of meteorological stations in the study area, G s represents the measured snow depth data of meteorological stations, and y s represents the daily snow depth data product after fusion.

[0033] The beneficial effects of the present invention are as follows: The present invention can effectively utilize multiple remote sensing data and achieve accurate fusion of snow depth data through deep learning methods, making up for the deficiencies of the existing technology in terms of data acquisition accuracy and timeliness, better serving fields such as water resource management, ecological monitoring, and climate change research, and providing a reliable scientific basis for decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] Embodiment, as Figure 1 shown, this embodiment takes the upper reaches of the Jinsha River basin as an example to provide a method for constructing a snow depth dataset in the upper reaches of the Jinsha River based on deep learning. The construction method includes:

[0037] First: Collect remote sensing snow depth data and ground station snow depth data.

[0038] Among them, the remote sensing snow depth data includes AMSR-E microwave snow cover snow depth products and SSMI / S remote sensing snow depth data; the ground station snow depth data is the measured snow depth data on a daily scale.

[0039] Second: Match the remote sensing snow depth data with unified resolution and the ground station snow depth data through geographic coordinates to generate snow depth matrix sample data and time series sample data.

[0040] Among them, unified resolution means unifying the spatial resolution of AMSR-E microwave snow cover snow depth products and SSMI / S remote sensing snow depth data using the bilinear interpolation method; geographic coordinate matching means that the remote sensing snow depth data after unified resolution is based on the ground station, and the nearest grid point is selected as the snow depth data for subsequent calculations.

[0041] Third: Use a multi-layer perceptron to learn the snow depth matrix sample data to obtain a snow depth matrix feature vector, and use a one-dimensional convolutional neural network model to learn the time series sample data to obtain a time series feature vector.

[0042] Among them, the specific formula for using a multi-layer perceptron to learn the snow depth matrix sample data is:

[0043]

[0044] In the formula: FV m is the feature vector in the matrix modal sample data of the m-th month; is the connection weight between the k-th neuron in the (l - 1)-th layer and the z-th neuron in the l-th layer; is the bias of the z-th neuron in the l-th layer; x Z is the z-th input variable, and L is the number of neural network layers.

[0045] The specific process of using a one-dimensional convolutional neural network model to learn the time series sample data is to input the time series sample data into the one-dimensional convolutional neural network model. The input time series sample data is a one-dimensional time series vector, denoted as X = [X1, X2, …, X n , where X i is the snow depth value at the i-th moment. Local time features of the time series sample data are extracted through convolution operations, and then time series feature values are output;

[0046] Among them, the formula of the one-dimensional convolutional neural network model is:

[0047]

[0048] In the formula: Z t is the output of the convolution operation, w k is the convolution kernel weight, b is the bias term, K is the size of the convolution kernel, t represents the time step; f is the activation function ReLU, A t is the output of the convolutional layer; P t is the output of the pooling layer, and S is the pooling window size, indicating that the maximum value is selected from consecutive S time steps as the output;

[0049] Concatenate or weighted fuse the time series feature values extracted by the convolutional layer and the pooling layer to form the final time series feature vector. The specific formula is:

[0050] F final = [P1, P2, ..., P m

[0051] In the formula: F final is the finally output feature vector, and m is the number of time steps of the pooling layer output.

[0052] Fourth: Input the snow depth matrix feature vector and the time series feature vector into a one-dimensional convolutional neural network model for data fusion to obtain a unified snow depth dataset.

[0053] Specifically, data fusion is to use the Concat layer to splice the snow depth matrix feature vector and the time series feature vector, and then input them into the fully connected layer for fusion.

[0054] Fifth: Compare the snow depth dataset with the observed data of ground stations, evaluate the snow depth dataset after data fusion, and ensure the accuracy and reliability of the snow depth dataset.

[0055] Among them, the evaluation of the snow depth dataset after fusion uses the root mean square error RMSE and the mean absolute error MAE metrics to evaluate its accuracy and reliability;

[0056] Among them, the formula for the root mean square error RMSE is:

[0057]

[0058] The formula for the mean absolute error MAE is:

[0059]

[0060] In the formula: n is the number of meteorological stations in the study area, G s represents the measured snow depth data of meteorological stations, and y s represents the daily snow depth data product after fusion.

[0061] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0062] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for constructing a snow depth dataset in the upper reaches of the river basin based on deep learning, characterized in that, The construction method includes the following steps: S1. Collect remote sensing snow depth data and ground station snow depth data; S2. Match the remote sensing snow depth data with the same resolution and the ground station snow depth data through geographic coordinates to generate snow depth matrix sample data and time series sample data; S3. Use a multi-layer perceptron to learn the snow depth matrix sample data to obtain a snow depth matrix feature vector, and use a one-dimensional convolutional neural network model to learn the time series sample data to obtain a time series feature vector; S4. Input the snow depth matrix feature vector and the time series feature vector into the one-dimensional convolutional neural network model for data fusion to obtain a unified snow depth data set; S5. Compare the snow depth data set with the observation data of the ground station, evaluate the snow depth data set after data fusion, and ensure the accuracy and reliability of the snow depth data set.

2. The method for constructing the snow depth dataset in the upper reaches of the river domain according to claim 1, wherein In S1, the remote sensing snow depth data includes AMSR-E microwave snow cover snow depth products and SSMI / S remote sensing snow depth data, and the ground station snow depth data is the measured snow depth data on a daily scale.

3. The method for constructing the snow depth dataset in the upper reaches of the river domain according to claim 1, wherein, In S2, the same resolution means that the AMSR-E microwave snow cover snow depth products and SSMI / S remote sensing snow depth data are unified in spatial resolution using the bilinear interpolation method.

4. The method for constructing the snow depth data set in the upper reaches of the river domain according to claim 1, wherein In S2, the geographic coordinate matching means that the remote sensing snow depth data after unified resolution is based on the ground station, and the nearest grid point is selected as the snow depth data.

5. The method for constructing the snow depth data set in the upper reaches of the river domain according to claim 1, characterized in that, In S3, the specific formula for using a multi-layer perceptron to learn the snow depth matrix sample data is: Where: FV m is the eigenvector in the matrix modal sample data of the m-th month; is the connection weight between the k-th neuron in the (l - 1)-th layer and the z-th neuron in the l-th layer; is the bias of the z-th neuron in the l-th layer; x Z is the z-th input variable, and L is the number of layers of the neural network.

6. The method for constructing the snow depth data set in the upper reaches of the river domain according to claim 1, wherein In S3, the specific process of using the one-dimensional convolutional neural network model to learn the time series sample data is to input the time series sample data into the one-dimensional convolutional neural network model. The input time series sample data is a one-dimensional time series vector, denoted as X = [X1, X2, …, X n , where X i is the snow depth value at the i-th moment. The local time features of the time series sample data are extracted through convolutional operations, and then the time series feature values are output; Among them, the formula for the one-dimensional convolutional neural network model is: Where: Z ％ is the output of the convolution operation, w k is the convolution kernel weight, b is the bias term, K is the size of the convolution kernel, t represents the time step; f is the activation function ReLU, A ％ is the output of the convolutional layer; P ％ is the output of the pooling layer, S is the pooling window size, indicating that the maximum value is selected from consecutive S time steps as the output; The time series feature values extracted by the convolutional layer and the pooling layer are concatenated or weighted and fused to form the final time series feature vector. The specific formula is: F 5in6l = [P1, P2,..., P m ​ Where: F 5in6l is the finally output feature vector, and m is the number of time steps output by the pooling layer.

7. The method for constructing the snow depth dataset in the upper reaches of the river domain according to claim 1, wherein, In S4, the data fusion specifically uses the Concat layer to concatenate the snow depth matrix feature vector and the time series feature vector, and then inputs it into the fully connected layer for fusion.

8. The method for constructing the snow depth data set in the upper reaches of the river domain according to claim 1, characterized in that, In S5, the evaluation of the snow depth data set after fusion uses the root mean square error RMSE and the mean absolute error MAE indicators to evaluate its accuracy and reliability; Among them, the formula for the root mean square error RMSE is: The formula for the mean absolute error MAE is: where: n is the number of meteorological stations in the study area, G s represents the measured snow depth data of meteorological stations, y s represents the daily snow depth data product after fusion.

Images