Double-branch network soil moisture inversion method based on TM-01 data

Abstract

The invention discloses a double-branch network soil moisture inversion method based on TM-01 data, and relates to the technical field of soil moisture remote sensing inversion, and the method comprises the basic steps: carrying out the time and space matching of TM-01 satellite data and SMAP satellite soil moisture product data; performing normalization and quality control on the matched training sample and verification sample; analyzing the importance of data features by using an XGBoost machine learning method, and eliminating three features with low importance; inputting the optimized sample data and verification data into a full-connection neural network and a random forest model, establishing the model and inverting a soil moisture result; and finally, fusing the soil moisture inversion results of the two models through a weighted average method to obtain a final soil moisture inversion result. The method has the advantages of objective and reasonable satellite observation data, high soil moisture inversion precision and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing inversion of soil moisture, and particularly to a method for inverting soil moisture by a dual-branch network based on TM-01 data. Background Art

[0002] Soil moisture is a key variable connecting the water cycle and ecological energy exchange between land and atmosphere, and is closely related to agricultural production, climate regulation and hydrological processes. Accurately monitoring soil moisture is of great significance for drought detection, climate change prediction and water resource management. Currently, optical remote sensing and passive microwave remote sensing are the main technical means for large-scale soil moisture observation. Optical remote sensing satellite data provides high spatial resolution, but is vulnerable to cloud and vegetation interference; passive microwave remote sensing data has high temporal resolution, but low spatial resolution. Passive microwave remote sensing technology represented by the Soil Moisture Active Passive (SMAP) satellite can provide global soil moisture data with a resolution of 36 km and a revisit period of 2-3 days. However, for some hydrological and climate applications, soil moisture data with higher spatio-temporal resolution is still urgently needed.

[0003] With the development of GNSS-R technology, as an alternative to traditional soil moisture monitoring methods, it has shown great potential. GNSS-R has the advantages of global coverage, low cost, insensitivity to weather conditions, near-real-time data acquisition and short satellite revisit time. In soil moisture inversion, characteristics such as surface vegetation cover and roughness lead to a non-linear relationship between GNSS-R signals and surface soil moisture. Compared with linear models, machine learning and deep learning models can better capture these complex relationships. Currently, most GNSS-R soil moisture inversion studies are based on TechDemoSat-1 (TDS-1) and Cyclone Global Navigation Satellite System (CYGNSS) satellite data, and no relevant research based on TM-01 satellite data has been reported. Therefore, the present invention uses TM-01 satellite data and proposes a method for inverting soil moisture by a dual-branch network based on TM-01 data, providing technical support for soil moisture monitoring based on TM-01 data. Summary of the Invention

[0004] The present invention provides a method for inverting soil moisture by a dual-branch network based on TM-01 data, effectively improving the accuracy of soil moisture inversion under complex surface conditions and providing a new technical path and data support for soil moisture monitoring.

[0005] According to one aspect of the present disclosure, there is provided a method for inverting soil moisture by a dual-branch network based on TM-01 data, comprising the following steps: (1) Obtain TM-01 satellite data and SMAP satellite soil moisture product data; match the TM-01 satellite data and SMAP satellite soil moisture product data in terms of time and space to generate training samples and validation samples with a ratio of 7:3. Among them, the time match is to match the data of the same date of the TM-01 satellite data and SMAP satellite soil moisture product data, and the space match is to match the TM-01 satellite data and SMAP satellite soil moisture product data using the nearest neighbor method, that is, retain the point with the closest longitude and latitude between the specular reflection point of the TM-01 satellite data and the grid point of the SMAP satellite soil moisture product data for spatial matching; (2) Perform preprocessing operations on the training samples and validation samples, including: normalization and quality control processing; among them, the quality control processing includes: removing low-quality reflection points and water body areas; (3) Use XGBoost machine learning to analyze the importance of information features in the training samples and validation samples, and remove the three features with low importance; (4) Establish a fully connected neural network model. By selecting the relu activation function, data batch normalization, L2 regularization, Dropout, and setting the parameters of the fully connected network, input the training sample data and validation sample data after removing the features with low importance into the fully connected neural network. After multiple iterations, obtain the soil moisture inversion result y1 and the root mean square error RMSE1; (5) Establish a random forest model. By setting the number of decision trees, the maximum depth of the decision tree, the maximum number of features, the minimum number of samples at internal nodes, the minimum number of samples at leaf nodes, the evaluation criterion, and the sample sampling method, input the training sample data and validation samples after removing the features with low importance into the random forest model to obtain the soil moisture inversion result y2 and the root mean square error RMSE2; (6) Use the weighted average method to calculate the soil moisture inversion results y1, y2 and the root mean square errors RMSE1, RMSE2 to obtain the final soil moisture inversion result.

[0006] In a possible implementation, in step (6), use the weighted average method to calculate according to the soil moisture inversion results y 1 , y 2 and the root mean square errors RMSE 1 , RMSE 2 to obtain the final soil moisture inversion result y. The calculation formula is: ; Among them, the unit of the inversion result y is cm 3 / cm 3 .

[0007] Compared with the prior art, the beneficial effects of the present invention are: A dual-branch network soil moisture inversion method based on TM-01 data according to an embodiment of the present disclosure. Currently, most of the research on soil moisture inversion based on GNSS-R focuses on TDS-1 and CYGNSS satellite data, and there is no relevant research on using TM-01 satellite data. By using TM-01 satellite data, the present invention fully exploits the potential of TM-01 satellite data based on a dual-branch network, effectively improving the accuracy of soil moisture inversion under complex surface conditions, and providing a new technical path and data support for soil moisture monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart showing a dual-branch network soil moisture inversion method based on TM-01 data according to an embodiment of the present disclosure.

[0009] Figure 2 A soil moisture inversion accuracy graph showing an embodiment of the present disclosure. DETAILED DESCRIPTION

[0010] Various exemplary embodiments, features, and aspects of the present disclosure will be described in detail below with reference to the drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0011] The specifically used term "exemplary" herein means "serving as an example, embodiment, or illustrative". Any embodiment described herein as "exemplary" is not necessarily to be construed as superior or better than other embodiments.

[0012] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following detailed description. Those skilled in the art should understand that the present disclosure can be implemented without some of these specific details. In some instances, methods, means, elements, and circuits well known to those skilled in the art are not described in detail so as to highlight the gist of the present disclosure.

[0013] According to one aspect of the present disclosure, a dual-branch network soil moisture inversion method based on TM-01 data is provided, including the following steps: (1) Acquire TM-01 satellite data and SMAP satellite soil moisture product data; match the TM-01 satellite data and SMAP satellite soil moisture product data in time and space to generate training samples and validation samples with a ratio of 7:3. Temporal matching refers to matching the data of the same date of the TM-01 satellite data and the SMAP satellite soil moisture product data. Spatial matching refers to matching the TM-01 satellite data and the SMAP satellite soil moisture product data using the nearest neighbor method, that is, retaining the point with the closest latitude and longitude between the mirror reflection point of the TM-01 satellite data and the grid point of the SMAP satellite soil moisture product data for spatial matching; (2) Preprocessing the training samples and validation samples, including normalization and quality control. Quality control includes removing low-quality reflection points and water areas. (3) Use XGBoost machine learning to analyze the importance of information features in training samples and validation sample data, and remove three features with low importance; (4) Establish a fully connected neural network model FCNN. By selecting the relu activation function, data batch normalization, L2 regularization, Dropout and fully connected network parameter settings, the training sample data with low-importance features removed and the validation sample data are input into the fully connected neural network. After multiple iterations, the soil moisture inversion result y1 and the root mean square error RMSE1 are obtained. (5) Establish a random forest model RF. By setting the number of decision trees, the maximum depth of decision trees, the maximum number of features, the minimum number of samples for internal nodes, the minimum number of samples for leaf nodes, the evaluation criteria and the sample sampling method, the training sample data and the validation samples with low-importance features are removed and input into the random forest model to obtain the soil moisture inversion result y2 and the root mean square error RMSE2. (6) The weighted average method is used to calculate the soil moisture inversion results y1, y2 and the root mean square errors RMSE1 and RMSE2 to obtain the final soil moisture inversion results.

[0014] The present invention discloses a dual-branch network soil moisture inversion method based on TM-01 data. Currently, most of the research on soil moisture inversion based on GNSS-R focuses on TDS-1 and CYGNSS satellite data, and there is no related research using TM-01 satellite data. By using TM-01 satellite data, the present invention fully explores the potential of TM-01 satellite data based on a dual-branch network, effectively improves the accuracy of soil moisture inversion under complex surface conditions, and provides a new technical path and data support for soil moisture monitoring.

[0015] In a possible implementation, in step (6), a weighted average method is used to invert the soil moisture y 1, y 2 and the root mean square error RMSE 1 , RMSE 2 are calculated to obtain the final soil moisture inversion result y, and the calculation formula is: ; wherein, the unit of the inversion result y is cm 3 / cm 3 .

[0016] As Figure 2 shown, it is a scatter plot of the soil moisture inversion result obtained by the dual-branch network soil moisture inversion method based on TM-01 data described in this embodiment. As Figure 2 shown, the abscissa (x-axis) in the figure is the SMAP satellite soil moisture, and the ordinate (y-axis) is the predicted soil moisture. The diagonal line is y = x, the number of samples is 83156, the root mean square error RMSE is 0.0409, and the goodness-of-fit R for measuring linear regression is 0.84. It can be obtained that the function of the goodness-of-fit is y = 0.703x + 0.03.

[0017] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other ordinary technical personnel in the technical field to understand the embodiments disclosed herein.

Claims

1. A dual-branch network soil moisture inversion method based on TM-01 data, characterized in that: The following steps are involved: (1) Acquire TM-01 satellite data and SMAP satellite soil moisture product data; match the TM-01 satellite data and SMAP satellite soil moisture product data in time and space to generate training samples and validation samples with a ratio of 7:

3. Temporal matching refers to matching the data of the same date of the TM-01 satellite data and the SMAP satellite soil moisture product data. Spatial matching refers to matching the TM-01 satellite data and the SMAP satellite soil moisture product data using the nearest neighbor method, that is, retaining the point with the closest latitude and longitude between the mirror reflection point of the TM-01 satellite data and the grid point of the SMAP satellite soil moisture product data for spatial matching; (2) Preprocessing the training samples and validation samples, including normalization and quality control. Quality control includes removing low-quality reflection points and water areas. (3) Use XGBoost machine learning to analyze the importance of information features in training samples and validation sample data, and remove three features with low importance; (4) Establish a fully connected neural network model. By selecting the relu activation function, data batch normalization, L2 regularization, Dropout and fully connected network parameter settings, the training sample data and validation sample data with low-importance features removed are input into the fully connected neural network. After multiple iterations, the soil moisture inversion result y1 and the root mean square error RMSE1 are obtained. (5) Establish a random forest model. By setting the number of decision trees, the maximum depth of decision trees, the maximum number of features, the minimum number of samples for internal nodes, the minimum number of samples for leaf nodes, the evaluation criteria and the sample sampling method, the training sample data and the validation samples with low-importance features are removed and input into the random forest model to obtain the soil moisture inversion result y2 and the root mean square error RMSE2. (6) The weighted average method is used to calculate the soil moisture inversion results y1, y2 and the root mean square errors RMSE1 and RMSE2 to obtain the final soil moisture inversion results.

2. A dual-branch network soil moisture inversion method based on TM-01 data as described in claim 1, characterized in that: In step (6), the weighted average method is used to calculate the soil moisture inversion results y1, y2 and the root mean square errors RMSE1 and RMSE2 to obtain the final soil moisture inversion result y. The calculation formula is: ； Among them, the unit of the inversion result y is cm 3 / cm 3 .

Images