Soil humidity inversion method based on normalization processing of multi-source remote sensing data

Abstract

The invention discloses a soil humidity inversion method based on normalization processing of multi-source remote sensing data, and belongs to the technical field of soil data processing, and the method comprises the steps: obtaining environmental meteorological data, topographic data and actually measured soil humidity data of a high-grass grassland region of the high-grass grassland region of the high-grass grassland region of the high-grass grassland region of the high-grass grassland region of the high-grass grassland; acquiring multi-source remote sensing data corresponding to the high grass steppe region; performing space-time normalization on the multi-source remote sensing data through Z-Score standardization, a Loess curve and inverse Z-Score standardization; and matching the environmental meteorological data of the high-grass grassland area, the topographic data of the high-grass grassland area, the remote sensing data after time-space normalization and the actually measured soil humidity data of the high-grass grassland area to establish a database, and predicting the soil humidity through an XGBoost model. According to the soil humidity inversion method, the problem that the accuracy and the reliability of a soil humidity inversion model are reduced due to the fact that the data consistency is affected by the difference between different optical sensors in an existing method is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of soil data processing, and particularly relates to a soil moisture inversion method based on multi-source remote sensing data normalization processing. Background Art

[0002] Soil moisture is a core parameter in hydrology and ecology, and has a dual regulatory effect on the carbon cycle, energy flow, and water, carbon, and energy exchanges between the earth's surface and the atmosphere. As an important indicator and driving factor of climate change, its spatial heterogeneity characteristics are significant. It not only regulates the runoff response of the basin through evapotranspiration (such as the infiltration and runoff distribution of rainfall), but also profoundly affects the soil carbon cycle, vegetation productivity, and the occurrence and development of extreme climate events (droughts, floods, heatwaves). In grassland ecosystems, this parameter exhibits multi-dimensional value: it is not only a key indicator for evaluating biomass, monitoring the impact of droughts and disasters, but also dominates the evolution of grazing systems, the long-term interaction between large herbivores and vegetation, and at the same time has strategic significance for agricultural and pastoral production, ecological protection, and sustainable development. The breakthrough of high-resolution soil moisture inversion technology will directly promote applications such as environmental precise monitoring, crop assessment, and flood warning, while the establishment of a long-term dynamic monitoring system provides a scientific basis for analyzing the co-evolution law of the "soil-vegetation-animal" system under the background of climate change, and has irreplaceable decision-making support value for global ecological governance and resource management.

[0003] Understanding and constructing soil moisture inversion models is a classic problem in the field of remote sensing detection. Traditional soil moisture measurement methods, such as the constant temperature weighing method, time domain reflectometer, and resistance method, although having high measurement accuracy and usually regarded as "ground truth", have a very small measurement range and are difficult to meet the needs of environmental monitoring and disaster prevention. In contrast, satellite remote sensing technology can provide soil moisture data with a large range and high spatio-temporal resolution. Among them, the red, green, and blue (RGB) and infrared methods are essential in soil moisture monitoring, which can capture high-resolution time and space data, thus providing a detailed understanding of the soil moisture dynamics in the study area. Usually in space, the soil moisture data measured on the ground in the tallgrass prairie area often takes the form of single-point measurement or regional measurement, and is usually sparsely distributed in space. Therefore, the problem of insufficient measured sample data volume is usually compensated by long-term monitoring in time series to obtain more continuous long-time series measured soil moisture samples. In addition, the multi-source remote sensing data corresponding to the tallgrass prairie area can make up for the problem of the low time resolution of a single sensor. However, for the multi-source remote sensing data corresponding to the tallgrass prairie area, in terms of time, different optical sensors have different revisit periods and different measurement time intervals for a certain point or area; there are spectral responses and internal differences within the sensors among different optical sensors, as well as differences in imaging time and environment, resulting in obvious differences in the spectral wavelength range and spectral response function curves between the images of different sensors. These differences will affect the data consistency, thus reducing the accuracy and reliability of the soil moisture inversion model. Summary of the Invention

[0004] In order to overcome the defects existing in the prior art, the present invention provides a soil moisture inversion method based on the normalization processing of multi-source remote sensing data to solve the above problems.

[0005] The technical solution adopted by the present invention to solve its technical problems is: a soil moisture inversion method based on the normalization processing of multi-source remote sensing data, including the following steps: S1: Obtain the environmental meteorological data of the tallgrass prairie area, the terrain data of the tallgrass prairie area, and the measured soil moisture data of the tallgrass prairie area; S2: Obtain the multi-source remote sensing data corresponding to the tallgrass prairie area; S3: Perform spatio-temporal normalization on the multi-source remote sensing data corresponding to the tallgrass prairie area through Z-Score standardization, Loess curve, and inverse Z-Score standardization to obtain the spatio-temporally normalized remote sensing data; S4: Match and establish a database with the environmental meteorological data of the tallgrass prairie area, the terrain data of the tallgrass prairie area, the spatio-temporally normalized remote sensing data, and the measured soil moisture data of the tallgrass prairie area, and predict the soil moisture of the tallgrass prairie area through the XGBoost model.

[0006] Preferably, in the step S2, remote sensing data corresponding to the tallgrass prairie area on the corresponding date is obtained according to the acquisition time of the measured soil moisture data in the tallgrass prairie area; the vegetation index NDVI, the two-band enhanced vegetation index EVI2, and the water index NDWI of each remote sensing data are calculated.

[0007] Optionally, the multi-source remote sensing data corresponding to the tallgrass prairie area includes Landsat-5, Landsat-7, Landsat-8, Sentinel-2, and Planetscope multi-band optical satellite image data. After obtaining the remote sensing data, images with cloud cover less than 30% are selected and resampled to a spatial resolution of 30 meters.

[0008] Specifically, in the step S3, the average value μ and the standard deviation σ of all pixel values of the single image corresponding to the vegetation index NDVI, the single image corresponding to the two-band enhanced vegetation index EVI2, and the single image corresponding to the water index NDWI calculated from the multi-source remote sensing data corresponding to the original tallgrass prairie area in the same year are calculated, and each image is normalized using Z-Score normalization. Then, Loess curves of the average value μ and the standard deviation σ changing with time are established respectively.

[0009] It should be noted that in the step S3, the spatio-temporally normalized average value is obtained according to the date on the Loess curve corresponding to the average value μ. The spatio-temporally normalized standard deviation is obtained according to the date on the Loess curve corresponding to the standard deviation σ. The inverse Z-Score normalization is used to obtain the single images corresponding to the spatio-temporally normalized corrected vegetation index NDVI, the two-band enhanced vegetation index EVI2, and the water index NDWI; the single images corresponding to the Loess-Z-Score spatio-temporally normalized corrected original vegetation index NDVI, the two-band enhanced vegetation index EVI2, and the water index NDWI are used as the spatio-temporally normalized remote sensing data.

[0010] Optionally, the step S32 is specifically: the sample composed of all average values μ or the sample composed of all standard deviations σ is divided into multiple intervals, polynomial fitting is performed on the samples in the intervals, and this process is repeated continuously to obtain weighted regression curves in different intervals. Finally, the centers of these weighted regression curves are connected together to synthesize a complete Loess curve.

[0011] Specifically, in the step S4, the database is randomly divided into a training set and a validation set at a ratio of 7:3. The XGBoost model is trained using the training set. Then, the training set and the validation set are input into the trained XGBoost model to obtain the predicted soil moisture values in the tallgrass prairie area. The correlation coefficient R, root mean square error RMSE, bias, and unbiased root mean square error ubRMSE between the predicted soil moisture values in the tallgrass prairie area and the measured soil moisture data in the tallgrass prairie area are calculated for accuracy evaluation.

[0012] Specifically, in the step S4, SHAP values are used to analyze the influence of the environmental meteorological data, topographic data, and remotely sensed data after spatio-temporal normalization in the tallgrass prairie area on the measured soil moisture data in the tallgrass prairie area in the trained XGBoost model. The importance of the influence is represented by the magnitude of the SHAP value. A positive value indicates that the feature contributes to the measured soil moisture data in the tallgrass prairie area. For positive values, the larger the SHAP value, the greater the contribution to the measured soil moisture data in the tallgrass prairie area. A negative value indicates that the feature has a negative contribution to the measured soil moisture data in the tallgrass prairie area. For negative values, the smaller the SHAP value, the greater the negative contribution to the measured soil moisture data in the tallgrass prairie area.

[0013] The beneficial effects of the present invention are as follows: In the soil moisture inversion method based on the normalization processing of multi-source remote sensing data, by adopting the idea of Z-Score standardization combined with Loess curve regression, the obvious differences in the spectral wavelength range and spectral response function curve between different sensor images caused by the spectral response between different optical sensors, the internal differences of the sensors, as well as the differences in imaging time and environment are eliminated. Combining the characteristics of the XGBoost model that can capture and learn complex patterns and non-linear relationships in the data, by inputting topographic, environmental climate, remotely sensed data after spatio-temporal normalization, and measured soil moisture data in the tallgrass prairie area into the XGBoost model for training, predicted soil moisture data with high accuracy, high spatio-temporal resolution, and significant spatial heterogeneity can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is the basic framework of the soil moisture inversion method based on the normalization processing of multi-source remote sensing data in an embodiment of the present invention; Figure 2 is the flowchart of the soil moisture inversion method based on the normalization processing of multi-source remote sensing data in an embodiment of the present invention; Figure 3 is the Loess-Z-Score spatio-temporal normalization preprocessing process taking NDVI as an example in an embodiment of the present invention; Figure 4Schematic diagram for evaluating the modeling accuracy after Loess-Z-Score spatio-temporal normalization preprocessing in an embodiment of the present invention; Figure 5 Schematic diagram for evaluating the modeling accuracy without Loess-Z-Score spatio-temporal normalization preprocessing in an embodiment of the present invention; Figure 6 Schematic diagram for evaluating the hierarchical modeling accuracy after Loess-Z-Score spatio-temporal normalization preprocessing in an embodiment of the present invention; Figure 7 Schematic diagram for evaluating the hierarchical modeling accuracy without Loess-Z-Score spatio-temporal normalization preprocessing in an embodiment of the present invention; Figure 8 Landsat 8 OLI true color image in an embodiment of the present invention; Figure 9 Comparison example of soil moisture distribution maps without and with Loess-Z-Score spatio-temporal normalization preprocessing at a depth of 10 cm in an embodiment of the present invention; Figure 10 Comparison example of soil moisture distribution maps without and with Loess-Z-Score spatio-temporal normalization preprocessing at a depth of 30 cm in an embodiment of the present invention; Figure 11 SHAP analysis diagram of modeling without Loess-Z-Score spatio-temporal normalization preprocessing in an embodiment of the present invention; Figure 12 SHAP analysis diagram of modeling with Loess-Z-Score spatio-temporal normalization preprocessing in an embodiment of the present invention. Detailed implementation manners

[0015] The following further describes the detailed implementation manners of the present invention with reference to the accompanying drawings. It should be noted here that the description of these implementation manners is used to help understand the present invention, but does not constitute a limitation to the present invention. In addition, the technical features involved in the various implementation manners of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0016] As Figures 1-12 shown, a soil moisture inversion method based on multi-source remote sensing data normalization processing includes the following steps: S1: Obtain the environmental meteorological data of the tallgrass prairie area, the terrain data of the tallgrass prairie area, and the measured soil moisture data of the tallgrass prairie area; In this embodiment, the environmental meteorological data of the tallgrass prairie area includes the air temperature, air relative humidity, wind speed, wind direction, maximum wind speed, solar radiation, soil temperature, vegetation evapotranspiration, rainfall data, and digital elevation model (DEM) data corresponding to the tallgrass prairie area; The DEM data is analyzed by Arcgis 10.8.1 to obtain elevation, slope, aspect, and topographic wetness index (TWI), and the spatial resolution is unified to 30 meters; At the same time, download the measured soil moisture data of the tallgrass prairie area for subsequent model accuracy verification analysis; S2: Obtain the multi-source remote sensing data corresponding to the tallgrass prairie area; S3: Perform spatio-temporal normalization on the multi-source remote sensing data corresponding to the tallgrass prairie area through Z-Score standardization, Loess curve, and inverse Z-Score standardization to obtain the remotely sensed data after spatio-temporal normalization correction; S4: Match and establish a database with the environmental meteorological data of the tallgrass prairie area, the terrain data of the tallgrass prairie area, the remotely sensed data after spatio-temporal normalization, and the measured soil moisture data of the tallgrass prairie area, and predict the soil moisture of the tallgrass prairie area through the XGBoost model.

[0017] In the soil moisture inversion method based on the normalization processing of multi-source remote sensing data, by adopting the idea of combining Z-Score standardization with Loess curve regression, the obvious differences in the spectral wavelength range and spectral response function curve between different sensor images caused by the spectral response between different optical sensors, the internal differences of the sensors, as well as the differences in imaging time and environment are eliminated. Combining the characteristics that the XGBoost model can capture and learn the complex patterns and nonlinear relationships in the data, by inputting the terrain, environmental climate, remotely sensed data after spatio-temporal normalization, and the measured soil moisture data of the tallgrass prairie area into the XGBoost model for training, the predicted soil moisture data of the tallgrass prairie area with high precision, high spatio-temporal resolution, and significant spatial heterogeneity can be obtained.

[0018] It should be noted that in step S2, according to the acquisition time of the measured soil moisture data of the tallgrass prairie area, obtain the corresponding remote sensing data of the tallgrass prairie area on the corresponding date; Calculate the vegetation index NDVI, the two-band enhanced vegetation index EVI2, and the water body index NDWI of each remote sensing data. Vegetation index NDVI = (NIR - RED) / (NIR + RED); Two-band enhanced vegetation index EVI2 = 2.5*(NIR - RED) / (NIR + 2.4*RED + 1); Water body index NDWI = (GREEN - NIR) / (GREEN + NIR).

[0019] Preferably, the multi-source remote sensing data corresponding to the tallgrass prairie region include Landsat-5, Landsat-7, Landsat-8, Sentinel-2, and Planetscope multi-band optical satellite image data. In this embodiment, the multi-band optical satellite image data of Landsat-5, Landsat-7, Landsat-8, Sentinel-2, and Planetscope are specifically downloaded from the USGS, Copernicus Data Center, and Planet Labs website for Landsat-5, Landsat-7, Landsat-8, Sentinel-2, and Planetscope multi-band optical satellite images. After obtaining the remote sensing data, images with cloud cover less than 30% are selected and resampled to a spatial resolution of 30 meters. Then, these remote sensing data are preprocessed, and the preprocessing includes radiometric calibration and atmospheric correction to obtain the GREEN data, RED data, and NIR data corresponding to each remote sensing data. The preprocessing process is completed on the ENVI5.6 software.

[0020] Optionally, in the step S3, calculate the mean μ and standard deviation σ of all pixel values of the single-image corresponding to the vegetation index NDVI, the single-image corresponding to the two-band enhanced vegetation index EVI2, and the single-image corresponding to the water index NDWI of the multi-source remote sensing data corresponding to the tallgrass prairie region in the same year, and perform normalization processing on each image using Z-Score normalization; specifically, the calculation formula of Z-Score normalization is ; where X is the pixel value of the original single image, and Z is the intermediate value after Z-Score normalization, that is, the pixel value of the image after Z-Score normalization; in this embodiment, the NDVI single-image, EVI2 single-image, and NDWI single-image corresponding to the multi-source remote sensing data corresponding to the tallgrass prairie region can be calculated through the vegetation index NDVI, the two-band enhanced vegetation index EVI2, and the water index NDWI; then, the Loess curves (the x-axis is the year, month, and day, and the y-axis is the pixel value) of the mean μ and standard deviation σ changing with time are established respectively. Specifically, if there are remote sensing data from multiple sensors on the same day, the mean μ and standard deviation σ of the remote sensing data from all sensors are then averaged respectively as the mean μ and standard deviation σ of that day.

[0021] Specifically, in the step S3, as Figure 3 shown, obtain the spatio-temporally normalized mean according to the date on the Loess curve corresponding to the mean μ, and obtain the spatio-temporally normalized standard deviation ; Using inverse Z-Score normalization to obtain single images corresponding to the vegetation index NDVI after spatio-temporal normalization and correction, single images corresponding to the two-band enhanced vegetation index EVI2, and single images corresponding to the water body index NDWI; Taking the single images corresponding to the vegetation index NDVI after spatio-temporal normalization and correction, single images corresponding to the two-band enhanced vegetation index EVI2, and single images corresponding to the water body index NDWI as remotely sensed data after spatio-temporal normalization; Specifically, the calculation formula for inverse Z-Score normalization is , where is the image pixel value obtained after Loess-Z-Score correction, that is, the remotely sensed data after spatio-temporal normalization.

[0022] It should be noted that the step S32 is specifically: dividing the sample composed of all means μ or the sample composed of all standard deviations σ into multiple intervals, performing polynomial fitting on the samples in the intervals, continuously repeating this process to obtain weighted regression curves in different intervals, and finally connecting the centers of these weighted regression curves together to synthesize a complete Loess curve.

[0023] Loess is a non-parametric method for local regression analysis, and its specific process is as follows: S321: Determine the number and positions of the fitting points; S322: With the fitting points as the center, determine k closest points; S323: Calculate the weights of these k closest points through the weight function; S324: Perform polynomial fitting (linear or quadratic) through weighted linear regression; S325: Repeat the above steps for all fitting points.

[0024] For the fitting point , first determine its neighborhood range, which is usually controlled by a smoothing parameter defined by the user (using span or frac to control the neighborhood size, with a value range of 0 to 1. The smaller the smoothing parameter, the more the fitting depends on local data, and the stronger the curve volatility. The larger the smoothing parameter, the more significant the smoothing effect, but details trends may be ignored). The data points within the neighborhood of the fitting point are assigned weights, and the weight function is calculated using the tricube kernel function. The formula for the tricube kernel function is: , when , where is the weighted value of the fitting point ; is the distance from the farthest point within the neighborhood to the fitting point , which is determined by the smoothing parameter; the weight decays as the distance increases, and the weight of points outside the neighborhood is 0.

[0025] Within the neighborhood, the weighted least squares method is used to fit a low-order polynomial (usually linear or quadratic). Taking linear regression as an example, the model is: ; The goal is to minimize the weighted sum of squared residuals: ; By solving this goal, local parameter estimates and are obtained, and then the smoothed value of the fitted point is predicted. Among them, this smoothed value is the center of a weighted regression curve corresponding to a segment, that is, the relationship between the fitted point and y, which is a small straight line. The final Loess curve is formed by combining the weighted regression curves established at the centers of all weighted regression curves.

[0026] Repeat the above steps for all fitted points in the dataset, and connect the smoothed values corresponding to each fitted point to form a complete Loess curve.

[0027] Preferably, in the step S4, the database is randomly divided into a training set and a validation set in a ratio of 7:3. The XGBoost model is trained using the training set; then the training set and the validation set are input into the trained XGBoost model to obtain the predicted soil moisture values in the tallgrass prairie area, and the correlation coefficient R, root mean square error RMSE, bias, and unbiased root mean square error ubRMSE between the predicted soil moisture values in the tallgrass prairie area and the measured soil moisture data in the tallgrass prairie area are calculated for accuracy assessment. Then the trained XGBoost model is applied to draw soil moisture distribution maps at different depths with a spatial resolution of 30 meters.

[0028] During the process of training the XGBoost model, the environmental meteorological data in the tallgrass prairie area, the topographic data in the tallgrass prairie area, and the remotely sensed data after spatio-temporal normalization are used as the input of the XGBoost model, and the measured soil moisture data in the tallgrass prairie area is used as the output of the XGBoost model. In this embodiment, the correlation coefficient ; The root mean square error ; The bias ; The unbiased root mean square error ; Among them is the average operator, is the measured soil moisture data in the tallgrass prairie area, is the predicted soil moisture value in the tallgrass prairie area, is the standard deviation of the predicted soil moisture value in the tallgrass prairie area and is the standard deviation of the measured soil moisture data in the tallgrass prairie area.

[0029] The correlation coefficient R, root mean square error RMSE, bias, and unbiased root mean square error ubRMSE are all common metrics for evaluating the performance of regression models. They can provide information on the goodness of fit of the model from different perspectives. The correlation coefficient R measures the statistical index of the closeness of the correlation between variables, with a value range from 0 to 1. The higher the value, the better the model fits the data. The root mean square error RMSE measures the standard deviation of the difference between the model's predicted values and the actual observed values. Its value is always non - negative. The closer the value is to 0, the smaller the prediction error of the model, and the better the model's performance. Bias is the average error between the predicted value and the true value, that is, the average of all prediction errors. The smaller the absolute value of bias, the closer the model's prediction results are to the true value as a whole, the smaller the bias of the model, and the relatively better the performance. The unbiased root mean square error ubRMSE is the root mean square error after considering bias correction. It eliminates the influence of the model's systematic bias on error evaluation and can more accurately reflect the random error part of the model prediction. Similar to the root mean square error RMSE, the smaller the value of the unbiased root mean square error ubRMSE, the higher the accuracy of the model prediction and the more reliable the prediction results.

[0030] Optionally, in step S4, SHAP values are used to analyze the influence of the environmental meteorological data in the tall - grass prairie area, the topographic data in the tall - grass prairie area, and the remotely sensed data after spatio - temporal normalization on the measured soil moisture data in the tall - grass prairie area. The SHAP value framework regards the environmental meteorological data in the tall - grass prairie area, the topographic data in the tall - grass prairie area, and the remotely sensed data after spatio - temporal normalization as members of the XGBoost model, and measures their influence on the XGBoost model prediction by calculating the average marginal contribution of the environmental meteorological data in the tall - grass prairie area, the topographic data in the tall - grass prairie area, and the remotely sensed data after spatio - temporal normalization. Among them, the importance of the influence is represented by the magnitude of the SHAP value. A positive value indicates that the feature contributes to the measured soil moisture data in the tall - grass prairie area. For positive values, the larger the SHAP value, the greater the contribution to the measured soil moisture data in the tall - grass prairie area. A negative value indicates that the feature has a negative contribution to the measured soil moisture data in the tall - grass prairie area. For negative values, the smaller the SHAP value, the greater the negative contribution to the measured soil moisture data in the tall - grass prairie area.

[0031] In the calculation of SHAP values, all feature combinations are traversed, and the new contribution after the feature is added to the combination is calculated, where the features include the environmental meteorological data in the tall - grass prairie area, the topographic data in the tall - grass prairie area, and the remotely sensed data after spatio - temporal normalization. Then, the new contributions of all combinations are weighted and averaged, where the weights are determined according to the size of the combination.

[0032] Among them, the SHAP value of feature i can be calculated as: ; Denote the SHAP value of feature i, that is, the contribution degree of feature i; N represents all combinations of feature i; Denote the revenue of all combinations S, The combination represents the revenue after removing feature i, Denote the marginal contribution of feature i in the coalition, that is, the marginal revenue of feature i; Denote the weighting factor, that is, the weight, ; Finally, the contribution degree of feature i can be sorted out as: .

[0033] In this solution, the Konza tallgrass prairie with an area of 34.87 km² in the Flint Hills of Manhattan, Kansas, USA is taken as an example. The environmental meteorological data, topographic data, and measured soil moisture data of the tallgrass prairie area in this example are provided by the Konza Prairie Biological Station. The main data include: (1) The environmental meteorological data of the tallgrass prairie area recorded daily, including air temperature (°C), relative air humidity (%), wind speed (m / s), wind direction (angle), maximum wind speed (m / s, recorded every ten seconds), solar radiation (joules / m²), soil temperature (°C), the most recent rainfall (mm), the time interval since the last rainfall (days), and the daily evapotranspiration of prairie grass (mm / d), which are collected by the meteorological micro-recorder at the headquarters of the Konza Prairie Biological Station. (2) The topographic data of the tallgrass prairie area is a 2-meter resolution digital elevation model (DEM) generated from LiDAR DEM data collected according to USGS standard specifications. The elevation, slope, aspect, and topographic wetness index (TWI) are obtained through data analysis using Arcgis 10.8.1, and the spatial resolution is unified to 30 meters. (3) The measured soil moisture data of the tallgrass prairie area are the daily soil moisture data (m³ / m³) at depths of 10 and 30 cm, measured using the neutron probe method. (4) The corresponding multi-source remote sensing data of the tallgrass prairie area are the multi-band optical satellite image data of Landsat-5, Landsat-7, Landsat-8, Sentinel-2, and Planetscope, which are downloaded from the USGS, Copernicus Data Center, and Planet Labs websites respectively. Radiometric calibration, atmospheric correction, and flat-field method processing are completed on ENVI5.6, and the vegetation index NDVI, the two-band enhanced vegetation index EVI2, and the water body index NDWI are calculated. (5) Loess-Z-Score spatio-temporal normalization preprocessing is performed on the multi-source multi-band optical sensor images.(6) Match the remotely sensed data, environmental meteorological data of the tallgrass prairie area, and topographic data of the tallgrass prairie area after spatio-temporal normalization with the measured soil moisture data of the tallgrass prairie area to establish a database. Match the multi-source remotely sensed data, environmental meteorological data of the tallgrass prairie area, and topographic data of the tallgrass prairie area corresponding to the original tallgrass prairie area without spatio-temporal normalization preprocessing with the measured soil moisture data of the tallgrass prairie area to establish a database (so there are two sets of databases, namely the database corresponding to the spatio-temporal normalization preprocessing and the database corresponding to the non-spatio-temporal normalization preprocessing). Finally, the sorted air temperature (T_air), air relative humidity (RHUM), wind speed (WSPEED), wind direction (WDIR), maximum wind speed (WMAX), solar radiation (SRAD), soil temperature (STEMP), the most recent rainfall (precipitation(mm)), the time interval since the last rainfall (rain_time(day)), daily evapotranspiration of grassland grass (ET), elevation (elevation), slope (slope), aspect (aspect), topographic wetness index (TWI), NDVI, EVI2, NDWI, and soil depth (depth) data are used as the input data for machine learning. The target variables are the daily soil moisture data (SM) at 10 and 30 cm depths.

[0034] Machine learning model establishment and accuracy evaluation: Hyperparameter tuning is crucial for the performance of machine learning models as it directly controls the behavior and effectiveness of the training algorithm. In this embodiment, a five-fold cross-validation and PSO particle swarm optimization method are used to automatically tune the optimal hyperparameters of the model. The PSO particle swarm optimization algorithm is used to tune hyperparameters based on the pyswarm library in the Python language. The particle optimization swarm is affected by multiple control parameters, namely the fitness criteria (e.g., RMSE, MSE, and R2), local coefficient (c1), global coefficient (c2), inertia coefficient (ω), and population / swarm size (s). The specific parameters used in the PSO particle swarm optimization algorithm in this embodiment are as follows: fitness criteria (R2), local coefficient (c1 = 1.49618), global coefficient (c2 = 1.49618), inertia coefficient (ω = 0.7298), and population / swarm size (s = 100).

[0035] Both sets of databases are divided into training sets and validation sets in a ratio of 7:3. The training sets are input into the XGBoost model for training, and the validation sets are used to verify the effectiveness of the model prediction results. R, RMSE, bias, and ubRMSE are used as the accuracy evaluation indicators of the model on the validation sets. The accuracy evaluation results of the predicted values and measured values of the two sets of databases predicted using the XGBoost model are as Figure 4 and Figure 5 shown.

[0036] As Figure 3 and 4 shown, using PSO-XGBoost for soil moisture prediction has high accuracy on both the training set and the validation set, but there is not much change in the accuracy of the two sets of databases on the validation set. In this embodiment, June 28, 2022 is taken as an example for soil moisture prediction, and the accuracy evaluation results are as Figure 6 and 7 shown. It can be seen that using PSO-XGBoost has the same high accuracy when predicting soil moisture at different depths and with different extraction methods, but there is not much change in the accuracy of the validation set for the two sets of databases when performing stratified accuracy evaluation.

[0037] Machine learning model to draw soil moisture distribution map: In this embodiment, the XGBoost models trained using the two sets of databases are respectively used to draw the soil moisture distribution map with a 30-meter spatial resolution for the data on June 28, 2022, and a comparative analysis is carried out. The results are as Figures 8-10 shown. It can be found that the prediction effect of the XGBoost model trained using the database corresponding to Loess-Z-Score spatio-temporal normalization preprocessing is significantly better than that of the XGBoost model trained using the database without Loess-Z-Score spatio-temporal normalization preprocessing in some details. Combining with optical images, compared with the XGBoost model trained using the database without Loess-Z-Score spatio-temporal normalization preprocessing, for example, on the bare soil road surface, the XGBoost model trained using the database corresponding to Loess-Z-Score spatio-temporal normalization preprocessing can well reflect the distribution of soil moisture on the bare soil road surface; In some areas with severe burning conditions (such as Figure 8 ② in, the 2D watershed is significantly more affected by burning than the 1D watershed, with less vegetation cover and more exposed soil), the model after spatio-temporal normalization can respond more sensitively to these areas and more accurately predict the differences in these areas. Moreover, the closer the soil moisture is to the surface depth, the more obvious the difference before and after correction in areas with higher slopes and relatively severe burning conditions. This shows that adopting the Loess-Z-Score spatio-temporal normalization preprocessing method to preprocess multi-source optical remote sensing data can effectively help the XGBoost model eliminate the problem of weakened spatial heterogeneity caused by the true surface spatial distribution information masked by time changes in the optical images presented by different sensors.

[0038] SHAP value analysis: In this embodiment, SHAP value analysis is performed on the XGBoost models trained by two sets of databases. The overall analysis results of SHAP value contributions are as Figure 11 and 12 shown. Due to the differences in the spectral responses of different optical sensors, the values of NDVI, EVI2, and NDWI images are uneven. The rankings of NDVI and EVI2 show an obvious upward trend in the corrected model, especially NDVI. This indicates that the Loess-Z-Score spatio-temporal normalization preprocessing method can enable XGBoost to more effectively learn the differences in NDVI, EVI2, and NDWI at different time points, thus better explaining the mechanism by which NDVI, EVI2, and NDWI affect soil moisture.

[0039] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, without departing from the principle and spirit of the present invention, various changes, modifications, substitutions, and variations to these embodiments still fall within the protection scope of the present invention.

Claims

1. A soil moisture inversion method based on normalization processing of multi-source remote sensing data, characterized in that: The following steps are involved: S1: Obtain environmental meteorological data, topographic data and measured soil moisture data of the tallgrass prairie area; S2: Obtain multi-source remote sensing data corresponding to the tallgrass prairie area; S3: Performing spatiotemporal normalization on the multi-source remote sensing data corresponding to the tallgrass prairie area through Z-Score normalization, Loess curve and inverse Z-Score normalization to obtain spatiotemporal normalized remote sensing data; S4: The environmental meteorological data of the tallgrass prairie area, the terrain data of the tallgrass prairie area, the remote sensing data after time and space normalization, and the measured soil moisture data of the tallgrass prairie area are matched to establish a database, and the soil moisture of the tallgrass prairie area is predicted through the XGBoost model.

2. The soil moisture inversion method based on normalization processing of multi-source remote sensing data according to claim 1 is characterized in that: In step S2, remote sensing data corresponding to the tallgrass prairie area on the corresponding date is obtained according to the acquisition time of the measured soil moisture data of the tallgrass prairie area; and the vegetation index NDVI, the dual-band enhanced vegetation index EVI2 and the water body index NDWI of each remote sensing data are calculated.

3. The soil moisture inversion method based on normalization processing of multi-source remote sensing data according to claim 2 is characterized in that: The multi-source remote sensing data corresponding to the tallgrass prairie area include Landsat-5, Landsat-7, Landsat-8, Sentinel-2 and Planetscope multi-band optical satellite image data. After obtaining the remote sensing data, images with less than 30% cloud cover were screened out and resampled to a spatial resolution of 30 meters.

4. The soil moisture inversion method based on normalization processing of multi-source remote sensing data according to claim 3 is characterized in that: In the step S3, the average value μ and standard deviation σ of all pixel values ​​of a single image corresponding to the vegetation index NDVI, a single image corresponding to the dual-band enhanced vegetation index EVI2, and a single image corresponding to the water body index NDWI calculated from the multi-source remote sensing data corresponding to the original tallgrass prairie area in the same year are calculated, and each image is standardized by using Z-Score standardization; Then, the Loess curves of the mean value μ and standard deviation σ changing with time are established respectively.

5. The soil moisture inversion method based on normalization processing of multi-source remote sensing data according to claim 4 is characterized in that: In step S3, the time-space normalized average value is obtained according to the Loess curve corresponding to the average value μ of the date. , according to the Loess curve corresponding to the standard deviation σ of the date, the standard deviation after time and space normalization is obtained ; Use inverse Z-Score standardization to obtain the single image corresponding to the vegetation index NDVI after spatiotemporal normalization correction, the single image corresponding to the dual-band enhanced vegetation index EVI2, and the single image corresponding to the water body index NDWI; Use the single image corresponding to the original vegetation index NDVI corrected by Loess-Z-Score spatiotemporal normalization, the single image corresponding to the dual-band enhanced vegetation index EVI2, and the single image corresponding to the water body index NDWI as the remote sensing data after spatiotemporal normalization.

6. The soil moisture inversion method based on normalization processing of multi-source remote sensing data according to claim 5 is characterized in that: The step S32 is specifically as follows: dividing all samples composed of the mean value μ or all samples composed of the standard deviation σ into multiple intervals, performing polynomial fitting on the samples in the intervals, and continuously repeating this process to obtain weighted regression curves in different intervals, and finally connecting the centers of these weighted regression curves together to synthesize a complete Loess curve.

7. The soil moisture inversion method based on normalization processing of multi-source remote sensing data according to claim 6 is characterized in that: In step S4, the database is randomly divided into a training set and a validation set in a ratio of 7:3, wherein the training set is used to train the XGBoost model; then the training set and the validation set are input into the trained XGBoost model to obtain the predicted soil moisture value of the tallgrass prairie area, and the correlation coefficient R, root mean square error RMSE, bias and unbiased root mean square error ubRMSE between the predicted soil moisture value of the tallgrass prairie area and the measured soil moisture data of the tallgrass prairie area are calculated for accuracy assessment.

8. The soil moisture inversion method based on normalization processing of multi-source remote sensing data according to claim 7 is characterized in that: In step S4, the SHAP value is used to analyze the influence of the environmental meteorological data of the tall grass prairie area, the terrain data of the tall grass prairie area and the remote sensing data after time and space normalization in the trained XGBoost model on the measured soil moisture data of the tall grass prairie area; the importance of the influence is represented by the size of the SHAP value, a positive value indicates that the feature contributes to the measured soil moisture data of the tall grass prairie area, for a positive value, a larger SHAP value indicates a greater contribution to the measured soil moisture data of the tall grass prairie area, a negative value indicates that the feature has a negative contribution to the measured soil moisture data of the tall grass prairie area, and a smaller SHAP value indicates a greater negative contribution to the measured soil moisture data of the tall grass prairie area.

Images