Surface soil humidity simulation method and system considering soil freeze-thaw state

Abstract

The invention discloses a surface soil humidity simulation method and system considering a soil freeze-thaw state, and belongs to the technical field of soil humidity simulation. The method comprises the following steps: acquiring land surface environment data and actually measured soil humidity data of a plateau area; preprocessing the data; dynamically judging the freeze-thaw state of the soil, and generating a machine learning model training sample; training and comparing a plurality of machine learning models considering the soil freeze-thaw state, and selecting an optimal model; and reconstructing the plateau area surface soil humidity based on the optimal model to obtain a time-space continuous simulation result. By considering the soil freezing and thawing state, the overall simulation precision of the surface soil humidity of the plateau region is improved, and the humidity simulation precision of the soil freezing period is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of soil moisture simulation, and more particularly to a surface soil moisture simulation method and system considering soil freeze-thaw state. Background Art

[0002] Surface soil moisture, as a key parameter in the water cycle and energy exchange, has important value in the research field of earth sciences. In terms of surface evaporation, it affects the energy transfer process between the earth and the atmosphere by influencing the surface albedo and plant transpiration. In addition, soil moisture plays a key role in fields such as flood detection and early warning, drought monitoring, carbon cycle, and climate change. Therefore, studying surface soil moisture is of great significance for climate disasters, climatology, and the earth's energy and material cycle.

[0003] There are various ways to obtain soil moisture, and the main methods include in-situ measurement at stations, remote sensing inversion, and reanalysis data. Due to the special geographical environment and harsh climate conditions in the Qinghai-Tibet Plateau, it is difficult to establish observation stations in this area. Therefore, there has been a lack of long-term monitoring of the key surface parameter soil moisture in this area. Remote sensing inversion, especially microwave remote sensing, has become a powerful tool for obtaining soil moisture at the regional and even global scales due to its inversion characteristics. Therefore, a large amount of work has been done in the field of microwave remote sensing inversion of soil moisture. According to research, in the Qinghai-Tibet Plateau region, the accuracy of SMAP is the best, followed by AMSR2 (JAXA), FY3B, AMSR2 (LPRM), and SMOS-IC. However, the time series of SMOS data and SMAP data based on L-band inversion of soil moisture are relatively short and cannot support long-term soil moisture change research. ESA-CCI data is a commonly used remote sensing soil moisture product, which integrates soil moisture datasets of various sensors into three consistent long-term products through a fusion method. Although previous studies have evaluated that the performance of the ESACCI product is good in most regions of the world, due to the algorithm reasons of the original data, there are a large number of missing spatio-temporal sequences in the Qinghai-Tibet Plateau region, especially obvious during the non-growing season in the Qinghai-Tibet Plateau region, and the data is discontinuous in spatio-temporal distribution.

[0004] Low-frequency microwaves are affected by the dielectric properties of soil, and changes in soil moisture can alter the soil dielectric properties. Microwave remote sensing is precisely based on this characteristic to invert the changes in soil moisture. During the soil freezing period, part of the liquid water in the soil is converted into ice, resulting in a decrease in the content of liquid water in the soil. At the same time, the dielectric constant obtained by the remote sensing sensor is close to that of dry soil, leading to the soil moisture value obtained by inversion being close to the soil moisture value of dry soil. Existing research has shown that in soil freezing areas, neither active microwave remote sensing nor passive microwave remote sensing can well obtain soil moisture. It can be seen that the freeze-thaw process of the soil causes congenital defects in remote sensing inversion of soil moisture products.

[0005] Currently, some institutions have released reanalysis soil moisture products with a wide coverage and relatively high spatio-temporal resolution. However, due to biases in the assimilated variables, these products have obvious uncertainties. On the other hand, due to the particularity of the Qinghai-Tibet Plateau region, the performance of reanalysis soil moisture products in the Qinghai-Tibet Plateau region is not as excellent as that in other regions. The performance of GLDAS in Nagqu and Maqu regions has been evaluated in the existing technology, and it is found that in the surface soil part, although GLDAS can capture the temporal changes of soil moisture in Maqu and Nagqu regions, it significantly underestimates the soil moisture value. Lin et al (2023) evaluated the performance of ESACCI, GLDAS, ERA5, and CSSPv2 products from 2007 to 2019 at Maqu, Nagqu, Ali, Pari, and Heihe stations, and found that among these data, the highest average correlation coefficient is 0.61 and the lowest is 0.28, while the lowest root mean square error is 0.083 and the highest is 0.155 m 3 / m 3 . And Cheng et al (2019) found that the performance of the ERA5 reanalysis product in the Qinghai-Tibet Plateau region (ubRMSE = 0.067 m 3 / m 3 , R = 0.52) is inferior to that of the ESA-CCI-C remote sensing product (ubRMSE = 0.043 m 3 / m 3 , R = 0.66). In addition, the simulation results of soil moisture products only for China in snow-covered areas deviate greatly from the site measurement values. Generally speaking, the applicability of reanalysis soil moisture data in the Qinghai-Tibet Plateau region still has relatively large uncertainties.

[0006] In recent years, machine learning methods have been applied to the simulation and prediction of soil moisture. However, the impact of soil freezing has not been fully considered in the existing technology research. For the Qinghai-Tibet Plateau with extensive permafrost distribution, without targeted tests, uncertainties may occur if the freeze-thaw process is not considered.

[0007] Therefore, how to provide a method and system for simulating surface soil moisture considering the soil freeze-thaw state is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0008] In view of this, the present invention provides a method and system for simulating surface soil moisture considering the soil freeze-thaw state to solve the technical problems existing in the prior art.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A method for simulating surface soil moisture considering the soil freeze-thaw state includes:

[0011] Obtain various publicly released remote sensing and reanalysis surface environment data in the plateau region, as well as measured soil moisture data;

[0012] Preprocess the surface environment data;

[0013] Dynamically discriminate the freeze-thaw state of the soil, and based on the dynamic discrimination result, generate a training sample set of a machine learning model from the preprocessed surface environment data and the measured soil moisture data;

[0014] Based on the training sample set, complete the training and simulation of various machine learning models considering the soil freeze-thaw state, compare the simulation effects of different machine learnings, and select the optimal soil moisture inversion model;

[0015] Based on the optimal soil moisture inversion model, reconstruct the surface soil moisture in the plateau region to obtain a spatio-temporally continuous simulation result of the surface soil moisture in the plateau region.

[0016] Further, the reanalyzed surface environment data includes: surface brightness temperature, surface temperature, surface albedo, land cover type, normalized difference vegetation index NDVI, precipitation, clay proportion, sand proportion, and loam proportion.

[0017] Further, the preprocessing of the surface environment data includes: reprojection, resampling, maximum-minimum normalization, and data filling processing.

[0018] Further, the dynamic discrimination of the freeze-thaw state of the soil includes:

[0019] Based on the surface brightness temperature data, use the threshold index method to dynamically discriminate the freeze-thaw state of the soil, and the discrimination result includes the frozen state and the ablation state.

[0020] Further, respectively according to the frozen state and the ablation state, generating a training sample set of a machine learning model from the preprocessed surface environment data and the measured soil moisture data includes:

[0021] Calculate the 36.5 GHz vertically polarized data, and the formula is:

[0022]

[0023] In the formula, represents the vertically polarized brightness temperature of AMSR-2 data at 36.5 GHz frequency, represents the vertically polarized brightness temperature of AMSR-E data at 36.5 GHz frequency, and v represents vertical polarization;

[0024] Calculate the standard deviation index of the horizontally polarized brightness temperature data:

[0025]

[0026] In the formula, n is the number of channels at different frequencies, is the brightness temperature under horizontal polarization at frequency f i at that time, is the average value of horizontal polarization over all frequencies, and SDI is the standard deviation index of the horizontally polarized brightness temperature data;

[0027] When the standard deviation index SDI of the calculated horizontally polarized brightness temperature data and the 36.5 GHz vertically polarized data are both less than the preset threshold, it is regarded as the frozen state, otherwise it is the non-frozen state;

[0028] Generate a training sample set for the machine learning model from the preprocessed plateau surface environment data and the measured soil moisture data.

[0029] Furthermore, based on the training sample set, complete the training and simulation of multiple machine learning models considering the soil freeze-thaw state, compare the simulation effects of different machine learning methods, and select the optimal soil moisture inversion model, including:

[0030] Use the same plateau surface environment data as input features and the same observed soil moisture data as the training target to train the Random Forest (RF) model, Gradient Boosting Decision Tree (GBDT) model, and Extreme Gradient Boosting (XGBoost) model respectively;

[0031] Conduct a comparative evaluation of the Random Forest (RF) model, Gradient Boosting Decision Tree (GBDT) model, and Extreme Gradient Boosting (XGBoost) model, and select the optimal model as the soil moisture inversion model.

[0032] Furthermore, the comparative evaluation of the simulation effects of the Random Forest (RF) model, Gradient Boosting Decision Tree (GBDT) model, and Extreme Gradient Boosting (XGBoost) model includes:

[0033] The evaluation index adopted is the correlation coefficient R, and the formula is:

[0034]

[0035] In the formula, and represent the mean values, where x i is the soil moisture value predicted by any one of the random forest RF model, the gradient boosting decision tree GBDT model, and the extreme gradient boosting tree XGBoost model, represents the mean value of the predicted values, and y i represents the soil moisture value observed at the site, represents the mean value of the observed values;

[0036] Calculate the root mean square error RMSE, and the formula is:

[0037]

[0038] Calculate the mean absolute error MAE, and the formula is:

[0039]

[0040] In the formula, N is the number of valid data points.

[0041] A surface soil moisture simulation system considering the soil freeze-thaw state, comprising:

[0042] Data acquisition module: acquiring various publicly released remote sensing and reanalysis surface environment data in the plateau region, as well as measured soil moisture data;

[0043] Data processing module: preprocessing the surface environment data;

[0044] Sample generation module: dynamically discriminating the freeze-thaw state of the soil, and based on the dynamic discrimination result, generating a training sample set of the preprocessed surface environment data and measured soil moisture data;

[0045] Model construction module: based on the training sample set, completing the training and simulation of various machine learning models considering the soil freeze-thaw state, comparing the simulation effects of different machine learning, and selecting the optimal soil moisture inversion model;

[0046] Simulation reconstruction module: based on the optimal soil moisture inversion model, performing surface soil moisture reconstruction in the plateau region to obtain a spatio-temporally continuous surface soil moisture simulation result for the plateau region.

[0047] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for simulating surface soil moisture considering the soil freeze-thaw state. First, the soil freezing state is identified, and then models considering freezing and non-freezing states are respectively constructed based on various machine learning methods. On this basis, the error between the soil moisture simulated by the optimal solution and the measured values at the stations is evaluated, and a comparison is made with existing soil moisture data products. By considering the soil freeze-thaw state, the present invention improves the simulation accuracy of soil moisture. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0049] Figure 1 is a schematic flow chart of the method of the present invention;

[0050] Figure 2 is a schematic structural diagram of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the 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 of 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.

[0052] See Figure 1 , the embodiments of the present invention disclose a method for simulating surface soil moisture considering the soil freeze-thaw state, including:

[0053] Obtain various publicly released remote sensing and reanalysis surface environment data in the plateau region, as well as measured soil moisture data;

[0054] Preprocess the surface environment data;

[0055] Dynamically discriminate the freeze-thaw state of the soil, and based on the dynamic discrimination result, generate a training sample set of the machine learning model from the preprocessed surface environment data and the measured soil moisture data;

[0056] Based on the training sample set, complete the training and simulation of various machine learning models considering the soil freeze-thaw state, compare the simulation effects of different machine learning methods, and select the optimal soil moisture inversion model;

[0057] Based on the optimal soil moisture inversion model, the surface soil moisture in the plateau region is reconstructed to obtain the spatio-temporal continuous simulation results of the surface soil moisture in the plateau region.

[0058] Specifically, the surface environment data mentioned above are all input data for modeling training; the measured soil moisture data are used for training targets and simulation effect verification.

[0059] Measured soil data information:

[0060] In this study, the measured surface soil moisture data at stations are used for model training and verification. The measured data mainly come from the Observation Network of Soil Moisture and Soil Temperature in the Qinghai-Tibet Plateau (Su et al., 2013) and the dataset of the Pali Soil Temperature and Humidity Observation Network (Chen et al., 2017).

[0061] Specifically, taking the simulation of the surface soil moisture in the Qinghai-Tibet Plateau as an example, high-quality gridded soil moisture products can provide necessary basic guarantees for earth system science research. In the Qinghai-Tibet Plateau region, current research has carried out soil moisture simulation, but most studies only consider the growing season and insufficiently consider the impact of soil freeze-thaw state.

[0062] In a specific embodiment, the plateau surface soil environment data includes: surface brightness temperature, surface temperature, surface albedo, land cover type, normalized difference vegetation index NDVI, precipitation, clay proportion, sand proportion, loam proportion.

[0063] Specifically, the auxiliary environmental variables (plateau surface environment data) required for soil moisture simulation in this embodiment include the brightness temperature data of AMSR2, the surface temperature data, normalized vegetation index, land cover type, surface albedo product data sourced from MODIS, GPM precipitation data, and the soil texture type data (sand content Sand, silt content Silt, clay content Clay) of the Harmonized World Soil Database (HWSD).

[0064] Specifically, in order to further evaluate the soil moisture inversion scheme considering the freeze-thaw state, the best inversion results in this embodiment are compared with three publicly available datasets, namely: 1) the 1km daily-scale dataset of soil moisture in China based on station observations (SMCI1.0, Soil Moisture of China by in situ data, version 1.0); 2) the ERA5-LAND dataset; 3) the ESA-CCI dataset.

[0065] The SMCI1.0 dataset is obtained through machine learning, with the observed soil moisture at 10 layers from 1648 stations provided by the China Meteorological Administration as the benchmark, and using ERA5_Land meteorological forcing data, leaf area index (LAI), land cover types (Landtypes), terrain (DEM), and soil properties (Soilproperties) as covariates. The dataset provides soil moisture values at a daily scale, with a 10 - cm interval and 10 - layer depths, and the time range is from 2000 to 2020. Since SMCI1.0 is based on in - situ observed soil moisture, it can serve as an effective supplement to existing model - and satellite - based datasets (Shangguan et al, 2023). In this embodiment, the surface layer (0 - 10 cm) data from 2017 to 2018 of this dataset is used for comparative verification.

[0066] The ERA5 - Land product is produced within the framework of the Copernicus Climate Change Service (C3S), by driving a land surface model with the atmospheric variables of the land fields simulated by ERA5 ( - Sabater et al, 2021). ERA5 - Land provides data on surface energy fluxes and soil temperature and humidity above 289 cm of the surface globally from 1950 to the present. Its minimum time resolution is 1 h, and the spatial resolution is 0.1°×0.1°. Currently, the ERA5 - Land product has been widely applied in fields such as drought detection, soil moisture reconstruction, and future climate change (Lin et al, 2023, O. and Orth, 2021, Li et al, 2022). In this embodiment, its surface layer (0 - 7 cm) data is used for comparison with other soil moisture data, and the time range is from 2017 to 2018.

[0067] The ESA-CCI soil moisture product is released by the European Space Agency's Climate Change Initiative project (http: / / www.esasoilmosture-cci.org / ). It is produced based on active / passive microwave sensors and has the advantages of long time series and global scale coverage. The ESA-CCI product provides daily surface soil moisture (5 cm) information at a spatial resolution of 0.25°. This product has three types: active product (ESA-CCI-A), passive product (ESA-CCI-P), and combined product (ESA-CCI-C). Previous studies have shown that the ESA-CCI product has higher accuracy compared to other remote sensing products (Yu et al., 2023). Currently, the ESA-CCI product has a relatively wide range of applications in climate change, surface interaction, hydrological applications, etc. (Dorigo et al, 2017). In this embodiment, the data of ESA-CCI-C v0.6 from 2017 to 2018 was used for comparison and verification.

[0068] In a specific embodiment, the simulation of soil moisture considering the freeze-thaw state mainly includes four parts: data preparation, constructing a soil moisture model, selecting the optimal model, and comparing and evaluating with existing products. The data preparation part mainly preprocesses each original data to obtain the variable data required for constructing and validating the model. When constructing the soil moisture model, first, the soil freezing state is judged to form two schemes considering and not considering the freezing effect, and then three machine learning methods are respectively used for modeling. In total, six soil moisture simulation schemes are obtained. Subsequently, their effects are evaluated in terms of space (random sampling points) and time (test period) with reference to the measured values at the stations. Finally, based on the optimal model, the soil surface moisture of the Tibetan Plateau is simulated and reconstructed, and compared with existing common and publicly released products to obtain the results of the simulation of TP soil moisture considering the freeze-thaw state.

[0069] In a specific embodiment, the preprocessing of the plateau surface soil environment data includes: reprojection, resampling, maximum-minimum normalization, and data filling.

[0070] Specifically, in order to overcome the problem of inconsistent spatial resolution of environmental variable data, in this embodiment, all datasets are resampled to 0.1°. In addition, before model training, maximum-minimum normalization is performed on each environmental variable to ensure the consistency of the dimensions of all training data.

[0071] In a specific embodiment, the method includes generating a training sample set of a machine learning model based on the preprocessed plateau surface environment data and the measured soil moisture data according to the freezing state and the thawing state respectively; completing the training and simulation of multiple machine learning models considering the soil freezing and thawing states, comparing the simulation effects of different machine learning methods, and selecting the optimal soil moisture inversion model, including:

[0072] Using the same plateau surface soil environment data as input features and the same observed soil moisture data as the training target, training the Random Forest (RF) model, the Gradient Boosting Decision Tree (GBDT) model, and the Extreme Gradient Boosting (XGBoost) model respectively;

[0073] Comparatively evaluating the RF model, the GBDT model, and the XGBoost model, and selecting the optimal model as the soil moisture inversion model.

[0074] Specifically, the random forest is a classifier that uses multiple classification and regression trees and adopts the Bagging strategy. During the sampling process, each sample is classified according to the feature variables. After selecting n features (n < m) from the complete m feature sets, a decision tree is established using the data. After obtaining the forest, when a new input sample is input, each decision tree in the forest makes a judgment and regression respectively. Then, the average value of all the trees is taken as the final result. An important advantage of the random forest is that it does not require cross-validation, nor does it require an independent test set to obtain an unbiased estimate of the error. The error can be evaluated internally, that is, an unbiased estimate of the error can be established during the generation process. XGBoost is an ensemble learning method based on gradient boosting that adopts Boosting. Similar to RF, XGBoost combines a bunch of CART decision trees to generate a learner. The difference between RF and XGBoost is that each tree in the RF algorithm is trained in parallel, while the decision trees in XGBoost are not independent of each other. The first tree is built on the entire training set, and the second tree takes into account the residuals of the first tree. GBDT is another popular ensemble learning method based on boosting. It constructs multiple weak decision trees and finally combines them into a strong learner after multiple iterations, so it is robust to outliers and imbalanced data (Friedman, 2002). When constructing the three machine learning models, the same environmental variables are used as input features and the same soil moisture data are used as the training target to ensure a fair comparison of the effects of the three methods.

[0075] In a specific embodiment, the evaluation of the RF model, the GBDT model, and the XGBoost model includes:

[0076] The evaluation index adopted is the correlation coefficient R, and the formula is:

[0077]

[0078] In the formula, and represent the mean value. x i is the soil moisture value predicted by any one of the random forest RF model, the gradient boosting decision tree GBDT model, and the extreme gradient boosting tree XGBoost model. represents the mean value of the predicted values, and y i represents the soil moisture value observed at the station. represents the mean value of the observed values;

[0079] Calculate the root mean square error RMSE, and the formula is:

[0080]

[0081] Calculate the mean absolute error MAE, and the formula is:

[0082]

[0083] In the formula, N is the number of valid data points.

[0084] In a specific embodiment, the identification of the soil freeze-thaw state includes:

[0085] Calculate the 36.5 GHz vertical polarization data, and the formula is:

[0086]

[0087] In the formula, represents the vertical polarization brightness temperature of AMSR-2 data at 36.5 GHz frequency. represents the vertical polarization brightness temperature of AMSR-E data at 36.5 GHz frequency, and v represents vertical polarization.

[0088] Specifically, AMSR-E was launched in 2002 and carried on the American Aqua satellite.

[0089] AMSR-2 was launched in 2012 and carried on the Japanese GCOM-W1 satellite.

[0090] AMSR-2 has significantly improved in data quality and accuracy compared with AMSR-E and can provide higher-resolution observations.

[0091] Calculate the standard deviation index of the horizontal polarization brightness temperature data:

[0092]

[0093] In the formula, n is the number of channels with different frequencies, is the brightness temperature under horizontal polarization at frequency f i when, is the average value of horizontal polarization over all frequencies, and SDI is the standard deviation index of the horizontal polarization brightness temperature data;

[0094] When the standard deviation index SDI of the calculated horizontal polarization brightness temperature data and the 36.5 GHz vertical polarization data are both less than the preset threshold, it is regarded as the frozen state, otherwise it is the non-frozen state.

[0095] Specifically, in this embodiment, AMSR-2 data is used in the improved dual-index algorithm to calculate and set the threshold for judging the freeze-thaw state. At the same time, the algorithm of the present invention is different from the prior art. When calculating SDI, it is only calculated within the horizontal polarization band, and the systematic errors can cancel each other out. Therefore, AMSR-2 is used to replace AMSR-E data when calculating SDI.

[0096] Specifically, when the calculated SDI and the 36.5 GHz vertical polarization data are both less than the threshold, the pixel is regarded as the frozen state (Frozen), otherwise it is the non-frozen state (Unforzen). The brightness temperature data used is ascending orbit data. Therefore, the thresholds of SDI and the vertical polarization brightness temperature data are set to 3.5 K and 263 K respectively.

[0097] In a specific embodiment, considering the frozen state, the mean correlation coefficients of the RF, GBDT, and XGBoost models all exceed 0.6 (0.667, 0.641, 0.601), which are all improved compared to not considering the frozen state (0.550, 0.534, 0.454). In terms of RMSE, the values of these three machine learning models when considering the frozen state are 0.068, 0.07, and 0.073 respectively, which are at least 15% lower than those without considering the frozen state. For MAE, it also significantly decreases after considering the frozen state, that is, the mean MAE of the three machine learning models decreases from about 0.07 to about 0.05. These indicate that in the random point evaluation results, the simulation results of the models considering the frozen state are also better than those not considering the frozen state. According to the evaluation results between the simulated values and the measured values at the stations, it can also be seen that RF performs better than the other two machine learning models, that is, it has a higher correlation coefficient (0.667) and a lower RMSE (0.068). Therefore, the RF model considering the frozen state can be considered as the optimal model result.

[0098] In a specific embodiment, the performance differences of the RF model simulation results considering the freeze-thaw state and three soil moisture data relative to the network observations in 2017 and 2018 were compared. Generally speaking, all four soil moisture data can capture the changing characteristics in time series, but ERA5-LAND, ESA-CCI, and SMCI all have significant overestimations, while the RF freeze model is overall closest to the network observation sequence. Over the entire period, the RMSE of the RF model considering the freeze-thaw state is at least 60% lower than that of the other three products. During the non-freezing period, the average correlation coefficients of each dataset are not very different, but the RF model considering the freeze-thaw state shows a very obvious advantage in the bias index. It should be noted that this advantage of the RF freeze model is further improved during the freezing period. Taking the RMSE index as an example, during the freezing period, the RMSEs of the RF freeze model, ERA5-LAND, and SMCI data are 0.039, 0.238, and 0.22 respectively. The result of the RF freeze model is at least 80% lower in RMSE compared with the other two reanalysis products.

[0099] On the other hand, referring to Figure 2 , the present invention also provides a surface soil moisture simulation system considering the soil freeze-thaw state, including:

[0100] Data acquisition module: acquiring various publicly released remote sensing and reanalysis surface environment data in the plateau region, as well as measured soil moisture data;

[0101] Data processing module: preprocessing the surface environment data;

[0102] Sample generation module: dynamically discriminating the freeze-thaw state of the soil, and based on the dynamic discrimination result, generating a training sample set of the machine learning model from the preprocessed surface environment data and measured soil moisture data;

[0103] Model construction module: completing the training and simulation of various machine learning models considering the soil freeze-thaw state based on the training sample set, comparing the simulation effects of different machine learnings, and selecting the optimal soil moisture inversion model;

[0104] Simulation reconstruction module: reconstructing the surface soil moisture in the plateau region based on the optimal soil moisture inversion model to obtain the spatio-temporal continuous surface soil moisture simulation result of the plateau region.

[0105] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method part.

[0106] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for simulating surface soil moisture taking into account soil freeze-thaw conditions, characterized in that: include: Obtain a variety of publicly released remote sensing and reanalysis surface environmental data in the plateau area, as well as measured soil moisture data; Preprocessing the surface environment data; Dynamically identify the freeze-thaw state of the soil, and based on the dynamic identification results, generate a training sample set for the machine learning model using the pre-processed surface environment data and measured soil moisture data; Based on the training sample set, complete the training and simulation of various machine learning models considering the soil freeze-thaw state, compare the simulation effects of different machine learning models, and select the optimal soil moisture inversion model; Based on the optimal soil moisture inversion model, the surface soil moisture in the plateau area is reconstructed, and the temporally and spatially continuous simulation results of the surface soil moisture in the plateau area are obtained.

2. A method for simulating surface soil moisture taking into account soil freeze-thaw conditions according to claim 1, characterized in that: The surface environmental data of the reanalysis include: surface brightness temperature, surface temperature, surface albedo, land cover type, normalized difference vegetation index NDVI, precipitation, clay ratio, sand ratio, and loam ratio.

3. The method for simulating surface soil moisture considering soil freeze-thaw state according to claim 1, characterized in that: The preprocessing of the surface environment data includes: reprojection, resampling, maximum and minimum normalization and data filling processing.

4. The method for simulating surface soil moisture taking into account soil freeze-thaw state according to claim 2, characterized in that: Dynamic identification of soil freeze-thaw status includes: Based on the surface brightness temperature data, a threshold index method is used to dynamically distinguish the freezing and thawing state of the soil, and the distinction results include a freezing state and a thawing state.

5. The method for simulating surface soil moisture taking into account soil freeze-thaw state according to claim 4, characterized in that: According to the freezing state and the ablation state, respectively, the pre-processed surface environment data and the measured soil moisture data are used to generate a training sample set of a machine learning model, including: Calculate the 36.5GHz vertical polarization data using the formula: In the formula, It is expressed as the vertical polarization brightness temperature of AMSR-2 data at 36.5 GHz, It is expressed as the vertically polarized brightness temperature of AMSR-E data at 36.5 GHz, where v represents vertical polarization; Calculate the standard deviation exponent of horizontally polarized brightness temperature data: Where n is the number of channels with different frequencies, The frequency is f i The brightness temperature under horizontal polarization is is the average value of horizontal polarization under all frequencies, and SDI is the standard deviation index of horizontal polarization brightness temperature data; When the calculated standard deviation index SDI of the horizontal polarization brightness temperature data and the 36.5GHz vertical polarization data are both less than the preset threshold, it is considered to be in a frozen state, otherwise it is in a non-frozen state; The preprocessed plateau surface environment data and measured soil moisture data are used to generate a training sample set for the machine learning model.

6. The method for simulating surface soil moisture considering soil freeze-thaw state according to claim 1, characterized in that: Based on the training sample set, complete the training and simulation of various machine learning models considering the soil freeze-thaw state, compare the simulation effects of different machine learning, and select the optimal soil moisture inversion model, including: The same plateau surface environment data was used as input features and the same observed soil moisture data was used as training targets to train the random forest RF model, the gradient boosted decision tree GBDT model, and the extreme gradient boosted tree XGBoost model. The random forest RF model, gradient boosted decision tree GBDT model and extreme gradient boosted tree XGBoost model were compared and evaluated, and the optimal model was selected as the soil moisture inversion model.

7. The method for simulating surface soil moisture taking into account soil freeze-thaw state according to claim 6, characterized in that: The simulation effects of the random forest RF model, the gradient boosted decision tree GBDT model and the extreme gradient boosted tree XGBoost model are compared and evaluated, including: The evaluation index used is the correlation coefficient R formula: In the formula, and represents the mean, x i The soil moisture value predicted by any of the random forest RF model, gradient boosted decision tree GBDT model, and extreme gradient boosted tree XGBoost model, represents the mean of the predicted values, y i represents the soil moisture value observed at the site, represents the mean of the observed values; Calculate the root mean square error RMSE, the formula is: Calculate the mean absolute error MAE, the formula is: Where N is the number of valid data points.

8. A surface soil moisture simulation system considering soil freeze-thaw state using the surface soil moisture simulation method considering soil freeze-thaw state according to any one of claims 1 to 6, characterized in that: include: Data acquisition module: obtains a variety of publicly released remote sensing and reanalysis surface environmental data in the plateau area, as well as measured soil moisture data; Data processing module: pre-processing the surface environment data; Sample generation module: dynamically identify the freeze-thaw state of the soil, and based on the dynamic identification results, generate a training sample set for the machine learning model using the pre-processed surface environment data and measured soil moisture data; Model building module: Based on the training sample set, complete the training and simulation of multiple machine learning models considering the soil freeze-thaw state, compare the simulation effects of different machine learning models, and select the optimal soil moisture inversion model; Simulation and reconstruction module: Based on the optimal soil moisture inversion model, the surface soil moisture in the plateau area is reconstructed to obtain the temporally and spatially continuous simulation results of the surface soil moisture in the plateau area.

Images