Passive microwave load-based snow depth and snow water equivalent inversion method

Abstract

A snow depth and snow water equivalent inversion method based on a passive microwave load comprises the following steps: step 1, simulating a microwave load snow surface brightness temperature by adopting a microwave snow radiation transmission model, and obtaining an effective snow particle size by iterating the snow particle size and minimizing a model simulation brightness temperature difference and a load actual brightness temperature difference under a corresponding condition, the load brightness temperature and the snow depth are combined to complete the construction of a snow depth machine learning model data set; step 2, adopting adaptive LASSO regression to improve an adaptive penalty weight, constructing a machine learning model and solving an inversion formula coefficient by combining the machine learning model data set constructed in the step S1 and adopting a load multi-channel brightness temperature difference and an effective accumulated snow particle size; and step 3, calculating the snow density by adopting near-surface air temperature and wind speed data, and multiplying the snow density by the snow depth obtained by inversion in the step 2 to obtain the snow water equivalent. According to the invention, the snow depth and snow water equivalent inversion algorithm which is universally applicable to the satellite pixel scale can be provided.

Description

Technical Field

[0001] This invention belongs to the technical field of snow cover parameter calculation, and specifically relates to a method for inverting snow depth and snow water equivalent based on passive microwave load. Background Technology

[0002] Snow depth and snow water equivalent are core elements in snow cover observation, and their spatiotemporal distribution characteristics have a profound impact on global change, Earth system science, water cycle and water resources, as well as snow disaster and flood monitoring. Currently, the observation and inversion of snow cover elements mainly rely on field measurements, long-term observations from ground stations, and regional monitoring by satellite remote sensing. Field measurements and ground stations have limitations and cannot fully reflect the spatial distribution of snow cover; satellite remote sensing, with its advantages of long-term, large-scale, and dynamic monitoring of global snow cover, can provide information at the "area scale."

[0003] There are two main methods for satellite remote sensing monitoring of snow cover: optical snow cover monitoring and microwave inversion of snow depth and snow water equivalent. Among them, passive microwave remote sensing technology is currently the most effective method for inverting snow depth and snow water equivalent.

[0004] Currently, based on field measurements, long-term ground station observations, and satellite remote sensing data, and combined with in-depth research on the physical characteristics of snow cover and radiative transfer models, various passive microwave inversion algorithms for snow depth and snow water equivalent have been developed. These algorithms include semi-empirical models combining satellite and ground data, physical models based on the radiative scattering characteristics of snow cover, and machine learning algorithms.

[0005] Semi-empirical models typically rely on empirical parameters closely related to factors such as vegetation type, topography, and climate conditions. These parameters can vary significantly across different regions, leading to models performing well in some areas but with lower accuracy in others. Furthermore, the characteristics of snowfall and snow accumulation, such as snow particle size, density, and water content, change over time, while model parameters are often set to fixed values, making it difficult to capture these dynamic changes and resulting in uncertainties in the inversion results.

[0006] While physical models based on snow radiation scattering characteristics have a clear physical foundation, they often require a large number of input parameters, including snow particle size, shape, density, water content, temperature, surface roughness, and vegetation cover. These parameters are difficult to obtain accurately in practical applications, especially those with significant spatial variability. Furthermore, the models are complex and computationally expensive; solving the radiative transfer equation typically requires complex numerical calculations, resulting in high computational costs and long processing times, limiting their application over large areas and long time series. The effectiveness of machine learning algorithms heavily depends on the quality and quantity of training data. If the training data is noisy, biased, or insufficiently representative, model performance will be severely affected. Ground observation data on snow depth and snow water equivalent are often difficult to obtain, especially in high-altitude and remote areas, which limits the training and validation of machine learning models. Additionally, training data is often unevenly distributed in time and space, leading to poor model performance in certain regions or time periods.

[0007] In summary, different algorithms are suitable for different snow cover environments, and their inversion effects and accuracy vary. Currently, there is a lack of a universally applicable snow depth and snow water equivalent inversion algorithm at the satellite pixel scale. Summary of the Invention

[0008] This invention provides a method for inverting snow depth and snow water equivalent based on passive microwave load, in order to solve the problem that current snow depth and snow water equivalent inversion algorithms lack universality.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for inverting snow depth and snowmelt equivalent based on passive microwave loading includes the following steps: Step 1: Using the microwave snow radiation transmission model, the brightness temperature of the snow surface under microwave load is simulated. By iterating the snow particle size, the model simulates the brightness temperature difference and the actual brightness temperature difference under the corresponding conditions is minimized to obtain the effective snow particle size. Combined with the load brightness temperature and snow depth, the snow depth machine learning model dataset is constructed. Step 2: Adaptive LASSO regression is used to improve the adaptive penalty weights. Combined with the machine learning model dataset built in step S1, the machine learning model is constructed and the inversion formula coefficients are solved using load multi-channel brightness temperature difference and effective snow particle size. Step 3: Calculate the snow density using near-surface air temperature and wind speed data, and multiply it by the snow depth obtained in step S2 to obtain the snow water equivalent. Furthermore, in step one, the microwave snow radiation transfer model integrates a wealth of microstructures and electromagnetic models of snow particles, which can be used to simulate the brightness and temperature of snow under various environments; the microwave snow radiation transfer model establishes single-layer or multi-layer models of snow by inputting the physical parameters of each layer of snow.

[0010] Furthermore, in step one, when solving the microwave snow radiation transmission equation, the specific equation is as follows:

[0011] in, It is the defined reduced ratio of radiation intensity. I = I' / n² , I’ It is a comparison of radiation intensity. n It is the refractive index at the same location; It is a 4×4 phase matrix; k a and k e These are the absorption coefficient and the extinction coefficient, respectively. k e =k s +k a ,in k s It is the scattering coefficient; Vector 1 = (1, 1, 1, 1); direction determined by the cosine of the zenith angle. µ and azimuth The definition, the relevant solid angle is Oh; T B,p The brightness temperature is where p = H or V , and the reduced ratio of radiation intensity I p Proportional, that is I p =αT B,p ,in α= 2 ν²k / c 0 ² , k and c 0 represents Boltzmann's constant and the speed of light in a vacuum, respectively. n It is the frequency of the wave.

[0012] Furthermore, for the passive mode, and utilizing the linearity of the microwave snow radiative transfer equation, it is possible to... I p Replace with brightness temperature, and α Set to 1.

[0013] Furthermore, the microwave snow radiative transfer equation can be simplified to the following two assumptions: (i) the medium has azimuthal symmetry, and (ii) the medium consists of a homogeneous layered structure:

[0014] in, l The representative layer index has a value range of 1 to 1. L ( l = 1... L ),in l =1 represents the top level. l = L It represents the lower level.

[0015] Furthermore, in step one, the snow depth machine learning model is input with snow surface brightness temperature, snow density, snow depth, and snow particle size; the model simulation step size is 1K and 10kg / m², respectively. 3 The microstructure is 1 cm and 0.02 mm; the microstructure is selected from the viscous hard sphere model; the frequency is consistent with the microwave imager, including five frequencies: 10.65, 18.7, 23.8, 36.5 and 89 GHz, each frequency includes H and V polarization, and the incident angle is 53.1°; By selecting different parameter step sizes and other parameters, the snow radiation transfer model is run cyclically to obtain the simulated brightness temperature under different parameters; by iterating the snow particle size, the effective snow particle size is obtained by minimizing the simulated brightness temperature difference between the model and the actual brightness temperature difference under the corresponding load.

[0016] Furthermore, in step one, the specific formula for the snow depth machine learning model is as follows:

[0017] in, d snow particle size, TB 18.7 V,MODEL and TB 36.5 V,MODEL The brightness temperatures at 18.7 GHz and 36.5 GHz, respectively, are the simulated brightness temperatures under vertical polarization. TB 18.7 V,MWRI and TB 36.5 V,MWRI The brightness temperatures are 18.7 GHz and 36.5 GHz under vertical polarization of the load, respectively.

[0018] Furthermore, in step two, adaptive LASSO regression is used to improve the adaptive penalty weights. Combined with the machine learning model dataset constructed in step S1, the machine learning model is constructed and the coefficients of the inversion formula are solved. The final inversion relationship of snow depth is obtained as follows:

[0019] in, y Snow depth TB 10.65V 、TB 36.5V 、TB 18.7V 、TB 89V The vertical polarization brightness temperatures are 10.65 GHz, 36.5 GHz, 18.7 GHz, and 89 GHz, respectively. TB 10.65H The brightness temperature for the 10.65 GHz horizontal polarization of the payload is [not specified]. d For effective snow particle size, β 1. β 2. β 3. β 4 represents the coefficients obtained from the LASSO regression. intercept To solve for the intercept.

[0020] Furthermore, in step three, when calculating the snow water equivalent, based on the snow depth obtained in step two, the snow water equivalent can be considered as the product of snow density and snow depth per unit area, that is: SWE=ρ*SD Where ρ is the snow density and SD is the snow depth.

[0021] Furthermore, snow density can be expressed as a function of near-surface air temperature and wind speed, i.e.:

[0022] in, and These represent near-surface air temperature (K) and wind speed, respectively. Coefficient value = 109 , = 6 , = 26 ; This is the triple point temperature of water.

[0023] The present invention can achieve the following beneficial effects: This invention fully utilizes a microwave radiative transfer model to more precisely describe the scattering, absorption, and emission processes of microwaves within snow cover, taking into account the influence of key parameters such as snow particle size, snow density, and surface temperature on microwave radiation. This provides richer input features for machine learning algorithms and enhances the physical basis for inversion.

[0024] However, the microwave radiation transfer model itself also has parameter uncertainties, such as snow particle shape and roughness. Machine learning algorithms can learn the error patterns caused by these uncertainties and make effective compensations.

[0025] This invention utilizes an improved adaptive LASSO regression method to capture the complex mapping relationship between microwave radiation and snow depth and snow water equivalent, overcoming the limitations of traditional methods in modeling.

[0026] The model can learn from a large amount of observational data and automatically adapt to changes in snow conditions in different regions and seasons, thereby improving the model's versatility and reliability. Attached Figure Description

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the snow depth and snow water equivalent inversion based on passive microwave load in this invention; Figure 2 This is a schematic diagram of the microwave snow radiation transmission model of the present invention; Figure 3 Flowchart for constructing the dataset for the machine learning model of this invention; Figure 4 This invention presents the spatial distribution of snow depth on the Qinghai-Tibet Plateau from January to April 2020. Figure 5 This invention relates to the spatial distribution of snow depth on the Qinghai-Tibet Plateau from May to August 2020. Figure 6 This invention relates to the spatial distribution of snow depth on the Qinghai-Tibet Plateau from September to December 2020. Figure 7 This invention presents the monthly average snow depth statistics for the Qinghai-Tibet Plateau in 2020. Figure 8 This invention presents the spatial distribution of snow cover density on the Qinghai-Tibet Plateau from January to April 2020. Figure 9 This invention relates to the spatial distribution of snow cover density on the Qinghai-Tibet Plateau from May to August 2020. Figure 10 This invention relates to the spatial distribution of snow cover density on the Qinghai-Tibet Plateau from September to December 2020. Figure 11 This invention provides a monthly snow water equivalent statistic of the Qinghai-Tibet Plateau in 2020. Detailed Implementation

[0028] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0029] like Figure 1 to Figure 11 As shown, a method for inverting snow depth and snowmelt equivalent based on passive microwave load includes the following steps: Step 1: Using a microwave snow radiation transmission model, the brightness temperature of the snow surface under microwave load is simulated. By iterating the snow particle size, the model simulates the brightness temperature difference from the actual brightness temperature difference under the corresponding conditions to obtain the effective snow particle size. Combined with the load brightness temperature and snow depth, the snow depth machine learning model dataset is constructed.

[0030] Among them, the microwave snow radiation transfer model integrates rich microstructure and electromagnetic models of snow particles, and can be used to simulate the brightness and temperature of snow in various environments, such as land, ice shelves, and sea ice. This model can establish single-layer or multi-layer models of snow and ice by inputting the physical parameters of each layer of snow and ice.

[0031] In solving the radiative transfer equation, it is assumed that the equation is independent of time and that the medium is horizontally homogeneous. Furthermore, at the microscopic scale, the model also assumes that the snow cover is isotropic, with the specific equation being: (1) in, It is the defined reduced ratio of radiation intensity. I = I' / n² ,in I’ It is a comparison of radiation intensity. n It is the refractive index at the same location. It is a 4×4 phase matrix. k a and k e These are the absorption coefficient and extinction coefficient, respectively, and vector 1 = (1, 1, 1, 1).

[0032] Extinction coefficient is from k e =k s +k a Given, among which k s It is the scattering coefficient. Its direction is determined by the cosine of the zenith angle. µ and azimuth The definition, the relevant solid angle is Oh The z-axis points upwards, representing the incident beam and the downward radiation. µ<0 And upward radiation µ>0The z-axis, which points upwards, is a commonly used definition in Earth science.

[0033] This equation is valid in both active and passive modes within the microwave range. (Brightness temperature) T B,p ,in p = H (Horizontal polarization) or V (Vertical polarization), and reduced ratio of radiation intensity I p Proportional; I p =αT B,p ,in α= 2 ν²k / c 0 ² , k and c 0 represents Boltzmann's constant and the speed of light in a vacuum, respectively. n It is the frequency of the wave.

[0034] In practical applications, for passive modes, and utilizing the linear property of formula (1), it can be... I p Replace with brightness temperature, and α Set to 1.

[0035] After further assuming the following two conditions: (i) the medium has orientational symmetry, and (ii) the medium consists of a homogeneous layered structure, the equation can be simplified to: (2) l The representative layer index has a value range of 1 to 1. L ( l = 1... L ),in l =1 represents the top level. l = L Representing the bottom layer. The continuity conditions at the interfaces between layers and the boundary conditions of the bottom interface can be expressed as follows: (3) (4) z l It is the first l Bottom of the layer z The axis position, on the contrary, z l-1 It is the first l The height of the top of the floor. R and T These are the reflectance and transmittance matrices, respectively. "spec"The superscript indicates the specular component. "diff" The superscript indicates the diffuse component. For a perfectly flat interface, the diffuse component is zero, and the specular component is given by the Fresnel coefficient. "top" The superscript indicates the coefficient from the first floor to the floor above it. "bottom" Superscripts indicate coefficients up to the levels below them. Function S l1,l2 ( µ 1) Calculate the angle of incidence of the beam due to refraction from the first... l From the 1st floor to the 2nd floor l The change is two-layered. In this case, a single-layer model is used to construct the snow radiation transfer model, that is... l = L =1.

[0036] The dataset for the machine learning model is constructed as follows: The input to the snow radiative transfer model includes snow surface brightness temperature, snow density, snow depth, and snow grain size. The simulation step size is 1K and 10kg / m², respectively. 3 The snow grain sizes were 1 cm and 0.02 mm. A viscous hard sphere model was chosen for the microstructure, with frequencies consistent with the microwave imager, including five frequencies: 10.65 GHz, 18.7 GHz, 23.8 GHz, 36.5 GHz, and 89 GHz. Each frequency included H and V polarization, with an incident angle of 53.1°. The snow radiative transfer model was cyclically run using different parameter step sizes and other parameter selections to obtain the simulated brightness temperature under different parameters. The effective snow grain size was obtained by iteratively minimizing the difference between the simulated brightness temperature and the actual brightness temperature under the corresponding load conditions, using the following formula: (5) Where d is the snow particle size, and The brightness temperatures at 18.7 GHz and 36.5 GHz, respectively, are the simulated brightness temperatures under vertical polarization. and The brightness temperatures at 18.7 GHz and 36.5 GHz under vertical polarization are respectively. The effective snow grain size, load brightness temperature, and snow depth are obtained when the cost function (Cost) is minimized, thus completing the construction of the machine learning model dataset.

[0037] Step 2: Adaptive LASSO regression is used to improve the adaptive penalty weights. Combined with the machine learning model dataset built in Step 1, the machine learning model is constructed and the inversion formula coefficients are solved using load multi-channel brightness temperature difference and effective snow particle size.

[0038] The construction of machine learning models and the solution of inversion formula coefficients specifically include: LASSO regression aims to optimize convex functions, enabling simultaneous variable selection and parameter estimation. It is a commonly used and increasingly popular regularization method. LASSO estimates linear model parameters using absolute values ​​and minimizing a penalty function. LASSO contraction estimation attempts to preserve good variable selection by setting some coefficients to zero. It is widely used in environments with large amounts of data, where the number of features, p, is much larger than the number of observations.

[0039] Consider a linear regression model. ,in e 1,..., e n Full id (0, s 2 LASSO estimates Defined as: (6) In the formula The adjustment parameter is non-negative, and the penalty function is... ,generally There is no closed-form expression.

[0040] However, for a special case of orthogonal design, there exists an analytical solution. For For an orthogonal design of the identity matrix, the LASSO estimator is l n Functions > 0: (7) in j =0, 1, ..., p , ( z ) + =max{ z ,0}; if z If the expression is greater than 0, less than 0, or equal to 0, then we have: sgn ( z ) = +1, 0, -1.

[0041] LASSO exhibits favorable optimization characteristics, minimizing a convex objective function. Therefore, it avoids the problem of multiple local minima, and the global minima problem can be efficiently solved using various algorithms, including but not limited to LASSO quadratic programming, LASSO shooting algorithm, LASSO homotopy algorithm, minimum angle regression and contraction, and LASSO coordinate descent algorithm.

[0042] Adaptive LASSO aims to improve LASSO regression by introducing weighted coefficients and using adaptive weights to penalize different parameters in the penalty function. Modifying the penalty term allows weighting based on the absolute values ​​of the coefficients. It also possesses sparse solutions and oracle properties, meaning it has the same asymptotic distribution as if the actual parameters were known in advance.

[0043] The definition of adaptive LASSO estimation is: (8) The penalty function is:

[0044] Adaptive weight selection , j=1,2,...,p, γ>0, are generally estimated by LASSO or OLS methods; the positive value γ is a power of the adaptive weight and is related to the order.

[0045] Select adjustment parameters from cross-validation. By combining the adaptive weight order γ, we obtain adaptive LASSO. Adaptive LASSO penalizes the initial parameters with smaller weight coefficients for estimating larger variables, while larger weight coefficients penalize the initial estimates of smaller variables, thus preserving the advantages of the original LASSO estimation. This method can effectively correct the model's error, thereby improving the algorithm's convergence. In this invention, based on adaptive LASSO, we improve the adaptive penalty weights as follows: (9) The estimator for the improved adaptive LASSO method is defined as: (10) in l n It is an improvement on the penalty weights of adaptive LASSO. c 1 is the penalty parameter for the coefficient. c 2 is the penalty parameter for the lag order. Additionally, p It is AR ( p The maximum lag order of the model, For the middle term, The meaning is the first j The order is either positive or negative 1. , , j The smaller the value less than the middle term, the smaller the penalty; when... hour, , j The larger the value is than the middle term, the greater the penalty becomes.

[0046] Using the effective snow grain size, load brightness temperature, and snow depth obtained in step one, an improved adaptive LASSO regression model is constructed. In this case, y Snow depth x The effective snow grain size and load brightness temperature are used. The load brightness temperature is represented by multi-channel brightness temperature difference, and the final snow depth inversion relationship can be expressed as: (11) in, y Snow depth TB 10.65V 、TB 36.5V 、TB 18.7V 、TB 89V The vertical polarization brightness temperatures are 10.65, 36.5, 18.7, and 89 GHz, respectively. TB 10.65H The brightness temperature for the 10.65 GHz horizontal polarization of the payload is [not specified]. d For effective snow particle size, β 1. β 2. β 3. β 4 represents the coefficients obtained from the LASSO regression. intercept To solve for the intercept.

[0047] Step 3: Using near-surface air temperature and wind speed data, calculate the snow density and multiply it by the snow depth obtained in step 2 to obtain the snow water equivalent. The snow water equivalent calculation method is as follows: Based on the snow depth obtained in step two, the snow water equivalent can be considered as the product of snow density and snow depth per unit area. In this invention, snow density can be expressed as a function of near-surface air temperature and wind speed: (12) in, and These represent near-surface air temperature (K) and wind speed, respectively. Coefficient value = 109 , = 6 , = 26 . This is the triple point temperature of water. Snow density is limited to 50 to 450. between.

[0048] The equivalent of snow water can be expressed as: SWE=ρ*SD (13) in, r snow density, SD This represents the depth of the snow cover.

[0049] Furthermore, the aforementioned model and method were applied to the inversion of snow depth and snow water equivalent on the Qinghai-Tibet Plateau. Taking FY3D microwave imager data from 2020 as an example, the daily snow depth on the Qinghai-Tibet Plateau was calculated, and the results were obtained using the monthly average synthesis method. Figure 4-6 The data shows the spatial distribution of monthly snow depth, and regional mean statistics for the Qinghai-Tibet Plateau region are obtained. Figure 7 Monthly average snow depth data.

[0050] The results show that the snow depth conforms to a temporal distribution pattern, with higher depths in winter and lower depths in summer, and the highest and lowest values ​​occurring in February and July, respectively. Spatially, winter snow cover is mainly distributed in the southwestern and central parts of the Qinghai-Tibet Plateau, areas with high altitudes, some exceeding 5000 meters. Due to the increasing altitude, temperatures gradually decrease, resulting in relatively low temperatures on the Qinghai-Tibet Plateau, making the snow less prone to melting. Simultaneously, the southern region of the Qinghai-Tibet Plateau is located on the southern slope of the Himalayas, characterized by high mountains and deep valleys, where air currents are strongly lifted, resulting in abundant precipitation, primarily in the form of snow. These factors combined contribute to the high snow depth.

[0051] Monthly near-surface air temperature and wind speed in the Qinghai-Tibet Plateau region in 2020 were selected, and monthly snow cover density was calculated according to Formula 12. The specific spatial distribution results are as follows: Figure 8-10 As shown, similar to the overall spatial distribution pattern of snow depth, higher snow density values ​​are mainly distributed in the southwestern and central parts of the Qinghai-Tibet Plateau. These areas have higher altitudes and relatively lower temperatures, making it difficult for snow to melt, thus resulting in higher snow density. Unlike snow depth, some areas also experience higher snow density during the Spring Festival and summer. This is mainly due to rising temperatures in some areas, causing snow to begin melting. The melted snow re-condenses or compacts, increasing the density of the remaining snow. Furthermore, summer precipitation in some areas is primarily in the form of snow or hail, which also increases snow density.

[0052] Monthly snow cover density and Figure 4-6 The snow depth data are multiplied and then accumulated pixel by pixel, such as... Figure 11 The time-series results of monthly snow water equivalent on the Qinghai-Tibet Plateau in 2020 can be obtained. The time-series statistical results show that the overall results are highly consistent with the snow depth. The snow depth is higher in winter and lower in summer, and the highest and lowest values ​​occur in February and July, respectively. This indicates that the snow water equivalent is mainly related to the snow depth. The snow depth and snow water equivalent inversion model established in this invention has a very good inversion effect.

[0053] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A snow depth and snow water equivalent inversion method based on passive microwave payload, characterized in that, The method comprises the following steps: Step one, using a microwave snow radiation transfer model to simulate the microwave load snow surface brightness temperature, through iterative snow particle size, minimizing the difference between the model simulated brightness temperature and the actual brightness temperature of the corresponding condition load, to obtain the effective snow particle size, and combining the load brightness temperature and snow depth, to complete the construction of the snow depth machine learning model data set; Step two, using adaptive LASSO regression to improve the adaptive penalty weight, combining the machine learning model data set constructed in S1 step, using the load multi-channel brightness temperature difference and effective snow particle size, to construct the machine learning model and solve the inversion formula coefficient; Step three, using the near-surface air temperature and wind speed data to calculate the snow density, and multiplying the snow depth obtained in S2 step to obtain the snow water equivalent.

2. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 1, characterized in that: In step one, the microwave snow radiation transfer model integrates rich ice and snow particle microstructure and electromagnetic model, which can be used for snow brightness temperature simulation under various environments; the microwave snow radiation transfer model establishes an ice and snow single-layer or multi-layer model by inputting the physical parameters of each layer of ice and snow.

3. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 1, characterized in that: In step one, when solving the microwave snow radiation transfer equation, the specific equation is: ； wherein is the reduced specific intensity of radiation defined by I = I’ / n² , I’ is the specific intensity of radiation, n is the refractive index at the same location; is a 4x4 phase matrix; κ a and κ e are the absorption and extinction coefficients, respectively, and the extinction coefficient κ e =κ s +κ a where κ s is the scattering coefficient; Vector 1 = (1, 1, 1, 1); direction is given by the cosine of the zenith angle µ and the azimuth angle defined, the relevant solid angle is Ω; T B,p for luminance temperature, where p = H or V is proportional to the reduced specific intensity I p i.e. I p =αT B,p where α= 2 ν²k / c 0 ² , k and c 0 are the Boltzmann constant and the speed of light in vacuum, respectively, ν is the frequency of the wave.

4. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 3, characterized in that: For the passive mode, and using the linear property of the microwave snow radiative transfer equation, one can replace I p with the brightness temperature, and set α to 1.

5. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 3 or 4, characterized in that: The microwave snow radiation transfer equation can be simplified as: ； wherein l represents a layer index, and takes a value ranging from 1 to L ( l = 1... L ), wherein l =1 represents the top layer, l = L represents the bottom layer.

6. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 1, characterized in that: In step one, the snow depth machine learning model inputs the snow surface brightness temperature, snow density, snow depth, and snow particle size; the model simulation step is 1K, 10kg / m 3 , 1cm, and 0.02mm, respectively; the microstructure selects the viscous hard sphere model; the frequency is consistent with the microwave imager, including five frequencies of 10.65, 18.7, 23.8, 36.5, and 89GHz, each frequency including H and V polarization, and the incident angle is 53.1°; By selecting different parameter steps and selecting other parameters, the snow radiation transfer model is run in a loop to obtain simulated brightness temperatures under different parameters; through iterative snow particle size, the difference between the model simulated brightness temperature and the actual brightness temperature of the corresponding condition load is minimized to obtain the effective snow particle size.

7. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 6, characterized in that: In step one, the specific formula of the snow depth machine learning model is: ； wherein, d is the snow grain size, TB 18.7 V,MODEL and TB 36.5 V,MODEL are the model simulated brightness temperatures at 18.7 GHz and 36.5 GHz for vertical polarization, respectively, TB 18.7 V,MWRI and TB 36.5 V,MWRI are the load simulated brightness temperatures at 18.7 GHz and 36.5 GHz for vertical polarization, respectively.

8. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 1, characterized in that: In step two, using adaptive LASSO regression to improve the adaptive penalty weight, combining the machine learning model data set constructed in S1 step, to construct the machine learning model and solve the inversion formula coefficient, and the final snow depth inversion relationship obtained is: ； wherein, y is the snow depth, TB 10.65V , TB 36.5V , TB 18.7V , TB 89V is the load 10.65 GHz, 36.5 GHz, 18.7 GHz, 89 GHz vertical polarization brightness temperature, TB 10.65H is the load 10.65 GHz horizontal polarization brightness temperature, d is the effective snow grain size, β 1, β 2, β 3, β 4 is the LASSO regression to solve the coefficient, intercept is the intercept.

9. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 1, characterized in that: In step three, when calculating the snow water equivalent, based on the snow depth obtained in step two, the snow water equivalent can be regarded as the product of the snow density and the snow depth per unit area, that is: SWE=ρ*SD; Wherein, ρ is the snow density, and SD is the snow depth.

10. The snow depth and snow water equivalent inversion method based on passive microwave load according to claim 9, characterized in that: The snow density can be expressed as a function of the near-surface air temperature and wind speed, that is: ； wherein, and represent the near-surface air temperature (K) and wind speed ; the coefficient values = 109 , = 6 , = 26 ; is the triple point temperature of water.

Images