A method for estimating sea ice surface snow depth combination

Abstract

The present application relates to a kind of sea ice surface snow depth combined estimation method, based on the characteristics that microwave radiometer brightness temperature decreases with the increase of sea ice surface snow depth, FY brightness temperature data is used to calculate the brightness temperature of open water at different frequencies, then the brightness temperature of sea ice surface of 10 sub-channels and 20 polarization gradient rate characteristics are calculated using open water brightness temperature and FY brightness temperature data.The OIB airborne data is gridded, and the daily average OIB snow depth of 12.5 km resolution is obtained.The correlation between 30 brightness temperature characteristics and OIB snow depth is calculated, and the linear regression model and machine learning model of one-year ice and multi-year ice surface snow depth estimation are established by selecting 5 sub-features with high correlation, thereby developing the combined estimation model of snow depth.The above is the sea ice surface snow depth combined estimation method proposed by the present application, which improves the estimation accuracy of polar sea ice surface snow depth.

Description

Technical Field

[0001] This invention relates to a combined method for estimating the depth of snow accumulation on the surface of sea ice, belonging to the field of remote sensing geoscience application technology. Background Technology

[0002] Sea ice surface snow cover is a crucial component of the polar climate system, and its changes are closely related to sea ice variations. Compared to sea ice, surface snow cover has higher reflectivity and lower thermal conductivity, playing a role in moderating sea ice formation and dissipation. Snow depth not only participates in sea ice thickness calculations based on satellite altimeters but also plays an indispensable role in sea ice freshwater budget calculations. Accurate estimation of Arctic sea ice surface snow depth is vital for calculating the surface heat and freshwater budget of the sea-air-ice coupled system, analyzing rapid changes in the Arctic environment, improving the accuracy of sea ice thickness and volume estimations, and promoting the development of sea ice growth models.

[0003] Methods such as observation stations, buoys, underway surveys, and airborne surveys can only obtain snow depth data for local time periods and specific sea areas, and are costly. Satellite-based remote sensing observations can achieve large-scale, long-term Earth observations and are the primary means of observing polar ice and snow.

[0004] Passive microwave remote sensing is unaffected by weather conditions, and its brightness temperature data is one of the effective data sources for inverting snow depth on the surface of polar sea ice. Based on the principle that volume scattering increases with increasing frequency and snow depth, while the observed brightness temperature decreases, the relationship between passive microwave brightness temperature and snow depth on the sea ice surface is established. Brightness temperatures of open water at different frequencies and polarization modes suitable for Fengyun satellites are calculated and used to calculate sea ice surface brightness temperature and polarization gradient rate. Five brightness temperature combinations (features) most closely related to measured snow depth are identified, and a one-year ice surface snow depth estimation model is established using a multiple linear regression method. Five more brightness temperature features most closely related to multi-year ice surface measured snow depth are identified, and a multi-year ice surface snow depth estimation model is established using machine learning, thus developing a combined snow depth estimation model, effectively improving the accuracy of snow depth estimation. This has significant application value for analyzing changes in snow depth and sea ice. Summary of the Invention

[0005] The technical problem this invention aims to solve is: addressing the current limitations of single microwave remote sensing methods for obtaining snow depth, poor satellite applicability, and low accuracy. This invention proposes a combined method for estimating snow depth on the surface of sea ice. It uses brightness temperature observations from Fengyun satellites to sequentially calculate the brightness temperature of open water, the brightness temperature of the sea ice surface, and the polarization gradient rate. Finally, it establishes the relationship between the open water surface brightness temperature (OIB) and different brightness temperature characteristics, thereby achieving comprehensive estimation of snow depth on the surface of sea ice in the Arctic and improving estimation accuracy.

[0006] To solve the above-mentioned technical problems, the technical solution proposed by this invention is: a combined method for estimating the depth of snow accumulation on sea ice surface, comprising the following steps:

[0007] The first step is FY-3 data processing. The brightness temperature of all pixels with zero sea ice concentration during the freezing period is averaged to obtain the brightness temperature of open water with horizontal and vertical polarization at various frequencies of the FY-3 brightness temperature product.

[0008] The second step is to calculate the surface brightness temperature of sea ice in M ​​sub-channels using the brightness temperature of open water and the observed brightness temperature of FY-3 brightness temperature products. Then, the brightness temperatures of different channels are combined to calculate N polarization gradient rate combinations, resulting in a total of M+N brightness temperature features.

[0009] The third step is to project the OIB airborne data onto the polar stereo projection coordinate system and obtain the daily OIB gridded snow depth after gridding.

[0010] Step 4: Perform spatiotemporal matching of OIB airborne data and brightness temperature characteristics using the same grid number;

[0011] Step 5: Perform linear regression analysis on the M+N brightness temperature features with the OIB gridded snow depth, sort them from high to low according to the correlation coefficient, and select the top 5 brightness temperature features to establish a multiple linear regression model for estimating the annual snow depth on the ice surface.

[0012] Step 6: Calculate the correlation and linear deviation between M+N brightness temperature features on the ice over many years and OIB airborne data. Select the top 5 brightness temperature features and use machine learning methods to model the snow depth on the ice surface over many years. Finally, combine the snow depth estimation model on the ice surface over one year to obtain a combined snow depth estimation model. Use this combined snow depth estimation model to estimate the snow depth on the sea ice surface.

[0013] The combined method for estimating the depth of snow accumulation on sea ice surface in this invention also has the following characteristics:

[0014] 1. In the first step, the freezing period is from October of the current year to April of the following year. The acquisition of the brightness temperature value of open water is used to reduce the impact of open water on the snow depth inversion.

[0015] 2. In the second step, N=10, M=20, and the 30 brightness temperature features are TB. ice (11H), TB ice (19H), TB ice (24H), TB ice (37H), TB ice (89H), TB ice (11V), TB ice (19V), TB ice (24V), TBice (37V), TB ice (89V), GR v (19 / 11), GR v (24 / 11), GR v (37 / 11), GR v (89 / 11), GR v (24 / 19), GR v (37 / 19), GR v (89 / 19), GR v (37 / 24), GR v (89 / 24), GR v (89 / 37), GR h (19 / 11), GR h (24 / 11), GR h (37 / 11), GR h (89 / 11), GR h (24 / 19), GR h (37 / 19), GR h (89 / 19), GR h (37 / 24), GR h (89 / 24) and GR h (89 / 37); The formulas for calculating the surface brightness temperature and polarization gradient rate of sea ice are as follows:

[0016] TB ice (f,V)=(TB(f,V)-(1-SIC)×TB ow (f,V)) / SIC)

[0017] GR v (f1 / f2)=(TB ice (f1,V)-TB ice (f2,V)) / (TB ice (f1,V)+TB ice (f2,V))

[0018] In the formula TB ice For the surface brightness temperature of sea ice, TB ow The value represents the open water brightness temperature, TB represents the observed brightness temperature, H (or h) and V (or v) represent the horizontal and vertical polarization modes, respectively, f, f1 and f2 represent the observation frequency, and SIC represents the sea ice concentration.

[0019] 3. In the third step, the OIB airborne data time series used is from 2014 to 2019 (March to May), with a spatial resolution of 40m. After gridding, the OIB spatial resolution is 12.5km.

[0020] 4. In the fourth step, the requirement for the matching point is that it has both OIB airborne data and FY-3 brightness temperature characteristic value at the same time and location. Finally, 8428 matching points were selected.

[0021] 5. In the fifth step, the five modeling features identified are GR v (24 / 19), GR h (37 / 19), GR v (37 / 19), TB ice (37V) and GR h (24 / 19).

[0022] 6. Of the 8428 matching points, 6742 were used to train the model, and 1686 were used to test the model's accuracy. The equation for the annual snow depth on the ice surface determined by this method is:

[0023] SD = -220.992 × GR v (24 / 19)-274.935×GR h (37 / 19)

[0024] +278.923×GR v (37 / 19)-0.155×TB ice (37V)

[0025] -179.972×GR h (24 / 19)+52.891

[0026] In the formula, SD represents the annual snow depth on the ice surface.

[0027] 7. In step six, the five features identified for machine learning modeling are GR. h (24 / 19), GR v (24 / 19), GR h (37 / 19), TB ice (24V) and TB ice (37V).

[0028] 8. A total of 5763 matching points were used for estimating the multi-year snow depth on the ice surface. 4610 matching points were used for training, and 1153 were used for testing. The machine learning models employed included Ridge Regression, Support Vector Machines, Adaptive Boosting, Random Forest, and Extremely Randomized Trees (ExtraTree). After experimentation, the ExtraTree model was ultimately selected to determine the multi-year snow depth estimation model for the ice surface.

[0029] The beneficial effects of this invention are:

[0030] Achieving high-precision snow depth data acquisition using domestically developed Fengyun satellites is of great significance for broadening the application fields of domestic satellites, analyzing drastic changes in the Arctic environment, and improving the accuracy of sea ice thickness estimation. This invention, based on the good relationship between snow depth and brightness temperature, uses OIB observation data for modeling and determines a method for estimating sea ice surface snow depth based on Fengyun satellites. Specific beneficial effects are as follows:

[0031] First, this invention utilizes FY-3MWRI brightness temperature data, which can construct a brightness temperature dataset from 2008 to the present. Using FY-3 data to derive the brightness temperature of open water areas, various different brightness temperature characteristics can be further calculated. FY brightness temperature data offers fast updates, strong real-time performance, convenient acquisition, and easy processing.

[0032] Second, this invention establishes the relationship between different features and OIB snow depth based on 10 sub-channel sea ice surface brightness temperature and 20 polarization gradient rate features. The five features with the highest correlation are selected according to their correlation to establish an estimation model for sea ice surface snow depth. This method is highly universal and easy to implement and promote.

[0033] Third, this invention takes into account the different characteristics of snow accumulation on the surface of multi-year ice and one-year ice, and proposes to use machine learning methods to establish a snow accumulation depth estimation model for multi-year ice and use multiple linear regression methods to establish a snow accumulation depth estimation model for one-year ice, which is more consistent with the actual situation of snow accumulation.

[0034] Fourth, the data processing and calculation process of this invention is automatically implemented through Python programming, which has high computational efficiency and can be applied to the production of long-term Arctic sea ice surface snow depth datasets and the analysis of spatiotemporal changes in snow depth. Attached Figure Description

[0035] The invention will now be further described with reference to the accompanying drawings.

[0036] Figure 1 This is a flowchart of a combined method for estimating the depth of snow accumulation on the surface of sea ice.

[0037] Figure 2 This is a numerical map of brightness temperature in open water at different frequencies using the MWRI.

[0038] Figure 3 It is a spatial distribution map of OIB airborne data used for modeling and validation.

[0039] Figure 4 It is brightness temperature feature data that matches the snow depth at OIB.

[0040] Figure 5It is an OIB airborne data spatial distribution map used for multi-year ice surface snow depth modeling and validation.

[0041] Figure 6 It is the input feature for multi-year ice surface snow depth modeling.

[0042] Figure 7 This is a map showing the distribution of snow depth on the surface of sea ice during the Arctic freezing period in 2011. Detailed Implementation

[0043] The present invention will now be described in detail with reference to the accompanying drawings, which will make the steps and effects of the present invention clearer.

[0044] The example data used in this invention include FY-3B, FY-3C, and FY-3D Microwave Radiometer Imager (MWRI) brightness temperature products with a spatial resolution of 12.5 km, and OSI SAF sea ice concentration and type data with a spatial resolution of 10 km. The example data time series covers November 11, 2010 to December 31, 2019, and the imagery covers all areas north of 60°N. OIB airborne data was provided by the U.S. Snow and Ice Data Center and was acquired between 2014 and 2019.

[0045] Figure 1 The flowchart shows a combined method for estimating snow depth on sea ice surfaces. The specific implementation steps of this method include the following:

[0046] Step 1: FY-3 Data Processing. The brightness temperature values ​​of all pixels with zero sea ice concentration during the freezing period (October to April) are averaged to obtain the brightness temperatures of open water with horizontal and vertical polarization at various frequencies in FY-3, which are then used to acquire the brightness temperature values ​​of the sea ice surface. The values ​​are as follows: Figure 2 As shown.

[0047] The second step involves calculating the sea ice surface brightness temperature characteristics of 10 sub-channels using the brightness temperature values ​​from open water and FY-3 observations. The brightness temperatures from different channels are combined and calculated to obtain 20 polarization gradient rate characteristics, resulting in a total of 30 brightness temperature characteristics. In this embodiment, these 30 brightness temperature characteristics include TB... ice (11H), TB ice (19H), TB ice (24H), TB ice (37H), TB ice (89H), TB ice (11V), TB ice (19V), TB ice (24V), TB ice (37V), TB ice (89V), GR v(19 / 11), GR v (24 / 11), GR v (37 / 11), GR v (89 / 11), GR v (24 / 19), GR v (37 / 19), GR v (89 / 19), GR v (37 / 24), GR v (89 / 24), GR v (89 / 37), GR h (19 / 11), GR h (24 / 11), GR h (37 / 11), GR h (89 / 11), GR h (24 / 19), GR h (37 / 19), GR h (89 / 19), GR h (37 / 24), GR h (89 / 24) and GR h (89 / 37). The formulas for calculating the surface brightness temperature and polarization gradient rate of sea ice are as follows:

[0048] TB ice (f,V)=(TB(f,V)-(1-SIC)×TB ow (f,V)) / SIC)

[0049] GR v (f1 / f2)=(TB ice (f1,V)-TB ice (f2,V)) / (TB ice (f1,V)+TB ice (f2,V))

[0050] In the formula TB ice For the surface brightness temperature of sea ice, TB ow The value represents the open water brightness temperature, TB represents the observed brightness temperature, H (or h) and V (or v) represent the horizontal and vertical polarization modes, respectively, f, f1 and f2 represent the observation frequency, and SIC represents the sea ice concentration.

[0051] The third step involves projecting the OIB airborne data from 2014 to 2019 (March-May) onto the polar stereo projection coordinate system to calculate the daily OIB gridded snow depth at a resolution of 12.5 km.

[0052] Step 4: Spatiotemporal matching of OIB airborne data and FY-3 brightness temperature data using the same grid number. The matching points are required to have both OIB airborne data and FY-3 brightness temperature characteristic values ​​at the same time and location. A total of 8428 matching points were finally selected.

[0053] Step 5: Perform linear regression analysis on the 30 brightness temperature features and OIB snow depth respectively. Sort the features by correlation coefficient from high to low, and select the top 5 features to establish a multiple linear regression model for estimating annual ice surface snow depth. Of the 8428 matching points, 6742 matching points were used to train the model, and 1686 matching points were used to test the model accuracy. Figure 3 This is a spatial distribution diagram of the OIB airborne data used for modeling and validation. Figure 4 This is brightness temperature characteristic data that matches the snow depth at the OIB (Oasis Irrigation Surface). The final equation for estimating the annual snow depth at the ice surface is as follows:

[0054] SD = -220.992 × GR v (24 / 19)-274.935×GR h (37 / 19)

[0055] +278.923×GR v (37 / 19)-0.155×TB ice (37V)

[0056] -179.972×GR h (24 / 19)+52.891

[0057] In the formula, SD represents the annual snow depth on the ice surface.

[0058] Step 6: Calculate the correlation and linearity bias between 30 brightness temperature features on the ice over many years and OIB airborne data, and select the top 5 GR data. h (24 / 19), GR v (24 / 19), GR h (37 / 19), TB ice (24V) and TB ice(37V) Different machine learning methods (Ridge Regression, Support Vector Machines, Adaptive Boosting, Random Forest, and Extremely Randomized Trees (ExtraTree)) were used to model snow depth on the surface of ice over many years. A total of 5763 matching points were used for snow depth estimation on the surface of ice over many years, with 4610 points used for training and 1153 points used for testing. After experiments, the ExtraTree model was ultimately selected to determine the snow depth estimation model for the surface of ice over many years. Combined with the annual snow depth estimation model, a combined snow depth estimation model was obtained, and this combined snow depth estimation model was used to estimate the snow depth on the surface of sea ice. This embodiment generated a dataset of Arctic sea ice surface snow depth from 2010 to 2019 after data processing. Figure 5 Spatial distribution map of OIB airborne data used for multi-year modeling and validation of snow depth on ice surface. Figure 6 Input features for modeling the depth of snow accumulation on the surface of ice over many years. Figure 7 This is a map showing the distribution of snow depth on the surface of sea ice during the Arctic freezing period in 2011.

[0059] In addition to the embodiments described above, the present invention may have other implementations. All technical solutions formed by equivalent substitution or equivalent transformation fall within the protection scope claimed by the present invention.

Claims

1. A combined method for estimating the depth of snow accumulation on sea ice surface, comprising the following steps: The first step is FY-3 data processing. The brightness temperature of all pixels with zero sea ice concentration during the freezing period is averaged to obtain the brightness temperature of open water with horizontal and vertical polarization at various frequencies of the FY-3 brightness temperature product. The second step is to calculate the surface brightness temperature of sea ice in M ​​sub-channels using the brightness temperature of open water and the observed brightness temperature of FY-3 brightness temperature products. Then, the brightness temperatures of different channels are combined to calculate N polarization gradient rate combinations, resulting in a total of M+N brightness temperature features. The third step is to project the OIB airborne data onto the polar stereo projection coordinate system and obtain the daily OIB gridded snow depth after gridding. Step 4: Perform spatiotemporal matching of OIB airborne data and brightness temperature characteristics using the same grid number; Step 5: Perform linear regression analysis on the M+N brightness temperature features with the OIB gridded snow depth, sort them from high to low according to the correlation coefficient, and select the top 5 brightness temperature features to establish a multiple linear regression model for estimating the annual snow depth on the ice surface. Step 6: Calculate the correlation and linear deviation between M+N brightness temperature features on the ice over many years and OIB airborne data. Select the top 5 brightness temperature features and use machine learning methods to model the snow depth on the ice surface over many years. Finally, combine the snow depth estimation model on the ice surface over one year to obtain a combined snow depth estimation model. Use this combined snow depth estimation model to estimate the snow depth on the sea ice surface.

2. The combined estimation method for sea ice surface snow depth according to claim 1, characterized in that: In the first step, the freezing period is from October of the current year to April of the following year. The acquisition of the brightness temperature value of open water is used to reduce the impact of open water on the snow depth inversion.

3. The combined estimation method for sea ice surface snow depth according to claim 1, characterized in that: In the second step, N=10, M=20, and the 30 brightness temperature features are TB. ice (11H), TB ice (19H), TB ice (24H), TB ice (37H), TB ice (89H), TB ice (11V), TB ice (19V), TB ice (24V), TB ice (37V), TB ice (89V), GR v (19 / 11), GR v (24 / 11), GR v (37 / 11), GR v (89 / 11), GR v (24 / 19), GR v (37 / 19), GR v (89 / 19), GR v (37 / 24), GR v (89 / 24), GR v (89 / 37), GR h (19 / 11), GR h (24 / 11), GR h (37 / 11), GR h (89 / 11), GR h (24 / 19), GR h (37 / 19), GR h (89 / 19), GR h (37 / 24), GR h (89 / 24) and GR h (89 / 37); The formulas for calculating the surface brightness temperature and polarization gradient rate of sea ice are as follows: TB ice (f,V)=(TB(f,V)-(1-SIC)×TB ow (f,V)) / SIC) GR v (f1 / f2)=(TB ice (f1,V)-TB ice (f2,V)) / (TB ice (f1,V)+TB ice (f2,V)) In the formula TB ice For the surface brightness temperature of sea ice, TB ow The values ​​represent the brightness temperature of open water, TB represents the observed brightness temperature, H and h represent the horizontal polarization, V and v represent the vertical polarization, f, f1 and f2 represent the observation frequency, and SIC represents the sea ice concentration.

4. The combined estimation method for sea ice surface snow depth according to claim 1, characterized in that: In the third step, the time series of the OIB airborne data is from March to May 2014 to 2019, with a spatial resolution of 40m; after gridding, the OIB spatial resolution is 12.5km.

5. The combined estimation method for sea ice surface snow depth according to claim 1, characterized in that: In the fourth step, the matching point is required to have both OIB airborne data and FY-3 brightness temperature characteristic value at the same time and location.

6. The combined estimation method for sea ice surface snow depth according to claim 1, characterized in that: In the fifth step, the five modeling features identified are GR v (24 / 19), GR h (37 / 19), GR v (37 / 19), TB ice (37V) and GR h (24 / 19).

7. The combined estimation method for sea ice surface snow depth according to claim 6, characterized in that: Of the matched points, a portion is used to train the model, and the remaining matched points are used to test the model's accuracy; the determined equation for the annual snow depth on the ice surface is: SD=-220.992×GR v (24 / 19)-274.935×GR h (37 / 19)+278.923×GR v (37 / 19)-0.155×TB ice (37V)-179.972×GR h (24 / 19)+52.891 In the formula, SD represents the annual snow depth on the ice surface.

8. The combined estimation method for sea ice surface snow depth according to claim 1, characterized in that: In the sixth step, the five brightness temperature features identified for machine learning modeling are GR. h (24 / 19), GR v (24 / 19), GR h (37 / 19), TB ice (24V) and TB ice (37V).

9. The combined estimation method for sea ice surface snow depth according to claim 8, characterized in that: Of the matching points for multi-year ice surface snow depth, a portion are used for training, and the remaining matching points are used for testing. The machine learning models used are Ridge Regression, Support Vector Machines, Adaptive Boosting, Random Forest, or Extremely Randomized Trees.

10. The combined estimation method for sea ice surface snow depth according to claim 8, characterized in that: The machine learning model uses an Extremely Randomized Trees model to determine the multi-year ice surface snow depth estimation model.

Images