Sea-gas turbulence heat flux collaborative inversion method and system

Abstract

The invention belongs to the technical field of heat flux remote sensing inversion, and relates to a sea-air turbulence heat flux collaborative inversion method and system. The method comprises the following steps: collecting observation data required by sea-gas turbulence heat flux remote sensing inversion; performing pretreatment; selecting variables as a model driving data set; calculating a sea-air turbulence wave-text ratio; constructing a sea-air turbulence heat flux collaborative inversion multivariable random forest model; determining an optimal hyper-parameter combination to obtain a target sea-air turbulence heat flux collaborative inversion multivariable random forest model; and the collaborative inversion multivariable random forest model is used for collaborative inversion of the sea-air turbulence heat flux. According to the method, the problem that the sea-gas turbulence wave-text ratio obtained by estimation is not accurate enough due to the fact that an existing data driving model ignores the cooperative change relation of the sea-gas turbulence sensible heat flux, the sea-gas turbulence latent heat flux and the sea-gas turbulence wave-text ratio can be solved, and the estimation precision of the sea-gas turbulence wave-text ratio is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of heat flux remote sensing inversion, and in particular, relates to a sea-air turbulent heat flux collaborative inversion method and system based on a multivariate random forest model. Background Art

[0002] The heat of the ocean and atmosphere will have a profound impact on the global meteorological and climate systems. As the main process at the sea-air interface, the sea-air turbulent heat flux includes the sea-air turbulent incoming heat flux and the sea-air turbulent sensible heat flux. The accurate estimation of the sea-air turbulent heat flux is of great significance for a better understanding of the interaction between the ocean and the atmosphere, and is helpful for the prediction of climate change and extreme meteorological events. However, the currently widely used overall aerodynamic model based on the Morning-Obukhov similarity theory has a simple parameterization scheme and a certain uncertainty in the model-driven data, which leads to deviations in the estimation of the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux, and further leads to errors and uncertainties in the estimated sea-air turbulent Bowen ratio. The existing data-driven methods estimate the sea-air turbulent incoming heat flux or the sea-air turbulent latent heat flux separately. Although the estimation errors of each are reduced, the coordinated variation relationship between the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux is ignored, which still leads to unreasonable estimation results of the sea-air turbulent Bowen ratio, bringing certain uncertainties to the research on global climate change projections, meteorological forecasts, and global energy and water cycles. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a sea-air turbulent heat flux collaborative inversion method and system based on a multivariate random forest model, aiming to solve the problem that the existing data-driven model ignores the coordinated variation relationship between sea-air turbulent incoming heat flux, sea-air turbulent latent heat flux and sea-air turbulent Bowen ratio, resulting in the estimated sea-air turbulent Bowen ratio being inaccurate, thereby introducing uncertainty into global climate change predictions, extreme meteorological event predictions and global energy and water cycle research.

[0004] In a first aspect, the present invention provides a sea-air turbulent heat flux collaborative inversion method, comprising:

[0005] Collect observation data required for remote sensing inversion of sea-air turbulent heat flux; observation data include near-sea meteorological observation data, sea surface observation data and sea-air turbulent heat flux observation data; sea-air turbulent heat flux includes sea-air turbulent current heat flux and sea-air turbulent latent heat flux;

[0006] Preprocess the collected observation data;

[0007] All available variables in the preprocessed observation data are sorted according to their contribution, and the variables are selected as the model-driven data set;

[0008] The sea-air turbulence Bowen ratio is calculated based on the sea-air turbulence current heat flux and the sea-air turbulence latent heat flux;

[0009] Taking all variables in the model-driven dataset as input, the sea-air turbulent Bowen ratio as the internal constraint, and the sea-air turbulent incoming heat flux, sea-air turbulent latent heat flux and sea-air turbulent Bowen ratio as output, a multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux is constructed.

[0010] The grid retrieval method is used to train the multivariate random forest model for the collaborative inversion of sea-air turbulent heat flux, determine the optimal hyperparameter combination, and obtain the target multivariate random forest model for the collaborative inversion of sea-air turbulent heat flux;

[0011] The target sea-air turbulent heat flux collaborative inversion multivariate random forest model is used to collaboratively invert the sea-air turbulent heat flux.

[0012] In a second aspect, the present invention provides a sea-air turbulent heat flux collaborative inversion system, comprising a collection unit, a preprocessing unit, a selection unit, a first processing unit, a model building unit, a model training unit and a second processing unit;

[0013] A collection unit is used to collect observation data required for remote sensing inversion of sea-air turbulent heat flux; the observation data include near-sea meteorological observation data, sea surface observation data and sea-air turbulent heat flux observation data; the sea-air turbulent heat flux includes sea-air turbulent current heat flux and sea-air turbulent latent heat flux;

[0014] A preprocessing unit, used for preprocessing the collected observation data;

[0015] A selection unit is used to sort all available variables in the preprocessed observation data according to their contribution and select variables as model-driven data sets;

[0016] The first processing unit is used to calculate the sea-air turbulent Bowen ratio according to the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux;

[0017] A model building unit is used to take all variables in the model-driven data set as input, take the sea-air turbulent Bowen ratio as an internal constraint, and take the sea-air turbulent incoming heat flux, the sea-air turbulent latent heat flux and the sea-air turbulent Bowen ratio as outputs to build a sea-air turbulent heat flux collaborative inversion multivariate random forest model;

[0018] A model training unit is used to train a multivariate random forest model for collaborative inversion of sea-air turbulent heat flux using a grid retrieval method, determine an optimal hyperparameter combination, and obtain a target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux;

[0019] The second processing unit is used to collaboratively invert the sea-air turbulent heat flux using a target sea-air turbulent heat flux collaborative inversion multivariate random forest model.

[0020] Based on the above technical solution, the present invention can also be improved as follows.

[0021] Furthermore, the near-sea surface observation data include sea surface air temperature, sea surface air specific humidity and sea surface wind speed, and the sea surface observation data include sea surface temperature and sea surface specific humidity.

[0022] Furthermore, the collected observation data are preprocessed, including: performing height correction on the collected observation data; averaging the observation data with a sampling time frequency higher than the set time length, determining the validity of the observation data according to the proportion of high-frequency data within the set time length, and discarding the observation data that is not valid; deleting the sea-air observation data and the sea-air turbulent heat flux observation data under ship maneuvers and plume conditions; deleting the sea-air observation data and the sea-air turbulent heat flux observation data with a sea-air turbulent Bowen ratio less than a first set value or greater than a second set value; and calculating the temperature gradient and the specific humidity gradient.

[0023] Furthermore, the sea-air turbulent Bowen ratio is calculated based on the sea-air turbulent current heat flux and the sea-air turbulent latent heat flux, including:

[0024] Assume SHF is the sea-air turbulent incoming heat flux, LHF is the sea-air turbulent latent heat flux, β is the sea-air turbulent Bowen ratio, then the sea-air turbulent Bowen ratio is:

[0025]

[0026] Furthermore, a multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux is constructed, including: assuming that SHF represents the sea-air turbulent incoming heat flux, LHF represents the sea-air turbulent latent heat flux, β represents the sea-air turbulent Bowen ratio, and the multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux is expressed as BrTHF(SHF, LHF, β), f() represents the functional relationship of the multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux, T s represents the sea surface temperature of the ocean, q s represents the surface specific humidity of the ocean, T a represents the sea surface air temperature at a set altitude, q a Indicates the specific humidity of sea surface air at a set altitude, W s Indicates the sea surface wind speed at a set height, diff T represents the calculated temperature gradient, diff q represents the calculated specific humidity gradient, then:

[0027] BrTHF(SHF,LHF,β)=f(T s,q s ,T a ,q a ,W s ,diff T ,diff q ).

[0028] Furthermore, a grid retrieval method is used to train a multivariate random forest model for collaborative inversion of sea-air turbulent heat flux, and the optimal hyperparameter combination is determined to obtain a target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux, including: presetting multiple sets of hyperparameter combinations, the hyperparameters include the number of decision trees, the minimum number of sample splits for each regression tree node, and the minimum number of samples for leaf nodes, and the range of each hyperparameter is determined by comparing the accuracies of each multivariate random forest model for collaborative inversion of sea-air turbulent heat flux under the preset multiple sets of hyperparameter combinations, and then the grid retrieval method is used to train the optimal hyperparameter combination to obtain the optimal target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux.

[0029] Furthermore, when training the optimal hyperparameter combination, the cost function of each regression tree node is determined by the Mahalanobis distance; let cost tree Represents the cost function of each regression tree node, y pre_i represents the predicted sea-air turbulent heat flux, y mea_i represents the observed sea-air turbulent heat flux, Σ i -1 Represents the inverse matrix of the covariance matrix of the output response matrix; when i = 1, y mea_1 represents the observed value of the sea-air turbulent heat flux, y pre_1 represents the predicted value of the sea-air turbulent heat flux; when i = 2, y mea_2 represents the observed value of the sea-air turbulent latent heat flux, y pre_2 represents the predicted value of the sea-air turbulent latent heat flux; when i = 3, y mea_3 represents the observed value of the sea-air turbulence Bowen ratio, y pre_3 represents the predicted value of the sea-air turbulent Bowen ratio; the cost function of the sea-air turbulent heat flux collaborative inversion multivariate random forest model with the sea-air turbulent Bowen ratio as the internal constraint is expressed as:

[0030]

[0031] The beneficial effect of the present invention is that the present invention can solve the problem that the existing data-driven model ignores the coordinated change relationship between the sea-air turbulent incoming heat flux, the sea-air turbulent latent heat flux and the sea-air turbulent Bowen ratio, resulting in the estimated sea-air turbulent Bowen ratio being inaccurate, thereby introducing uncertainty into global climate change predictions, extreme meteorological event predictions and global energy and water cycle research, reduce the deviation in the estimation of the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux, and improve the accuracy of the estimation of the sea-air turbulent Bowen ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of the sea-air turbulent heat flux collaborative inversion method provided in Example 1 of the present invention;

[0033] Figure 2 Schematic diagram of the sea-air turbulent heat flux collaborative inversion system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0035] Example 1

[0036] As an example, Figure 1 As shown, in order to solve the above technical problems, this embodiment provides a sea-air turbulent heat flux collaborative inversion method, including:

[0037] Collect observation data required for remote sensing inversion of sea-air turbulent heat flux; observation data include near-sea meteorological observation data, sea surface observation data and sea-air turbulent heat flux observation data; sea-air turbulent heat flux includes sea-air turbulent current heat flux and sea-air turbulent latent heat flux;

[0038] Preprocess the collected observation data;

[0039] All available variables in the preprocessed observation data are sorted according to their contribution, and the variables are selected as the model-driven data set;

[0040] The sea-air turbulence Bowen ratio is calculated based on the sea-air turbulence current heat flux and the sea-air turbulence latent heat flux;

[0041] Taking all variables in the model-driven dataset as input, the sea-air turbulent Bowen ratio as the internal constraint, and the sea-air turbulent incoming heat flux, sea-air turbulent latent heat flux and sea-air turbulent Bowen ratio as output, a multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux is constructed.

[0042] The grid retrieval method is used to train the multivariate random forest model for the collaborative inversion of sea-air turbulent heat flux, determine the optimal hyperparameter combination, and obtain the target multivariate random forest model for the collaborative inversion of sea-air turbulent heat flux;

[0043] The target sea-air turbulent heat flux collaborative inversion multivariate random forest model is used to collaboratively invert the sea-air turbulent heat flux.

[0044] Optionally, the near-sea surface observation data include sea surface air temperature, sea surface air specific humidity and sea surface wind speed, and the sea surface observation data include sea surface temperature and sea surface specific humidity.

[0045] Optionally, the collected observation data are preprocessed, including: performing altitude correction on the collected observation data; averaging the observation data with a sampling time frequency higher than a set duration, determining the validity of the observation data based on the proportion of high-frequency data within the set duration, and discarding observation data that is not valid; deleting sea-air observation data and sea-air turbulent heat flux observation data under ship maneuvers and plume conditions; deleting sea-air observation data and sea-air turbulent heat flux observation data with a sea-air turbulent Bowen ratio less than a first set value or greater than a second set value; and calculating the temperature gradient and specific humidity gradient.

[0046] Specifically, the preprocessing process includes, for example: correcting the sea surface air temperature, sea surface air specific humidity and sea surface wind speed collected at an observation height of 11-25m to the sea surface air temperature, sea surface air specific humidity and sea surface wind speed at an altitude of 10m; correcting the sea surface temperature and sea surface specific humidity collected at a measurement depth of 0.05-5m to the sea surface temperature and sea surface specific humidity of the ocean surface; averaging the observation data with a sampling time frequency higher than 1 hour as the observation data for 1 hour, and if the high-frequency observation data within 1 hour is not less than 75%, the 1-hour observation data is valid, otherwise it is discarded; quality control is performed on the observation data, and meteorological, oceanographic observation data and sea-air turbulent heat flux observation data under ship maneuvers and plume conditions are deleted, and meteorological, oceanographic observations and sea-air turbulent heat flux observations with sea-air turbulent Bowen ratio less than -1 or greater than 2 that are beyond the sea-air interaction range are deleted; and temperature gradient and specific humidity gradient are calculated.

[0047] Optionally, the sea-air turbulent Bowen ratio is calculated based on the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux, including:

[0048] Assume SHF is the sea-air turbulent incoming heat flux, LHF is the sea-air turbulent latent heat flux, β is the sea-air turbulent Bowen ratio, then the sea-air turbulent Bowen ratio is:

[0049]

[0050] Optionally, a multivariate random forest model for coordinated inversion of sea-air turbulent heat flux is constructed, including: let SHF represent the sea-air turbulent current heat flux, LHF represent the sea-air turbulent latent heat flux, β represent the sea-air turbulent Bowen ratio, and the multivariate random forest model for coordinated inversion of sea-air turbulent heat flux is expressed as BrTHF(SHF , LHF,β), f() represents the functional relationship of the multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux, T s represents the sea surface temperature of the ocean, q s represents the surface specific humidity of the ocean, T a represents the sea surface air temperature at a set altitude, q a Indicates the specific humidity of sea surface air at a set altitude, W s Indicates the sea surface wind speed at a set height, diff T represents the calculated temperature gradient, diff q represents the calculated specific humidity gradient, then:

[0051] BrTHF(SHF,LHF,β)=f(T s ,q s ,T a ,q a ,W s ,diff T ,diff q ).

[0052] In actual application, the setting height of sea surface air temperature, sea surface air specific humidity and sea surface wind speed is selected as 10m.

[0053] Optionally, a multivariate random forest model for collaborative inversion of sea-air turbulent heat flux is trained using a grid retrieval method to determine the optimal hyperparameter combination and obtain a target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux, including: presetting multiple sets of hyperparameter combinations, the hyperparameters include the number of decision trees, the minimum number of sample splits for each regression tree node and the minimum number of samples for leaf nodes, determining the range of each hyperparameter by comparing the accuracy of each multivariate random forest model for collaborative inversion of sea-air turbulent heat flux under the preset multiple sets of hyperparameter combinations, and then using a grid retrieval method to train the optimal hyperparameter combination to obtain the optimal target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux.

[0054] Optionally, when training the optimal hyperparameter combination, the cost function of each regression tree node is determined by the Mahalanobis distance; let cost tree Represents the cost function of each regression tree node, y pre_i represents the predicted sea-air turbulent heat flux, y mea_i represents the observed sea-air turbulent heat flux, ∑ i -1 represents the inverse of the covariance matrix of the output response matrix; when i = 1, y mea_1 represents the observed value of the sea-air turbulent heat flux, y pre_1 represents the predicted value of the sea-air turbulent heat flux; when i = 2, y mea_2 represents the observed value of the sea-air turbulent latent heat flux, y pre_2 represents the predicted value of the sea-air turbulent latent heat flux; when i = 3, y mea_3 represents the observed value of the sea-air turbulence Bowen ratio, y pre_3 represents the predicted value of the sea-air turbulent Bowen ratio; the cost function of the sea-air turbulent heat flux collaborative inversion multivariate random forest model with the sea-air turbulent Bowen ratio as the internal constraint is expressed as:

[0055]

[0056] The present invention can solve the problem that the existing data-driven model ignores the coordinated change relationship between the sea-air turbulent incoming heat flux, the sea-air turbulent latent heat flux and the sea-air turbulent Bowen ratio, resulting in the estimated sea-air turbulent Bowen ratio being inaccurate, thereby introducing uncertainty into the global climate change prediction, extreme meteorological event prediction and global energy and water cycle research, reduce the deviation in the estimation of the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux, and improve the accuracy of the estimation of the sea-air turbulent Bowen ratio.

[0057] Example 2

[0058] Based on the same principle as the method shown in Example 1 of the present invention, as shown in the attached Figure 1 As shown, an embodiment of the present invention also provides a sea-air turbulent heat flux collaborative inversion system, including a collection unit, a preprocessing unit, a selection unit, a first processing unit, a model building unit, a model training unit and a second processing unit;

[0059] A collection unit is used to collect observation data required for remote sensing inversion of sea-air turbulent heat flux; the observation data include near-sea meteorological observation data, sea surface observation data and sea-air turbulent heat flux observation data; the sea-air turbulent heat flux includes sea-air turbulent current heat flux and sea-air turbulent latent heat flux;

[0060] A preprocessing unit, used for preprocessing the collected observation data;

[0061] A selection unit is used to sort all available variables in the preprocessed observation data according to their contribution and select variables as model-driven data sets;

[0062] The first processing unit is used to calculate the sea-air turbulent Bowen ratio according to the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux;

[0063] A model building unit is used to take all variables in the model-driven data set as input, take the sea-air turbulent Bowen ratio as an internal constraint, and take the sea-air turbulent incoming heat flux, the sea-air turbulent latent heat flux and the sea-air turbulent Bowen ratio as outputs to build a sea-air turbulent heat flux collaborative inversion multivariate random forest model;

[0064] A model training unit is used to train a multivariate random forest model for collaborative inversion of sea-air turbulent heat flux using a grid retrieval method, determine an optimal hyperparameter combination, and obtain a target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux;

[0065] The second processing unit is used to collaboratively invert the sea-air turbulent heat flux using a target sea-air turbulent heat flux collaborative inversion multivariate random forest model.

[0066] Optionally, the near-sea surface observation data include sea surface air temperature, sea surface air specific humidity and sea surface wind speed, and the sea surface observation data include sea surface temperature and sea surface specific humidity.

[0067] Optionally, the collected observation data are preprocessed, including: performing altitude correction on the collected observation data; averaging the observation data with a sampling time frequency higher than a set duration, determining the validity of the observation data based on the proportion of high-frequency data within the set duration, and discarding observation data that is not valid; deleting sea-air observation data and sea-air turbulent heat flux observation data under ship maneuvers and plume conditions; deleting sea-air observation data and sea-air turbulent heat flux observation data with a sea-air turbulent Bowen ratio less than a first set value or greater than a second set value; and calculating the temperature gradient and specific humidity gradient.

[0068] Optionally, the sea-air turbulent Bowen ratio is calculated based on the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux, including:

[0069] Assume SHF is the sea-air turbulent incoming heat flux, LHF is the sea-air turbulent latent heat flux, β is the sea-air turbulent Bowen ratio, then the sea-air turbulent Bowen ratio is:

[0070]

[0071] Optionally, a multivariate random forest model for coordinated inversion of sea-air turbulent heat flux is constructed, including: let SHF represent the sea-air turbulent current heat flux, LHF represent the sea-air turbulent latent heat flux, β represent the sea-air turbulent Bowen ratio, and the multivariate random forest model for coordinated inversion of sea-air turbulent heat flux is expressed as BrTHF(SHF , LHF,β), f() represents the functional relationship of the multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux, T s represents the sea surface temperature of the ocean, q s represents the surface specific humidity of the ocean, T a represents the sea surface air temperature at a set altitude, q a Indicates the specific humidity of sea surface air at a set altitude, W s Indicates the sea surface wind speed at a set height, diff T represents the calculated temperature gradient, diff q represents the calculated specific humidity gradient, then:

[0072] BrTHF(SHF,LHF,β)=f(T s ,q s ,T a ,q a ,W s ,diff T ,diff q ).

[0073] Optionally, a multivariate random forest model for collaborative inversion of sea-air turbulent heat flux is trained using a grid retrieval method to determine the optimal hyperparameter combination and obtain a target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux, including: presetting multiple sets of hyperparameter combinations, the hyperparameters include the number of decision trees, the minimum number of sample splits for each regression tree node and the minimum number of samples for leaf nodes, determining the range of each hyperparameter by comparing the accuracy of each multivariate random forest model for collaborative inversion of sea-air turbulent heat flux under the preset multiple sets of hyperparameter combinations, and then using a grid retrieval method to train the optimal hyperparameter combination to obtain the optimal target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux.

[0074] Optionally, when training the optimal hyperparameter combination, the cost function of each regression tree node is determined by the Mahalanobis distance; let cost tree Represents the cost function of each regression tree node, y pre_i represents the predicted sea-air turbulent heat flux, y mea_i represents the observed sea-air turbulent heat flux, ∑ i -1 represents the inverse of the covariance matrix of the output response matrix; when i = 1, y mea_1 represents the observed value of the sea-air turbulent heat flux, ypre_1 represents the predicted value of the sea-air turbulent heat flux; when i = 2, y mea_2 represents the observed value of the sea-air turbulent latent heat flux, y pre_2 represents the predicted value of the sea-air turbulent latent heat flux; when i = 3, y mea_3 represents the observed value of the sea-air turbulence Bowen ratio, y pre_3 represents the predicted value of the sea-air turbulent Bowen ratio; the cost function of the sea-air turbulent heat flux collaborative inversion multivariate random forest model with the sea-air turbulent Bowen ratio as the internal constraint is expressed as:

[0075]

[0076] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A collaborative inversion method for sea-air turbulent heat flux, characterized in that: include: Collect observational data required for remote sensing inversion of sea-air turbulent heat flux; The observation data include near-sea meteorological observation data, sea surface observation data and sea-air turbulent heat flux observation data; The sea-air turbulent heat flux includes the sea-air turbulent current heat flux and the sea-air turbulent latent heat flux; Preprocess the collected observation data; All available variables in the preprocessed observation data are sorted according to their contribution, and the variables are selected as the model-driven data set; The sea-air turbulence Bowen ratio is calculated based on the sea-air turbulence current heat flux and the sea-air turbulence latent heat flux; Taking all variables in the model-driven dataset as input, the sea-air turbulent Bowen ratio as the internal constraint, and the sea-air turbulent incoming heat flux, sea-air turbulent latent heat flux and sea-air turbulent Bowen ratio as output, a multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux is constructed. The grid retrieval method is used to train the multivariate random forest model for the collaborative inversion of sea-air turbulent heat flux, determine the optimal hyperparameter combination, and obtain the target multivariate random forest model for the collaborative inversion of sea-air turbulent heat flux; The target sea-air turbulent heat flux collaborative inversion multivariate random forest model is used to collaboratively invert the sea-air turbulent heat flux.

2. The sea-air turbulent heat flux collaborative inversion method according to claim 1, characterized in that: Near-sea surface observation data include sea surface air temperature, sea surface air specific humidity and sea surface wind speed, and sea surface observation data include sea surface temperature and sea surface specific humidity.

3. The sea-air turbulent heat flux collaborative inversion method according to claim 1 is characterized in that: The collected observation data are preprocessed, including: performing altitude correction on the collected observation data; averaging the observation data with a sampling time frequency higher than the set time length, determining the validity of the observation data according to the proportion of high-frequency data within the set time length, and discarding the observation data that is not valid; deleting the sea-air observation data and sea-air turbulent heat flux observation data under ship maneuvers and plume conditions; deleting the sea-air observation data and sea-air turbulent heat flux observation data with a sea-air turbulent Bowen ratio less than a first set value or greater than a second set value; and calculating the temperature gradient and specific humidity gradient.

4. The sea-air turbulent heat flux collaborative inversion method according to claim 1, characterized in that: The sea-air turbulent Bowen ratio is calculated based on the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux, including: Assume SHF is the sea-air turbulent incoming heat flux, LHF is the sea-air turbulent latent heat flux, β is the sea-air turbulent Bowen ratio, then the sea-air turbulent Bowen ratio is:

5. The sea-air turbulent heat flux collaborative inversion method according to claim 1, characterized in that: A multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux is constructed, including: assuming that SHF represents the sea-air turbulent incoming heat flux, LHF represents the sea-air turbulent latent heat flux, β represents the sea-air turbulent Bowen ratio, the multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux is expressed as BrTHF(SHF, LHF, β), f() represents the functional relationship of the multivariate random forest model for the coordinated inversion of sea-air turbulent heat flux, T s represents the sea surface temperature, i.e. the ocean surface temperature, q s represents the sea surface specific humidity, that is, the specific humidity of the ocean surface, T a represents the sea surface air temperature at a set altitude, q a Indicates the specific humidity of sea surface air at a set altitude, W s Indicates the sea surface wind speed at a set height, diff T represents the calculated temperature gradient, diff q represents the calculated specific humidity gradient, then: BrTHF(SHF,LHF,β)=f(T s ,q s ,T a ,q a ,W s ,diff T ,diff q )。 6. The sea-air turbulent heat flux collaborative inversion method according to claim 1, characterized in that: A multivariate random forest model for collaborative inversion of sea-air turbulent heat flux is trained by using a grid retrieval method to determine the optimal hyperparameter combination, and obtain a target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux, including: presetting multiple sets of hyperparameter combinations, the hyperparameters include the number of decision trees, the minimum number of sample splits for each regression tree node, and the minimum number of samples for leaf nodes, and determining the range of each hyperparameter by comparing the accuracy of each multivariate random forest model for collaborative inversion of sea-air turbulent heat flux under the preset multiple sets of hyperparameter combinations, and then using a grid retrieval method to train the optimal hyperparameter combination to obtain the optimal target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux.

7. The sea-air turbulent heat flux collaborative inversion method according to claim 6, characterized in that: When training the optimal hyperparameter combination, the cost function of each regression tree node is determined by the Mahalanobis distance; let cost tree Represents the cost function of each regression tree node, y pre_i represents the predicted sea-air turbulent heat flux, y mea_i represents the observed sea-air turbulent heat flux, Σ i -1 represents the inverse of the covariance matrix of the output response matrix; when i = 1, y mea_1 represents the observed value of the sea-air turbulent heat flux, y pre_1 represents the predicted value of the sea-air turbulent heat flux; when i = 2, y mea_2 represents the observed value of the sea-air turbulent latent heat flux, y pre_2 represents the predicted value of the sea-air turbulent latent heat flux; when i = 3, y mea_3 represents the observed value of the sea-air turbulence Bowen ratio, y pre_3 represents the predicted value of the sea-air turbulent Bowen ratio; the cost function of the sea-air turbulent heat flux collaborative inversion multivariate random forest model with the sea-air turbulent Bowen ratio as the internal constraint is expressed as:

8. The sea-air turbulent heat flux collaborative inversion system is characterized by: It includes a collection unit, a preprocessing unit, a selection unit, a first processing unit, a model building unit, a model training unit and a second processing unit; A collection unit is used to collect observation data required for remote sensing inversion of sea-air turbulent heat flux; the observation data include near-sea meteorological observation data, sea surface observation data and sea-air turbulent heat flux observation data; The sea-air turbulent heat flux includes the sea-air turbulent current heat flux and the sea-air turbulent latent heat flux; A preprocessing unit, used for preprocessing the collected observation data; A selection unit is used to sort all available variables in the preprocessed observation data according to their contribution and select variables as model-driven data sets; The first processing unit is used to calculate the sea-air turbulent Bowen ratio according to the sea-air turbulent incoming heat flux and the sea-air turbulent latent heat flux; A model building unit is used to take all variables in the model-driven data set as input, take the sea-air turbulent Bowen ratio as an internal constraint, and take the sea-air turbulent incoming heat flux, the sea-air turbulent latent heat flux and the sea-air turbulent Bowen ratio as outputs to build a sea-air turbulent heat flux collaborative inversion multivariate random forest model; A model training unit is used to train a multivariate random forest model for collaborative inversion of sea-air turbulent heat flux using a grid retrieval method, determine an optimal hyperparameter combination, and obtain a target multivariate random forest model for collaborative inversion of sea-air turbulent heat flux; The second processing unit is used to collaboratively invert the sea-air turbulent heat flux using a target sea-air turbulent heat flux collaborative inversion multivariate random forest model.

Images