Ocean internal temperature inversion method fusing PIES and sea surface multi-source parameters

Abstract

The invention relates to an ocean internal temperature inversion method fusing PIES and sea surface multi-source parameters, and belongs to the technical field of ocean environment parameter inversion and ocean dynamics observation. Comprising the following steps: constructing a nonlinear empirical model (Multi-RF) of seawater temperature inversion by using a random forest algorithm and fusing data such as a historical observation temperature-salinity profile observed by a temperature-salinity-depth meter (CTD) and sea surface multi-source parameters; and inverting the seawater temperature through a Multi-RF model by using the sound propagation time and multi-source sea surface data calculated by the PIES. Compared with a traditional empirical mode mapping (GEM) model, the Mull-RF model provided by the invention has the advantages that on one hand, the Mull-RF model utilizes fewer samples to establish a mapping relation between the seawater temperature and multiple parameters, and the problem of sample shortage in construction of a seawater temperature inversion model fusing sea surface multiple parameters and PIES observation sound propagation time is solved; on the other hand, compared with polynomial fitting in a GEM model, the random forest adopted by the Multi-RF model can achieve nonlinear fitting from multiple parameters to the seawater temperature, and therefore the temperature inversion precision is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of ocean internal temperature detection, and specifically relates to a method for inversely calculating seawater temperature using a PIES device. Background Art

[0002] The accurate determination of the seawater temperature field has important scientific significance for deeply understanding ocean dynamic processes, material transport mechanisms, and their responses to ecosystem and climate change. However, affected by the comprehensive effects of tides, wind stress, river input, and terrain complexity, the temperature structure of the ocean shows significant spatio-temporal heterogeneity. This heterogeneity not only directly affects ocean circulation, nutrient distribution, and biological productivity, but also has an important impact on the acoustic propagation characteristics in ocean engineering, shipping safety, and military applications. The ocean thermal environment is characterized by high dynamics, frequent small-scale processes, and complex terrain conditions. Traditional observation methods such as CTD profiles, satellite remote sensing, and Argo floats have obvious deficiencies in terms of spatial coverage, temporal continuity, and spatial resolution, and it is difficult to meet the requirements for high-precision inversion of ocean internal temperature. To address this problem, it is urgent to develop new observation technologies and inversion methods to improve the spatio-temporal resolution and accuracy of ocean temperature observations and provide reliable data support for related research.

[0003] The Pressure Inverse Echo Sounder equipped with a pressure sensor (PIES) provides a new approach for the observation of the ocean temperature field. The PIES device records the round-trip acoustic travel time (τ) from the seabed to the sea surface. This time depends on the sound speed distribution of the water body along the signal propagation path. According to the sound speed equation (such as the Del Grosso equation), the seawater sound speed is closely related to temperature, salinity, and pressure. That is, the PIES signal travel time is closely related to the temperature and salinity of the water body through which the signal passes. Therefore, the temperature-salinity field of seawater can be inversely calculated through the acoustic travel time observed by PIES, so as to achieve the purpose of detecting the vertical structure of temperature and salinity from the seabed to the sea surface. Compared with traditional observation methods, PIES has the advantages of high temporal resolution, long-term stable observation, and adaptability to complex sea conditions, and has more application value especially in shallow sea environments where it is impossible to deploy profiling instruments frequently.

[0004] At present, the GEM method is mainly used for the inversion of temperature and salinity profiles based on the PIES observation system. The essence of this method is to establish an empirical regression relationship model between the sound propagation time and the temperature and salinity fields at different depths, so as to realize the inversion of the temperature in the whole water depth. However, this method uses polynomial fitting to establish the empirical regression model, which will cause certain deviations in fitting the non-linear relationship between the sound propagation time and the temperature and salinity fields at different depths. In addition, it is difficult to incorporate a variety of other parameters into the construction of the empirical model to improve the inversion accuracy. In recent years, deep learning and machine learning methods have shown significant advantages in the research of ocean dynamic element detection, which also provides a new technical approach to solve the limitations of the GEM method in temperature field inversion.

[0005] Therefore, there is an urgent need in this field to propose a method for accurately inverting the seawater temperature based on PIES observation. Summary of the Invention

[0006] To solve the above problems, the present invention proposes a method for accurately inverting the seawater temperature based on PIES observation.

[0007] The technical solution adopted by the present invention to achieve the above object is: an ocean internal temperature inversion method integrating PIES and multi-source parameters on the sea surface, including the following steps:

[0008] S1. Obtain the round-trip sound wave propagation time τ measured by PIES from the seabed to the sea surface b and the seabed pressure data p b ;

[0009] S2. Preprocess the τ measured by PIES b and correct τ b to the sound propagation time τ from the bottom of the thermocline to the sea surface tb-pies ;

[0010] S3. Obtain the CTD-observed temperature and salinity profiles, and calculate the sound propagation time τ from the bottom of the thermocline to the sea surface for each temperature and salinity profile tn-ctd ;

[0011] S4. Match τ tb-ctd with the ocean multi-parameters according to the principle of spatio-temporal proximity;

[0012] S5. Use the combination of τ tb-ctd ocean multi-parameters, time, and location multi-parameters as the input data of the inversion model, and the temperatures measured by CTD at different depths as the output data of the inversion model. Use the input data and output data to construct a training data set for training the inversion model; use the random forest algorithm to build the inversion model, and train the inversion model with the training data set, and finally obtain an inversion model that can reflect the non-linear relationship between the multi-parameter input of the model and the ocean temperature of the model output;

[0013] S6. For the ocean area to be inverted, obtain the sound propagation time τ from the bottom of the thermocline to the sea surface of the ocean area to be inverted through steps S1 - S2 tb-pies , and match τ tb-pies with the ocean multi - parameters of the ocean area to be inverted, and combine τ tb-pies , ocean multi - parameters, time, and location multi - parameters into a PIES inversion dataset, and input it into the trained inversion model to output the seawater temperature profile corresponding to the corresponding time and corresponding location, so as to realize the inversion of seawater temperature.

[0014] In step S3, to obtain the CTD - observed temperature - salinity profile, select the CTD - observed data within 2° of the PIES observation station location.

[0015] The ocean multi - parameters include sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth.

[0016] In step S3, the sound propagation time τ tb-ctd from the bottom of the thermocline to the sea surface is calculated as follows:

[0017]

[0018] where c is the sound speed, ρ is the seawater density calculated from the CTD temperature and salinity, g is the acceleration due to gravity, p ref is the depth of the bottom of the thermocline, and p represents pressure.

[0019] In steps S4 and S6, the ocean multi - parameters are grid data, and the spatial resolution is less than or equal to 0.25°, and the time resolution is 1 day.

[0020] In steps S4 and S6, the matching according to the spatio - temporal proximity principle is as follows: for the time and location of each CTD profile observation or each PIES observation, select the values of sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth products that are at the closest distance and on the same day for matching.

[0021] In step S5, the inputs of the non - linear mapping model include time parameters and location information;

[0022] The time parameters include year, month, day, hour, and the day of the year, and also include sin(Month*π / 12), cos(Month*π / 12), sin(DOY*π / 365), and cos(DOY*π / 365) obtained by performing trigonometric function processing on the month and DOY parameters respectively, a total of 9 parameters; where Month represents the month and DOY represents the day of the year;

[0023] The position information includes longitude and latitude, and also includes sin(Lat*π / 180), cos(Lat*π / 180), sin(Lon*π / 180), and cos(Lon*π / 180) obtained by performing trigonometric function processing on longitude and latitude respectively, for a total of 6 parameters; where Lon represents longitude and Lat represents latitude.

[0024] In step S5, the input of the non-linear mapping model includes time parameters, position parameters, and ocean multi-parameters; the output of the non-linear mapping model is the temperature at different depths, and the input and output correspond one-to-one according to time and position; the depth division of the temperature output by the non-linear mapping model refers to the division standard of the WOA18 climatological temperature depth.

[0025] In step S5, both the non-linear mapping model and the temperature inversion model are constructed using the random forest algorithm; among them, the non-linear mapping model is constructed in layers, and independent non-linear mapping models are established at different depths to be applicable to the inversion of temperatures at different depths.

[0026] An ocean internal temperature inversion system that fuses PIES and multi-source sea surface parameters, including:

[0027] A data acquisition module for acquiring the round-trip acoustic propagation time τ measured by PIES from the seabed to the sea surface b and the seabed pressure data p b ;

[0028] A data preprocessing module for preprocessing the τ measured by PIES b and correcting τ b to the acoustic propagation time τ from the bottom of the thermocline to the sea surface tb-pies ;

[0029] An acoustic propagation time calculation module for obtaining the CTD observed temperature-salinity profile and calculating the acoustic propagation time τ from the bottom of the thermocline to the sea surface for each temperature-salinity profile tb-ctd ;

[0030] A parameter matching module for matching τ tb-ctd with ocean multi-parameters according to the principle of spatio-temporal proximity;

[0031] A model training module that uses the combination of τ tb-ctd , ocean multi-parameters, time, and position multi-parameters as the input data of the inversion model, and the temperatures measured by CTD at different depths as the output data of the inversion model, constructs a training dataset for training the inversion model using the input data and output data; builds an inversion model using the random forest algorithm and trains the inversion model with the training dataset, and finally obtains an inversion model that can reflect the non-linear relationship between the multi-parameter input of the model and the ocean temperature of the model output;

[0032] A temperature inversion module, which is used for the ocean area to be inverted, and obtains the acoustic propagation time τ from the bottom of the thermocline to the sea surface of the ocean area to be inverted through steps S1 to S2 tb-pies , and matches τ according to the principle of spatio-temporal proximity tb-pies with the ocean multi-parameters of the ocean area to be inverted, and combines τ tb-pies , ocean multi-parameters, time, and position multi-parameters into a PIES inversion data set, and inputs it into the trained inversion model to output the seawater temperature profile corresponding to the time and position, so as to realize the inversion of seawater temperature.

[0033] The present invention has the following beneficial effects and advantages:

[0034] 1. The present invention incorporates information such as sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth into the seawater temperature inversion based on PIES by using the random forest algorithm, overcomes the dilemma of insufficient data in the construction process of traditional multi-parameter models, and improves the accuracy of the PIES inversion temperature;

[0035] 2. The present invention uses a random forest to construct the non-linear relationship between acoustic propagation time, sea surface observations, etc. and seawater temperature, and realizes a higher-precision capture of the non-linear variation of seawater temperature;

[0036] 3. The seawater temperature inversion based on PIES in the present invention can realize the observation of the long-term, continuous and multi-scale ocean thermal structure compared with traditional observation means, thereby promoting the long-term, efficient and fine monitoring and research of the internal dynamic processes of the ocean. Description of the Drawings

[0037] Figure 1 The inversion flow chart of the present invention.

[0038] Figure 2 The comparison chart of the inversion temperature accuracy between the GEM method and the Mult-RF method. Specific Embodiments

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] The present invention relates to a method for inverting the ocean internal temperature by integrating a Pressure Inverse Echo Sounder (PIES) with multi-source sea surface parameters, belonging to the technical fields of ocean environmental parameter inversion and ocean dynamics observation. The present invention includes the following steps: using the random forest algorithm, integrating historical observed temperature and salinity profiles measured by a Conductivity Temperature Depth (CTD) instrument and multi-source sea surface parameter data, etc., to construct a non-linear empirical model (Mult-RF) for seawater temperature inversion; using the sound propagation time calculated by the PIES and multi-source sea surface data, and inversing the seawater temperature through the Mult-RF model. Compared with the traditional Empirical Mode Mapping (GEM) model, on the one hand, the Mult-RF model in the present invention uses fewer samples to establish the mapping relationship between seawater temperature and multi-parameters, overcoming the problem of lack of samples in constructing a seawater temperature inversion model that integrates multi-source sea surface parameters and the sound propagation time observed by the PIES. On the other hand, the random forest adopted by the Mult-RF model can achieve non-linear fitting from multi-parameters to seawater temperature compared with polynomial fitting in the GEM model, thereby significantly improving the accuracy of temperature inversion.

[0041] As Figure 1 shown, a method for inverting the ocean internal temperature by integrating a Pressure Inverse Echo Sounder (PIES) with multi-source sea surface parameters includes the following steps:

[0042] S1: Obtain the round-trip acoustic propagation time (τ b ) from the seabed to the sea surface measured by the PIES and the seabed pressure data (p b ), with a time resolution of 1 hour.

[0043] S2: Referring to the method of Kennelly et al., perform operations such as quality control, removing the barotropic effect, and removing seasonal variations on the τ b obtained in S1, and finally correct the τ b to the acoustic propagation time τ tb-pies from the bottom of the thermocline to the sea surface.

[0044] S3: Collect the CTD observed temperature and salinity profiles within a 2° range of the PIES observation station, and calculate the acoustic propagation time τ tb-ctd from the bottom of the thermocline to the sea surface for each temperature and salinity profile using the acoustic propagation calculation formula; the calculation formula is as follows:

[0045]

[0046] where c is the sound speed, ρ is the seawater density calculated from the CTD temperature and salinity, g is the acceleration due to gravity, and p ref is the depth of the bottom of the thermocline.

[0047] Collect grid data of sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth, with the requirement that its spatial resolution is less than or equal to 0.25°, and the time resolution is 1 day. Then, match it with τ tb-ctd according to the principle of spatio-temporal proximity, that is, for the time and location of each CTD profile observation or each PIES observation, select the values of sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth products that are at the closest distance and on the same day.

[0048] Use the combination of τ tb-ctd , sea surface height, sea surface temperature, sea surface wind field, mixed layer depth, time, and location multi-parameters as the model input, and the temperature measured by CTD at different depths as the model output to construct a training dataset. Among them, the time parameters of the model input include 5 parameters: year, month (Month), day, hour, and day of the year (DOY), and also include 4 parameters obtained by trigonometric function processing of the month and DOY parameters respectively: sin(Month*π / 12), cos(Month*π / 12), sin(DOY*π / 365), and cos(DOY*π / 365), for a total of 9 parameters. The location parameters of the model input include 2 parameters: longitude (Lon) and latitude (Lat), and also include 4 parameters obtained by trigonometric function processing of the longitude and latitude parameters respectively: sin(Lat*π / 180), cos(Lat*π / 180), sin(Lon*π / 180), and cos(Lon*π / 180), for a total of 6 parameters. The depth division of the output temperature of the model refers to the standardized depth division with reference to the division standard of the WOA18 climatological temperature depth (57 depth layers within 1500 meters and 47 depth layers within 1000 meters). Match the model input data and the model output data according to time and location respectively to construct the model training dataset. Finally, use the random forest algorithm to hierarchically construct a non-linear mapping model (Mult-RF) from the model input to the model output, where the hyperparameters of the random forest are set as follows: the number of trees is 100, and the number of leaves of the tree is 8.

[0049] S4: Match τ tb-pies with multi-parameters (including sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth) according to the principle of spatio-temporal proximity.

[0050] S5: Use the combination of τ tb-ctd , sea surface height, sea surface temperature, sea surface wind field, mixed layer depth, time, and location multi-parameters as the model input, and the temperature measured by CTD at different depths as the model output to construct the model training dataset, and use the random forest algorithm to hierarchically construct a non-linear mapping model (Mult-RF) from the model input to the model output;

[0051] S6: Match τ according to the principle of spatio-temporal proximitytb-pies with multiple parameters (including sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth), and using τ tb-pies , sea surface height, sea surface temperature, sea surface wind field, mixed layer depth, time, and location to form a PIES inversion dataset, and input it into Mult-RF to output the seawater temperature profile corresponding to the time and location, thus realizing the inversion of seawater temperature.

[0052] In the actual application process, the method for inverting the ocean internal temperature by fusing the pressure-sensing inverse echo sounder (PIES) and multi-source sea surface parameters is as follows:

[0053] Collect the two-way acoustic propagation time (τ b ) from the seabed to the sea surface and the seabed pressure data (p b ) observed by PIES at 56 stations in the Kuroshio Extension region. The time resolution is 1 hour, and the time coverage is from April 2004 to July 2006. Referring to the method of Kennelly et al., perform quality control, remove the barotropic effect, remove seasonal variations, etc. on the τ b measured by PIES at each station, and finally correct τ b to the acoustic propagation time τ tb-pies from the bottom of the thermocline to the sea surface.

[0054] Collect the CTD observed temperature and salinity profiles within a 2° range of each PIES observation station, a total of 10,309 profiles, and use the acoustic propagation calculation formula to calculate the acoustic propagation time τ tb-ctd from the bottom of the thermocline to the sea surface for each temperature and salinity profile.

[0055] Collect the grid data of sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth. Among them, the sea surface height data is the AVISO+ daily height grid data, the sea surface temperature is the daily sea surface temperature grid product measured by microwave of the Remote Sensing System (RSS), the sea surface wind field is the CCMP daily sea surface wind field fusion product, and the mixed layer depth is the daily average mixed layer depth grid product of CMEMS (Global Monitoring and Forecasting Centre). The spatial resolution of these products is 0.25 degrees, and then match τ tb-ctd with multiple parameters (including sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth) according to the principle of spatio-temporal proximity.

[0056] Use the τ tb-ctdThe combination of the corresponding sea surface height, sea surface wind field, sea surface temperature, mixed layer depth, time, and position parameters is used as the model input, and the temperatures at different depths are used as the model output. The inversion models at different depths are trained to obtain the Mult-RF models for each depth.

[0057] Similarly, 56 stations are matched according to the principle of spatio-temporal proximity for τ tb-pies with multiple parameters (including sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth), and τ tb-pies , the combination of sea surface height, sea surface temperature, sea surface wind field, mixed layer depth, time, and position multi-parameters is input into Mult-RF, and the sea water temperatures inverted at 56 PIES stations are output. It is the monitoring of the long-term state change of the sea water temperature at a certain time period at the positions of each PIES station.

[0058] To compare the advantages and disadvantages of the Mult-RF method of the present invention with the traditional GEM method, the GEM method is similarly used for inversion, and the CTD temperature profiles (a total of 582 profiles) near 56 stations are selected to compare the inversion results of the two methods, as Figure 2 shown. The Mult-RF method proposed by the present invention shows higher inversion accuracy than the GEM method at each depth. RMSE represents the root mean square error.

[0059] In this example, taking the Kuroshio Extension as an example, first, a Mult-RF model of sound propagation time + sea surface multi-parameters and sea water temperatures at different depths is constructed through CTD temperature-salinity profiles and sea surface observations, and then the sound propagation time + sea surface multi-parameters are combined with the PIES array in the study area to achieve high-precision inversion of the temperature in the study area, which has high application value and popularization prospects.

[0060] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An ocean internal temperature inversion method integrating PIES and multi-source sea surface parameters, characterized in that, It includes the following steps: S1. Obtain the round-trip acoustic wave propagation time τ from the seabed to the sea surface measured by PIES b and the seabed pressure data p b ; S2. Preprocess the τ measured by PIES b and correct τ b to the acoustic propagation time τ from the bottom of the pycnocline to the sea surface tb-pies ; S3. Obtain the CTD-observed temperature-salinity profile and calculate the sound propagation time τ from the bottom of the thermocline to the sea surface for each temperature-salinity profile tb-ctd ; S4. Match τ according to the spatio-temporal proximity principle tb-ctd With ocean multi-parameters; S5. Utilize τ tb-ctd The combination of ocean multi-parameters, time, and location multi-parameters is used as the input data of the inversion model, and the temperature measured by CTD at different depths is used as the output data of the inversion model. The training data set for training the inversion model is constructed using the input data and output data; the inversion model is built using the random forest algorithm and trained with the training data set to finally obtain an inversion model that can reflect the non-linear relationship between the multi-parameter input of the model and the ocean temperature of the model output; S6. For the ocean area to be inverted, obtain the sound propagation time τ from the bottom of the thermocline to the sea surface of the ocean area to be inverted through steps S1 - S2 tb-pies , and match τ tb-pies with the ocean multi-parameters of the ocean area to be inverted, and combine τ tb-pies , ocean multi-parameters, time, and location multi-parameters into a PIES inversion dataset, and input it into the trained inversion model to output the seawater temperature profile corresponding to the corresponding time and corresponding location, so as to realize the inversion of the seawater temperature.

2. The ocean internal temperature inversion method integrating PIES and multi-source sea surface parameters according to claim 1, wherein In step S3, for obtaining the CTD observed temperature and salinity profile, select the CTD observation data within 2° of the PIES observation station location.

3. The method for inverting the ocean internal temperature by integrating PIES and multi-source sea surface parameters according to claim 1, characterized in that, The marine multi-parameters include sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth.

4. A method for retrieving ocean internal temperature by fusing PIES and multi-source sea surface parameters according to claim 1, characterized in that, In step S3, the acoustic propagation time τ from the bottom of the thermocline to the sea surface tb-ctd is calculated as follows: Among them, c is the speed of sound, ρ is the seawater density calculated from the CTD temperature and salinity, g is the acceleration due to gravity, p ref is the depth at the bottom of the thermocline, and p represents the pressure.

5. A method for retrieving ocean internal temperature by fusing PIES and multi-source sea surface parameters as described in claim 1, characterized in that, In steps S4 and S6, the marine multi-parameters are grid data, and the spatial resolution is less than or equal to 0.25°, and the time resolution is 1 day.

6. The ocean internal temperature inversion method integrating PIES and multi-source sea surface parameters according to claim 1, characterized in that In steps S4 and S6, the matching according to the spatio-temporal proximity principle is as follows: for the time and location of each CTD profile observation or each PIES observation, select the values of the sea surface height, sea surface temperature, sea surface wind field, and mixed layer depth products that are at the closest distance and on the same day for matching.

7. A method for inverting the ocean internal temperature by integrating PIES and multi-source sea surface parameters according to claim 1, characterized in that, In step S5, the input of the non-linear mapping model includes time parameters and location information; The time parameters include year, month, day, hour, and the day of the year, and also include sin(Month*π / 12), cos(Month*π / 12), sin(DOY*π / 365), and cos(DOY*π / 365) obtained by performing trigonometric function processing on the month and DOY parameters respectively, a total of 9 parameters; where Month represents the month and DOY represents the day of the year; The location information includes longitude and latitude, and also includes sin(Lat*π / 180), cos(Lat*π / 180), sin(Lon*π / 180), and cos(Lon*π / 180) obtained by performing trigonometric function processing on the longitude and latitude respectively, a total of 6 parameters; where Lon represents longitude and Lat represents latitude.

8. A method for inverting the ocean internal temperature by integrating PIES and multi-source sea surface parameters according to claim 1, characterized in that, In step S5, the input of the non-linear mapping model includes time parameters, location parameters, and marine multi-parameters; the output of the non-linear mapping model is the temperature at different depths, and the input and output correspond one by one according to time and location; the depth division of the temperature output by the non-linear mapping model refers to the division standard of the WOA18 climatological temperature depth.

9. The ocean internal temperature inversion method integrating PIES and multi-source sea surface parameters according to claim 1, characterized in that, In step S5, both the non-linear mapping model and the temperature inversion model are constructed using the random forest algorithm; among them, the non-linear mapping model is constructed in layers, and independent non-linear mapping models are established at different depths to be applicable to the inversion of temperatures at different depths.

10. An ocean internal temperature inversion system integrating PIES and multi-source sea surface parameters, characterized in that, It includes: A data acquisition module for acquiring the round-trip acoustic wave propagation time τ measured by PIES from the seabed to the sea surface b and the seabed pressure data p b ; A data preprocessing module for preprocessing τ measured by PIES b and correcting τ b to the acoustic propagation time τ from the bottom of the pycnocline to the sea surface tb-pies ; The sound propagation time calculation module is used to obtain the CTD observed temperature-salinity profile and calculate the sound propagation time τ from the bottom of the thermocline to the sea surface for each temperature-salinity profile tb-ctd ; A parameter matching module for matching τ according to the principle of spatio-temporal proximity tb-ctd and ocean multi-parameters; Model training module, using τ tb-ctd , the combination of ocean multi-parameters, time, and position multi-parameters as the input data of the inversion model, and the temperature measured by CTD at different depths as the output data of the inversion model. A training data set for training the inversion model is constructed using the input data and output data; a random forest algorithm is used to build the inversion model, and the inversion model is trained with the training data set. Finally, an inversion model that can reflect the non-linear relationship between the multi-parameter input of the model and the ocean temperature of the model output is obtained; A temperature inversion module, which is used for an ocean area to be inverted, to obtain the sound propagation time τ from the bottom of the thermocline to the sea surface of the ocean area to be inverted through steps S1 to S2 tb-pies , and match τ according to the principle of spatio-temporal proximity tb-pies with the ocean multi-parameters of the ocean area to be inverted, and combine τ tb-pies , the ocean multi-parameters, time, and location multi-parameters into a PIES inversion data set, and input it into the trained inversion model to output the sea water temperature profile corresponding to the time and the location, so as to realize the inversion of the sea water temperature.

Images