Ocean sea surface temperature prediction method based on optimal interpolation in combination with ConvLSTM

Abstract

The invention discloses an ocean sea surface temperature prediction method based on optimal interpolation in combination with ConvLSTM, and relates to the technical field of ocean remote sensing detection. The invention provides a sea surface temperature prediction method fusing an OI method and a ConvLSTM model, the OI method solves the problem of spatial integrity of data and provides high-quality input for ConvLSTM, and the ConvLSTM model exerts the learning ability of the ConvLSTM model on spatio-temporal characteristics on the basis of the data processed by the OI method and realizes accurate time sequence prediction; the sea surface temperature prediction method and the sea surface temperature prediction device are matched with each other, so that the method can not only ensure the continuity and the physical rationality of sea surface temperature data in space, but also effectively predict the change of the sea surface temperature data in a time sequence, and a more reliable and more accurate solution is provided for sea surface temperature prediction under the conditions that the marine environment is complex and the data is difficult to acquire.

Description

Technical Field

[0001] The present invention relates to the technical field of marine remote sensing detection, and particularly relates to a method for predicting the sea surface temperature of the ocean based on optimal interpolation combined with ConvLSTM. Background Technique

[0002] The sea surface temperature (SST) is the temperature at the interface between the ocean and the atmosphere. As a core variable in the air-sea interaction, its change directly affects the atmospheric circulation, extreme weather events (such as typhoons, heatwaves), and the stability of the marine ecosystem. The accurate prediction of SST is crucial for climate research, disaster warning, and fishery resource management.

[0003] In recent years, with the progress of satellite remote sensing technology, remote sensing observation has become the main means of obtaining SST. Currently, satellite observation of SST mainly relies on infrared sensors and microwave sensors. Infrared sensors have the advantage of high spatial resolution but are vulnerable to aerosol and atmospheric cloud interference; microwave sensors have strong cloud penetration ability and can achieve all-day and all-weather observation, but the spatial resolution is relatively low. Given the limitations of the observation capabilities of single sensors, satellite multi-source data prediction technology has emerged, aiming to integrate the advantages of different sensors, improve the spatio-temporal integrity and accuracy of SST products, and better meet the actual application requirements.

[0004] Optimal Interpolation (OI) is a method for predicting SST data. This method fully considers the spatial distribution characteristics of each pixel, interpolates unevenly distributed data to the corresponding grid points, and thus obtains the optimal estimated value, effectively solving the problem of sparse spatial distribution of ocean data. However, in the process of constructing the background field and weighting in the traditional OI method, the spatio-temporal correlation of the observed data has not been fully considered, which limits its applicability and accuracy improvement in high-dynamic sea areas.

[0005] In recent years, the application of deep learning in the field of marine remote sensing has gradually emerged. Among them, the Convolutional Long Short-Term Memory (ConvLSTM) has the ability to extract spatial features of convolutional neural networks and the ability to model time series of LSTM at the same time, and is suitable for processing remote sensing data with strong spatio-temporal dependence.

[0006] In view of this, the problem faced by the existing technology is how to combine the optimal interpolation method with ConvLSTM, give full play to the advantages of OI in constructing a physically consistent background field and suppressing observation noise, and at the same time utilize the spatio-temporal learning ability of ConvLSTM to dynamically capture the spatio-temporal evolution law of SST, so as to improve the prediction accuracy and stability of the missing measurement area. By combining these two, it is expected to solve the limitations of traditional OI methods in the application of high-dynamic sea areas and improve the accuracy and stability of SST data prediction.

[0007] The information disclosed in this background art section is only intended to increase the understanding of the overall background of the present invention and should not be regarded as an admission or any form of suggestion that this information constitutes prior art already known to those of ordinary skill in the art. Summary of the Invention

[0008] In view of the above technical problems, an embodiment of the present invention provides an ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM to solve the problems mentioned in the above background art.

[0009] The present invention provides the following technical solutions: An ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM, comprising the following steps: Step 1: Obtain AMSR2, MODIS, HY1C, and HY1D satellite data, as well as OSTIA reanalysis data and iQuam buoy data in sequence; construct a multi-source sea surface temperature data set; Step 2: Perform quality control (preprocessing) on the multi-source sea surface temperature data set to eliminate abnormal or invalid sea surface temperature data; Step 3: Use the OSTIA reanalysis data as the initial background field, and use the optimal interpolation algorithm to interpolate the sea surface temperature data of AMSR2, MODIS, HY1C, and HY1D in sequence to generate a standard grid sea surface temperature product; Step 4: Divide the sea surface temperature grid product after optimal interpolation into a training set, a validation set, and a test set, and perform normalization processing; Step 5: Construct an OI-ConvLSTM model, use the normalized training set and validation set to train the model, and obtain the optimal OI-ConvLSTM model; Step 6: Use the optimal OI-ConvLSTM model to predict the sea surface temperature of the normalized test set, and perform anti-normalization processing to obtain real sea surface temperature prediction data; Step 7: Use the iQuam buoy data to evaluate the accuracy of the predicted sea surface temperature product.

[0010] Preferably, the quality control method includes: reading the sea surface temperature variables in AMSR2, MODIS, HY1C, and HY1D, removing the sea surface temperature values less than 0°C or greater than 40°C, setting the filling values of -999.0 and -32767.0 to 0, and using the variable representing the quality level to mask the pixel values with poor quality.

[0011] Preferably, in step 3, the optimal interpolation algorithm mainly includes: the background field error covariance matrix B and the observation field error covariance matrix R.

[0012] Preferably, the background field error covariance matrix consists of a diagonal matrix D composed of background field errors and a background field horizontal correlation matrix ; The formula expression is: Among them, D is the diagonal matrix composed of background field errors, is the forecast background field horizontal correlation matrix, a and b are the correlation scales in the longitude and latitude directions respectively, are the correlation distances in the longitude and latitude directions respectively.

[0013] Preferably, the same method as the background field error covariance matrix is adopted for the observation field error covariance matrix; the calculation formula is: Among them, i and j represent different observation points.

[0014] Preferably, in step 4, the gridded product of the sea surface temperature after optimal interpolation is divided into a training set, a validation set, and a test set according to the ratio of 7:1:2; The input data dimension of the model is 5, including: batch size, time step, number of channels, latitude, and longitude; The output dimension of the label data is 4, including: batch size, latitude value, longitude value, and channel size.

[0015] Preferably, in step 4, the normalization processing formula is as follows: ; In the formula, is the normalized value, is the maximum value of the sea surface temperature in the sample, is the minimum value of the sea surface temperature in the sample, is the true sea surface temperature value in the sample.

[0016] Preferably, the OI-ConvLSTM model structure includes: three stacked ConvLSTMs; each layer uses a 3×3 convolutional kernel and 64 filters to capture the spatio-temporal evolution characteristics of SST through temporal information; each layer of the ConvLSTM model includes: an input gate, a forget gate, a memory unit, and an output gate.

[0017] Specifically, the input gate, forget gate, memory cell, and output gate are expressed as follows: Among them, is the input gate, is the forget gate, is the memory cell, is the output gate, is the sigmoid activation function, is the hyperbolic tangent activation function, is the input at the current time step, and are the hidden states at times t and t-1 respectively, is the memory cell at the previous time step, W and b are the weight matrix and bias matrix parameters respectively, represents the product, and * represents the convolution operation.

[0018] Preferably, in step 6, the predicted sea surface temperature is de-normalized to obtain the true sea surface temperature prediction data; the de-normalization formula is: ; In the formula, is the true sea surface temperature value in the sample, is the value after normalization, is the maximum value of the sea surface temperature in the sample, is the minimum value of the sea surface temperature in the sample.

[0019] Preferably, in step 7, the accuracy of the predicted sea surface temperature product is evaluated. The accuracy evaluation indicators include: correlation coefficient , mean relative error , mean bias , mean standard deviation ; the specific formulas are as follows: Among them, X is the observed value, X ’ is the reference value, and N is the number of matching data points.

[0020] The ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM provided by the embodiments of the present invention has the following beneficial effects: (1) The present invention proposes a sea surface temperature prediction method that combines the OI method and the ConvLSTM model. The OI method solves the problem of spatial integrity of data and provides high-quality input for ConvLSTM. Based on the data processed by the OI method, the ConvLSTM model exerts its learning ability for spatio-temporal features to achieve accurate time series prediction; (2) The two cooperate with each other, enabling the method to not only ensure the spatial continuity and physical rationality of sea surface temperature data, but also effectively predict its changes in the time series. In the case of complex marine environments and difficult data acquisition, it provides a more reliable and accurate solution for sea surface temperature prediction. Description of the Drawings

[0021] Figure 1 It is a flowchart of the steps of a method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM provided by an embodiment of the present invention; Figure 2 It is a structural diagram of the optimal interpolation algorithm model described in an embodiment of the present invention; Figure 3 It is a structural diagram of the OI-ConvLSTM algorithm model described in an embodiment of the present invention. Specific Embodiments

[0022] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0023] Design principle of a method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM of the present invention: (1) Limitations of multi-source data: When satellite remote sensing is used to obtain sea surface temperature (SST), the infrared sensor has high resolution but is easily interfered by clouds, while the microwave sensor can observe all-weather but has low resolution, resulting in discontinuous and unevenly distributed data in space. For example, in areas covered by clouds, the infrared sensor cannot obtain effective data, resulting in data missing.

[0024] (2) Role of optimal interpolation (OI): The OI method uses OSTIA reanalysis data as the initial background field. The OSTIA data itself has integrated various data and has certain spatial continuity and physical basis; on this basis, the OI method calculates the background field error covariance matrix and the observation field error covariance matrix, fully considering the spatial distribution characteristics of each pixel, and interpolates the unevenly distributed data in multi-source observation data such as AMSR2 and MODIS to the corresponding grid points according to a certain weight; thereby constructing a physically consistent and spatially complete background field, effectively alleviating the spatial missing problem in multi-source remote sensing data, and making the originally discrete and discontinuous data form a continuous gridded SST product.

[0025] (3)Features of ConvLSTM: ConvLSTM combines the spatial feature extraction ability of convolutional neural network (CNN) and the temporal modeling ability of long short-term memory network (LSTM); when processing sea surface temperature data, its convolutional structure can extract the spatial features of SST data, such as the shape of temperature distribution, gradient change, etc.; the LSTM structure is good at processing time series data, can remember the information of past time steps, and update the memory according to the current input, so as to capture the evolution law of SST over time, such as seasonal changes, short-term fluctuations, etc.

[0026] (4)Advantages of combining ConvLSTM with OI data: Using the spatially continuous gridded SST product generated by the OI method as the input of ConvLSTM provides high-quality, continuous spatio-temporal data for the model; ConvLSTM learns and trains based on these data, can better mine the dependence relationship of SST in the time dimension, and achieve accurate prediction of SST at future time points, thus endowing the method with better time series prediction ability.

[0027] Second, in view of the problems mentioned in the above background technology, the embodiments of the present invention provide an ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM to solve the above technical problems, and its technical solutions are as follows: The following combines the attached Figures 1-3 , and the specific implementation manners to further illustrate the present invention.

[0028] First, install the python programming software on the user terminal, and it is necessary to equip the Python-3.10, tensoR, flow-2.16.1, keras-3.0.5 environment packages. tensoR, flow, and keras are open-source software libraries for deep learning; tensoR, flow provides a flexible platform that can be used to build, train, and deploy various complex neural network models, and it supports multiple programming languages; keras is a high-level neural network API that can provide a more concise and friendly interface based on tensoR, flow, allowing users to quickly build and run common neural network models.

[0029] Figure 1 As shown; an ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM; includes the following steps: Step S1, obtain AMSR2, MODIS, HY1C, HY1D satellite data, OSTIA reanalysis data, iQuam buoy data, and construct a multi-source sea surface temperature data set; Step S2, perform quality control on the obtained multi-source sea surface temperature data set; Step S3: Using the OSTIA reanalysis data as the initial background field for the preprocessed multi-source sea surface temperature data set, interpolation is performed on the ocean temperature observation data using the optimal interpolation method to generate a standard gridded sea surface temperature product; Step S4: The gridded sea surface temperature product after optimal interpolation is divided into a training set, a validation set, and a test set; Step S5: Build a ConvLSTM model and use the training set and the validation set to train the model to obtain the optimal ConvLSTM model; Step S6: Use the optimal ConvLSTM model to predict the sea surface temperature using the test set, and use the iQuam buoy data to evaluate the accuracy of the predicted sea surface temperature product.

[0030] In this embodiment, in step S1, AMSR2, MODIS, HY1C, HY1D satellite data, OSTIA reanalysis data, and iQuam buoy data are obtained to construct a multi-source sea surface temperature data set. Specifically as follows: Regarding MODIS satellite data, MODIS is a moderate-resolution imaging spectrometer carried on the AQUA\TERRA satellites. TERRA is the morning satellite with a transit time of around 10:30 am, and AQUA is the afternoon satellite with a transit time of around 1:30 pm. It captures data in 36 spectral bands at various spatial resolutions. For the SST product, it is generated using an improved non-linear SST algorithm with the 4, 11, and 12-micron infrared bands. The MODIS sea surface temperature data used in the present invention is from Ocean Color (https: / / oceancolor.gsfc.nasa.gov), the product level is L3 Mapped data, the spatial resolution is 4 km, this product is global single-day data that has been calibrated and radiometrically corrected, has been projected onto the WGS84 coordinate system, and has undergone strict quality control.

[0031] Regarding AMSR-2 satellite data, AMSR-2 is an advanced microwave scanning radiometer carried on the Global Change Observation Mission - Water (GCOM-W) satellite developed by JAXA. The transit time is 1:30 pm, the antenna rotates once every 1.5 seconds, it can obtain data with a swath width of 1450 km, and it can obtain day and night data covering more than 99% of the Earth every two days. The AMSR-2 sea surface temperature data used in the present invention comes from Remote Sensing Systems (RSS) in the United States, the product level is L3, and the spatial resolution is 0.25°X 0.25°.

[0032] Regarding HY1C and HY1D satellite data, HY1C / 1D is an ocean remote sensing satellite of my country's HY-1 series satellites in orbit, equipped with an ocean color scanner COCTS. The two thermal infrared channels on this sensor can be used for sea surface temperature SST observation. The HY1C\1D sea surface temperature data used in this paper comes from the National Satellite Ocean Application Center (https: / / osdds.nsoas.org.cn / ), which can be obtained free of charge, with a product level of L3A and a spatial resolution of 4km.

[0033] For iQuam data, iQuam data is developed by NOAA, and the data set includes: Argo buoys, traditional drifting, high-resolution drifting, tropical mooring, coastal mooring, coral reef observation buoys, traditional ships, and ships of the International Maritime Organization. iQuam data is stored in NetCDF4 format, stored as a data file every month, and updated every 24 hours. The data file is divided into two parts: global attributes and data variables. The present invention selects data with observation types of ships, drifting buoys, tropical moored buoys, coastal moored buoys, and high-resolution drifting buoys to perform accuracy verification of fusion products.

[0034] For OSTIA analysis data (sea surface temperature and ice analysis data), this data is a global sea surface temperature and sea ice analysis product developed by the UK Met Office, which combines satellite data (AVHRR, AMSR, TMI, AATSR and SEVIRI) and field observation data to generate high-resolution daily analysis data of sea surface temperature and sea ice area fraction, with a grid spacing of 1 / 20°. The data is obtained from the website https: / / www.ncei.noaa.gov / data / oceans / ghrsst / L4 / GLOB / UKMO / OSTIA / .

[0035] In this embodiment, in step S2, the quality control is performed on the acquired multi-source sea surface temperature dataset. The specific method of quality control is as follows: Based on the L3 sea surface temperature data from AMSR2, MODIS, HY1C, and HY1D, the time range is June 1, 2024 - December 31, 2024, and the spatial range is 104°E-122°E, 10°N-26°N. First, read the sea surface temperature variables in each data file, remove the sea surface temperature values less than 0°C or greater than 35°C, and then set the -999.0 and -32767.0 fill values to 0. Next, use the variable representing the quality level to mask the pixel values with poor quality, and finally obtain a high-quality single-day multi-source sea surface temperature dataset.

[0036] In this embodiment, in the step S3, for the preprocessed multi-source sea surface temperature data set, the OSTIA reanalysis data is used as the initial background field, and the optimal interpolation method is used to interpolate the ocean temperature observation data to generate a standard gridded sea surface temperature product. The optimal interpolation method is an objective analysis method. In the process, the analysis value at the spatial grid point is obtained by weighting the deviation of the observation point relative to the background field, and the weight coefficient should minimize the analysis error of the grid point and is not arbitrarily selected. The specific method is as follows: In the present invention, the preprocessed 0.1°X0.1° OSTIA reanalysis data is used as the initial background field, the 0.1°X0.1° AMSR2 sea surface temperature data is used as the observation field, the study area is 104°E - 122°E, 10°N - 26°N, and optimal interpolation fusion is performed to finally generate a 0.1°X0.1° sea surface temperature product (OI-SST) for sea surface temperature prediction. The specific formula for optimal interpolation is: ; Wherein, represents the analysis value of the variable at the spatial grid point, represents the background field value of the variable at the spatial grid point, is the weight coefficient, represents the observed value of the variable at the observation point, represents the background field value of the variable at the observation point, is the observation operator. The weight coefficient matrix has the expression: ; Wherein, represents the transpose of the matrix, represents the background field error covariance matrix, represents the observation error covariance matrix. To calculate the weight coefficient , it is necessary to first estimate the background field error covariance matrix and the observation error covariance matrix .

[0037] (1) Calculate the background field error covariance matrix

[0038] The background field error covariance matrix consists of the diagonal matrix D composed of the background field errors and the background field horizontal correlation matrix and is expressed by the formula: and is expressed by the formula: Wherein, D is the diagonal matrix composed of the background field errors, is the forecast background field horizontal correlation matrix, a and b are the correlation scales in the longitude and latitude directions respectively, are the correlation distances in the longitude and latitude directions respectively.

[0039] (2) Calculate the observation field error covariance matrix The same method as that for the background field error covariance matrix is adopted for the observation field error covariance matrix. It is generally considered that the observation errors at different positions are uncorrelated, and the calculation formula is as follows: where i and j represent different observation points.

[0040] (3) Calculate the weights Based on the obtained background field error covariance matrix and the observation field error covariance matrix The coefficient matrix for optimal interpolation is obtained by solving equations , and the sea surface temperature after fusing OSTIA and AMSR2 is calculated.

[0041] Subsequently, the result of this optimal interpolation is used as the background field for the next optimal interpolation. The sea surface temperature data of MODIS, HY1C, and HY1D are sequentially introduced as the observation field, and the optimal interpolation process is repeated to gradually optimize the spatial distribution and accuracy of the sea surface temperature field, and the final sea surface temperature fusion product is obtained.

[0042] In this embodiment, in the step S4, the gridded product of the sea surface temperature after optimal interpolation is divided into a training set, a validation set, and a test set, and normalization processing is performed. Specifically as follows: First, select the OI-SST data from January 1, 2024 to December 31, 2024, and divide it into a training set, a validation set, and a test set according to the ratio of 7:1:2. The input data of the model has 5 dimensions, including: batch size B, time step T, number of channels C, latitude H, and longitude W; the output dimension of the label data is 4, namely batch size B, channel size C, latitude value H, and longitude value W, to ensure that 5 consecutive time series output the prediction effect of 1 time series.

[0043] Secondly, normalization processing is performed on the divided data set before inputting it into the model. The purpose is to accelerate convergence, avoid gradient disappearance, prevent the feature dominant effect, improve numerical stability, and better adapt to the weight initialization of the network. The formula is as follows: In the formula, is the value after normalization, is the maximum value of the sea surface temperature in the sample, is the minimum value of the sea surface temperature in the sample, is the true sea surface temperature value in the sample.

[0044] Specifically, in the step S5, an OI-ConvLSTM model is constructed and trained using a training set and a validation set to obtain an optimal OI-ConvLSTM model. Specifically as follows: (1) Construct the OI-ConvLSTM model As Figure 3 shown, the OI-ConvLSTM model developed by the present invention using the pytorch framework in the python 3.10 environment takes the optimally interpolated sea surface temperature (OI-SST) as the input, and is composed of three stacked ConvLSTMs. Each layer uses a 3×3 convolutional kernel and 64 filters to capture the spatio-temporal evolution characteristics of SST through temporal information. Each layer of the ConvLSTM model includes: an input gate, a forget gate, a memory unit, and an output gate. The expressions are as follows: Among them, is the input gate, is the forget gate, is the memory unit, is the output gate, is the sigmoid activation function, is the hyperbolic tangent activation function, is the input at the current time step, and are the hidden states at times t and t-1 respectively, is the memory unit at the previous time step, W and b are the weight matrix and bias matrix parameters respectively, represents the product, and * represents the convolution operation.

[0045] (2) Model training Take the divided training set as the input of the OI-ConvLSTM model. The input tensor size is (B, T, C, H, W), and the network output is a single SST image predicted at the next moment (the (T+1)-th day). The output tensor size is (B, C, H, W). During the training process of this model, the adaptive moment estimation (Adam) optimization algorithm is adopted, and the mean square error MSE is selected as the loss function. Record the training loss of each Epoch, and terminate the training when the loss function does not improve. The obtained model is used as the optimal model for sea surface temperature prediction. The formula for the mean square error is: Among them, is the true value of the i-th data in a batch, is the predicted value given by the model.

[0046] The other parameter settings are as follows: sequence length seq_length = 7, prediction length pred_len = 1, batch size batch_size = 4, number of training epochs num_epochs = 500, learning rate learning_rate = 0.001.

[0047] Specifically, in step S6, the optimal ConvLSTM model is used to predict the sea surface temperature using the test set, and the iQuam buoy data is used to evaluate the accuracy of the predicted sea surface temperature product. The specific steps for accuracy evaluation are as follows: (1) Model prediction First, the divided test set is used as the input of the optimal ConvLSTM model to obtain the normalized predicted sea surface temperature, and then the denormalization operation is performed and the projection output is carried out to obtain the final sea surface temperature data. The denormalization formula is: In the formula, is the true sea surface temperature value in the sample, is the value after normalization, is the maximum value of the sea surface temperature in the sample, is the minimum value of the sea surface temperature in the sample.

[0048] (2) Accuracy evaluation Using the iQuam data, the accuracy of the predicted sea surface temperature is tested. The spatio-temporal matching principle is as follows: ① The time window is set to the current day; ② The spatial window is 3×3 pixels, and the mean value of the adjacent 3×3 pixels of the sampling point is taken as the sea surface temperature value of this point; ③ The effective pixels in the window are greater than 50%; ④ The pixels outside X±(3×σ) are excluded; ⑤ Calculate the coefficient of variation (CV) CV = σ / X. If CV≤0.15, then the uniformity identification is passed. The following 4 evaluation indicators are used, namely the correlation coefficient r, the mean absolute relative error MARE, the mean bias Bias, and the mean standard deviation STD. Their calculation formulas are as follows: Among them, Bias is the mean bias, STD is the mean standard deviation, r is the correlation coefficient, MARE is the mean absolute relative error, X is the observed value, X ’ is the reference value, and N is the number of matching data points.

[0049] II. Test results Based on the optimal sea surface temperature fusion method obtained from the above training, in this test, AMSR2, MODIS, HY1C, and HY1D satellite data in the range of 104°E - 122°E and 10°N - 26°N in January 2024, as well as OSTIA reanalysis data, were selected for fusion. The iQuam buoy data in January 2024 was used to evaluate the accuracy of the sea surface temperature fusion results. Following the above spatio-temporal matching rules, the final number of matched data points between the iQuam buoy data and the sea surface temperature fusion data was 188,183, the correlation coefficient r was 0.99, the average relative error was 1.45%, the average deviation was -0.21°C, and the standard deviation was 0.48°C. This indicates that the sea surface temperature fusion product obtained by using the sea surface temperature fusion algorithm provided by the present invention has high accuracy and meets the requirements of operationalization.

[0050] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or deformations that can be made without creative efforts by those skilled in the art are still within the protection scope of the present invention.

Claims

1. A method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM, characterized in that, Including the following steps: Step 1: Successively obtain AMSR2, MODIS, HY1C, and HY1D satellite data, as well as OSTIA reanalysis data and iQuam buoy data; construct a multi-source sea surface temperature dataset; Step 2: Conduct quality control on the multi-source sea surface temperature dataset to eliminate abnormal or invalid sea surface temperature data; Step 3: Using the OSTIA reanalysis data as the initial background field, sequentially interpolate the sea surface temperature data of AMSR2, MODIS, HY1C, and HY1D using the optimal interpolation algorithm to generate a standard gridded sea surface temperature product; Step 4: Divide the sea surface temperature gridded product after optimal interpolation into a training set, a validation set, and a test set, and perform normalization processing; Step 5: Construct an OI-ConvLSTM model, use the normalized training set and validation set to train the model, and obtain the optimal OI-ConvLSTM model; Step 6: Use the optimal OI-ConvLSTM model to predict the sea surface temperature of the normalized test set, and perform denormalization processing to obtain the true sea surface temperature prediction data; Step 7: Use the iQuam buoy data to evaluate the accuracy of the predicted sea surface temperature product.

2. The method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM according to claim 1, wherein In Step 2, the methods of quality control include: reading the sea surface temperature variables in AMSR2, MODIS, HY1C, and HY1D, eliminating sea surface temperature values less than 0°C or greater than 40°C, setting the filling values of -999.0 and -32767.0 to 0, and using the variable representing the quality level to mask the pixel values with poor quality.

3. The ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM according to claim 1, wherein In Step 3, the optimal interpolation algorithm mainly includes: the background field error covariance matrix B and the observation field error covariance matrix R.

4. The ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM according to claim 3, wherein The background field error covariance matrix B consists of the diagonal matrix D composed of background field errors and the background field horizontal correlation matrix and is expressed by the formula: ; ; where D is a diagonal matrix composed of background field errors, is the horizontal correlation matrix of the forecast background field, and a and b are the correlation scales in the longitude and latitude directions respectively, , are the correlation distances in the longitude and latitude directions respectively.

5. The method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM according to claim 3, wherein The observation field error covariance matrix R adopts the same method as the background field error covariance matrix B; the calculation formula is: ; Among them, i and j represent different observation points.

6. The method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM according to claim 1, characterized in that In Step 4, the sea surface temperature gridded product after optimal interpolation is divided into a training set, a validation set, and a test set according to the ratio of 7:1:2; The input data dimension of the model is 5, including: batch size, time step, number of channels, latitude, and longitude; The output dimension of the label data is 4, including: batch size, latitude value, longitude value, and channel size.

7. The method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM according to claim 1, wherein In step 4, the normalization formula is as follows: ; In the formula, is the normalized value, is the maximum value of the sea surface temperature in the sample, is the minimum value of the sea surface temperature in the sample, is the true sea surface temperature value in the sample.

8. The method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM according to claim 1, characterized in that In Step 5, the OI-ConvLSTM model structure includes: three stacked ConvLSTMs; each layer uses a 3×3 convolutional kernel and 64 filters to capture the spatio-temporal evolution characteristics of SST through temporal information; each layer of the ConvLSTM model includes: an input gate, a forget gate, a memory unit, and an output gate.

9. The method for predicting ocean sea surface temperature based on optimal interpolation combined with ConvLSTM according to claim 1, characterized in that In Step 6, perform denormalization processing on the predicted sea surface temperature to obtain the true sea surface temperature prediction data; The anti-normalization formula is as follows: ; In the formula, is the true sea surface temperature value in the sample, is the normalized value, is the maximum value of the sea surface temperature in the sample, is the minimum value of the sea surface temperature in the sample.

10. The ocean sea surface temperature prediction method based on optimal interpolation combined with ConvLSTM according to claim 1, wherein, In step 7, the accuracy of the predicted sea surface temperature product is evaluated. The accuracy evaluation indicators include: correlation coefficient , mean relative error , mean bias , mean standard deviation ; The specific formula is: where X is the observed value, X ’ is the reference value, and N is the number of matched data points.

Images