Three-dimensional sea temperature space-time prediction method and El Nino event prediction method

Abstract

The invention is suitable for the field of marine meteorology, and provides a three-dimensional sea temperature space-time prediction method, which comprises the following steps: obtaining reanalysis data and numerical mode prediction data; and inputting the reanalysis data and the numerical mode prediction data into a pre-trained three-dimensional sea temperature intelligent prediction model to obtain a three-dimensional sea temperature space-time prediction result. According to the method, the spatial-temporal characteristics of the sea temperature change can be effectively captured, so that the prediction precision is improved, a high-precision sea temperature prediction result can be provided, good interpretability is achieved, and more powerful support is provided for scientific research and climate change response.

Description

Technical Field

[0001] This application belongs to the field of marine meteorological technology, and in particular relates to a three-dimensional sea surface temperature spatiotemporal prediction method and an El Niño event prediction method. Background Technology

[0002] Variations in the upper sea surface temperature of the tropical Pacific have significant impacts on marine ecosystems and the global climate system, particularly in relation to El Niño and the Southern Oscillation (ENSO). Currently, the scientific community has established a system for regular monitoring and real-time forecasting of ENSO status, primarily relying on dynamical and statistical models. These models, through physical equations and high-performance computing, can provide short-term climate predictions. However, despite significant progress, the complexity and diversity of ENSO still lead to considerable uncertainty and error in real-time forecasts, especially in predicting extreme events.

[0003] In recent years, with the development of big data and machine learning, deep learning-based methods have gradually gained attention. These methods have shown good performance in meteorological element prediction, automatically mining data features and establishing mapping relationships. However, research on ENSO prediction based on deep learning is still relatively scarce and lacks a complete theoretical framework. Existing models often only target single variables and cannot fully reflect the complex evolution of ENSO. In addition, deep learning also faces challenges in terms of interpretability and generalization ability.

[0004] Therefore, a new three-dimensional spatiotemporal prediction method for sea surface temperature is needed, which can combine reanalysis data and numerical model forecast data and utilize deep learning technology to improve the prediction capability of ENSO events. Summary of the Invention

[0005] This application provides a three-dimensional sea surface temperature spatiotemporal prediction method and an El Niño event prediction method, which can solve the above-mentioned problems.

[0006] In a first aspect, embodiments of this application provide a three-dimensional sea surface temperature spatiotemporal prediction method, including the following steps: Acquire reanalysis data and numerical model forecast data; wherein, the reanalysis data is upper sea surface temperature observation data within a preset first time period obtained from the global ocean data assimilation system, and the numerical model forecast data is upper sea surface temperature forecast data within a preset second time period predicted based on a fully coupled global climate model. The reanalysis data and the numerical model forecast data are input into a pre-trained three-dimensional intelligent sea surface temperature (SST) prediction model to obtain a three-dimensional SST spatiotemporal prediction result. The pre-trained three-dimensional intelligent SST prediction model includes at least a data fusion input module and a spatiotemporal recurrent neural network module. The spatiotemporal recurrent neural network module is trained based on a target pre-training algorithm, a target transfer learning algorithm, and a target loss function. The target loss function combines the target error of the upper ocean temperature anomaly field and the average SST anomaly index of the target ocean region.

[0007] Secondly, embodiments of this application provide an El Niño event prediction method, including the following steps: Based on the three-dimensional sea surface temperature spatiotemporal prediction method described in the first aspect above, the three-dimensional sea surface temperature spatiotemporal prediction results are obtained; Based on the three-dimensional sea surface temperature spatiotemporal prediction results and the preset El Niño event prediction algorithm, El Niño events are predicted.

[0008] Thirdly, embodiments of this application provide a three-dimensional sea surface temperature spatiotemporal prediction device, comprising: The first acquisition unit is used to acquire reanalysis data and numerical model forecast data; wherein, the reanalysis data is upper sea surface temperature observation data within a preset first time period acquired from the global ocean data assimilation system, and the numerical model forecast data is upper sea surface temperature forecast data within a preset second time period predicted based on a fully coupled global climate model. The first processing unit is used to input the reanalysis data and the numerical model forecast data into a pre-trained three-dimensional intelligent sea surface temperature prediction model to obtain a three-dimensional sea surface temperature spatiotemporal prediction result. The pre-trained three-dimensional intelligent sea surface temperature prediction model includes at least a data fusion input module and a spatiotemporal recurrent neural network module. The spatiotemporal recurrent neural network module is trained based on a target pre-training algorithm, a target transfer learning algorithm, and a target loss function. The target loss function combines the target error of the upper ocean temperature anomaly field and the average sea surface temperature anomaly index of the target ocean region.

[0009] Fourthly, embodiments of this application provide an El Niño event prediction device, comprising: The first processing unit is used to obtain the three-dimensional sea surface temperature spatiotemporal prediction result according to the three-dimensional sea surface temperature spatiotemporal prediction method described in the first aspect above. The second processing unit is used to predict El Niño events based on the three-dimensional sea surface temperature spatiotemporal prediction results and the preset El Niño event prediction algorithm.

[0010] Fifthly, embodiments of this application provide a three-dimensional sea surface temperature spatiotemporal prediction device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in the first aspect above.

[0011] In a sixth aspect, embodiments of this application provide an El Niño event prediction device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in the second aspect above.

[0012] In a seventh aspect, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0013] Eighthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the second aspect above.

[0014] In this application embodiment, the present invention provides a three-dimensional sea surface temperature (SST) spatiotemporal prediction method, aiming to address the problems of insufficient SST prediction accuracy, inadequate data utilization, and weak model generalization ability in existing technologies. By acquiring reanalysis data and numerical model forecast data, this method can fully integrate multi-source information from the global ocean data assimilation system and a fully coupled global climate model, enhancing the basic data support for the prediction model. Inputting this data into a pre-trained three-dimensional intelligent SST prediction model, utilizing a data fusion input module and a spatiotemporal recurrent neural network module, effectively captures the spatiotemporal characteristics of SST changes, thereby improving prediction accuracy. This intelligent prediction model employs a target pre-training algorithm, a target transfer learning algorithm, and a target loss function combining the target error of the upper ocean temperature anomaly field and the average SST anomaly index of the target ocean region for training, significantly enhancing the model's generalization ability in the face of scarce data and uncertainty. This series of innovative designs enables the model to not only provide high-precision SST prediction results but also possess good interpretability, providing stronger support for scientific research and climate change response. Furthermore, the real-time data input and intelligent modeling methods give this approach a significant advantage in responding to extreme climate events and achieving rapid response, further promoting the development of marine climate prediction technology.

[0015] On the other hand, the technical solution of this invention achieves efficient prediction of El Niño events by combining three-dimensional spatiotemporal sea surface temperature (SST) prediction results with a pre-defined El Niño event prediction algorithm. The core of this method lies in utilizing the accurate SST spatiotemporal prediction results obtained in previous steps. These results fully reflect the spatiotemporal variation characteristics of ocean temperature, thus providing reliable data support for the occurrence and development of El Niño events. By combining SST variation information with the established El Niño prediction algorithm, the imminent occurrence of El Niño phenomena can be more accurately identified and predicted. This integrated prediction method not only improves the accuracy and timeliness of El Niño event prediction but also provides a scientific basis for climate monitoring, disaster early warning, and the formulation of response measures, thereby effectively reducing the potential risks to the ecological environment and economic activities caused by climate anomalies resulting from El Niño. In summary, the technical solution of this invention improves the accuracy of SST and El Niño event prediction while promoting the further development of climate prediction technology. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of a three-dimensional sea surface temperature spatiotemporal prediction method provided in the first embodiment of this application; Figure 2 This is a schematic flowchart of S103~S105 in a three-dimensional sea surface temperature spatiotemporal prediction method provided in the first embodiment of this application; Figure 3 This is a schematic flowchart of steps S1021~S1022 in a three-dimensional sea surface temperature spatiotemporal prediction method provided in the first embodiment of this application; Figure 4 This is a schematic flowchart of S10211~S10215 in a three-dimensional sea surface temperature spatiotemporal prediction method provided in the first embodiment of this application; Figure 5 This is a schematic flowchart of steps S106 to S109 in a three-dimensional sea surface temperature spatiotemporal prediction method provided in the first embodiment of this application; Figure 6 This is a schematic flowchart of S111~S112 in a three-dimensional sea surface temperature spatiotemporal prediction method provided in the first embodiment of this application; Figure 7 This is a schematic flowchart of an El Niño event prediction method provided in the second embodiment of this application; Figure 8This is a schematic diagram of the three-dimensional sea surface temperature spatiotemporal prediction device provided in the third embodiment of this application; Figure 9 This is a schematic diagram of the El Niño event prediction device provided in the fourth embodiment of this application; Figure 10 This is a schematic diagram of the three-dimensional sea surface temperature spatiotemporal prediction device provided in the fifth embodiment of this application; Figure 11 This is a schematic diagram of the El Niño event prediction device provided in the third embodiment of this application. Detailed Implementation

[0018] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0019] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0020] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0021] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0022] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0024] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating a three-dimensional sea surface temperature (SST) spatiotemporal prediction method provided in the first embodiment of this application. In this embodiment, the executing entity of the three-dimensional SST spatiotemporal prediction method is a device with three-dimensional SST spatiotemporal prediction capabilities, such as a desktop computer, server, mobile device, etc. Figure 1 The three-dimensional sea surface temperature spatiotemporal prediction method shown may include: S101: Acquire reanalysis data and numerical model forecast data; wherein, the reanalysis data is upper sea surface temperature observation data within a preset first time period obtained from the global ocean data assimilation system, and the numerical model forecast data is upper sea surface temperature forecast data within a preset second time period predicted based on a fully coupled global climate model.

[0025] The equipment acquires reanalysis data. This reanalysis data consists of upper sea surface temperature (SST) observation data for a predetermined first time period, obtained from the Global Ocean Data Assimilation System. In other words, the reanalysis data is a comprehensive climate dataset generated by assimilating meteorological and oceanographic observation data. It contains long-term observation records of multiple meteorological elements, particularly upper sea surface temperature (SST) data.

[0026] Set a preset first time period, such as monthly sea surface temperature data within the past 12 months, to ensure the timeliness and continuity of the data.

[0027] The equipment acquires numerical model forecast data. This numerical model forecast data is upper sea surface temperature (SST) forecast for a predetermined second time period, based on a fully coupled global climate model. The numerical model forecast data is simulated and predicted using global climate models (such as CMIP and GFDL) to provide SST forecasts for a certain future time range.

[0028] In one implementation, the reanalysis data is the tropical Pacific upper sea surface temperature observation data over the past 12 months, and the numerical model prediction data is the tropical Pacific upper sea surface temperature forecast data for the next 20 months.

[0029] The device can select a fully coupled global climate model to ensure high simulation accuracy. A preset second time period can be set, such as sea surface temperature forecast data for the next 20 months.

[0030] In one embodiment, before S101, steps S103 to S105 are further included, such as... Figure 2 As shown, S103~S105 are as follows: S103: Obtain subsurface sea surface temperature observation data, historical restart files, and historical sea surface temperature observation data within a preset third time period from the global ocean data assimilation system.

[0031] The equipment acquires the required data from the Global Ocean Data Assimilation System, including subsurface sea surface temperature observation data for the preset third time period, historical restart files, and historical sea surface temperature observation data.

[0032] A preset third time period (e.g., data from the past 30 days) is set to ensure the timeliness of the data. Subsurface sea surface temperature (SST) observation data is typically obtained through buoys, satellite remote sensing, etc. Historical restart files generally refer to system state files used to initialize or restart numerical simulations during ocean data reanalysis. Historical SST observation data, which records ocean temperature data, can be obtained from meteorological agencies or oceanographic research institutions.

[0033] S104: Based on the subsurface sea surface temperature observation data within the preset third time period, the historical restart file, the historical sea surface temperature observation data, and the preset fully coupled global climate model, the subsurface sea surface temperature observation data is assimilated into the ocean model to obtain the assimilation result; wherein, the preset fully coupled global climate model is assimilated based on the Newton relaxation approximation method.

[0034] The acquired subsurface sea surface temperature observation data are assimilated into a pre-defined fully coupled global climate model. A suitable fully coupled global climate model (such as the CESM series models) should be selected, which should be able to handle the interactions of multiple elements such as ocean, land, and atmosphere.

[0035] The pre-defined fully coupled global climate model is assimilated using the Newtonian relaxation approximation method. This method linearizes the model and uses observational data to correct the model state. The model is linearized to obtain the state equation. Based on the error in the observational data, the model's predicted state is updated to minimize the prediction error. This update process is repeated until convergence or a pre-defined number of iterations is reached.

[0036] Based on subsurface sea surface temperature (SST) observation data within a pre-defined third time period, historical restart files, historical SST observation data, and a pre-defined fully coupled global climate model, the subsurface SST observation data is assimilated into the ocean model to obtain the assimilation results. Specifically, nudging assimilation is a data assimilation method that adds a relaxation term to the model equations to allow the model solution to gradually approximate the observed values. Assimilating subsurface GODAS SST into the ocean model refers to using this method to integrate subsurface SST observation data provided by GODAS into the ocean model to improve the model's initial field and enhance simulation and prediction capabilities.

[0037] Nudging assimilation uses the Newton relaxation approximation method to continuously adjust model variables (such as sea surface temperature) during model integration, causing them to converge to the observed values. This method is computationally efficient, suitable for operational system applications, and ensures the dynamic consistency of the analyzed field.

[0038] First, long-term integration (e.g., 100 years) is performed based on the 20th-century climatological background field to bring the atmosphere and upper ocean into equilibrium. Then, using the Nudging scheme, subsurface GODAS sea surface temperature data (e.g., 15–400 meters deep) is assimilated into the ocean model, with an assimilation period typically lasting several years (e.g., 1981–2011). Finally, the assimilation results are used as the initial field for post-reporting or forecasting (e.g., 6 months later).

[0039] By fusing observational data and optimizing the initial field of ocean models, the ability to simulate key climate variables such as sea surface temperature and sea surface temperature anomalies (SSTA) is improved, especially the prediction skills for climate modes such as ENSO are significantly improved.

[0040] S105: Using the assimilation result as the initial field, make predictions to obtain upper sea surface temperature forecast data for a preset second time period.

[0041] Using the assimilation results as the initial field, sea surface temperature (SST) is predicted to obtain upper-layer SST forecast data for a pre-defined second time period. Predictions are made based on the model's time step (e.g., hourly, daily). The model continuously advances according to the time step, calculating SST changes at each moment. During the prediction process, intermediate results are periodically output to monitor model performance and make necessary adjustments.

[0042] Finally, the model will output upper sea surface temperature forecast data for a preset second time period. Depending on the needs, it can output prediction results for specific time points or generate sea surface temperature variation curves for the entire time period.

[0043] S102: Input the reanalysis data and the numerical model forecast data into the pre-trained three-dimensional sea surface temperature intelligent prediction model to obtain the three-dimensional sea surface temperature spatiotemporal prediction result; wherein, the pre-trained three-dimensional sea surface temperature intelligent prediction model includes at least a data fusion input module and a spatiotemporal recurrent neural network module, the spatiotemporal recurrent neural network module is trained based on a target pre-training algorithm, a target transfer learning algorithm and a target loss function, the target loss function combines the target error of the upper ocean temperature anomaly field and the average sea surface temperature anomaly index of the target ocean area.

[0044] The device pre-stores a pre-trained 3D sea surface temperature intelligent prediction model. The pre-trained 3D sea surface temperature intelligent prediction model includes at least a data fusion input module and a spatiotemporal recurrent neural network module.

[0045] The data fusion input module is responsible for receiving and processing the fused data, ensuring that the input data format meets the model requirements. The spatiotemporal recurrent neural network module (ST-RNN) uses recurrent neural networks (RNNs) to process time-series data and capture the variation patterns of sea surface temperature over time and space.

[0046] The spatiotemporal recurrent neural network module is trained based on a target pre-training algorithm, a target transfer learning algorithm, and a target loss function. The target loss function combines the target error of the upper ocean temperature anomaly field and the average sea temperature anomaly index of the target ocean region.

[0047] Specifically, the model is initially trained using historical data. During training, a target pre-training algorithm is employed to continuously adjust model parameters and improve prediction accuracy. A target transfer learning algorithm is then combined to transfer learned feature knowledge to the new dataset, accelerating the model's adaptability and learning efficiency in new environments. A target loss function is defined, combining the target error of the upper ocean temperature anomaly field with the average sea surface temperature anomaly index of the target ocean region, to help the model effectively evaluate the accuracy of the prediction results.

[0048] The target error of the upper ocean temperature anomaly field can be the root mean square error (RMSE) of the upper ocean temperature anomaly field, and the average sea temperature anomaly index of the target ocean area is the Niño 3.4 index.

[0049] The reanalysis data and the numerical model forecast data are input into the pre-trained three-dimensional sea surface temperature intelligent prediction model to obtain the three-dimensional sea surface temperature spatiotemporal prediction results.

[0050] Before inputting the pre-trained 3D sea surface temperature intelligent prediction model, the device can employ data fusion technology to combine reanalysis data with numerical model forecast data, forming a unified dataset. Various methods, such as weighted averaging and interpolation, can be used for data fusion to ensure compatibility and consistency of data from different sources.

[0051] The model outputs three-dimensional spatiotemporal prediction results of sea surface temperature, including sea surface temperature data at different time periods and depths, which can be presented in the form of a three-dimensional grid for easy subsequent analysis and visualization.

[0052] The specific form can be described as follows: Where Ci represents the numerical model forecast data for month i, Ai represents the reanalysis data for month i, Bi represents the spatiotemporal prediction result of 3D sea surface temperature for month i, t represents the current month, and F represents the pre-trained 3D intelligent sea surface temperature prediction model.

[0053] In one embodiment, S102 may include S1021~S1022, such as Figure 3 As shown, the details are as follows: S1021: Input the reanalysis data and the numerical model forecast data into the data fusion input module to obtain fused data; wherein, the data fusion input module includes a reanalysis data preprocessing submodule, a numerical model forecast data preprocessing submodule, and a data fusion submodule.

[0054] The reanalysis data and numerical model forecast data are input into the data fusion input module to obtain fused data.

[0055] The data fusion input module includes a reanalysis data preprocessing submodule, a numerical model forecast data preprocessing submodule, and a data fusion submodule.

[0056] The data preprocessing submodule can clean the data, remove missing and outliers to ensure data quality; normalize the data to the same scale to reduce the impact of different units on the model; and generate features by extracting physical features such as sea temperature, salinity, and current velocity.

[0057] The numerical model forecast data preprocessing submodule can be used for format conversion, converting the data output by the numerical model into a format suitable for further processing, such as gridding; and for timeliness adjustment, aligning the data in time to ensure the consistency between reanalysis data and forecast data in time.

[0058] The data fusion submodule is used to fuse preprocessed reanalysis data with numerical model forecast data using methods such as weighted averaging and principal component analysis (PCA) to obtain comprehensive fused data.

[0059] Alternatively, a weighted method can be used, setting weights for the reanalysis data and forecast data, and adjusting them based on the accuracy of historical forecasts to obtain more reliable fusion results.

[0060] In one embodiment, S1021 may include S10211~S10215, such as Figure 4As shown, S10211~S10215 are as follows: S10211: Interpolate the reanalysis data and the numerical model forecast data respectively to obtain interpolated reanalysis data and interpolated numerical model forecast data.

[0061] Collect reanalysis data and select an appropriate interpolation algorithm, such as linear interpolation, spline interpolation, or kriging interpolation, choosing the optimal algorithm based on the data distribution and characteristics. Apply the interpolation algorithm to transform the irregular reanalysis data into regular grid data, ensuring the spatial continuity of the data.

[0062] Acquire sea surface temperature forecast data from numerical models, which are typically already on a regular grid. Use the same interpolation method (as described above) to further smooth or adjust these data, ensuring consistency in spatial distribution with the reanalysis data.

[0063] S10212: The interpolated reanalysis data is spliced ​​together to obtain the first spliced ​​data.

[0064] Define the temporal and spatial range for concatenation. Concatenate the interpolated reanalysis data from different time points in chronological order to ensure the continuity of the time series. Data frames (such as NumPy arrays, Pandas DataFrames, etc.) can be used to manage the concatenation operation.

[0065] S10213: The interpolated numerical model forecast data is spliced ​​together to obtain the second spliced ​​data.

[0066] Similarly, the temporal and spatial scope of the stitching is determined. The interpolated numerical model forecast data are stitched together in chronological order to form a continuous forecast dataset, resulting in the second stitched data.

[0067] S10214: Normalize the first spliced ​​data and the second spliced ​​data.

[0068] Different normalization techniques can be used, such as min-max normalization, Z-score normalization (mean normalization), or other normalization methods suitable for the characteristics of sea surface temperature data.

[0069] For the first concatenated data (the interpolated reanalysis data), the selected normalization method is applied to standardize the data values ​​to a specific range (e.g., 0 to 1) or adjust them to a mean of 0 and a standard deviation of 1.

[0070] Repeat the above process to normalize the second stitched data (the interpolated numerical model forecast data) as well, ensuring that the parameters used (such as minimum, maximum or mean, standard deviation) are calculated based on the training set data to avoid data leakage.

[0071] S10215: Merge the first spliced ​​data and the second spliced ​​data after normalization processing to obtain the fused data.

[0072] The first and second concatenated data, after normalization, are then merged. For example, if a weighted average is chosen, weights can be assigned to each dataset based on its historical performance, and then a weighted average can be calculated.

[0073] Fusion methods can include weighted averages, principal component analysis (PCA), deep learning models (such as neural networks), or simple linear combinations.

[0074] Generate fused data, which will include sea surface temperature forecasts that take into account both reanalysis and numerical model forecasts.

[0075] For example, in one implementation, the upper sea surface temperature (SST) data for the past 12 months from the reanalysis data are first stitched together, with a data dimension of B×12×7×40×160, where B is the sample size, 12 represents the past months, 7 represents the SST layer number, and 40 and 160 represent latitude and longitude, respectively. Then, the upper SST data for the next 20 months from the numerical model forecast data are stitched together, with a data dimension of B×20×7×40×160, where B is the sample size, 20 represents the future forecast months, 7 represents the SST layer number, and 40 and 160 represent latitude and longitude, respectively. Subsequently, normalization is performed separately before data fusion, resulting in a fused data dimension of B×32(12+20)×7×40×16.

[0076] In this embodiment, interpolation and stitching processes generate a high-resolution spatiotemporal dataset of sea surface temperature (SST), enhancing the spatial and temporal continuity of the data. Fusion of reanalysis data and numerical model forecast data provides more accurate SST predictions than using either data source alone, especially in the context of frequent climate events. The model design and data processing workflow exhibit good scalability, adapting to new data sources and technological advancements. Data fusion methods and model parameters can be flexibly adjusted according to different research needs and environmental conditions to address various climate prediction challenges.

[0077] S1022: The fused data is input into the spatiotemporal recurrent neural network module to obtain the three-dimensional sea surface temperature spatiotemporal prediction result; wherein, the spatiotemporal recurrent neural network module is a deep learning module that combines a moving window attention mechanism and a long short-term memory network.

[0078] The fused data is then input into the spatiotemporal recurrent neural network module, which is a deep learning module that combines a moving window attention mechanism and a long short-term memory network.

[0079] Long Short-Term Memory (LSTM) networks are recurrent neural networks suitable for processing time-series data, effectively capturing long-term dependent features. In this method, an LSTM model is used to capture dynamic changes in the time series. The moving window mechanism, by setting a moving window (e.g., sea surface temperature data from the past 7 days), allows the LSTM to learn the trends and periodic changes in the time series. During prediction, an attention mechanism allows the model to focus on more important parts of the input data, thereby improving model accuracy. The moving window attention mechanism can assign different weights to different regions of the input sequence, optimizing prediction performance.

[0080] The trained model is applied to new fused data to predict future three-dimensional sea surface temperatures. The output can include sea surface temperature trends and spatial distribution over future time periods.

[0081] In one embodiment, before S102, steps S106 to S109 may be included, such as... Figure 5 As shown, S106~S109 are detailed below: S106: Obtain reanalysis training data and numerical model forecast training data.

[0082] The equipment acquires reanalysis training data and numerical model forecast training data, and can clean, standardize, and handle missing values ​​in the acquired data. It also converts the data into a format suitable for model training, such as time series or spatial grid formats.

[0083] S107: Input the reanalysis training data and the numerical model forecast training data into the initialized three-dimensional sea surface temperature intelligent prediction model to obtain the current three-dimensional sea surface temperature spatiotemporal prediction result.

[0084] Choose a suitable 3D sea surface temperature (SST) prediction model architecture (such as 3D convolutional neural network, LSTM, etc.). Input the processed reanalysis data and numerical model forecast data into the model in temporal and spatial order. Data normalization can be performed to improve model convergence. Run the model to make predictions and obtain the current 3D SST spatiotemporal prediction results.

[0085] S108: Based on the reanalysis training data, the numerical model forecast training data, the current three-dimensional sea surface temperature spatiotemporal prediction results, the target pre-training algorithm, and the target loss function, iteratively optimize the trainable parameters in the three-dimensional sea surface temperature intelligent prediction model until the loss value meets the preset termination condition, and save the trainable parameters; wherein, the target error of the upper ocean temperature anomaly field in the target loss function is the root mean square error of the upper ocean temperature anomaly field, and the average sea surface temperature anomaly index of the target ocean region is the Niño 3.4 exponent.

[0086] In this embodiment, the target error of the upper ocean temperature anomaly field in the target loss function is set as the root mean square error (RMSE) of the upper ocean temperature anomaly field. This index is used to evaluate the difference between the predicted and the actual values. The mean sea surface temperature anomaly index for the target ocean area, such as the Niño 3.4 index, also needs to be calculated to monitor temperature anomalies in specific sea areas.

[0087] Use backpropagation and optimization algorithms (such as Adam, SGD, etc.) to update the trainable parameters of the model.

[0088] In each iteration, the gradient is calculated based on the current loss value, and parameters are adjusted to minimize the loss. Preset termination conditions are set, such as when the loss value no longer decreases significantly or when a preset maximum number of iterations is reached. The model's performance on the validation set is monitored during training to prevent overfitting. After the termination conditions are met, the trainable parameters of the current model are saved for later use or further training.

[0089] S109: Adjust the trainable parameters according to the reanalysis training data, the numerical model prediction training data, and the target transfer learning algorithm to obtain the pre-trained spatiotemporal recurrent neural network module.

[0090] In this embodiment, different transfer learning strategies can be selected, such as: fixed feature extractor, which only fine-tunes the last few layers of the model while keeping the weights of the previous layers unchanged, in order to utilize the learning ability of the existing model; full model fine-tuning, which fine-tunes the entire model on new data, suitable for situations where the reanalysis data and numerical pattern data are similar to the original training data.

[0091] The trainable parameters are tuned using transfer learning algorithms. The performance of the transfer learning model is monitored during training, and a validation set is used to evaluate the model's accuracy.

[0092] After training, the model is comprehensively evaluated using a test set to calculate its predictive performance, such as RMSE, MAE, and R². Based on the evaluation results, it may be necessary to readjust the model parameters or try different transfer learning strategies.

[0093] For example, fused data is acquired, and a pre-trained spatiotemporal recurrent neural network module is constructed using a SwinLSTM network. The model is trained through pre-training and transfer learning. During pre-training, multiple sample assimilation and corresponding prediction data are used to train the model, providing a high-quality initial parameter state. Pre-training allows the model to learn initially, adapting more quickly to task requirements and laying the foundation for subsequent fine-tuning through transfer learning. The pre-training iteration period (epochs) is set to 100, and the learning rate is set to 10⁻³. Subsequently, transfer learning is used to fine-tune the parameters of the pre-trained model, using GODAS and future prediction data as data samples. The learning rate is set to 10⁻⁵, and early stopping is used to prevent overfitting and obtain the optimal prediction results. The batch size is uniformly set to 8. To enhance ENSO index prediction, the root mean square error (RMSE) of the upper ocean temperature anomaly field is combined with the Niño 3.4 exponent as a loss function to measure the deviation between the prediction and the target label. Finally, the predicted upper ocean temperature of the tropical Pacific is output after testing.

[0094] In this embodiment, a deep learning model is employed for high-precision prediction: Deep learning models effectively capture complex spatiotemporal features, thereby improving the accuracy of sea surface temperature (SST) prediction. By combining reanalysis data and numerical model forecast data, historical data and the latest meteorological information are fully utilized to enhance the model's training foundation. Transfer learning strategies enable the model to maintain good predictive performance under different climatic conditions and in various ocean regions, especially in situations where data is scarce. Real-time input of new meteorological data allows the model to quickly update its predictions, making it suitable for climate monitoring and early warning. The model architecture can be adjusted according to needs, and different deep learning techniques (such as CNN, LSTM, etc.) can be selected to adapt to specific application scenarios. With the continuous accumulation of new data, the model can be continuously retrained and optimized to improve its long-term predictive capabilities.

[0095] In one embodiment, the pre-trained three-dimensional sea surface temperature intelligent prediction model further includes a verification and evaluation module. This embodiment may also include S111~S112, such as... Figure 6 As shown, S111~S112 are as follows: S111: Input the three-dimensional sea surface temperature spatiotemporal prediction results into the inspection and evaluation module to obtain the inspection and evaluation results; wherein, the inspection and evaluation results include the Niño 3.4 index, ocean surface temperature, and equatorial ocean temperature profile inspection results.

[0096] Collect three-dimensional spatiotemporal prediction data for sea surface temperature (SST), ensuring the data covers SST information at multiple time points and spatial latitude and longitude. Select appropriate ocean monitoring data sources, such as satellite remote sensing data and buoy observation data, to ensure the accuracy and timeliness of the data.

[0097] The design includes an evaluation module capable of processing and analyzing sea surface temperature (SST) data. Module functions include calculating the Niño 3.4 index, SST, and equatorial ocean temperature profile (T-profile) results.

[0098] Input the prepared 3D sea surface temperature spatiotemporal prediction data into the verification and evaluation module. Ensure that the data format is consistent with the module requirements; format conversion (such as CSV, NetCDF, etc.) may be necessary.

[0099] Sea surface temperature (SST) data for the Niño 3.4 region (5°S-5°N, 170°W-120°W) is extracted from the input data, and the average SST for this region is calculated. The upper surface temperature is extracted directly from the input SST data and statistically analyzed. Based on the input data, a temperature profile of the equatorial region is plotted, and temperature variations at different depths are analyzed.

[0100] S112: Visualize the test and evaluation results to display the image corresponding to the test and evaluation results.

[0101] Choose appropriate visualization tools and software, such as Python's Matplotlib, Seaborn, and Plotly, or more professional GIS tools, such as ArcGIS and QGIS.

[0102] Identify the target audience for the visualization and select the appropriate graphic type (such as line chart, heatmap, cross-sectional view, etc.) to facilitate understanding and analysis.

[0103] The Niño 3.4 index visualization can generate a line chart showing the trend of the Niño 3.4 index over time. The horizontal axis represents time, and the vertical axis represents the Niño 3.4 index value. Key events (such as the occurrence of El Niño and La Niña phenomena) can be marked on the chart for easier analysis.

[0104] Ocean surface temperature (SST) visualization can use heat maps to show the temperature distribution of the ocean surface in different regions. Different shades of color can be used to represent temperature levels. Overlaying isotherms can highlight temperature gradients and ocean current directions.

[0105] Equatorial ocean temperature profile visualization can generate cross-sectional maps showing temperature variations at different depths in the equatorial region. The horizontal axis represents depth (possibly in meters), and the vertical axis represents temperature (in degrees Celsius). Comparisons of profiles at different time points can be added to the map to demonstrate the dynamic process of temperature change.

[0106] In this embodiment, by verifying and evaluating the spatiotemporal prediction results of three-dimensional sea surface temperature (SST), deficiencies in the prediction model can be identified in a timely manner, improving the model's adjustment and optimization capabilities, thereby enhancing the accuracy of SST prediction. Visualization processing makes complex ocean temperature data more intuitive and understandable, helping researchers and policymakers quickly grasp SST change trends and spatial distribution, thus supporting scientific research and policy formulation.

[0107] This application provides a three-dimensional sea surface temperature (SST) spatiotemporal prediction method, aiming to address the problems of insufficient prediction accuracy, inadequate data utilization, and weak model generalization ability in existing technologies. By acquiring reanalysis data and numerical model forecast data, this method can fully integrate multi-source information from global ocean data assimilation systems and fully coupled global climate models, enhancing the basic data support for the prediction model. Inputting this data into a pre-trained three-dimensional intelligent SST prediction model, utilizing a data fusion input module and a spatiotemporal recurrent neural network module, effectively captures the spatiotemporal characteristics of SST changes, thereby improving prediction accuracy. This intelligent prediction model employs a target pre-training algorithm, a target transfer learning algorithm, and a target loss function combining the target error of the upper ocean temperature anomaly field and the average SST anomaly index of the target ocean region for training, significantly enhancing the model's generalization ability in the face of scarce data and uncertainty. This series of innovative designs enables the model to not only provide high-precision SST prediction results but also possess good interpretability, providing stronger support for scientific research and climate change response. Furthermore, the real-time data input and intelligent modeling methods give this approach a significant advantage in responding to extreme climate events and achieving rapid response, further promoting the development of marine climate prediction technology.

[0108] Please see Figure 7 , Figure 7 This is a schematic flowchart illustrating an El Niño event prediction method provided in the second embodiment of this application. In this embodiment, the executing entity of the El Niño event prediction method is a device with El Niño event prediction functionality, such as a desktop computer, server, mobile device, etc. Figure 7 The El Niño event prediction methods shown may include: S201: The three-dimensional sea surface temperature spatiotemporal prediction method described in the first embodiment yields the three-dimensional sea surface temperature spatiotemporal prediction result.

[0109] In this embodiment, the three-dimensional sea surface temperature spatiotemporal prediction result is obtained according to the three-dimensional sea surface temperature spatiotemporal prediction method described in the first embodiment. Specific details can be found in the relevant description in the first embodiment, and will not be repeated here.

[0110] S202: Based on the three-dimensional sea surface temperature spatiotemporal prediction results and the preset El Niño event prediction algorithm, predict the El Niño event.

[0111] Choose a suitable El Niño event prediction algorithm as needed, for example: Statistical models: such as the Autoregressive Moving Average (ARIMA) model, regression analysis, etc.

[0112] Machine learning models, such as random forests and support vector machines (SVM), are used to make predictions by combining meteorological and oceanographic data.

[0113] Deep learning models: These models use recurrent neural networks (RNNs) or LSTMs to capture the dynamic changes in time series data.

[0114] The three-dimensional spatiotemporal prediction results of sea surface temperature obtained in step S201 are used as input data, especially surface temperature data of El Niño regions (such as the equatorial Pacific). The model will output information about the probability, intensity, and duration of El Niño events. For example, the model may output "The probability of a strong El Niño event occurring within the next 6 months is 70%".

[0115] The technical solution in this application achieves efficient prediction of El Niño events by combining three-dimensional spatiotemporal sea surface temperature (SST) prediction results with a pre-defined El Niño event prediction algorithm. The core of this method lies in utilizing the accurate SST spatiotemporal prediction results obtained in previous steps. These results fully reflect the spatiotemporal variation characteristics of ocean temperature, thus providing reliable data support for the occurrence and development of El Niño events. By combining SST variation information with the established El Niño prediction algorithm, the imminent occurrence of El Niño phenomena can be more accurately identified and predicted. This integrated prediction method not only improves the accuracy and timeliness of El Niño event prediction but also provides a scientific basis for climate monitoring, disaster early warning, and the formulation of response measures, thereby effectively reducing the potential risks to the ecological environment and economic activities caused by climate anomalies resulting from El Niño. In summary, the technical solution of this invention improves the accuracy of SST and El Niño event prediction while promoting the further development of climate prediction technology.

[0116] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0117] Please see Figure 8 , Figure 8 This is a schematic diagram of the three-dimensional sea surface temperature spatiotemporal prediction device provided in the third embodiment of this application. The included units are used for execution... Figures 1-6 The steps in the corresponding embodiments. Please refer to the details. Figures 1-6 The relevant descriptions in the corresponding embodiments are shown below. For ease of explanation, only the parts relevant to this embodiment are shown. See also... Figure 8The three-dimensional sea surface temperature spatiotemporal prediction device 8 includes: The first acquisition unit 810 is used to acquire reanalysis data and numerical model forecast data; wherein, the reanalysis data is upper sea surface temperature observation data within a preset first time period acquired from the global ocean data assimilation system, and the numerical model forecast data is upper sea surface temperature forecast data within a preset second time period predicted based on a fully coupled global climate model. The first processing unit 820 is used to input the reanalysis data and the numerical model forecast data into a pre-trained three-dimensional sea surface temperature intelligent prediction model to obtain a three-dimensional sea surface temperature spatiotemporal prediction result; wherein, the pre-trained three-dimensional sea surface temperature intelligent prediction model includes at least a data fusion input module and a spatiotemporal recurrent neural network module, the spatiotemporal recurrent neural network module is trained based on a target pre-training algorithm, a target transfer learning algorithm and a target loss function, the target loss function combines the target error of the upper ocean temperature anomaly field and the average sea surface temperature anomaly index of the target ocean area.

[0118] Furthermore, the three-dimensional sea surface temperature spatiotemporal prediction device 8 also includes: The second acquisition unit is used to acquire subsurface sea surface temperature observation data, historical restart files and historical sea surface temperature observation data within a preset third time period from the global ocean data assimilation system. The second processing unit is used to assimilate the subsurface sea surface temperature observation data into the ocean model based on the subsurface sea surface temperature observation data within the preset third time period, the historical restart file, the historical sea surface temperature observation data, and the preset fully coupled global climate model, to obtain the assimilation result; wherein, the preset fully coupled global climate model is used for assimilation processing based on the Newton relaxation approximation method. The third processing unit is used to make predictions using the assimilation results as the initial field to obtain upper sea surface temperature forecast data for a preset second time period.

[0119] The first processing unit in the advanced area is specifically used for: The reanalysis data and the numerical model forecast data are input into the data fusion input module to obtain fused data; wherein, the data fusion input module includes a reanalysis data preprocessing submodule, a numerical model forecast data preprocessing submodule, and a data fusion submodule; The fused data is input into the spatiotemporal recurrent neural network module to obtain the three-dimensional sea surface temperature spatiotemporal prediction result; wherein, the spatiotemporal recurrent neural network module is a deep learning module that combines a moving window attention mechanism and a long short-term memory network.

[0120] In the first processing unit of the advanced area, it is also used for: The reanalysis data and the numerical model forecast data are respectively interpolated to obtain interpolated reanalysis data and interpolated numerical model forecast data; The interpolated reanalysis data is spliced ​​together to obtain the first spliced ​​data; The interpolated numerical model forecast data is spliced ​​together to obtain the second spliced ​​data; The first and second spliced ​​data are normalized. The first spliced ​​data and the second spliced ​​data after normalization are merged to obtain the fused data.

[0121] Furthermore, the three-dimensional sea surface temperature spatiotemporal prediction device 8 also includes: The third acquisition unit is used to acquire reanalysis training data and numerical model prediction training data. The fourth processing unit is used to input the reanalysis training data and the numerical model forecast training data into the initialized three-dimensional sea surface temperature intelligent prediction model to obtain the current three-dimensional sea surface temperature spatiotemporal prediction result. The fifth processing unit is used to iteratively optimize the trainable parameters in the three-dimensional sea surface temperature intelligent prediction model based on the reanalysis training data, the numerical model forecast training data, the current three-dimensional sea surface temperature spatiotemporal prediction results, the target pre-training algorithm, and the target loss function, until the loss value meets the preset termination condition, and save the trainable parameters; wherein, the target error of the upper ocean temperature anomaly field in the target loss function is the root mean square error of the upper ocean temperature anomaly field, and the average sea surface temperature anomaly index of the target ocean region is the Niño 3.4 exponent; The sixth processing unit is used to adjust the trainable parameters based on the reanalysis training data, the numerical model prediction training data, and the target transfer learning algorithm to obtain the pre-trained spatiotemporal recurrent neural network module.

[0122] Furthermore, the pre-trained three-dimensional sea surface temperature intelligent prediction model also includes a verification and evaluation module, and the three-dimensional sea surface temperature spatiotemporal prediction device 8 also includes: The seventh processing unit is used to input the three-dimensional sea surface temperature spatiotemporal prediction results into the inspection and evaluation module to obtain the inspection and evaluation results; wherein, the inspection and evaluation results include Niño 3.4 index, ocean surface temperature, and equatorial ocean temperature profile inspection results; The eighth processing unit is used to visualize the test and evaluation results and display the image corresponding to the test and evaluation results.

[0123] Furthermore, the reanalysis data consists of tropical Pacific upper sea surface temperature observation data over the past 12 months, and the numerical model prediction data consists of tropical Pacific upper sea surface temperature forecast data for the next 20 months.

[0124] Please see Figure 9 , Figure 9 This is a schematic diagram of the El Niño event prediction device provided in the fourth embodiment of this application. The included units are used for performing... Figure 7 The steps in the corresponding embodiments. Please refer to the details. Figure 7 The relevant descriptions in the corresponding embodiments are shown below. For ease of explanation, only the parts relevant to this embodiment are shown. See also... Figure 9 The El Niño event prediction device 9 includes: The first processing unit 910 is used to obtain the three-dimensional sea surface temperature spatiotemporal prediction result according to the three-dimensional sea surface temperature spatiotemporal prediction method described in the first embodiment above. The second processing unit 920 is used to predict El Niño events based on the three-dimensional sea surface temperature spatiotemporal prediction results and the preset El Niño event prediction algorithm.

[0125] Figure 10 This is a schematic diagram of the three-dimensional sea surface temperature spatiotemporal prediction device provided in the fifth embodiment of this application. Figure 10 As shown, the three-dimensional sea surface temperature spatiotemporal prediction device 10 of this embodiment includes: a processor 100, a memory 101, and a computer program 102 stored in the memory 101 and executable on the processor 100, such as a three-dimensional sea surface temperature spatiotemporal prediction program. When the processor 100 executes the computer program 102, it implements the steps in the various three-dimensional sea surface temperature spatiotemporal prediction method embodiments described above, for example... Figure 1 Steps 101 to 102 are shown. Alternatively, when the processor 100 executes the computer program 102, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 8 The functions of modules 810 to 820 are shown.

[0126] For example, the computer program 102 can be divided into one or more modules / units, which are stored in the memory 101 and executed by the processor 100 to complete this application. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 102 in the three-dimensional sea surface temperature spatiotemporal prediction device 10. For example, the computer program 102 can be divided into a first acquisition unit and a first processing unit, with the specific functions of each unit as follows: The first acquisition unit is used to acquire reanalysis data and numerical model forecast data; wherein, the reanalysis data is upper sea surface temperature observation data within a preset first time period acquired from the global ocean data assimilation system, and the numerical model forecast data is upper sea surface temperature forecast data within a preset second time period predicted based on a fully coupled global climate model. The first processing unit is used to input the reanalysis data and the numerical model forecast data into a pre-trained three-dimensional intelligent sea surface temperature prediction model to obtain a three-dimensional sea surface temperature spatiotemporal prediction result. The pre-trained three-dimensional intelligent sea surface temperature prediction model includes at least a data fusion input module and a spatiotemporal recurrent neural network module. The spatiotemporal recurrent neural network module is trained based on a target pre-training algorithm, a target transfer learning algorithm, and a target loss function. The target loss function combines the target error of the upper ocean temperature anomaly field and the average sea surface temperature anomaly index of the target ocean region.

[0127] The three-dimensional sea surface temperature spatiotemporal prediction device may include, but is not limited to, a processor 100 and a memory 101. Those skilled in the art will understand that... Figure 10 This is merely an example of the three-dimensional sea surface temperature spatiotemporal prediction device 10 and does not constitute a limitation on the three-dimensional sea surface temperature spatiotemporal prediction device 10. It may include more or fewer components than shown, or combine certain components, or different components. For example, the three-dimensional sea surface temperature spatiotemporal prediction device may also include input / output devices, network access devices, buses, etc.

[0128] The processor 100 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0129] The memory 101 can be an internal storage unit of the three-dimensional sea surface temperature (SST) spatiotemporal prediction device 10, such as a hard disk or memory of the three-dimensional SST spatiotemporal prediction device 10. The memory 101 can also be an external storage device of the three-dimensional SST spatiotemporal prediction device 10, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the three-dimensional SST spatiotemporal prediction device 10. Furthermore, the three-dimensional SST spatiotemporal prediction device 10 can include both internal storage units and external storage devices. The memory 101 is used to store the computer program and other programs and data required by the three-dimensional SST spatiotemporal prediction device. The memory 101 can also be used to temporarily store data that has been output or will be output.

[0130] Figure 11 This is a schematic diagram of the El Niño event prediction device provided in the sixth embodiment of this application. Figure 11 As shown, the El Niño event prediction device 11 of this embodiment includes: a processor 110, a memory 111, and a computer program 112 stored in the memory 111 and executable on the processor 110, such as an El Niño event prediction program. When the processor 110 executes the computer program 112, it implements the steps in the various El Niño event prediction method embodiments described above, for example... Figure 7 Steps 201 to 202 are shown. Alternatively, when the processor 110 executes the computer program 112, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 9 The functions of modules 910 to 920 are shown.

[0131] For example, the computer program 112 can be divided into one or more modules / units, which are stored in the memory 111 and executed by the processor 110 to complete this application. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 112 in the El Niño event prediction device 11. For example, the computer program 112 can be divided into a first processing unit and a second processing unit, with the specific functions of each unit as follows: The first processing unit is used to obtain the three-dimensional sea surface temperature spatiotemporal prediction result according to the three-dimensional sea surface temperature spatiotemporal prediction method described in the first aspect above. The second processing unit is used to predict El Niño events based on the three-dimensional sea surface temperature spatiotemporal prediction results and the preset El Niño event prediction algorithm.

[0132] The El Niño event prediction device may include, but is not limited to, a processor 110 and a memory 111. Those skilled in the art will understand that... Figure 11 This is merely an example of the El Niño event prediction device 11 and does not constitute a limitation on the El Niño event prediction device 11. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the El Niño event prediction device may also include input / output devices, network access devices, buses, etc.

[0133] The processor 110 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0134] The memory 111 can be an internal storage unit of the El Niño event prediction device 11, such as a hard drive or memory of the El Niño event prediction device 11. The memory 111 can also be an external storage device of the El Niño event prediction device 11, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the El Niño event prediction device 11. Furthermore, the El Niño event prediction device 11 can include both internal storage units and external storage devices. The memory 111 is used to store the computer program and other programs and data required by the El Niño event prediction device. The memory 111 can also be used to temporarily store data that has been output or will be output.

[0135] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0136] This application also provides a network device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0137] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0138] This application provides a computer program product that, when run on a mobile terminal, enables the mobile terminal to implement the steps described in the above-described method embodiments.

[0139] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0140] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0141] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0142] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0143] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0144] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A three-dimensional spatiotemporal prediction method for sea surface temperature, characterized in that, Including the following steps: Acquire reanalysis data and numerical model forecast data; wherein, the reanalysis data is upper sea surface temperature observation data within a preset first time period obtained from the global ocean data assimilation system, and the numerical model forecast data is upper sea surface temperature forecast data within a preset second time period predicted based on a fully coupled global climate model. The reanalysis data and the numerical model forecast data are input into a pre-trained three-dimensional intelligent sea surface temperature (SST) prediction model to obtain a three-dimensional SST spatiotemporal prediction result. The pre-trained three-dimensional intelligent SST prediction model includes at least a data fusion input module and a spatiotemporal recurrent neural network module. The spatiotemporal recurrent neural network module is trained based on a target pre-training algorithm, a target transfer learning algorithm, and a target loss function. The target loss function combines the target error of the upper ocean temperature anomaly field and the average SST anomaly index of the target ocean region.

2. The three-dimensional sea surface temperature spatiotemporal prediction method according to claim 1, characterized in that, Before acquiring reanalysis data and numerical model forecast data, the following steps are included: Obtain subsurface sea surface temperature observation data, historical restart files, and historical sea surface temperature observation data within a preset third time period from the global ocean data assimilation system; Based on the subsurface sea surface temperature observation data within the preset third time period, the historical restart file, the historical sea surface temperature observation data, and the preset fully coupled global climate model, the subsurface sea surface temperature observation data is assimilated into the ocean model to obtain the assimilation result; wherein, the preset fully coupled global climate model is assimilated based on the Newton relaxation approximation method. Using the assimilation result as the initial field, predictions are made to obtain upper sea surface temperature forecast data for a preset second time period.

3. The three-dimensional sea surface temperature spatiotemporal prediction method according to claim 1 or 2, characterized in that, The step of inputting the reanalysis data and the numerical model forecast data into a pre-trained three-dimensional sea surface temperature (SST) intelligent prediction model to obtain a three-dimensional SST spatiotemporal prediction result includes the following steps: The reanalysis data and the numerical model forecast data are input into the data fusion input module to obtain fused data; wherein, the data fusion input module includes a reanalysis data preprocessing submodule, a numerical model forecast data preprocessing submodule, and a data fusion submodule; The fused data is input into the spatiotemporal recurrent neural network module to obtain the three-dimensional sea surface temperature spatiotemporal prediction result; wherein, the spatiotemporal recurrent neural network module is a deep learning module that combines a moving window attention mechanism and a long short-term memory network.

4. The three-dimensional sea surface temperature spatiotemporal prediction method according to claim 3, characterized in that, The step of inputting the reanalysis data and the numerical model forecast data into the data fusion input module to obtain fused data includes the following steps: The reanalysis data and the numerical model forecast data are interpolated to obtain interpolated reanalysis data and interpolated numerical model forecast data, respectively. The interpolated reanalysis data is spliced ​​together to obtain the first spliced ​​data; The interpolated numerical model forecast data is spliced ​​together to obtain the second spliced ​​data; The first and second spliced ​​data are normalized. The first spliced ​​data and the second spliced ​​data after normalization are merged to obtain the fused data.

5. The three-dimensional sea surface temperature spatiotemporal prediction method according to claim 1 or 2, characterized in that, Before inputting the reanalysis data and the numerical model forecast data into the pre-trained three-dimensional sea surface temperature intelligent prediction model to obtain the three-dimensional sea surface temperature spatiotemporal prediction results, the following steps are included: Acquire reanalysis training data and numerical model prediction training data; The reanalysis training data and the numerical model forecast training data are input into the initialized three-dimensional sea surface temperature intelligent prediction model to obtain the current three-dimensional sea surface temperature spatiotemporal prediction result. Based on the reanalysis training data, the numerical model forecast training data, the current 3D sea surface temperature spatiotemporal prediction results, the target pre-training algorithm, and the target loss function, the trainable parameters in the 3D sea surface temperature intelligent prediction model are iteratively optimized until the loss value meets the preset termination condition, and the trainable parameters are saved; wherein, the target error of the upper ocean temperature anomaly field in the target loss function is the root mean square error of the upper ocean temperature anomaly field, and the average sea surface temperature anomaly index of the target ocean region is the Niño 3.4 exponent; The trainable parameters are adjusted based on the reanalysis training data, the numerical model prediction training data, and the target transfer learning algorithm to obtain the pre-trained spatiotemporal recurrent neural network module.

6. The three-dimensional sea surface temperature spatiotemporal prediction method according to claim 1 or 2, characterized in that, The pre-trained three-dimensional sea surface temperature intelligent prediction model also includes a verification and evaluation module, and the method further includes the following steps: The three-dimensional sea surface temperature spatiotemporal prediction results are input into the inspection and evaluation module to obtain the inspection and evaluation results; wherein, the inspection and evaluation results include the Niño 3.4 index, the ocean surface temperature, and the equatorial ocean temperature profile inspection results; The test and evaluation results are visualized to display the corresponding images.

7. The three-dimensional sea surface temperature spatiotemporal prediction method according to claim 1 or 2, characterized in that, The reanalysis data consists of tropical Pacific upper sea surface temperature observation data from the past 12 months, and the numerical model prediction data consists of tropical Pacific upper sea surface temperature forecast data for the next 20 months.

8. A method for predicting El Niño events, characterized in that, Including the following steps: The three-dimensional sea surface temperature spatiotemporal prediction method according to any one of claims 1 to 7 yields a three-dimensional sea surface temperature spatiotemporal prediction result; Based on the three-dimensional sea surface temperature spatiotemporal prediction results and the preset El Niño event prediction algorithm, El Niño events are predicted.

9. A three-dimensional sea surface temperature spatiotemporal prediction device, comprising: A processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the steps of the method as claimed in any one of claims 1 to 7.

10. An El Niño event prediction device, comprising: A processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the steps of the method as claimed in claim 8.

Images