Marine bottom layer heat wave prediction method and system and computer equipment

Abstract

The invention relates to the technical field of ocean observation, in particular to an ocean bottom layer heat wave prediction method and system based on a generative adversarial network and computer device.The method comprises the steps that a data set is established, satellite remote sensing data and reanalysis data of the ocean surface are preprocessed based on longitude and latitude grids, and a multi-dimensional data set is constructed; and a bottom layer heat wave prediction step: based on a pre-trained heat wave prediction model, feature extraction is performed on the multi-dimensional data set, the heat wave prediction model outputs a prediction result by using a generator, and the generator is obtained by training confrontation based on historical heat wave phenomenon data of the ocean bottom layer and a discriminator. The method can effectively predict ocean bottom layer heat waves, thus improving deep sea heat wave phenomenon perception capability and response efficiency.

Description

Technical Field

[0001] The present application relates to the field of ocean observation technology, and in particular to a method, system, and computer equipment for predicting ocean bottom heat waves based on a generative adversarial network. Background Art

[0002] Marine heatwaves are abnormally high temperatures that persist for a sustained period in the ocean. These phenomena not only directly damage the habitats of marine life but can also lead to coral bleaching, habitat destruction, and shifts in species distribution. As a key indicator of climate change, the study of marine heatwaves can help better predict and address global climate change. By monitoring and analyzing marine heatwaves, scientists can better understand changes in ocean circulation, which is crucial for predicting future climate change and sea level rise.

[0003] Currently, research and monitoring of marine heatwaves primarily rely on ocean surface temperature data. Due to technical bottlenecks such as complex ocean conditions and difficulty acquiring data, existing observation technologies are mostly focused on the sea surface or shallow waters, utilizing satellite remote sensing, buoy observations, and shipborne equipment to conduct long-term monitoring and analysis of sea surface temperature. These methods have achieved some success in analyzing the frequency, intensity, and spatial distribution of marine heatwaves, providing important insights for global climate change research.

[0004] However, existing technologies still have significant shortcomings in identifying and predicting the vertical structure of ocean heatwaves, particularly those occurring in the bottom ocean. Because deep ocean temperature variations are often difficult to directly detect through remote sensing, and the deployment of deep-sea observation equipment is costly and time-consuming, there is currently a lack of effective technical solutions for real-time and high-precision analysis and prediction of bottom-ocean heatwaves. Summary of the Invention

[0005] The embodiments of the present application provide a method, apparatus, system, and computer equipment for predicting ocean bottom heat waves, which are intended to effectively predict ocean bottom heat waves to improve the perception and response efficiency of deep-sea heat wave phenomena.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting ocean bottom heat waves, comprising: The dataset establishment step is to pre-process the satellite remote sensing data and reanalysis data of the ocean surface based on the latitude and longitude grid to construct a multidimensional dataset; The bottom layer heat wave prediction step includes training a heat wave prediction model based on the multidimensional data set. The heat wave prediction model uses a generator to extract features from the multidimensional data set and output a prediction result. The generator is obtained by training and competing with a discriminator based on historical heat wave phenomenon data of the bottom layer of the ocean. The trained heat wave prediction model is used to predict heat wave phenomena in the bottom layer of the ocean.

[0007] In some embodiments, the historical heat wave phenomenon data is calculated based on the following steps: The sea temperature data of multiple years of historical reanalysis data are used as a climate reference period, a time window is selected, and for each calendar day, the daily average sea temperature value within the time window of the climate reference period is extracted, and the climate state value and threshold value of the calendar day are calculated based on the daily average sea temperature value; The sea temperature on the target date is compared with the threshold of that calendar day to determine whether it exceeds the threshold continuously for a set period. If so, it is considered that an ocean heat wave event has occurred on that day, and it is marked as a heat wave phenomenon at the corresponding latitude and longitude grid points, thereby obtaining the historical heat wave phenomenon data of the ocean bottom layer at the corresponding latitude and longitude grid points on each historical day.

[0008] In some embodiments, the underlying heat wave prediction step further comprises: Establish a heat wave prediction model based on conditional generative adversarial networks (CGANs); Inputting the multidimensional dataset into a generator of the heat wave prediction model, the generator being configured to generate classification categories corresponding to grid points of the latitude and longitude grid based on the multidimensional dataset, wherein the classification categories include: presence of a heat wave, absence of a heat wave, and belonging to a land point; The multidimensional data set, classification categories, and historical heat wave phenomenon data are spliced together and input into the discriminator for recognition. Through continuous training and confrontation between the generator and the discriminator until the preset conditions are met, the generator can generate prediction results that are closer to the heat wave phenomenon data.

[0009] During the training process, the generator tries to generate data that is increasingly difficult for the discriminator to recognize, while the discriminator strives to improve its recognition ability. When the discriminator judges that the probability of the generated image is 0.5, it is considered that the discriminator cannot determine whether the data is real data or generated data, that is, the prediction results generated by the generator are difficult for the discriminator to recognize, and the training adversary is terminated, that is, the generator has learned how to generate high-quality classification categories of ocean bottom heat waves based on the input data, and the discriminator has reached its performance limit.

[0010] In some embodiments, the step of establishing a data set further comprises: Extracting multidimensional data from the satellite remote sensing data and reanalysis data, wherein the satellite remote sensing data includes sea surface height (SSH) and sea surface current, which are used to characterize the water flow state on the ocean surface. The sea surface current includes a u component and a v component, where the u component is the eastward component and the v component is the northward component. The reanalysis data includes ocean depth data, sea surface temperature (SST), sea surface salinity (SSS), and ocean surface atmospheric data, which are used to characterize the impact of ocean surface environmental conditions on heat waves. The meteorological elements of the ocean surface atmospheric data include total precipitation, sea surface pressure, 2-meter air temperature, 2-meter dew point temperature, and 10-meter wind (u and v components). The radiation parameters of the ocean surface atmospheric data include surface shortwave radiation and surface longwave radiation. Each dimension of data generates a two-dimensional matrix based on the latitude and longitude grid. Each element value in the matrix represents the physical quantity value of the corresponding latitude and longitude point. The distribution of each physical quantity at the latitude and longitude position is visualized as a two-dimensional matrix to facilitate further analysis and processing. Normalization processing is performed on the plurality of two-dimensional matrices, wherein the normalization processing adopts maximum and minimum normalization processing and is performed based on the following calculation model:

[0011] in, is the data value to be processed, min and max are the minimum and maximum values of the dimension data to which it belongs.

[0012] Through normalization operations, data of different dimensions are converted to a unified numerical range, eliminating the impact of different dimensions and value differences on subsequent model training, and ensuring the comparability and stability of the data.

[0013] Considering that land areas are considered invalid in ocean predictions, they are masked to distinguish them. This results in the original 0-1 normalization potentially generating many zero values, which can cause problems in numerical calculations, such as numerical stability and accuracy. Based on the above steps, compressing the normalized values to 0.1-1 can remove these zero values and help alleviate the vanishing gradient problem caused by small gradient values during subsequent model training.

[0014] In some embodiments, the generator adopts a U-NET network, utilizes the encoder downsampling process to gradually extract deep features in the multidimensional data, and then uses the decoder to combine the key features extracted by the encoder with the shallow features extracted by the decoder through upsampling and jump connections to restore the situation of the ocean bottom heat wave.

[0015] In some embodiments, a residual module (RESNET) is used to replace the two-layer convolutional structure of the U-NET network. By adding direct connection paths, the residual module (RESNET) allows gradient signals to bypass some layers and propagate directly. This not only improves the gradient flow problem in deep networks, but also reduces the gradient vanishing phenomenon that is common during training. This makes the training of deep networks more stable because gradients can be more effectively backpropagated to early layers of the network, thereby updating network weights.

[0016] In some embodiments, the discriminator adopts a PatchGAN classifier. This application uses the PatchGAN classifier to output a discriminant matrix. The discriminant matrix is an N×N image area of the longitude and latitude image, where N is greater than 1 and less than the image size. Its value is true or false, indicating whether there is a heat wave phenomenon in the area. The local area judgment method can effectively improve the model's sensitivity to the local area, ensuring that the prediction results can reflect the spatial distribution characteristics of the heat wave.

[0017] In some embodiments, the loss function of the discriminator adopts the FocalLoss function. The function is based on the following calculation model: ,in, is the weight of the c-th class sample, is the probability value of the cth class output by the softmax function of the discriminator, It is an adjustable factor, configured as 2, which is used to reduce the contribution of easy-to-classify samples to the loss function and enhance the model's attention to difficult-to-classify samples. The cth class can be positive (heat wave occurs) or negative (heat wave does not occur).

[0018] In a second aspect, an embodiment of the present application provides an ocean bottom heat wave prediction system for executing the ocean bottom heat wave prediction method described in the first aspect above, the system comprising: The dataset building unit pre-processes the satellite remote sensing data and reanalysis data of the ocean surface based on the latitude and longitude grid to construct a multidimensional dataset; A bottom-layer heat wave prediction unit trains a heat wave prediction model based on the multidimensional dataset. The heat wave prediction model uses a generator to extract features from the multidimensional dataset and output a prediction result. The generator is obtained by training and competing with a discriminator based on historical heat wave phenomenon data of the bottom ocean. The trained heat wave prediction model is used to predict heat wave phenomena in the bottom ocean.

[0019] In a third aspect, an embodiment of the present application provides a computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for predicting ocean bottom heat waves as described in the first aspect above is implemented.

[0020] Datasets were constructed using Aviso satellite remote sensing data and the ERA5 reanalysis dataset. These datasets contain multiple meteorological and oceanographic features, comprehensively reflecting the conditions for heatwave occurrence. In the heatwave prediction step, a generative adversarial network (GAN) was used to train and optimize the prediction model. The generator extracts features from the multidimensional dataset and generates heatwave predictions, while the discriminator verifies these predictions to ensure their accuracy. The adversarial training between the generator and discriminator continuously optimizes the generator's predictive capabilities, improving the accuracy and reliability of heatwave predictions.

[0021] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 is a flow chart of a method for predicting ocean bottom heat waves according to an embodiment of the present application; Figure 2 is a schematic diagram of the principle of a heat wave prediction model according to an embodiment of the present application; Figure 3 is a schematic diagram of the principle of a discriminator according to an embodiment of the present application; Figure 4 This is a structural block diagram of an ocean bottom heat wave prediction system according to an embodiment of the present application.

[0023] In the figure: 1. Dataset creation unit; 2. Underlying heat wave prediction unit. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0025] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0026] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0027] This embodiment provides a method for predicting ocean bottom heat waves. Figure 1 : is a flow chart of the method for predicting ocean bottom heat waves according to an embodiment of the present application. Figure 1 As shown, the process includes the following steps: Dataset creation step S1: preprocessing Aviso's satellite remote sensing data and ERA5's reanalysis data of the ocean surface based on the latitude and longitude grid to construct a multidimensional dataset; In step S2 of predicting a bottom-layer heat wave, a heat wave prediction model is trained based on the multidimensional dataset. The heat wave prediction model uses a generator to extract features from the multidimensional dataset and output a prediction result. The generator is obtained by training and competing with a discriminator based on historical heat wave phenomenon data of the bottom ocean. The trained heat wave prediction model is used to predict heat wave phenomena in the bottom ocean.

[0028] In some embodiments, the historical heat wave phenomenon data is calculated based on the following steps: The sea temperature data of the historical reanalysis data of ERA5 for multiple years is used as the climate reference period. A time window is selected, and the daily average sea temperature value within the time window of the climate reference period is extracted for each calendar day. The climate state value and threshold of the calendar day are calculated based on the daily average sea temperature value. For example, but not limited to, the historical reanalysis data includes multiple years, such as 2019 to 2010. The time window is set to the calendar day and the 5 days before and after the calendar day. An 11-day window is used to reduce short-term fluctuation interference and more accurately reflect the climate state characteristics. The climate state value is the mean or median, and the threshold is the 90th percentile value. In addition, different window lengths can also be used according to the temperature fluctuation characteristics of the target area, such as 7 days before and after or 3 days before and after, to improve the sensitivity of local area identification; The sea temperature on the target date is compared with the threshold for that calendar day to determine whether it has continuously exceeded the threshold for a set period. If so, a marine heatwave event is considered to have occurred on that day, and a heatwave phenomenon is marked at the corresponding latitude and longitude grid point. This generates historical heatwave data for the bottom ocean floor at the corresponding latitude and longitude grid point for each day. The set period is at least five days. In this embodiment, a marine heatwave is defined as an event in which the daily sea temperature in the ocean exceeds the local seasonal threshold (i.e., the 90th percentile of the daily sea temperature during the same period of the climate reference period) for five consecutive days.

[0029] In one embodiment, ERA5 sea temperature data from 2019 to 2010 is used as a climate reference period. The daily average sea temperature over an 11-day window (five days before and after, plus the current day) is used to calculate the daily climate state and threshold values during the same period. Daily temperature anomalies from 2019 to 2023 are calculated based on a set threshold. When the temperature anomaly exceeds the set threshold and persists for at least five days, a marine heatwave event is defined for that grid point. The historical ocean floor heatwave data calculated from 2019 to 2023 is then used as the ground truth for training the discriminator.

[0030] Based on the above steps, by clearly defining the climate reference period and temperature threshold, and combining it with set periodic judgment, historical ocean bottom heat wave events can be identified to ensure the scientific nature of heat wave identification. Heat wave events in different years and regions can also be compared and analyzed, thereby assisting subsequent model training.

[0031] In some of these embodiments, reference Figure 2 As shown, the bottom-level heat wave prediction step S2 further includes: Establish a heat wave prediction model based on conditional generative adversarial networks (CGANs); Inputting the multidimensional dataset into a generator of the heat wave prediction model, the generator being configured to generate classification categories corresponding to grid points of the latitude and longitude grid based on the multidimensional dataset, wherein the classification categories include: presence of a heat wave, absence of a heat wave, and belonging to a land point; The multidimensional data set, classification categories, and historical heat wave phenomenon data are spliced together and input into the discriminator for recognition. Through continuous training and confrontation between the generator and the discriminator until the preset conditions are met, the generator can generate prediction results that are closer to the heat wave phenomenon data.

[0032] During the training process, the generator tries to generate data that is increasingly difficult for the discriminator to recognize, while the discriminator strives to improve its recognition ability. When the discriminator judges that the probability of the generated image is 0.5, it is considered that the discriminator cannot determine whether the data is real data or generated data, that is, the prediction results generated by the generator are difficult for the discriminator to recognize, and the training adversary is terminated, that is, the generator has learned how to generate high-quality classification categories of ocean bottom heat waves based on the input data, and the discriminator has reached its performance limit.

[0033] Undersea heatwaves involve heterogeneous data from multiple sources (such as water temperature, ocean currents, and salinity), and deep-sea observational data is sparse, posing a challenge to GAN training. This embodiment utilizes conditional generative adversarial networks (CGANs) to enhance the adaptability and accuracy of heatwave prediction models by incorporating multidimensional datasets and historical heatwave data. This not only accurately predicts the occurrence of ocean heatwave events, but also clearly distinguishes between land points and non-heatwave areas, further improving the model's applicability.

[0034] In some embodiments, the dataset creation step S1 further includes: Extract multidimensional data from satellite remote sensing data and reanalysis data. The satellite remote sensing data includes sea surface height (SSH) and sea surface current, which are used to characterize the water flow state on the ocean surface. The sea surface current includes a u component and a v component, where the u component is the eastward component and the v component is the northward component. The reanalysis data includes ocean depth data, sea surface temperature (SST), sea surface salinity (SSS), and ocean surface atmospheric data, which are used to characterize the impact of ocean surface environmental conditions on heat waves. The meteorological elements of the ocean surface atmospheric data include total precipitation, sea surface pressure, 2-meter air temperature, 2-meter dew point temperature, and 10-meter wind (u and v components). The radiation parameters of the ocean surface atmospheric data include surface shortwave radiation and surface longwave radiation. The above data are based on Aviso satellite remote sensing data, which directly observes sea surface dynamic parameters (height and flow field) with high spatial resolution but limited coverage. In combination with ERA5 reanalysis data, the model assimilates multi-source data to provide continuous atmospheric and surface ocean parameters and indirectly inferred parameters (such as salinity).

[0035] Each dimension of data is mapped to a two-dimensional matrix based on a longitude and latitude grid. Each element in the matrix represents the value of a physical quantity at the corresponding longitude and latitude point. The distribution of each physical quantity at each longitude and latitude location is visualized as a two-dimensional matrix, facilitating further analysis and processing. For example, but not limited to, the numpy functions arange and meshgrid are used to generate a two-dimensional longitude and latitude coordinate grid. Each value in the matrix represents the measured value at that location, such as temperature or salinity. Based on this, each data point has a corresponding geographic location, creating a unified gridded representation for geospatial data (ocean / atmospheric parameters). Each parameter value is precisely associated with its corresponding longitude and latitude location, facilitating subsequent analysis, model input, or visualization.

[0036] Normalization processing is performed on the plurality of two-dimensional matrices, wherein the normalization processing adopts maximum and minimum normalization processing and is performed based on the following calculation model:

[0037] in, is the data value to be processed, and min and max are the minimum and maximum values of the dimension data to which it belongs. In another embodiment, the normalization processing method can be adjusted according to specific needs, such as using Z-score normalization. The specific method selected should be optimized based on the distribution characteristics of different data and the requirements of subsequent models.

[0038] Through normalization operations, data of different dimensions are converted to a unified numerical range, eliminating the impact of different dimensions and value differences on subsequent model training, and ensuring the comparability and stability of the data.

[0039] Considering that land areas are considered invalid in ocean predictions, they are masked to distinguish them. This results in the original 0-1 normalization potentially generating many zero values, which can cause problems in numerical calculations, such as numerical stability and accuracy. Based on the above steps, compressing the normalized values to 0.1-1 can remove these zero values and help alleviate the vanishing gradient problem caused by small gradient values during subsequent model training.

[0040] In some embodiments, the dataset creation step S1 further includes filling missing values in the data to ensure the integrity of the dataset.

[0041] In some embodiments, the generator adopts a U-NET network, utilizes the encoder downsampling process to gradually extract deep features in the multidimensional data, and then uses the decoder to combine the key features extracted by the encoder with the shallow features extracted by the decoder through upsampling and jump connections to restore the situation of the ocean bottom heat wave.

[0042] The above embodiment uses a U-NET network as a generator, fully leveraging the contextual information of the encoder and decoder for forecasting sub-ocean heat waves. The encoder extracts deep data features that provide powerful abstraction capabilities for the prediction model, while the decoder incorporates shallow features through skip connections, enabling the model to simultaneously preserve global information and local details. This results in more accurate spatial predictions, particularly for ocean data with high spatial resolution. It can effectively recover sub-ocean heat wave conditions, improving the model's accuracy and reliability.

[0043] In some embodiments, a residual module (RESNET) is used to replace the two-layer convolutional structure of the U-NET network. By adding direct connection paths, the residual module (RESNET) allows gradient signals to bypass some layers and propagate directly. This not only improves the gradient flow problem in deep networks, but also reduces the gradient vanishing phenomenon that is common during training. This makes the training of deep networks more stable because gradients can be more effectively backpropagated to early layers of the network, thereby updating network weights.

[0044] In some of these embodiments, reference Figure 3 As shown, the discriminator adopts the PatchGAN classifier. This application uses the PatchGAN classifier to process through the convolution layer and the activation layer to output a discriminant matrix. The discriminant matrix is an N×N image area of the longitude and latitude image, N is greater than 1 and less than the image size, and its value is true or false, respectively indicating whether there is a heat wave phenomenon in the area. The local area judgment method can effectively improve the model's sensitivity to the local area, ensuring that the prediction results can reflect the spatial distribution characteristics of the heat wave.

[0045] In some embodiments, N is configured as 3. This embodiment uses the PatchGAN classifier to classify 3×3 overlapping blocks, and the discriminator performs classification region by region, which can focus on analyzing local areas of the data, capture detailed features of the image, and maintain a small model size. Therefore, in processing the ocean bottom heat wave calculation task, it can effectively analyze large areas of ocean data and identify local heat wave phenomena while maintaining high processing.

[0046] During generative adversarial training, the generator outputs heat wave predictions, while the discriminator, based on the PatchGAN architecture, determines whether a heat wave actually exists in each 3×3 region. Through adversarial training, the generator and discriminator are continuously optimized, ultimately reaching a state of equilibrium where the generator can accurately output heat wave predictions and the discriminator can precisely determine the heat wave situation in each region.

[0047] In the above embodiment, considering that the ocean heat wave phenomenon is a small sample phenomenon, the loss function of the discriminator adopts the FocalLoss function. This function is based on the following calculation model: ,in, is the weight of the c-th class sample, is the probability value of the cth class output by the softmax function of the discriminator, This is an adjustable factor, configured to 2, that reduces the contribution of easily classified samples to the loss function and increases the model's focus on difficult-to-classify samples. Class c can be either positive (heat wave occurrence) or negative (heat wave non-occurrence). By introducing this adjustable factor, the contribution of easily classified samples to the loss function is significantly reduced, while the weight of difficult-to-classify samples is increased. This helps the model focus on small, difficult-to-classify samples, improving detection accuracy. Especially when heat wave events are rare, Focal Loss can significantly improve the model's classification accuracy for heat wave samples, thereby enhancing heat wave prediction performance.

[0048] A dataset was established by utilizing Aviso satellite remote sensing data and the ERA5 reanalysis dataset. These datasets contain multiple meteorological and ocean-related features and can comprehensively reflect the conditions for the occurrence of heat wave phenomena. In the heat wave prediction step, a generative adversarial network (GAN) is used to train and optimize the prediction model. The generator is responsible for extracting features from the multidimensional dataset and generating heat wave predictions, while the discriminator verifies these predictions to ensure the accuracy of the prediction results. The adversarial training method between the generator and the discriminator continuously optimizes the generator's prediction ability and improves the accuracy and reliability of heat wave predictions. This method can effectively predict bottom-ocean heat waves, provide accurate prediction information to marine researchers and relevant departments, and assist in disaster prevention and emergency response. By combining satellite remote sensing data and reanalysis data, the method can comprehensively reflect the various factors affecting marine heat waves, and the adversarial training of the generative adversarial network improves the accuracy of the prediction, which has significant advantages over traditional heat wave prediction methods.

[0049] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0050] This embodiment also provides an ocean bottom heatwave prediction system, which is used to implement the above-mentioned embodiments and preferred embodiments. Details already described are omitted. As used below, terms such as "module," "unit," and "subunit" may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0051] Figure 4 is a structural block diagram of the ocean bottom heat wave prediction system according to an embodiment of the present application, such as Figure 4 As shown, the system includes: Dataset building unit 1, preprocessing the satellite remote sensing data and reanalysis data of the ocean surface based on the latitude and longitude grid to construct a multidimensional dataset; The bottom-layer heat wave prediction unit 2 trains a heat wave prediction model based on the multidimensional dataset. The heat wave prediction model uses a generator to extract features from the multidimensional dataset and output a prediction result. The generator is obtained by training and competing with a discriminator based on historical heat wave phenomenon data of the bottom ocean. The trained heat wave prediction model is used to predict heat wave phenomena in the bottom ocean.

[0052] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0053] In addition, combined Figure 1 The ocean bottom heat wave prediction method described in the embodiment of the present application can be implemented by a computer device. The computer device may include a processor and a memory storing computer program instructions.

[0054] Specifically, the processor may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0055] The memory may include a large-capacity memory for data or instructions. By way of example, and not limitation, the memory may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory may include removable or non-removable (or fixed) media. Where appropriate, the memory may be internal or external to the data processing device. In a specific embodiment, the memory is non-volatile memory. In a specific embodiment, the memory includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data out dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0056] The memory may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor.

[0057] The processor implements any one of the ocean bottom heat wave prediction methods in the above embodiments by reading and executing computer program instructions stored in the memory.

[0058] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0059] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for predicting ocean bottom heat waves, characterized in that: include: The dataset establishment step is to pre-process the satellite remote sensing data and reanalysis data of the ocean surface based on the latitude and longitude grid to construct a multidimensional dataset; The bottom layer heat wave prediction step includes training a heat wave prediction model based on the multidimensional data set. The heat wave prediction model uses a generator to extract features from the multidimensional data set and output a prediction result. The generator is obtained by training and competing with a discriminator based on historical heat wave phenomenon data of the bottom layer of the ocean. The trained heat wave prediction model is used to predict heat wave phenomena in the bottom layer of the ocean.

2. The method for predicting ocean bottom heat waves according to claim 1, wherein: The historical heat wave data is calculated based on the following steps: The sea temperature data of multiple years of historical reanalysis data are used as a climate reference period, a time window is selected, and for each calendar day, the daily average sea temperature value within the time window of the climate reference period is extracted, and the climate state value and threshold value of the calendar day are calculated based on the daily average sea temperature value; The sea temperature on the target date is compared with the threshold of that calendar day to determine whether it exceeds the threshold continuously for a set period. If so, it is considered that an ocean heat wave event has occurred and is marked as a heat wave phenomenon at the corresponding latitude and longitude grid points, thereby obtaining the historical heat wave phenomenon data of the ocean bottom layer at the corresponding latitude and longitude grid points on each historical day.

3. The method for predicting ocean bottom heat waves according to claim 2, characterized in that: The bottom-level heat wave prediction step further comprises: Establish a heat wave prediction model based on conditional generative adversarial networks (CGANs); Inputting the multidimensional dataset into a generator of the heat wave prediction model, the generator being configured to generate classification categories corresponding to grid points of the latitude and longitude grid based on the multidimensional dataset, wherein the classification categories include: presence of a heat wave, absence of a heat wave, and belonging to a land point; The multidimensional dataset, classification categories, and historical heat wave phenomenon data are spliced together and input into the discriminator for recognition. The generator and the discriminator are continuously trained and confronted until the preset conditions are met.

4. The method for predicting ocean bottom heat waves according to claim 2 or 3, characterized in that: The data set creation step further comprises: Extracting multidimensional data from the satellite remote sensing data and reanalysis data; Each dimension of data generates a two-dimensional matrix based on the latitude and longitude grid. Each element value in the matrix represents the physical value of the corresponding latitude and longitude point. Normalization is performed on the plurality of two-dimensional matrices.

5. The method for predicting ocean bottom heat waves according to claim 3, characterized in that: The generator adopts U-NET network.

6. The method for predicting ocean bottom heat waves according to claim 5, characterized in that: The residual module RESNET is used to replace the double-layer convolutional structure of the U-NET network.

7. The method for predicting ocean bottom heat waves according to claim 3, wherein: The discriminator adopts PatchGAN classifier.

8. The method for predicting ocean bottom heat waves according to claim 3, characterized in that: The loss function of the discriminator adopts the FocalLoss function.

9. An ocean bottom heat wave prediction system, characterized in that: The system is used to execute the ocean bottom heat wave prediction method according to any one of claims 1 to 8, comprising: The dataset building unit pre-processes the satellite remote sensing data and reanalysis data of the ocean surface based on the latitude and longitude grid to construct a multidimensional dataset; A bottom-layer heat wave prediction unit trains a heat wave prediction model based on the multidimensional dataset. The heat wave prediction model uses a generator to extract features from the multidimensional dataset and output a prediction result. The generator is obtained by training and competing with a discriminator based on historical heat wave phenomenon data of the bottom ocean. The trained heat wave prediction model is used to predict heat wave phenomena in the bottom ocean.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the ocean bottom heat wave prediction method according to any one of claims 1 to 8 is implemented.

Images