A sea surface temperature sensitive area identification method, device, equipment and medium

Abstract

The application discloses a sea surface temperature sensitive area identification method and device, equipment and medium, and relates to the technical field of weather forecasting. The method comprises the following steps: acquiring model input data, and constructing a target function for measuring the heat wave event intensity of a target area; freezing the model parameters of a climate prediction model, setting the sea surface temperature field as a unique derivable variable, and inputting the model input data into the climate prediction model for forward propagation to obtain a temperature forecast result; calculating the target function values corresponding to the target function in different time periods according to the temperature forecast result, and determining the sea temperature gradient sensitivity fields corresponding to the different time periods based on the target function values; and generating a sea surface temperature sensitive area mask according to the sea temperature gradient sensitivity fields. Through the generation of the sea surface temperature sensitive area mask according to the sea temperature gradient sensitivity fields, the problem that it is difficult to determine which sea surface temperature changes of the ocean grid points will affect the prediction of the heat wave event is solved.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting technology, and in particular to a method, apparatus, equipment and medium for identifying sea surface temperature-sensitive areas. Background Technology

[0002] Sea surface temperature (SST) is a crucial slowly varying boundary forcing source in subseasonal forecasting. SST anomalies in key sea areas can influence the occurrence and development of East Asian heat waves through processes such as air-sea interactions, anomalous convective activity, teleconnection wave trains, and subtropical high-pressure anomalies. Therefore, identifying the SST regions most sensitive to East Asian heat wave forecasts on a subseasonal scale is essential for improving the interpretability of extreme heat wave forecasts and supporting target observations.

[0003] Current methods for identifying sea surface temperature rely on statistical correlation analysis, composite analysis, and ensemble sensitivity analysis. These methods struggle to determine which ocean grid points have sea surface temperature changes that will most significantly affect the model's prediction of target heat wave events. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for identifying sea surface temperature sensitive areas. This method generates a sea surface temperature sensitive area mask based on the sea temperature gradient sensitivity field, solving the problem of difficulty in determining which ocean grid points' sea surface temperature changes will affect the forecasting of visible wave events. The specific solution is as follows:

[0005] In a first aspect, this application provides a method for identifying sea surface temperature-sensitive areas, including:

[0006] Obtain the target model input data corresponding to the pre-trained target climate prediction model, and construct an objective function to measure the intensity of heat wave events in the target region; wherein, the target model input data includes the atmospheric state field, the sea surface temperature field, and the corresponding time coding information;

[0007] The model parameters of the target climate prediction model are frozen, the sea surface temperature field is set as the only differentiable variable, and the target model input data is input into the target climate prediction model for forward propagation to obtain the temperature forecast results within the target forecast period.

[0008] Based on the temperature forecast results, the objective function values ​​corresponding to the objective function in different time periods are calculated, and the sea surface temperature gradient sensitivity fields corresponding to different time periods are determined based on the objective function values; wherein, the sea surface temperature gradient sensitivity field is a spatial distribution field characterizing the degree of influence of sea surface temperature changes at each ocean grid point on the intensity of heat wave events;

[0009] Based on the sea surface temperature gradient sensitivity fields described above, a sea surface temperature sensitive area mask is generated to complete the identification of sea surface temperature sensitive areas.

[0010] Optionally, the atmospheric state field includes an atmospheric temperature field, an atmospheric humidity field, and a sea level pressure field; the sea surface temperature field includes ocean surface temperature variables.

[0011] Optionally, the objective function for measuring the intensity of heat wave events in the target area includes:

[0012] A first target term is constructed to characterize the degree to which the temperature in the target area exceeds a preset temperature threshold, and a second target term is constructed to characterize the proportion of the target area that reaches the preset temperature threshold;

[0013] A third target term is constructed to characterize the duration of high temperature corresponding to a preset high temperature within the target region, and the target function is constructed using the first target term, the second target term, and the third target term; wherein, the preset high temperature is a temperature exceeding the preset temperature threshold.

[0014] Optionally, before inputting the target model input data into the target climate prediction model for forward propagation, the method further includes:

[0015] Spatial interpolation is performed on each modal data in the input data of the target model to align the spatial scale of each modal data, and temporal alignment and missing value filling operations are performed on each modal data after spatial scale alignment.

[0016] Optionally, determining the sea surface temperature gradient sensitivity field corresponding to different time periods based on the objective function value includes:

[0017] The objective function value is backpropagated along the forward propagation computation link of the target climate prediction model to obtain the partial derivatives of the objective function value with respect to each ocean grid point in the sea surface temperature field, and the sea surface temperature gradient sensitivity field is determined based on each partial derivative; wherein, the sea surface temperature field is the spatial distribution field corresponding to the sea surface temperature variable in the input data of the target model, which is composed of the sea surface temperature values ​​of several grid points covering the global ocean area, and the partial derivatives characterize the impact of the sea surface temperature rise at each grid point on the intensity of heat wave events.

[0018] Optionally, generating a sea surface temperature sensitive area mask based on each of the sea surface temperature gradient sensitivity fields includes:

[0019] The target weights corresponding to each of the sea surface temperature gradient sensitivity fields are determined, and the sea surface temperature gradient sensitivity fields are weighted and fused according to the target weights to obtain a comprehensive sensitivity field. The sea surface temperature sensitive area mask is generated based on the comprehensive sensitivity field. The comprehensive sensitivity field represents the overall impact of sea surface temperature on heat wave events in different forecast time periods. Each ocean grid point in the comprehensive sensitivity field corresponds to a sensitivity value, and the sensitivity value represents the impact of the sea surface temperature change at the current ocean grid point on the intensity of the heat wave event.

[0020] Optionally, generating the sea surface temperature sensitive area mask based on the integrated sensitivity field includes:

[0021] The gradient sign consistency test is performed on the comprehensive sensitivity field. Grid points with consistent gradient signs in different time periods are retained, while noisy grid points with changing signs in different time periods are removed to obtain the corresponding comprehensive sensitivity field after the test. Gradient sign consistency characterizes whether the direction of the effect of the same ocean grid point on the heat wave event is stable in different forecast time periods.

[0022] Spatial connectivity screening is performed on each candidate grid point in the comprehensive sensitivity field after the test. Spatially adjacent candidate grid points are divided into the same connected region, and isolated connected regions with an area smaller than a preset threshold are removed to obtain the sea surface temperature sensitive area mask; wherein, the candidate grid point is a grid point with a sensitivity value not less than a preset score threshold.

[0023] Secondly, this application provides a sea surface temperature-sensitive area identification device, comprising:

[0024] The function construction module is used to obtain the target model input data corresponding to the pre-trained target climate prediction model and construct an objective function to measure the intensity of heat wave events in the target region; wherein, the target model input data includes atmospheric state field, sea surface temperature field and corresponding time coding information;

[0025] The parameter freezing module is used to freeze the model parameters of the target climate prediction model, set the sea surface temperature field as a unique differentiable variable, and input the target model input data into the target climate prediction model for forward propagation to obtain the temperature forecast results within the target forecast period.

[0026] The function value calculation module is used to calculate the objective function values ​​corresponding to the objective function in different time periods based on the temperature forecast results, and to determine the sea surface temperature gradient sensitivity field corresponding to different time periods based on the objective function values; wherein, the sea surface temperature gradient sensitivity field is a spatial distribution field characterizing the degree of influence of sea surface temperature changes at each ocean grid point on the intensity of heat wave events;

[0027] The sensitive area identification module is used to generate a sea surface temperature sensitive area mask based on each of the sea temperature gradient sensitivity fields, so as to complete the identification of sea surface temperature sensitive areas.

[0028] Thirdly, this application provides an electronic device, comprising:

[0029] Memory, used to store computer programs;

[0030] A processor is used to execute the computer program to implement the aforementioned method for identifying sea surface temperature-sensitive areas.

[0031] Fourthly, this application provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the aforementioned method for identifying sea surface temperature-sensitive areas.

[0032] This application first obtains the target model input data corresponding to the pre-trained target climate prediction model and constructs an objective function to measure the intensity of heat wave events in the target area. The target model input data includes an atmospheric state field, a sea surface temperature field, and corresponding time-coded information. Then, the model parameters of the target climate prediction model are frozen, the sea surface temperature field is set as a unique differentiable variable, and the target model input data is input into the target climate prediction model for forward propagation to obtain temperature forecast results within the target forecast period. Subsequently, based on the temperature forecast results, the objective function values ​​corresponding to the objective function in different time periods are calculated, and the sea surface temperature gradient sensitivity fields corresponding to different time periods are determined based on the objective function values. The sea surface temperature gradient sensitivity field is a spatial distribution field characterizing the degree of influence of sea surface temperature changes at each ocean grid point on the intensity of heat wave events. Finally, a sea surface temperature sensitive area mask is generated based on each sea surface temperature gradient sensitivity field to complete the identification of sea surface temperature sensitive areas. Therefore, this application addresses the problem of lacking target identification of event-oriented sensitive areas in existing technologies by constructing an objective function to measure the intensity of heat wave events; it solves the problems of high computational cost and low efficiency of traditional perturbation experiments by freezing model parameters and setting the sea surface temperature field as the only differentiable variable, and combining forward and backward propagation to calculate the sea surface temperature gradient sensitivity field; and it solves the problem of existing methods having difficulty in determining which ocean grid points' sea surface temperature changes will most significantly affect the model's prediction of target heat wave events by generating a sea surface temperature sensitive area mask based on the sea surface temperature gradient sensitivity field. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0034] Figure 1 This is a schematic flowchart of a method for identifying sea surface temperature-sensitive areas disclosed in this application.

[0035] Figure 2 This application discloses a flowchart for identifying sea surface temperature-sensitive areas.

[0036] Figure 3 This is a schematic diagram of a sea surface temperature-sensitive area identification device disclosed in this application.

[0037] Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] See Figure 1 As shown, this embodiment of the invention discloses a method for identifying sea surface temperature-sensitive areas, including:

[0040] Step S11: Obtain the target model input data corresponding to the pre-trained target climate prediction model, and construct an objective function to measure the intensity of heat wave events in the target area; wherein, the target model input data includes the atmospheric state field, the sea surface temperature field, and the corresponding time coding information.

[0041] The core process of this solution is as follows: Figure 2As shown, this includes: acquiring the atmospheric state field, sea surface temperature field, and time-coded information required for the FuXi-S2S model (i.e., the target climate prediction model). In one specific implementation, a differentiable objective function containing heat wave temperature intensity, heat wave coverage area, and heat wave duration can be constructed for extreme heat wave events in East Asia during the second to fourth weeks of the future. With the FuXi-S2S model parameters frozen, gradient calculation is enabled only for the input sea surface temperature field. The gradient sensitivity of the objective function to the sea surface temperature field is obtained through automatic differentiation and backpropagation. Further, multi-forecast time-weighted fusion, gradient sign consistency constraints, and spatial connectivity filtering are used to generate a sea surface temperature sensitive area mask, sensitivity level map, action type identifier, and priority ranking results for extreme heat wave events in East Asia. The target area in this embodiment can be set according to actual application needs, preferably 20°N-55°N and 100°E-145°E, but can also be set to North China, Northeast China, East China, South China, and the middle and lower reaches of the Yangtze River according to operational needs. The preferred lead time for target forecasts is from the 2nd to the 4th week, and daily, pentad, or weekly average forecast results can be used.

[0042] The specific process of the solution is as follows:

[0043] (1) Obtain the input data of the FuXi-S2S model. The input data includes at least the atmospheric state field, sea surface temperature field and time coding information, and the input data is spatially aligned, temporally aligned and standardized.

[0044] (2) Determine the target area and target forecast lead time for the East Asian extreme heat wave event, wherein the target area is the East Asian land area and the target forecast lead time includes one or more forecast periods from the second to the fourth week of the next season forecast;

[0045] (3) Construct an objective function for extreme heat wave events in East Asia based on the 2 m air temperature, daily maximum air temperature or air temperature anomaly output by the FuXi-S2S model. The objective function includes at least two of the following: heat wave temperature intensity term, heat wave coverage area term and heat wave duration term.

[0046] (4) Load the pre-trained FuXi-S2S model and freeze the model parameters, setting only the input sea surface temperature field as a differentiable variable;

[0047] (5) Input the input data obtained in step (1) into the FuXi-S2S model for forward forecasting to obtain the East Asian temperature forecast results within the target forecast period;

[0048] (6) Calculate the target value of the East Asian extreme heat wave event according to the objective function described in step (3), and calculate the gradient of the objective function with respect to the input sea surface temperature field through backpropagation to obtain the sea surface temperature gradient sensitivity field;

[0049] (7) Perform time-weighted fusion of sea surface temperature gradient sensitivity fields corresponding to multiple forecast lead times to generate a comprehensive sea surface temperature sensitivity field;

[0050] (8) Based on gradient sign consistency constraints and spatial connectivity constraints, the comprehensive sea surface temperature sensitivity field is screened to generate a sea surface temperature sensitive area mask;

[0051] (9) Output the sea surface temperature sensitive area mask, sensitivity level map, sensitive area action type and sensitive area priority ranking results.

[0052] It should be noted that the atmospheric state field in this embodiment includes one or more of the following: geopotential height field, temperature field, zonal wind field, meridional wind field, humidity field, sea level pressure field, 2 m air temperature field, and 10 m wind field; the sea surface temperature field is the sea surface temperature channel in the FuXi-S2S input, or the ocean state variable channel corresponding to the ocean surface temperature in the model input; furthermore, the sea surface temperature sensitive area in this embodiment includes positive and negative sensitive areas; a positive sensitive area indicates that the increase in sea surface temperature in the region enhances the objective function of the East Asian extreme heat wave event, while a negative sensitive area indicates that the increase in sea surface temperature in the region weakens the objective function of the East Asian extreme heat wave event. In this embodiment, the sea surface temperature sensitive area mask is used for the deployment of ocean target observations, key correction of the sea surface temperature boundary field, ensemble perturbation construction, forecast uncertainty diagnosis, or extreme high temperature risk warning in the sub-seasonal forecast of the East Asian extreme heat wave event.

[0053] In this embodiment, the process of constructing an objective function to measure the intensity of a heat wave event in a target area includes: constructing a first objective term characterizing the degree to which the temperature in the target area exceeds a preset temperature threshold, and constructing a second objective term characterizing the proportion of the target area that reaches the preset temperature threshold; constructing a third objective term characterizing the duration of high temperature corresponding to a preset high temperature in the target area, and constructing the objective function using the first objective term, the second objective term, and the third objective term; wherein, the preset high temperature is the temperature exceeding the preset temperature threshold.

[0054] The above objective function Specifically:

[0055] ;

[0056] Among them, This indicates the temperature intensity of the heat wave. This indicates the area covered by the heat wave. Indicates the duration of the heat wave. These are the weighting coefficients.

[0057] The heatwave temperature intensity term is constructed based on the positive temperature anomaly in the target area that exceeds the 90th or 95th percentile climatological threshold; the heatwave coverage area term is constructed based on the proportion of the grid area exceeding the high temperature threshold to the total area of ​​the target area; the heatwave duration term is constructed based on the number of forecast days that continuously exceed the high temperature threshold, and the threshold for continuous days is 3 to 6 days.

[0058] In addition, to make the objective function of the heat wave event applicable to back propagation, this embodiment uses the sigmoid function to approximate the binary determination of exceeding the high temperature threshold and the softplus function to approximate the positive anomaly truncation of exceeding the high temperature threshold, so that the heat wave temperature intensity term, heat wave coverage area term, and heat wave duration term are all differentiable functions.

[0059] The heat wave objective function (i.e., the objective function) in this embodiment is constructed based on the 2 m air temperature, daily maximum air temperature, or air temperature anomaly output by FuXi-S2S. The objective function is used to measure the intensity, spatial range, and duration of extreme high temperatures within the target area for a specified period in the future.

[0060] Step S12: Freeze the model parameters of the target climate prediction model, set the sea surface temperature field as the only differentiable variable, and input the target model input data into the target climate prediction model for forward propagation to obtain the temperature forecast results within the target forecast period.

[0061] In this embodiment, a pre-trained FuXi-S2S model is loaded and all model parameters are frozen, preventing the model parameters from participating in gradient updates. Gradient calculation is only enabled for the input sea surface temperature field; other input variables participate in forward propagation as a fixed background field.

[0062] In this embodiment, before inputting the target model input data into the target climate prediction model for forward propagation, the method further includes: performing spatial interpolation on each modal data in the target model input data to align the spatial scale of each modal data, and performing temporal alignment and missing value supplementation on the spatially scale aligned modal data.

[0063] In other words, before inputting the data into the model, spatial interpolation, time alignment, missing measurement processing, land and sea masking, and variable-wise standardization need to be performed on the data.

[0064] In this embodiment, the preprocessed input field is fed into the FuXi-S2S model for forward forecasting to obtain the East Asian temperature forecast results for the second to fourth weeks, and the target event intensity value is calculated based on the objective function of the heat wave event.

[0065] Step S13: Calculate the objective function values ​​corresponding to the objective function in different time periods based on the temperature forecast results, and determine the sea surface temperature gradient sensitivity field corresponding to different time periods based on the objective function values; wherein, the sea surface temperature gradient sensitivity field is a spatial distribution field characterizing the degree of influence of sea surface temperature changes at each ocean grid point on the intensity of heat wave events.

[0066] This embodiment uses the objective function of a heatwave event as the backpropagation target. An automatic differentiation framework is used to calculate the gradient of the objective function with respect to the input SST (Sea Surface Temperature) field, obtaining the SST gradient sensitivity for each ocean grid point. The absolute value of the gradient represents the sensitivity, and the gradient sign indicates whether the increase in SST in that region enhances or weakens the objective function of the heatwave.

[0067] In this embodiment, determining the sea surface temperature gradient sensitivity field corresponding to different time periods based on the objective function value includes: backpropagating the objective function value along the forward propagation calculation link of the target climate prediction model to obtain the partial derivative of the objective function value with respect to each ocean grid point in the sea surface temperature field, and determining the sea surface temperature gradient sensitivity field based on each partial derivative; wherein, the sea surface temperature field is the spatial distribution field corresponding to the sea surface temperature variable in the input data of the target model, which is composed of the sea surface temperature values ​​of several grid points covering the global ocean area, and the partial derivatives characterize the impact of the sea surface temperature rise at each grid point on the intensity of heat wave events.

[0068] That is, the sea surface temperature gradient sensitivity field is the partial derivative of the objective function with respect to the input sea surface temperature field:

[0069] ;

[0070] Among them, For the target forecast lead time, i and j are spatial grid indexes. The absolute value of represents the sensitivity of the corresponding ocean grid point to the objective function of the East Asian extreme heat wave event.

[0071] Step S14: Generate a sea surface temperature sensitive area mask based on each of the sea surface temperature gradient sensitivity fields to complete the identification of sea surface temperature sensitive areas.

[0072] In this embodiment, generating a sea surface temperature sensitive area mask based on each sea surface temperature gradient sensitivity field includes: determining the target weights corresponding to each sea surface temperature gradient sensitivity field; performing weighted fusion of each sea surface temperature gradient sensitivity field according to the target weights to obtain a comprehensive sensitivity field; and generating a sea surface temperature sensitive area mask based on the comprehensive sensitivity field. The comprehensive sensitivity field characterizes the overall impact of sea surface temperature on heatwave events within different forecast time periods. Each ocean grid point in the comprehensive sensitivity field corresponds to a sensitivity value, and the sensitivity value characterizes the impact of the sea surface temperature change at the current ocean grid point on the intensity of the heatwave event.

[0073] The process of generating the sea surface temperature sensitive area mask based on the comprehensive sensitivity field can specifically include: performing a gradient sign consistency test on the comprehensive sensitivity field, retaining grid points whose gradient signs remain consistent across different time periods, and removing noisy grid points whose signs change across different time periods to obtain the corresponding post-test comprehensive sensitivity field; wherein, gradient sign consistency characterizes whether the direction of the effect of the same ocean grid point on the heat wave event is stable across different forecast time periods; performing spatial connectivity screening on each candidate grid point in the post-test comprehensive sensitivity field, dividing spatially adjacent candidate grid points into the same connected region, and removing isolated connected regions with an area smaller than a preset threshold to obtain the sea surface temperature sensitive area mask; wherein, candidate grid points are grid points with sensitivity values ​​not less than a preset score threshold.

[0074] The formula for time-weighted fusion of the sensitivity fields of each sea surface temperature gradient is as follows:

[0075] ;

[0076] Where Ω represents the target forecast lead time set. The forecast lead time weights for the target are determined based on the forecast focus period or historical verification accuracy.

[0077] Furthermore, the aforementioned gradient sign consistency constraint is as follows: For the same ocean grid point across multiple forecast lead times, a consistency check is performed on the gradient sign, and only grid points with a sign consistency index greater than a preset threshold are retained (in some specific implementations, only grid points whose gradient signs remain consistent across different time periods may be retained). The sign consistency index is:

[0078] ;

[0079] in The sign function is defined as follows: when the input value is greater than 0, it takes the value 1; when the input value is equal to 0, it takes the value 0; and when the input value is less than 0, it takes the value −1. This function characterizes the direction of the effect of sea surface temperature rise on the objective function of the heat wave event at the corresponding ocean grid point. Furthermore, the spatial connectivity constraint is as follows: Connectivity component identification is performed on the ocean grid points that have passed the sensitivity threshold and sign consistency screening. Discrete grid points or small patches with an area smaller than a preset area threshold are removed, retaining only spatially continuous sea surface temperature sensitive areas.

[0080] The process of symbol consistency constraint and spatial connectivity screening mentioned above is to check the consistency of gradient symbols of the same ocean grid point across multiple forecast lead times, and eliminate noisy grid points with frequently changing symbols; further, to identify connected components of highly sensitive grid points, eliminate isolated patches with areas smaller than the threshold, and retain SST sensitive areas that are spatially continuous and have strong physical interpretability.

[0081] In this embodiment, the SST sensitive area mask is generated based on the overall sensitivity, symbol consistency and connectivity area. The sensitive areas are divided into high, medium and low levels, and the priority ranking of the sensitive areas is output according to the average sensitivity, area and stability of the region.

[0082] In summary, the beneficial effects of this embodiment include: First, it eliminates the need to retrain the FuXi-S2S model or change its structure; event-oriented SST sensitive areas can be obtained simply by utilizing its differentiable forecasting link. Second, it focuses on the target region as the analysis object, making it more event-specific than traditional correlation analysis. Third, it transforms the temperature threshold, coverage area, and duration in the definition of heat waves into differentiable objective functions, making it compatible with backpropagation calculations. Fourth, through time-weighted calculations, sign consistency constraints, and spatial connectivity filtering, the original gradient field is transformed into a directly usable sensitive area mask. Fifth, the sensitive area mask, action type, and priority ranking output by this invention can serve marine target observation, key correction of sea surface temperature boundary fields, ensemble forecast perturbation design, and extreme heat wave risk warning.

[0083] The solution in this embodiment can be deployed in sub-seasonal climate prediction operational systems, extreme high-temperature risk early warning platforms, marine target observation design systems, and power meteorological service systems. This method requires only one or a small number of FuXi-S2S forward forecasts and backpropagation calculations to obtain the target event-oriented SST sensitive area, and features high computational efficiency, high degree of automation, intuitive output results, and strong engineering usability.

[0084] In one specific implementation, the target event is the extreme summer heatwave event in East Asia, with the target area set at 20°N-55°N and 100°E-145°E, and the target lead time set at the second, third, and fourth weeks of the FuXi-S2S forecast. Input data includes the atmospheric state field, sea surface temperature field, and time-coded information at the initial time of the FuXi-S2S forecast.

[0085] First, the input data is preprocessed. All variables are interpolated to the grid required by the FuXi-S2S model; each variable is standardized using historical climate mean and standard deviation; a land-sea mask is applied to the sea surface temperature field, preserving the SST gradient only at the ocean grid points; and the forecast initialization date is encoded as intra-annual day sequence and seasonal cycle information.

[0086] Secondly, determine the high-temperature threshold. Let... The temperature at 2 m or the maximum temperature on forecast day t is the temperature of the i-th land grid point in the target area. This refers to the climatological high temperature threshold corresponding to this grid point and day sequence. The high temperature threshold can be the 90th or 95th percentile of the climatological data, or a fixed high temperature threshold can be used according to operational standards. To adapt to different climatological backgrounds in East Asia, a grid-by-grid, day-by-day percentile threshold is preferred.

[0087] Furthermore, to ensure the objective function can be used for backpropagation, the binary decision function exceeding the high-temperature threshold is transformed into a sigmoid smoothing function, and the positive anomaly truncation function exceeding the threshold is transformed into a softplus smoothing function. Specifically, a soft high-temperature indicator is defined. and soft overthreshold strength as follows:

[0088] ;

[0089] ;

[0090] in This represents the sigmoid function. To smooth out the temperature scale parameter, the preferred value range is 0.5-2.0K. The closer the value is to 1, the more likely that the grid point is above the high temperature threshold; This represents the differentiable positive anomaly intensity exceeding the high temperature threshold.

[0091] Then, construct the heat wave temperature intensity term. This item represents the area-weighted average of the high-temperature exceedance intensity within the target area and the target duration, calculated using the following formula:

[0092] ;

[0093] Where D represents the set of land grid points in the target area of ​​East Asia, and Ω represents the target forecast lead time set. This represents the area weight of each grid point.

[0094] Constructing the heat wave coverage area item This item represents the proportion of soft area within the target region that reaches the high-temperature threshold, and the calculation formula is:

[0095] ;

[0096] Constructing heat wave sustainability items If the duration of a heat wave is determined by exceeding the high-temperature threshold for m consecutive days, then the area-weighted average of the product of soft high-temperature indicators within a rolling window can be used as the duration term. The calculation formula is as follows:

[0097] ;

[0098] Where m is the threshold for the number of consecutive days, preferably 3-6 days; This is the set of forecast days that can form an m-day rolling window. This definition preserves the information of "continuous high temperatures" during backpropagation, while avoiding the non-differentiability problem caused by hard thresholding and discrete counting.

[0099] Combining the above three factors, the objective function for the East Asian extreme heat wave event is constructed as follows:

[0100] ;

[0101] in These are weighting coefficients for temperature intensity, coverage area, and duration, respectively. Preferably, when the focus is on high-temperature intensity, the following settings are used: =0.5、 =0.3、 =0.2; When focusing on the risk of persistent heat waves, increase The value of .

[0102] Load the pre-trained FuXi-S2S model and freeze all weight parameters. Set the input SST field as a differentiable variable with `requires_grad=True`, and set other input variables as a fixed background field. After performing forward prediction, calculate the heatwave objective function values ​​for weeks 2, 3, and 4. , and .

[0103] Using the objective function value of each forecast lead time as the object of backpropagation, the sea surface temperature gradient sensitivity is calculated respectively:

[0104] ;

[0105] Where i and j represent the positions of ocean grid points. If If it is greater than zero, it means that increasing the SST at that grid point will enhance the target heat wave function; if If the value is less than zero, it means that increasing the SST at that grid point will weaken the target heat wave function.

[0106] Multi-time-weighted fusion is performed. The comprehensive sensitivity field is defined as:

[0107] ;

[0108] in, , and These represent the objective functions for the ocean grid points in weeks 2, 3, and 4, respectively. Sensitivity to sea surface temperature gradients; , and These represent the timeliness weights for weeks 2, 3, and 4, respectively. , If the focus is on the risk of a stable heatwave in weeks 3-4, settings can be configured. =0.25、 =0.35、 =0.40; if you are focusing on more recent forecasts, you can increase the weight of the second week.

[0109] Perform a sign consistency test. Define the sign consistency index:

[0110] ;

[0111] in, , and The meaning is the same as before; This represents the sign function, which takes values ​​of 1, 0, and -1 when the input value is greater than 0, equal to 0, and less than 0, respectively. , , Representing symbolic function pairs respectively , and The possible values ​​of ; Representing ocean grid points The gradient sign consistency index across different forecast lead times ranges from 0 to 1. A value closer to 1 indicates a more stable influence of the grid point on the heatwave event across different forecast lead times. Less than the preset threshold If the grid point is deemed to have unstable sensitivity, it will be removed. The preferred value range is 0.5-0.8.

[0112] Spatial connectivity screening is performed. First, grid points whose comprehensive sensitivity field S(i,j) is located in the top 5%-15% quantile of the ocean grid points are selected as candidate sensitive grid points; then, eight-neighbor connectivity component identification is performed on the candidate grid points; finally, grid points with an area smaller than [a certain value] are removed. Connected patches. It can be set according to the grid resolution. Other business thresholds.

[0113] Generate SST sensitive region masks. For the preserved connected components, comprehensively score them based on average regional sensitivity, area, symbol consistency index, and duration, classifying them into high-sensitivity, medium-sensitivity, and low-sensitivity regions, and outputting the sensitivity region's effect type. Positively affected sensitive regions indicate that SST increases in these regions tend to enhance the East Asian heatwave objective function, while negatively affected sensitive regions indicate that SST increases in these regions tend to weaken the East Asian heatwave objective function.

[0114] The final output includes: a mask file for sensitive areas of the East Asian extreme heat wave (SST) event, a sensitivity level map, a map of the types of effects of sensitive areas, boundary coordinates of each sensitive area, average sensitivity of the region, symbol consistency index, connectivity area, and a priority ranking table.

[0115] Therefore, this application addresses the problem of lacking target identification of event-oriented sensitive areas in existing technologies by constructing an objective function to measure the intensity of heat wave events; it solves the problems of high computational cost and low efficiency of traditional perturbation experiments by freezing model parameters and setting the sea surface temperature field as the only differentiable variable, and combining forward and backward propagation to calculate the sea surface temperature gradient sensitivity field; and it solves the problem of existing methods having difficulty in determining which ocean grid points' sea surface temperature changes will most significantly affect the model's prediction of target heat wave events by generating a sea surface temperature sensitive area mask based on the sea surface temperature gradient sensitivity field.

[0116] See Figure 3 As shown, an embodiment of the present invention discloses a device for identifying sea surface temperature-sensitive areas, comprising:

[0117] The function construction module 11 is used to obtain the target model input data corresponding to the pre-trained target climate prediction model and construct an objective function to measure the intensity of heat wave events in the target area; wherein, the target model input data includes atmospheric state field, sea surface temperature field and corresponding time coding information;

[0118] The parameter freezing module 12 is used to freeze the model parameters of the target climate prediction model, set the sea surface temperature field as a unique differentiable variable, and input the target model input data into the target climate prediction model for forward propagation to obtain the temperature forecast results within the target forecast period.

[0119] The function value calculation module 13 is used to calculate the target function values ​​corresponding to the target function in different time periods according to the temperature forecast results, and to determine the sea surface temperature gradient sensitivity field corresponding to different time periods based on the target function values; wherein, the sea surface temperature gradient sensitivity field is a spatial distribution field characterizing the degree of influence of sea surface temperature changes at each ocean grid point on the intensity of heat wave events;

[0120] The sensitive area identification module 14 is used to generate a sea surface temperature sensitive area mask based on each of the sea temperature gradient sensitivity fields, so as to complete the identification of sea surface temperature sensitive areas.

[0121] In some specific embodiments, the function construction module 11 may specifically include:

[0122] The target item construction unit is used to construct a first target item representing the degree to which the temperature in the target area exceeds a preset temperature threshold, and to construct a second target item representing the proportion of the target area that reaches the preset temperature threshold.

[0123] The function construction unit is used to construct a third target term representing the duration of high temperature corresponding to a preset high temperature in the target region, and to construct the target function using the first target term, the second target term, and the third target term; wherein, the preset high temperature is a temperature exceeding the preset temperature threshold.

[0124] In some specific embodiments, the parameter freezing module 12 further includes:

[0125] The data alignment unit is used to perform spatial interpolation on each modal data in the input data of the target model to align the spatial scale of each modal data, and to perform temporal alignment and missing value filling operations on the spatially aligned modal data.

[0126] In some specific embodiments, the function value calculation module 13 may specifically include:

[0127] The sensitivity field determination unit is used to backpropagate the objective function value along the forward propagation calculation link of the target climate prediction model to obtain the partial derivatives of the objective function value with respect to each ocean grid point in the sea surface temperature field, and to determine the sea surface temperature gradient sensitivity field based on each partial derivative; wherein, the sea surface temperature field is the spatial distribution field corresponding to the sea surface temperature variable in the input data of the target model, which is composed of the sea surface temperature values ​​of several grid points covering the global ocean area, and the partial derivatives characterize the impact of the sea surface temperature rise at each grid point on the intensity of heat wave events.

[0128] In some specific embodiments, the sensitive area identification module 14 may specifically include:

[0129] The mask generation submodule is used to determine the target weights corresponding to each of the sea surface temperature gradient sensitivity fields, perform weighted fusion of each sea surface temperature gradient sensitivity field according to the target weights to obtain a comprehensive sensitivity field, and generate the sea surface temperature sensitive area mask based on the comprehensive sensitivity field; wherein, the comprehensive sensitivity field characterizes the overall influence of sea surface temperature on heat wave events in different forecast time periods, each ocean grid point in the comprehensive sensitivity field corresponds to a sensitivity value, and the sensitivity value characterizes the influence of the sea surface temperature change at the current ocean grid point on the intensity of the heat wave event.

[0130] In some specific embodiments, the mask generation submodule may specifically include:

[0131] The regional division unit is used to perform gradient sign consistency test on the comprehensive sensitivity field, retaining grid points whose gradient signs remain consistent in different time periods and removing noisy grid points whose signs change in different time periods, so as to obtain the corresponding comprehensive sensitivity field after the test; wherein, gradient sign consistency characterizes whether the direction of the effect of the same ocean grid point on the heat wave event is stable in different forecast time periods.

[0132] The mask generation unit is used to perform spatial connectivity screening on each candidate grid point in the post-tested comprehensive sensitivity field, divide each spatially adjacent candidate grid point into the same connected region, and remove isolated connected regions with an area smaller than a preset threshold, so as to obtain the sea surface temperature sensitive area mask; wherein, the candidate grid point is a grid point with a sensitivity value not less than a preset score threshold.

[0133] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0134] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the sea surface temperature sensitive area identification method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0135] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0136] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0137] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the sea surface temperature sensitive area identification method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0138] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned method for identifying sea surface temperature-sensitive areas. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0139] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0140] Those skilled in the art will further 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, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. 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.

[0141] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0142] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0143] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying sea surface temperature-sensitive areas, characterized in that, include: Obtain the target model input data corresponding to the pre-trained target climate prediction model, and construct an objective function to measure the intensity of heat wave events in the target region; wherein, the target model input data includes the atmospheric state field, the sea surface temperature field, and the corresponding time coding information; The model parameters of the target climate prediction model are frozen, the sea surface temperature field is set as the only differentiable variable, and the target model input data is input into the target climate prediction model for forward propagation to obtain the temperature forecast results within the target forecast period. Based on the temperature forecast results, the objective function values ​​corresponding to the objective function in different time periods are calculated, and the sea surface temperature gradient sensitivity fields corresponding to different time periods are determined based on the objective function values; wherein, the sea surface temperature gradient sensitivity field is a spatial distribution field characterizing the degree of influence of sea surface temperature changes at each ocean grid point on the intensity of heat wave events; Based on the sea surface temperature gradient sensitivity fields described above, a sea surface temperature sensitive area mask is generated to complete the identification of sea surface temperature sensitive areas.

2. The method for identifying sea surface temperature-sensitive areas according to claim 1, characterized in that, The atmospheric state field includes the atmospheric temperature field, the atmospheric humidity field, and the sea level pressure field; the sea surface temperature field includes ocean surface temperature variables.

3. The method for identifying sea surface temperature-sensitive areas according to claim 1, characterized in that, The objective function for measuring the intensity of heat wave events in the target area includes: A first target term is constructed to characterize the degree to which the temperature in the target area exceeds a preset temperature threshold, and a second target term is constructed to characterize the proportion of the target area that reaches the preset temperature threshold; A third target term is constructed to characterize the duration of high temperature corresponding to a preset high temperature within the target region, and the target function is constructed using the first target term, the second target term, and the third target term; wherein, the preset high temperature is a temperature exceeding the preset temperature threshold.

4. The method for identifying sea surface temperature-sensitive areas according to claim 1, characterized in that, Before inputting the target model input data into the target climate prediction model for forward propagation, the method further includes: Spatial interpolation is performed on each modal data in the input data of the target model to align the spatial scale of each modal data, and temporal alignment and missing value filling operations are performed on each modal data after spatial scale alignment.

5. The method for identifying sea surface temperature-sensitive areas according to claim 1, characterized in that, The step of determining the sea surface temperature gradient sensitivity field corresponding to different time periods based on the objective function value includes: The objective function value is backpropagated along the forward propagation computation link of the target climate prediction model to obtain the partial derivatives of the objective function value with respect to each ocean grid point in the sea surface temperature field, and the sea surface temperature gradient sensitivity field is determined based on each partial derivative; wherein, the sea surface temperature field is the spatial distribution field corresponding to the sea surface temperature variable in the input data of the target model, which is composed of the sea surface temperature values ​​of several grid points covering the global ocean area, and the partial derivatives characterize the impact of the sea surface temperature rise at each grid point on the intensity of heat wave events.

6. The method for identifying sea surface temperature-sensitive areas according to claim 1, characterized in that, The process of generating a sea surface temperature sensitive area mask based on each of the sea temperature gradient sensitivity fields includes: The target weights corresponding to each of the sea surface temperature gradient sensitivity fields are determined, and the sea surface temperature gradient sensitivity fields are weighted and fused according to the target weights to obtain a comprehensive sensitivity field. The sea surface temperature sensitive area mask is generated based on the comprehensive sensitivity field. The comprehensive sensitivity field represents the overall impact of sea surface temperature on heat wave events in different forecast time periods. Each ocean grid point in the comprehensive sensitivity field corresponds to a sensitivity value, and the sensitivity value represents the impact of the sea surface temperature change at the current ocean grid point on the intensity of the heat wave event.

7. The method for identifying sea surface temperature-sensitive areas according to claim 6, characterized in that, The process of generating the sea surface temperature sensitive area mask based on the comprehensive sensitivity field includes: The gradient sign consistency test is performed on the comprehensive sensitivity field. Grid points with consistent gradient signs in different time periods are retained, while noisy grid points with changing signs in different time periods are removed to obtain the corresponding comprehensive sensitivity field after the test. Gradient sign consistency characterizes whether the direction of the effect of the same ocean grid point on the heat wave event is stable in different forecast time periods. Spatial connectivity screening is performed on each candidate grid point in the comprehensive sensitivity field after the test. Spatially adjacent candidate grid points are divided into the same connected region, and isolated connected regions with an area smaller than a preset threshold are removed to obtain the sea surface temperature sensitive area mask; wherein, the candidate grid point is a grid point with a sensitivity value not less than a preset score threshold.

8. A device for identifying sea surface temperature-sensitive areas, characterized in that, include: The function construction module is used to obtain the target model input data corresponding to the pre-trained target climate prediction model and construct an objective function to measure the intensity of heat wave events in the target region; wherein, the target model input data includes atmospheric state field, sea surface temperature field and corresponding time coding information; The parameter freezing module is used to freeze the model parameters of the target climate prediction model, set the sea surface temperature field as a unique differentiable variable, and input the target model input data into the target climate prediction model for forward propagation to obtain the temperature forecast results within the target forecast period. The function value calculation module is used to calculate the objective function values ​​corresponding to the objective function in different time periods based on the temperature forecast results, and to determine the sea surface temperature gradient sensitivity field corresponding to different time periods based on the objective function values; wherein, the sea surface temperature gradient sensitivity field is a spatial distribution field characterizing the degree of influence of sea surface temperature changes at each ocean grid point on the intensity of heat wave events; The sensitive area identification module is used to generate a sea surface temperature sensitive area mask based on each of the sea temperature gradient sensitivity fields, so as to complete the identification of sea surface temperature sensitive areas.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the sea surface temperature sensitive area identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the sea surface temperature sensitive area identification method as described in any one of claims 1 to 7.

Images