Method for predicting change of downward radiation of generic northern ice ocean and contribution of seasonal driving factor of generic northern ice ocean

Abstract

The invention discloses a method for predicting the change of downward radiation of a generic northern ice ocean and the contribution of a seasonal driving factor of the generic northern ice ocean. The method comprises the following steps: firstly, acquiring and preprocessing downlink radiation and related environmental factor data of a generic northern ice ocean area; then, a hybrid deep learning model CNN-STLSTM-CNN is constructed and trained to predict the change of downlink radiation; next, the trained model is used for predicting the downward radiation change of the next month, the contribution proportion of main environmental factors such as cold season and warm season temperature, sea ice concentration, relative humidity and latent heat flux to the downward radiation change is analyzed and quantified through model sensitivity, and finally the contrast effect of the sea ice concentration in different seasons is revealed. According to the method, the change of the downward radiation of the generic northern ice ocean can be predicted with relatively high precision, the seasonal driving factors are deeply analyzed, and a scientific basis is provided for understanding and coping with the climate change of the north pole.

Description

Technical Field

[0001] The present invention relates to the field of climate prediction, and particularly to a method for predicting the change trend of the downward radiation in the pan-Arctic Ocean and quantitatively analyzing the contribution of its seasonal driving factors. Background Art

[0002] The Arctic region is one of the most sensitive regions to global climate change, and its changes have an important impact on the global climate system. In recent years, with the intensification of global warming, the Arctic region has experienced significant changes, including a rapid increase in temperature and a substantial reduction in sea ice area, and the "Arctic amplification" phenomenon is significant. [1, 2] The downward radiation in the pan-Arctic Ocean (the sum of short-wave and long-wave radiation) is a key component of the energy balance in this region, and its changes directly affect processes such as sea ice melting, ocean temperature, and atmospheric circulation. Current research shows that environmental factors such as sea ice, cloud cover, humidity, and air temperature can all cause changes in the downward radiation in the pan-Arctic Ocean. [3, 4] .

[0003] Although traditional statistical methods can analyze historical data of the downward radiation in the pan-Arctic Ocean, it is difficult to capture complex non-linear relationships and make future predictions. Although climate models can simulate climate changes in the Arctic region, there is still room for improvement in the ability to analyze complex driving factors. [5-7] Currently, there is a lack of a comprehensive method that can accurately predict changes in the downward radiation in the pan-Arctic Ocean and effectively quantify the contribution of its seasonal driving factors.

[0004] References [1] Jahn A, Holland MM, Kay JE. Projections of an ice-free Arctic Ocean. Nature Reviews Earth & Environment, 2024, 5(3): 164-176. [2] Rantanen M, Karpechko AY, Lipponen A, et al. The Arctic has warmed nearly four times faster than the globe since 1979. Communications Earth & Environment, 2022, 3(1): 168. [3] Lu P, Cheng B, Leppäranta M, et al. Partitioning of solar radiation in Arctic sea ice during melt season. Oceanologia, 2018, 60(4): 464-477. [4] Gong T, Feldstein S, Lee S. The Role of Downward Infrared Radiation in the Recent Arctic Winter Warming Trend. Journal of Climate, 2017, 30(13): 4937-4949. [5] Goessling HF, Rackow T, Jung T. Recent global temperature surge intensified by record-low planetary albedo. Science, 2025, 387(6729): 68-73. [6] Kim D, Kang SM, Kim H, et al. Quantifying Changes in the Arctic Shortwave Cloud Radiative Effects. Journal of Geophysical Research: Atmospheres, 2024, 129(15): e2023J-e40707J. [7] Previdi M, Smith KL, Polvani LM. Arctic amplification of climate change: a review of underlying mechanisms. Environmental research letters, 2021, 16(9): 93003. Summary of the Invention

[0005] The object of the present invention is to overcome the deficiencies of the prior art and provide a method for predicting the changes in the downward radiation in the pan-Arctic Ocean and the contributions of its seasonal driving factors, which can more accurately predict the future change trend of the downward radiation in the pan-Arctic Ocean and quantitatively analyze the contribution ratios of the main driving factors in different seasons.

[0006] To achieve the above object, the present invention adopts the following technical solutions. A method for predicting the change of downward radiation in the pan-Arctic Ocean and the contribution of its seasonal driving factors includes the following steps:

[0007] (1) Obtain monthly downward radiation data and related environmental factor data in the pan-Arctic Ocean region over a long time range, conduct quality control on the data, and further select the main environmental factors affecting the change of downward radiation using the recursive feature elimination method;

[0008] (2) Construct a CNN-STLSTM-CNN hybrid deep learning model including a convolutional neural network (CNN) layer and a spatio-temporal long short-term memory network (STLSTM) layer. Use the processed environmental factor data as the input of the model and the historical downward radiation data as the output of the model for training, and further optimize the model parameters to improve the prediction accuracy of the model;

[0009] (3) Use the trained deep learning model, input the historical environmental factor data, and predict the change trend of downward radiation in the pan-Arctic Ocean for the next month;

[0010] (4) Based on the trained deep learning model, extract the weights of each main environmental factor in the cold season (October to March of the following year) and the warm season (April to September), conduct sensitivity perturbation analysis, and quantify the contribution ratio of each environmental factor to the change of downward radiation in the pan-Arctic Ocean in the cold and warm seasons;

[0011] Preferably, the radiation and environmental factor data can be obtained from the European Centre for Medium-Range Weather Forecasts ERA5 reanalysis product (https: / / www.ecmwf.int / en / forecasts / dataset / ecmwf-reanalysis-v5), and the sea ice concentration data can be obtained from the National Snow and Ice Data Center of the United States (https: / / nsidc.org / data / g02202 / versions / 4).

[0012] The above technical solutions can achieve the following beneficial effects:

[0013] (1) By constructing a hybrid deep learning model, the present invention can effectively capture the spatio-temporal complexity of the change of downward radiation in the pan-Arctic Ocean and the non-linear relationship with environmental factors, so as to achieve high-precision prediction.

[0014] (2) The present invention can quantitatively analyze the contribution ratio of key environmental factors to the change of downward radiation in different seasons (cold season and warm season), which helps to deeply understand the driving mechanism of Arctic climate change and provides an important basis for understanding the impact of sea ice change on the Arctic energy balance.

[0015] (3) The method of the present invention is based on advanced deep learning technology and combines the understanding of Arctic climate characteristics, with high scientificity and practical application value, and can provide strong technical support for climate change research and Arctic environmental management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the CNN-STLSTM-CNN model architecture;

[0017] Figure 2 is the comparison between the prediction results of the CNN_STLSTM_CNN model and the test set;

[0018] Figure 3 is the spatial contribution map of the main environmental factors in the warm and cold seasons under the sensitivity test;

[0019] Figure 4 is the illustration of the contribution difference of the main environmental factors in the warm and cold seasons to the downward radiation. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The following further describes the specific technical solutions of the present invention with reference to the drawings, so as to facilitate those skilled in the art to further understand the present invention without restricting its rights.

[0021] In order to fully disclose the present invention, the embodiments take the data of the downward radiation and related environmental factors in the pan-Arctic Ocean from 1980 to 2023 as examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] Embodiment 1, referring to Figures 1-4 , a method for predicting the change of the downward radiation in the pan-Arctic Ocean and the contribution of its seasonal driving factors, comprising the following steps,

[0023] (1) Data acquisition and preprocessing: Obtain the monthly total downward radiation data (the sum of short-wave and long-wave radiation) in the pan-Arctic Ocean region from 1980 to 2023, as well as the related monthly environmental factor data, from the ERA5 reanalysis dataset. Perform necessary quality control on the obtained data, such as checking for physically unreasonable values or obvious outliers, and process or eliminate them. For data points with missing values, use the bilinear interpolation method to fill them to ensure the spatial integrity of the data. Use the recursive feature elimination method to screen out the four environmental impact factors most relevant to the downward radiation in the pan-Arctic Ocean, which are atmospheric temperature, latent heat flux (negative (positive) values indicate that the ocean releases (absorbs) heat to the atmosphere), relative humidity, and sea ice concentration (SIC).

[0024] (2) Model construction and training: The temperature, latent heat flux, relative humidity, and SIC data for each month are used as the four feature channels of the model. A CNN-STLSTM-CNN model ( Figure 1 ) is constructed, with a time step of 3 and a convolutional kernel size of 5×5. The architecture starts with a 2D convolutional layer containing 32 convolutional kernels for extracting spatial features. Then, there are two stacked ST-LSTM units, which contain convolutional layers and dropout layers (Dropout = 0.3). Through the convolutional layers, the input 128 feature channels are processed again to further capture the spatial and temporal dependencies in the data. The output of the STLSTM layer is input into another convolutional layer to further extract features. Finally, the predicted radiation value is obtained. In the experiment, the dataset is randomly divided into a training set and a test set in a ratio of 80%:20% to improve the generalization ability of the model. The learning rate of the model is set to 0.001, the batch size is 15, and the training period is 400 rounds. In addition, to prevent overfitting, an early stopping mechanism is adopted during training, with a patience value set to 10 rounds, that is, training is stopped early when the validation set error does not decrease for more than 10 rounds. During the model training process, the mean squared error (MSE) is used as the loss function, and the change trends of the training loss and validation loss with the number of iterations are recorded.

[0025] (3) Prediction of downward radiation change: Based on the trained CNN-STLSTM-CNN model, by inputting the temperature, latent heat flux, relative humidity, and SIC data for the previous three months, the downward radiation of the entire Arctic Ocean for the next month can be predicted. Figure 2 shows the prediction error of the CNN-STLSTM-CNN model on the test set. The fitting result between the predicted radiation value of the model and the true value is good, with a correlation reaching 0.98. The best root mean square errors of the model for the training set and the test set are 0.05 and 0.06 respectively, reflecting that the model has a relatively effective prediction ability for the downward radiation of the entire Arctic Ocean and can better capture the change law of the downward radiation of the entire Arctic Ocean in terms of spatial distribution. From the error distribution of the model prediction, the high error values are mainly in the central sea area of the entire Arctic Ocean covered by sea ice all year round, with a maximum underestimation of 1.79×10 6 J m -2 in the cold season and a maximum underestimation of 2.15×10 6 J m -2 in the warm season. The overall prediction result of the model in the Atlantic inflow area of the entire Arctic Ocean is good. Among them, there is a slight overestimation in the Barents Sea in the cold season, about 0.21×10 6 J m -2 , and there is an overestimation of about 0.68×10 6 J m -2 in the Atlantic inflow area in the warm season.

[0026] (4) Seasonal driving factor contribution analysis: Based on the trained model, sensitivity analysis is carried out on the weights of the main environmental factors in advance. The data of each environmental factor in the test set are perturbed by 10% respectively, and then input into the model to observe the change of the downward radiation prediction value. By comparing the change amplitude of the downward radiation caused by the perturbation of different environmental factors, the sensitivity of each factor to the change of the downward radiation is quantified, so as to evaluate its contribution ratio. The results show that, from the whole year, temperature, latent heat flux and SIC make positive contributions to the change of downward radiation, while relative humidity makes negative contributions. The absolute value ranking of their weights is temperature (40.29%) > SIC (28.44%) > latent heat flux (20.87%) > relative humidity (10.40%). The increase in temperature is the key to the increase of downward radiation in the pan-Arctic Ocean. Followed by SIC and latent heat flux, while the increase in relative humidity plays an inhibitory role in the growth of downward radiation.

[0027] Furthermore, the above sensitivity analysis is carried out for the cold season (October to March of the following year) and the warm season (April to September) respectively to obtain the contribution distribution and ratio of each environmental factor in different seasons. Refer to Figure 3 , which shows that the distribution of the contributions of each environmental factor to the downward radiation in the pan-Arctic Ocean in the cold and warm seasons is uneven, especially in the Barents Sea, which is the pioneer of the Arctic Ocean environment response. The contribution rates of relative humidity and SIC near the Barents Sea are relatively low in the cold season, while the contribution rates of temperature and latent heat flux are relatively high here. Generally speaking, in the cold season, the contributions of temperature and latent heat flux increase significantly compared with the whole year, rising by 10.09% and 5.56% year-on-year respectively. The contribution of SIC in the cold season changes from a positive contribution throughout the year to a negative contribution of -13.64%. The contributions of temperature and latent heat flux in the warm season decrease significantly compared with the cold season, decreasing by 21.04% and 11.57% respectively. The contribution of SIC changes from negative to positive and increases significantly to 46.76%. Relative humidity makes negative contributions in both the cold and warm seasons and fluctuates relatively smoothly ( Figure 4 ).

[0028] Refer to Figure 4 , by analyzing the different influence directions of SIC on the downward radiation prediction value in the cold and warm seasons, its seasonal contrast effect is revealed. In the cold season, the increase of SIC enhances the surface albedo, reflects solar radiation and inhibits downward radiation. While in the warm season, the downward radiation depends on the interaction between SIC and humidity and atmospheric stability. The growth of a small amount of SIC will lead to a significant increase in humidity, weakening its radiation effect, which highlights its transformation from an inhibitory factor to a catalytic factor in the melting state.

[0029] The present invention provides a new method for predicting the change of downward radiation in the pan-Arctic Ocean and the contribution of its seasonal driving factors, which can effectively predict the change trend of downward radiation in the pan-Arctic Ocean in different seasons and quantitatively analyze the contribution ratio of each driving factor. Through this method, not only the accuracy of predicting downward radiation in the pan-Arctic Ocean is improved, but also the differential effects of environmental factors such as temperature, sea ice concentration, relative humidity and latent heat flux on radiation change in cold and warm seasons can be revealed.

[0030] The above are all preferred embodiments of the present invention. For those of ordinary skill in the art, without departing from the principle of the present invention, the modifications of various equivalent forms of the present invention all fall within the protection scope of the appended claims of this application.

Claims

1. A method for predicting the changes in the downward radiation of the pan-Arctic Ocean and the contributions of its seasonal driving factors, characterized in that, It includes the following steps: (1) Obtain and preprocess the downward radiation data and related environmental factor data in the pan-Arctic Ocean region, and select temperature, latent heat flux, relative humidity, and sea ice concentration as key driving factors from the environmental factors through recursive feature elimination method; (2) Construct and train a hybrid deep learning model, which includes a convolutional neural network (CNN) layer and a spatio-temporal long short-term memory network (STLSTM) layer, and is trained with the key environmental factor data as the input and the historical downward radiation data as the output; (3) Use the trained deep learning model to input the data of 4 key environmental factors and predict the change trend of the downward radiation in the pan-Arctic Ocean in the next month; (4) Extract the weights of the main environmental factors of the model for sensitivity analysis, quantify the contribution ratios of the environmental factors to the change of the downward radiation in the cold season and the warm season, and analyze the contrast effect of the sea ice concentration on the downward radiation in different seasons.

2. The method according to claim 1, wherein The hybrid deep learning model is a CNN-STLSTM-CNN model, which includes at least one two-dimensional convolutional layer and at least one spatio-temporal long short-term memory network layer.

3. The method according to claim 1, characterized in that, The cold season is defined as from October to March of the following year, and the warm season is defined as from April to September.

4. The method according to claim 1, wherein In step (2), the mean square error (MSE) is used as one of the indexes to evaluate the prediction performance of the model.

5. The method according to claim 1, characterized in that In step (4), the sensitivity analysis is realized by perturbing the environmental factor data input into the model to increase by 10% and observing the change of the downward radiation output by the model to quantify the contribution ratio of each factor.

Images