Research On The Arctic Daily Sea Ice Thickness Estimation And The Drivers Of Its Anomalous Variability

Sea ice is an important component of the Arctic Ocean,affecting the climate,marine ecosystems,energy development,and shipping in the region.The sea ice thickness(SIT)is one of the important parameters in the Arctic,which is significant in exploring the life cycle and motion of sea ice,and its response to global climate change.Additionally,the Arctic SIT is a key indicator for evaluating changes in the Arctic marine environment and is of great importance to studying and understanding changes in the Arctic marine environment.Since the 1980 s,the SIT of the Arctic has shown a significant downward trend,with 2012 marking the historical minimum.Considering the factors of reduction in SIT and modeling SIT with thermodynamic parameters can help discover the mechanisms of sea ice change.Considering the variations in Arctic SIT at a finer granularity helps to discover more detailed patterns of Arctic sea ice variability.Considering the drivers of daily changes in Arctic SIT through Arctic meteorological conditions are beneficial to further explore the relationship between Arctic SIT and meteorology.This thesis focuses on these issues and conducts research in the following three aspects:1)Estimation of daily Arctic SIT based on a self-attention fully convolutional neural networkPolar SIT is significantly influenced by meteorological thermodynamic parameters,however,modeling the relationship between them is quite difficult due to the complexity of their mechanisms.This study proposed a self-attention convolutional neural network(Self Attention Convolutional Network,SAC-Net),which aims to model the relationship between thermodynamic parameters and SIT more parsimoniously,and directly estimate winter SIT based on these parameters.SAC-Net used a fully convolutional network as a baseline model to detect the spatial information of the thermodynamic parameters.Furthermore,a self-attention block was introduced to enhance the correlation among features.SAC-Net was trained on a dataset consisting of SIT products and thermodynamic data from the freeze-up period 2012–2019,including surface upward sensible heat flux,surface upward latent heat flux,2 m temperature,skin temperature,and surface snow temperature.The accuracy assessment of the SAC-Net Arctic SIT model shows that the hyperparameters of the model are optimal in the test dataset;the accuracy of this model is the highest when validated against SIMBA buoys,with correlation coefficients,root mean square error,and deviation of 0.58,0.43 m and 0.33 m,respectively;the spatial and temporal distributions of SIT derived from the model are highly agreed with CS2 SMOS,indicating a good fit of the model.In summer,the accuracy of SIT results of SAC-Net is slightly worse because SAC-Net is not trained using summer data and PIOMAS accuracy is relatively high.2)Spatial and temporal variability of the Arctic daily SITIn this study,the SAC-Net SIT data with higher accuracy in winter and PIOMAS SIT data with higher accuracy in summer were combined as a fused dataset,and the changes in SIT in the Arctic from 2012 to 2021 were analyzed.It is found that thick ice mainly locates in the north of Greenland and the Canadian Arctic Archipelago,with an average SIT of about 1.8 meters,while other sea areas are mainly covered with new ice,young ice,and first-year ice.The maximum SIT in most sea areas mainly occurs in April and May each year,while the minimum value often occurs in September and October each year.During the minimum SIT period,except for the Beaufort Sea and the central Arctic sea areas,almost all other sea areas are completely ice-free.The SIT in most sea areas showed a gradually decreasing trend over time,with a mean of about1.25 cm per year,and the largest significant downward trends in SIT were observed in the Chukchi Sea,with an annual rate of change of-3.5 cm/yr,followed by the East Siberian Sea,the central Arctic sea areas,the Beaufort Sea,the Laptev Sea,the Greenland Sea,the Norwegian Sea,and the Baffin Bay.In the entrance to the Northwest Passage of the Beaufort Sea,the Kara Sea,and the Barents Sea,the SIT showed a relatively abnormal thickening trend,with an increased rate of about 4 cm/yr,1 cm/yr,and 2 cm/yr,respectively.The first mode of Arctic SIT from 2012 to 2021 indicated that the trends of SIT are consistent in the entire study area3)Meteorological drivers of key periods in regions of anomalous Arctic SIT changeBased on previous research,2 m temperature,sea level pressure,surface sensible heat flux,surface net radiation,and wind speed were selected as meteorological factors of sea ice changes,and the spatiotemporal variations of each factor were analyzed.Then the areas of abnormal changes in SIT in the Beaufort and Barents Seas were extracted and their key abnormal change periods were analyzed by interannual monthly distance level.It is found that the SIT change in the Beaufort Sea was within [-4,4] meters from2012 to 2021,with the largest abnormal changes from 2015 to 2017.The SIT change in the Barents Sea is relatively gentle,ranging from-1 m to 1 m,with the highest abnormal changes from 2019 to 2021.Furthermore,the Singular Value Decomposition method was used to analyze the correlation between SIT and each meteorological field.It is found that the SIT change in the Beaufort Sea is highly correlated with 2 m air temperature with the largest correlation coefficient of 0.74,and the temporal changes are also highly consistent.The first spatial distribution pattern of SIT and 2 m air temperature shows a strong negative correlation zone,with a correlation coefficient exceeding 0.7,indicating that 2m air temperature is still the main forcing driver of sea ice changes in the Beaufort Sea abnormal area.The sea ice change in the Barents Sea is relatively highly correlated with surface sensible heat flux and 2 m air temperature,with values of 0.59 and 0.53,respectively.Sea ice changes had a strong response to changes in air temperature,and the first spatial distribution patterns of SIT,2 m air temperature,surface net thermal radiation,surface sensible heat flux,and air pressure fields had relatively strong negative correlation zones,with a correlation coefficient exceeding 0.7,indicating that the main meteorological forcing driver affecting sea ice changes in the abnormal area of the Barents Sea is 2 m air temperature,and the surface net thermal radiation is relatively slightly affecting the sea ice changes.