| The Arctic sea ice albedo is a key variable in the balance of surface radiation and energy,which determines the distribution and budget of polar radiation energy,and has important scientific research significance in global climate change and polar environment research.Although many Arctic sea ice albedo data sets have been obtained based on satellite remote sensing data,most of the data sets simplify the consideration of sea ice reflection anisotropy,and generally ignore the impact of melting pool on sea ice albedo,making the accuracy of the existing Arctic sea ice albedo remote sensing data sets relatively low.At the same time,due to the impact of cloud cover,the remote sensing data set of Arctic sea ice albedo based on optical data usually has a large number of data missing,which seriously restricts the development of research in related fields.In view of the above key issues,this study developed a method of generating Arctic sea ice albedo remote sensing data set considering the impact of melting pool.Through the remote sensing estimation method of Arctic sea ice albedo considering the impact of melting pool and the spatial-temporal filling method of Arctic sea ice albedo based on passive microwave data,the Arctic sea ice albedo remote sensing data set with long time series(2000-2020),continuous and complete spatial-temporal and high accuracy is generated.This study proposes a method for estimating the albedo of Arctic sea ice surface based on the Ensemble Back Propagation Neural Network(EBPNN)model.It improves the accuracy and calculation efficiency of the remote sensing data set of Arctic sea ice albedo by considering the impact of melting pool on the anisotropy of sea ice reflection.First,it uses the Asymptotic Analytical Radiative Transfer(AART)model to build snow The bidirectional reflection distribution function(BRDF)/albedo database of white ice and melting pool,and then use the EBPNN model to establish the nonlinear relationship between the surface reflectance product(MOD09GA)of the Moderate Resolution Imaging Spectrometer(MODIS)and the albedo of the Arctic sea ice surface.Finally,when the daily MOD09 GA data of the Arctic sea ice area is available,Based on the EBPNN model,the Arctic sea ice albedo estimation results with high accuracy and high time resolution are obtained.This study also proposes a spatial-temporal filling algorithm for the missing value of Arctic sea ice albedo based on passive microwave data.The missing value is filled by establishing the relationship between passive microwave brightness temperature data and Arctic sea ice albedo.Finally,a long time series of spatial-temporal continuous and complete Arctic sea ice albedo data set is generated,and the spatial-temporal change characteristics of Arctic sea ice albedo in the past 21 years are preliminarily analyzed.The main findings of this study include the following:(1)The remote sensing estimation method of Arctic sea ice albedo based on EBPNN proposed in this study has higher estimation accuracy and calculation efficiency,and has higher accuracy than the remote sensing estimation method of Arctic sea ice albedo that ignores the effects of reflection anisotropy and melting pool.EBPNN model has more stable estimation accuracy(R(correlation coefficient)=0.996,RMSE(Root Mean Square Error)=0.023,bias=0.000)than single BP(Back Propagation)neural network,and can be effectively applied to remote sensing estimation of Arctic sea ice albedo.Compared with the iterative procedure,the EBPNN model greatly improves the calculation efficiency of Arctic sea ice albedo.The EBPNN model prediction is highly consistent with the field measurements of Tara expedition(R=0.931,RMSE=0.052,bias=-0.027),PROMICE automatic weather station(R=0.956,RMSE=0.058,bias=-0.011),Polarstern cruise(bias=-0.006)and Barrow station(bias=0.001).The comparison with other sea ice albedo products shows that the sea ice albedo estimated by the EBPNN model is more consistent with the MERIS(Medium Resolution Imaging Spectrometer)product generated based on the physical model(R=0.916,RMSE=0.085,bias=0.001).If the influence of reflection anisotropy of Arctic sea ice is ignored,the albedo of Arctic sea ice will be underestimated under the condition of large solar zenith angle,and if the influence of melting pool is ignored,the albedo of Arctic sea ice will be significantly overestimated.Therefore,the remote sensing estimation method of Arctic sea ice albedo considering reflection anisotropy and the effect of melting pool proposed in this study has higher estimation accuracy.(2)The spatial-temporal filling algorithm based on passive microwave data proposed in this study can be effectively applied to fill the missing value of Arctic sea ice albedo,and finally generate a spatial-temporal continuous and complete remote sensing data set of Arctic sea ice albedo.The monthly spatiotemporal filling model further improves the accuracy and stability of the filling and reconstruction results(comprehensive R=0.949,RMSE=0.069,bias=0.000).The comparison with Polarstern and Barrow field measurements shows that the R,RMSE and bias of filling reconstruction results are 0.961(0.902),0.069(0.070)and-0.008(-0.003),respectively.The results of filling and reconstruction are in good agreement with the field measurement data,and have the same accuracy as the remote sensing estimation results.The validation of time series and missing areas proves that the daily seamless Arctic sea ice albedo data set has good spatiotemporal consistency and continuity with the original remote sensing estimation results.The comparison with the time average method shows that the spatiotemporal filling algorithm based on passive microwave data is more consistent with the original real value,and more accurate filling and reconstruction results are obtained.The original estimation results with missing data before filling can not clearly show the temporal and spatial distribution of the Arctic sea ice albedo,while the seamless reconstruction results after filling can more completely show the temporal and spatial variation characteristics of the Arctic sea ice albedo during the thaw-refreezing season.(3)The long time series Arctic sea ice albedo data set generated in this study can well show the rule of the Arctic sea ice albedo fluctuation falling first,then stabilizing and then rising during the melting-refreezing period,which corresponds well with the dynamic change of the melting pool coverage.In summer,the whole Arctic sea ice region and seven subregions show a downward trend of sea ice albedo.The albedo of the whole Arctic sea ice region in July and August has the most obvious downward trend,respectively-0.0062/year and-0.0050/year,and the overall downward trend in summer has reached-0.0039/year.Among the seven sub-sea areas,the Canadian archipelago has the most severe sea ice albedo reduction trend(-0.0062/year),followed by the Beaufort and Chukchi Sea(-0.0052/year)and Hudson Bay(-0.0051/year).The albedo of the entire Arctic sea ice region in the summer of2000-2020 showed a downward trend in the overall spatial distribution.The area of albedo decreasing trend was statistically significant,while the area of scattered albedo increasing trend was not statistically significant.In this study,a generation method of Arctic sea ice albedo remote sensing data set considering the impact of melting pool was proposed,and the remote sensing estimation method of Arctic sea ice albedo based on EBPNN and the spatial-temporal filling method of Arctic sea ice albedo based on passive microwave data were developed,and the remote sensing data set of Arctic sea ice albedo with long time series and continuous and complete spatial-temporal was generated.This study provides a new idea for remote sensing monitoring and analysis of Arctic sea ice albedo based on satellite observation data.Under the background of global climate warming and polar amplification effect,the research results can effectively promote the development of research on radiation and energy balance of the cryosphere,and provide important data support for global climate change and polar environment research. |
PDF Full Text Request