Uncertainty Analysis And Fusion Research Of Multi-source Snow Depth Data

As an important element of the earth's surface,snow cover plays an important role in the global terrestrial ecosystem,climate change,water cycle and energy cycle.Snow cover,storing abundant fresh water resources,is an important source of water supply for surface runoff and groundwater in arid-semi-arid regions.Abnormal changes of the snow cover would cause natural disasters,such as avalanches,snow melt floods,and wind-blown snow,which have an important impact on the sustainable development of regional ecology,economy,and society.Snow depth provides information on the spatial distribution and material energy of snow cover.It is one of the important parameters for characterizing snow cover.It is also an important parameter to study snow cover climatic effects,watershed water balance and snowmelt runoff simulation,and monitoring and assessment of snow disaster.Snow depth data has become an indispensable basic supporting data in multidisciplinary field research.However,the current snow depth data is relatively poor in integrity and consistency,and cannot meet the needs of related scientific research and industry applications.With the advent of the era of big data,the number of multi-source snow depth data is exploding.How to fully integrate and utilize the rich information contained in multi-source snow depth data to leverage the advantages of various data sources and improve the accuracy of snow depth data,has important research significance and value.This study selected 11 sets of snow depth data covering the China,including five types of passive microwave remote sensing snow depth data and six types of reanalysis data.Based on the uncertainty analysis of multi-source snow depth,machine learning and Triple Collocation methods were used to construct a fusion model of multi-source snow depth data,and a set of Chinese long-term snow depth data sets was formed based on the optimal fusion model.Finally,the generated snow depth data was used to analyze the temporal and spatial characteristics of snow cover in China and its impact on vegetation growth.Through the above research,some valuable insights and conclusions can be obtained.(1)Comprehensive analysis of the uncertainty of multi-source snow depth data in China.? The large uncertainty of multi-source snow depth data is mainly distributed in Inner Mongolia-Northeast China,Northern Xinjiang,and Qinghai-Tibet Plateau.?The correlation between multi-source snow depth data and ground observations on the Qinghai-Tibet Plateau is low.?NSIDC(National Snow and Ice Data Center)AMSR-E,CMA(China Meteorological Administration)FY3B and JAXA(Japan Aerospace Exploration Agency)AMSR2 snow depth data are overestimated in most areas of China,and CMC products are clearly overestimated in the Tibetan Plateau.(2)The performance of multi-source snow depth data under multiple factors and its error are evaluated.? Microwave remote sensing snow depth data performs well when the snow depth is less than 45cm.With the increase of snow depth,the microwave remote sensing inversion snow saturation value appears saturated.In comparison,CMC(Canadian Meteorological Centre)and MERRA2 reanalyzed the snow depth data to maintain good consistency with the ground observation snow depth when the snow depth is greater than 45cm.?Microwave remote sensing snow depth data has great uncertainty in the forest coverage area,and increases with the increase of forest coverage,and the accuracy of reanalyzing the snow depth data mainly depends on the model design and driving data.The impact is not consistent.? As the surface roughness increases,the uncertainty of multi-source snow depth data also gradually increases.? Microwave remote sensing snow depth data has greater uncertainty in the moss prototype and Taigalin type snow area,and then analysis of the CMC,GLDAS-NOAH and ERA5 data in the snow depth data has a larger Deviations,other snow depth data have greater uncertainty in the taiga snow regions.(3)Construct a multi-source snow depth data fusion model.Based on the results of uncertainty analysis,five snow depth products with high accuracy,such as WESTDC,GLDAS-NOAH,ERA-Interim,ERA5 and MERRA2,were selected as the data set to be fused.Combined with ground measurements,the fusion of multi-source data was used the four methods,including Support Vector Machine(SVM),Back Propagation Neural Networks(BPNN),Random Forest(RF)and Triple Collocation method.Through qualitative and quantitative performance evaluation,the results show that the snow depth of the three machines can improve the data quality and the saturation effect of microwave remote sensing data.Although Triple Collocation has improved the snow depth data quality,the improvement effect is not significant.In general,the RF fusion model performs best among all algorithms,with a correlation coefficient(R)of 0.87 and a root mean square error(RMSE)of 5.1cm;followed by BPNN and SVM fusion models with R values of 0.80 and 0.82,RMSE respectively The values are 6.38cm and 6.1cm,respectively;the R and RMSE values of the Triple Collocation method are 0.7 and 8.63cm,respectively.(4)The characteristics of spatiotemporal changes of snow cover and vegetation growth in China were analyzed from 1982 to 2015.? From 1982 to 2015,the average annual snow depth in China showed a decreasing trend at a rate of 0.21cm/10a;the snow cover onset date(SCOD)showed a significant delay trend with a delay rate of 0.2d/a;Snow covered days(SCD)showed an insignificant reduction trend with a reduction rate of only 0.09d/a;Snow cover ending date(SCED)was an insignificant delay Trend,the delay rate is only 0.2d/a.?The start of the growing season(SOS)showed a trend of delaying advance,the delay rate was 0.3 d/a,and the vegetation growth condition index(VCI)showed an increasing trend with a rate of 0.5/a(5)The characteristics of the response of the growth of different vegetation types to changes in snow cover were discussed.The SOS of cold temperate and temperate mountain coniferous forest vegetation has a significant negative correlation with snow depth,while VCI and snow depth show a significant positive correlation.In temperate deciduous broad-leaved forest,SOS showed a significant negative correlation with SCOD,and SOS showed a positive correlation with other snow parameters.Vegetation in temperate meadows is greatly affected by changes in snow cover.As snow depth increases,SCD increases,SCOD advances,and SCED advances.Vegetation SOS shows a delayed trend,while vegetation VCI and snow depth in this area Shows a positive correlation.The SOS of temperate grasslands has a significant negative correlation with SCOD and SCD,and the rich snow cover in winter helps to improve the vegetation growth in the next year.Compared to other vegetations,the alpine steppes and alpine meadows show opposite responses to the variability of snow cover.