【作者】 韩超；

【导师】 王晓艳；

【作者基本信息】 兰州大学 ， 地理学·地图学与地理信息系统， 2023， 硕士

【摘要】 积雪是气候变化的一个重要影响因子,其对地表辐射平衡、全球水文过程和生态环境变化具有重要的影响。同时,积雪也是重要的淡水资源,对人类生产生活具有重要意义。因此,准确监测积雪的分布和变化对全球气候变化的研究具有重要价值。东北地区是我国三大稳定积雪区之一,也是我国主要的粮食生产基地,研究其季节性积雪的积累、消融等变化对该地区农作物生产有着重要的意义。本研究首先对东北地区MODIS(Moderate Resolution Imaging Spectroradiometer)V6积雪产品Snow＿cover＿class数据集中的云像元进行了统计,发现在冬季(12月、1月和2月),大兴安岭、小兴安岭和长白山的森林地区存在明显的过度云掩膜问题,大量的森林积雪被错分为了云。这使得该地区冬季MODIS积雪产品的NDSI(Normalized differential snow index)数据存在大量的缺失,影响了积雪制图的精度。本研究发现选取合适的绿光波段阈值可以很好地区分森林积雪和云像元,有效减小MODIS积雪产品过度云掩膜现象。对于剩余的云像元,本文发展了基于NDSI相似度的时空立方体去云算法(Spatiotemporal cube cloud removal algorithm based on NDSI similarity,STNSI),对云下像元NDSI值进行重构填补。该算法首先统计出中心像元与邻域像元NDSI的标准化欧式距离,并以此作为相似度判别的标准,再通过邻域像元与中心像元NDSI的误差偏移量进行数据校正,从而生产出精度较高、逐日无云的长时间NDSI序列,然后基于站点雪深数据确定不同地表类型的NDSI最优阈值,生产出积雪二值产品。在此基础上,分析了中国东北地区积雪时空变化规律,并采用本文提出的积雪持续时间指数(Snow cover duration index,SDI)对东北地区积雪的稳定性进行了定量评价。本文主要结论如下:(1)当前MODIS V6积雪产品提供的归一化差值积雪指数(NDSI)在中国东北地区,尤其是东北森林地区存在明显的过度云掩膜问题,严重影响了该产品在积雪时空变化研究中的应用。本研究发现,绿光波段对云和森林积雪有较强的区分作用,云像元在绿光波段的反射率通常大于0.4,而森林积雪像元在绿光波段的反射率较低。(2)结合上下午星合成、决策树判别法和STNSI算法,能够生成高精度的长时间序列连续无云的NDSI积雪产品。采用云假设检验对去云过程进行精度评价,最后得到的无云影像,与原始真实影像的平均相关系数、平均均方根误差和平均绝对误差分别为0.96、0.10和0.08。基于站点雪深数据确定林地和非林地的最优NDSI阈值分别为0和0.09,并生产出逐日无云积雪二值产品。(3)中国东北地区的积雪从整体分布上看,大兴安岭、小兴安岭和长白山等山地地区的积雪覆盖天数(Snow Cover Days,SCD)较高,中部平原地区SCD值较低。SCD大于180天的地区主要集中在东北地区的最北部,中部平原地区的SCD大面积集中在60天以下。各年度积雪分布相对稳定,空间分布规律也十分类似。同时,研究发现积雪覆盖面积(Snow Cover area,SCA)在东北地区具有明显的纬度地带性差异,呈现高纬度SCA高,低纬度SCA低的趋势,纬度高的地区出现积雪覆盖的时间早于纬度较低的地区,且积雪融化的时间晚于纬度较低的地区。采用Mann-Kendall检验法和Theil-Sen Median分析的方法探究东北积雪的年际变化,结果表明,积雪覆盖天数呈增长趋势的地区面积占比为52.33%,呈减少趋势的地区面积占比为43.61%,变化不显著的地区面积占比为4.06%。(4)东北地区在9、10月份全域基本无积雪覆盖,月均SDI大多低于5天;从11月份开始积雪覆盖累计天数开始增多,月均SDI大多集中在5到10天之间,12月、1月和2月,为积雪稳定期,其中1月份积雪覆盖天数达到最高,且在1月份有10.64%的区域积雪覆盖持续天数达到一个整月,即SDI达到31天。3月份积雪覆盖逐渐减少,整个区域月均SDI普遍开始降低,直到4月份基本恢复全域没有积雪覆盖,月均SDI低于10天的地区面积占比恢复至96.74%。从不同下垫面来看,无论是SCD还是月均SDI森林地区都高于非森林地区。更多还原

【Abstract】 Snow is a significant climatic factor influencing various aspects such as surface radiation balance,global hydrological processes,and ecological environment.Additionally,snow serves as a crucial freshwater resource with essential implications for human livelihood and production.Therefore,precise monitoring of snow distribution and changes holds substantial value in studying global climate change.The Northeast region of China is recognized as one of the country’s primary stable snow regions and a key agricultural production area.Investigating the seasonal dynamics of snow accumulation and melt in this region is of utmost importance for crop production.In this study,an analysis was conducted on the cloud-contaminated pixels present in the MODIS(Moderate Resolution Imaging Spectroradiometer)V6 snow product,specifically the Snow＿cover＿class dataset,in the Northeast region.During the winter months(December,January,and February),noticeable issues of excessive cloud masking were observed in the forested regions of the Da Hinggan Mountains,Xiao Hinggan Mountains,and Changbai Mountains.This misclassification of significant forest snow cover as clouds resulted in significant data gaps in the Normalized Differential Snow Index(NDSI)of the MODIS snow product.To address this limitation,an optimal threshold for the green band was identified to effectively differentiate forest snow from cloud-contaminated pixels,thereby mitigating the problem of excessive cloud masking.For the remaining cloud-contaminated pixels,a novel Spatiotemporal Cube Cloud Removal Algorithm based on NDSI Similarity(STNSI)was developed to reconstruct and fill the missing NDSI values.The algorithm employed a standardized Euclidean distance measure between the NDSI of a central pixel and its neighboring pixels as a similarity criterion.This measure facilitated the adjustment of the NDSI values of the neighborhood pixels based on the error offset with the central pixel,resulting in the generation of high-precision,daily cloud-free NDSI time series.Furthermore,optimal NDSI thresholds for different land surface types were determined using ground-based snow depth data,enabling the production of binary snow cover products.The study analyzed the spatiotemporal variations of snow in the Northeast region of China and quantitatively evaluated snow cover stability using the proposed Snow Cover Duration Index(SDI).The key findings of this study are as follows:(1)The current MODIS V6 snow product,which utilizes the Normalized Differential Snow Index(NDSI),exhibits notable issues of excessive cloud masking in the Northeast region of China,particularly in forested areas.This significantly hampers the product’s applicability in studying snow spatiotemporal variations.The study identified the green band as a discriminative factor between clouds and forest snow,as cloud-contaminated pixels typically exhibit a reflectance greater than 0.4 in the green band,while forest snow pixels exhibit lower reflectance.(2)By integrating terra and aqua,decision tree classification,and the STNSI algorithm,it was possible to generate high-precision,cloud-free NDSI snow products with continuous long-term time series.Accuracy evaluation of the cloud removal process using cloud assumption tests revealed average correlation coefficients,root mean square errors,and absolute errors of 0.96,0.10,and 0.08,respectively,when comparing the cloud-free images with the original true images.Optimal NDSI thresholds of 0 for forest areas and 0.09 for non-forest areas were determined,resulting in the production of daily binary snow cover products.(3)Snow cover in Northeast China exhibits distinct spatial and temporal variations.The mountainous regions,including Da Hinggan Mountains,Xiao Hinggan Mountains,and Changbai Mountains,have higher Snow Cover Days(SCD),while the central plain region has lower SCD values.Regions with SCD exceeding 180 days are mainly concentrated in the northernmost part of Northeast China,while extensive areas of the central plain region have SCD below 60 days.Snow cover distribution remains relatively stable over the years,with similar spatial patterns.Additionally,snow cover area(SCA)in Northeast China shows significant latitudinal zonation,with higher SCA at higher latitudes and lower SCA at lower latitudes.Higher latitude areas experience earlier snow cover onset and later snowmelt compared to lower latitude regions.MannKendall and Theil-Sen Median analyses were employed to examine interannual variations in Northeast China’s snow cover.The results indicated that the area with an increasing trend in accumulated snow days accounts for 52.33%,while the area with a decreasing trend accounts for 43.61%,and the area with insignificant changes accounts for 4.06%.(4)In Northeast China,snow cover is virtually absent at the regional scale in September and October,with monthly mean SDI mostly below 5 days.From November onward,the accumulation of snow cover days increases,with the majority of monthly mean SDI falling between 5 and 10 days.December,January,and February are considered the stable snow period,with January having the highest snow cover days.In January,10.64% of the region experiences continuous snow cover for an entire month,resulting in an SDI of 31 days.Snow cover gradually decreases in March,and the monthly mean SDI begins to decline throughout the region,reaching nearly no snow cover by April,with the area where the monthly mean SDI is below 10 days recovering to 96.74%.Forested areas consistently exhibit higher SCD and monthly mean SDI compared to non-forested areas.更多还原