Pointner Georg, Bartsch Annett, Forbes Bruce C, Kumpula Timo
Austrian Polar Research Institute, Vienna, Austria.
b.geos, Korneuburg, Austria.
Int J Remote Sens. 2018 Sep 26;40(3):832-858. doi: 10.1080/01431161.2018.1519281. eCollection 2019.
In this study, we assess the effect of the lake size on the accuracy of a threshold-based classification of ground-fast and floating lake ice from Sentinel-1 Synthetic Aperture Radar (SAR) imagery. For that purpose, two new methods (flood-fill and watershed method) are introduced and the results between the three classification approaches are compared regarding different lake size classes for a study area covering most of the Yamal Peninsula in Western Siberia. The focus is on April, the stage of maximum lake ice thickness, for the years 2016 and 2017. The results indicate that the largest lakes are likely most prone to errors by the threshold classification. The newly introduced methods seem to improve classification results. The results also show differences in fractions of ground-fast lake ice between 2016 and 2017, which might reflect differences in temperatures between the winters with severe impact on wildlife and freshwater fish resources in the region. Patterns of low backscatter responsible for the classification errors in the centre of the lakes were investigated and compared to the optical Sentinel-2 imagery of late-winter. Strong similarities between some patterns in the optical and SAR data were identified. They might be zones of thin ice, but further research is required for clarification of this phenomenon and its causes.
在本研究中，我们评估了湖泊大小对基于阈值的 Sentinel-1 合成孔径雷达（SAR）图像中固定湖冰和漂浮湖冰分类精度的影响。为此，引入了两种新方法（泛洪填充法和分水岭法），并针对覆盖西西伯利亚亚马尔半岛大部分地区的研究区域，比较了三种分类方法在不同湖泊大小类别上的结果。重点关注 2016 年和 2017 年 4 月，即湖冰厚度最大的阶段。结果表明，最大的湖泊可能最容易因阈值分类而产生误差。新引入的方法似乎改善了分类结果。结果还显示了 2016 年和 2017 年固定湖冰比例的差异，这可能反映了冬季温度的差异，对该地区的野生动物和淡水鱼类资源产生了严重影响。研究了导致湖泊中心分类误差的低后向散射模式，并与冬季末期的 Sentinel-2 光学图像进行了比较。在光学数据和 SAR 数据中的一些模式之间发现了很强的相似性。它们可能是薄冰区域，但需要进一步研究来澄清这一现象及其成因。
