Seasonal dynamics of snow cover is an essential area of research for hydrological modelling and water resource management. With the increased availability of remote sensing data, the timely information of the spatiotemporal distribution of snow cover is feasible at regular intervals. The primary objective of this study is to assess the effect of pansharpening in the accuracy of snow cover change detection in mountainous regions using freely available Landsat-8 multispectral data. In mountainous regions at the medium resolution, the changes at the mountain ridges are seldom identified. The incorporation of pansharpening in the change detection framework facilitates an improvement in the snow cover change detection at the ridges. For pansharpening, the PanNet architecture based on convolutional neural networks was adopted. A study area around Dhundi in the state of Himachal Pradesh in India was selected for the analysis. The experiments were carried out using a subset of Landsat-8 multispectral data acquired in the autumn and the winter seasons of 2017 and 2018, respectively. An improvement of 0.184 and 0.267 in the kappa coefficient was observed for the overall changes in the snow cover and at the ridges, respectively, based on the results from the proposed approach.

中文翻译：

---

使用卷积神经网络的雪覆盖变化检测中的全色锐化效果

积雪的季节性动态是水文建模和水资源管理的重要研究领域。随着遥感数据可用性的增加，定期获得积雪时空分布的及时信息是可行的。本研究的主要目的是使用免费的 Landsat-8 多光谱数据评估全色锐化对山区积雪变化检测准确性的影响。在中等分辨率的山区，很少发现山脊的变化。在变化检测框架中加入全色锐化有助于改进山脊处的积雪变化检测。对于全色锐化，采用了基于卷积神经网络的 PanNet 架构。选择了印度喜马偕尔邦 Dhundi 周围的一个研究区域进行分析。这些实验是使用分别在 2017 年和 2018 年秋季和冬季获得的 Landsat-8 多光谱数据子集进行的。根据所提出方法的结果，观察到积雪和山脊的整体变化的 kappa 系数分别提高了 0.184 和 0.267。