Abstract Large, e.g., provincial or national, scale near-real-time surface water monitoring is an ambitious task, which can be accomplished by using Synthetic Aperture Radar (SAR) satellite data. SAR has demonstrated the ability to distinguish water and land, but there are many common errors of commission and omission that arise due to the side-looking nature of SAR and due to some landcover types with similar backscatter like roads and pasture. A method is proposed to fix/mitigate these errors through the use of combined ascending/descending RADARSAT-2 image pairs and ancillary data. The results of a corrected water/land binary image were, on average, 99.4% accurate for the Boreal Forest Region (Utikuma) of Alberta, Canada, while for the Rocky Mountain Region (Westcastle) also in Alberta, the results proved to be 99.9% accurate when distinguishing water from land. These accuracies were achieved through the reduction of the water false positive rate and a slight reduction in the water true positive rate. These high accuracy values can be partially attributed to the relative low ratios of water to land in the study regions. We hope that these methods can be used and improved in order to move towards large scale dynamic surface water and wetland mapping.

中文翻译：

---

协同使用 RADARSAT-2 上升和下降图像提高加拿大艾伯塔省地表水检测精度

摘要 大型（例如省级或国家级）近实时地表水监测是一项艰巨的任务，可以通过使用合成孔径雷达（SAR）卫星数据来完成。SAR 已经展示了区分水和土地的能力，但是由于 SAR 的侧视性质以及由于一些具有相似后向散射的土地覆盖类型（如道路和牧场），存在许多常见的错误和遗漏。提出了一种通过使用组合的上升/下降 RADARSAT-2 图像对和辅助数据来修复/减轻这些错误的方法。对于加拿大阿尔伯塔省的北方森林地区 (Utukuma)，校正后的水/陆地二值图像的结果平均准确率为 99.4%，而对于同样位于阿尔伯塔省的落基山脉地区 (Westcastle)，结果证明为 99 . 区分水和陆地时的准确率为 9%。这些准确度是通过降低水假阳性率和轻微降低水真阳性率来实现的。这些高精度值可以部分归因于研究区域中相对较低的水陆比。我们希望这些方法可以得到使用和改进，以实现大规模动态地表水和湿地测绘。