Retrogressive thaw slump activity and related lake colour change in five areas of the western Canadian Arctic

This database contains the spatial coordinates and years (timing and duration) of activity for more than 7400 retrogressive thaw slumps (RTS) in five areas of the western Canadian Arctic covering >150 000 km2. RTS are thermokarst landforms whose headwalls retreat due to thawing of exposed ice-rich permafrost or massive ice at rates that can exceed 10 m per year. Their initiation can relate to rapid ground thaw (e.g., due to a particularly warm summer) or from local erosion undercutting a slope, and once initiated, they may expand for several decades, exporting water, sediment, and dissolved solids downslope or into aquatic systems.
The five study areas and their periods of record are: (1) Banks Island (1984-2016), (2) northwest Victoria Island (1984-2018), (3) Bluenose moraine (1984-2018), (4) Paulatuk region (1984-2018), and (5) Richardson Mountains / Peel Plateau (2001-2018). The entries include a code to indicate where each RTS initiated in the landscape (at the coast, on a riverbank, on a lakeshore, or on a slope) and the locations of lakes which changed colour due to RTS activity, and the dates of those changes. The data on new RTS initiations commences one year after the start of record shown above and ends one year earlier than the final date. Some RTS were already active at the start of the record (denoted by the year 1900 in the dataset) and others were still active in the final year of each record (denoted by the year 2100).
The RTS activity and lake colour change data were generated from visual observation of Google Earth Timelapse videos (Lewkowicz and Way 2019; Lewkowicz under review) which are mainly derived from Landsat images. Active RTS were identified by the retrogression of their headwalls which appears as movement on the Timelapse videos. The location of the approximate centroid of the RTS was then recorded on Google Earth, together with the dates of activity and the initiation code. The visual identification technique had errors of omission of 8% and of commission (incorrect identification of features as active RTS) of 1% for Banks Island (Lewkowicz and Way 2019). Similar errors are expected for the other four study areas. The activity data are resolved annually, but the year of initiation shown in the database may lag the true date by one year because the Timelapse imagery pre-dated RTS initiation, or because the RTS was too small to be observed in the first year of its existence (Lewkowicz and Way 2019). The date of cessation of activity (if this occurred) may also be subject to a degree of imprecision because it usually takes place gradually over several years, rather than suddenly across the entire headwall.
The lakes listed are those that visually changed colour from deep blue to turquoise or beige due to sediment inputs from RTS activity on their shorelines, or up-basin and affecting inflowing rivers or other lakes. The lake data for each study area are provided for the same period as the RTS data. Colour variations not clearly associated with RTS activity were excluded. Lakes reverted to their original colour when RTS activity ceased, or when RTS headwalls receded sufficiently that sediment inputs to the lake declined. Some lakes changed colour more than once and these separate periods were recorded in the database. The use of 1900 and 2100 for dates represents, respectively, lakes that exhibited a colour associated with sediment inputs at the start of the record and those that remained changed at the end of the record.
Over 1984-2016, total RTS numbers in the first four study areas increased more than 50 times. The vast majority of the RTS were initiated in association with particularly warm summers (Lewkowicz under review). More than 500 lakes exhibited an altered colour due to inputs of sediment from RTS activity. The complete dataset should be useful for validation of machine-learning techniques which are currently being developed to map RTS and their impacts across the entire Arctic.

本数据集包含加拿大北极西部5个覆盖面积超15万平方千米的研究区内共计7400余处热融滑塌（retrogressive thaw slumps, RTS）的空间坐标及其活动年份（时间与持续时长）。热融滑塌属于热喀斯特地貌，其侧壁因暴露的富冰多年冻土或厚层冰融化而发生后退，后退速率可超过10米/年。热融滑塌的触发可与地面快速解冻（如异常暖夏）或局部侵蚀掏蚀边坡有关；一旦形成，其可扩张数十年，并向下坡或水生系统输送水体、沉积物与溶解固体。
本次研究涵盖5个区域及其观测时段：(1) 班克斯岛（1984-2016）；(2) 维多利亚岛西北部（1984-2018）；(3) 布卢诺斯冰碛区（1984-2018）；(4) 波拉图克地区（1984-2018）；(5) 理查森山脉/皮尔高原（2001-2018）。每条数据包含一项编码，用于标识每处热融滑塌在景观中的初始位置（海岸、河岸、湖岸或边坡），同时记录了因热融滑塌活动发生颜色变化的湖泊位置，以及这些颜色变化的发生日期。新增热融滑塌的记录起始于上述观测时段开始后一年，终止于观测时段结束前一年。部分热融滑塌在观测时段开始时已处于活动状态（数据集以1900年标注），另有部分在观测时段结束当年仍处于活动状态（以2100年标注）。
热融滑塌活动与湖泊颜色变化数据通过目视解译谷歌地球延时影像（Google Earth Timelapse）获取（Lewkowicz与Way 2019；Lewkowicz 待刊），此类影像主要源自陆地卫星（Landsat）影像。通过延时影像中侧壁后退的现象识别活动中的热融滑塌，随后在谷歌地球上记录该热融滑塌的近似质心位置、活动日期与初始位置编码。班克斯岛区域的目视解译方法存在8%的漏检误差与1%的误检误差（即将非热融滑塌特征错误识别为活动热融滑塌，Lewkowicz与Way 2019），其余四个研究区预计存在类似误差。活动数据按年度分辨率统计，但数据集记录的热融滑塌初始年份可能与真实年份存在一年的滞后：原因可能是延时影像获取早于热融滑塌发生，或是热融滑塌在形成当年规模过小无法被观测到（Lewkowicz与Way 2019）。活动终止日期（若存在）也存在一定精度误差，因为活动终止通常是数年的渐进过程，而非整个侧壁同时停止后退。
本次记录的湖泊均为因岸线、流域上游或入湖河流附近的热融滑塌活动输入沉积物，从而从深蓝色变为绿松石色或米色的湖泊。各研究区的湖泊数据与热融滑塌数据采用相同的观测时段。未明确与热融滑塌活动相关的颜色变化被排除。当热融滑塌活动停止，或其侧壁后退至向湖泊输入的沉积物量显著减少时，湖泊颜色会恢复至初始状态。部分湖泊颜色变化不止一次，这些独立的变化时段均被记录于数据集中。以1900年和2100年标注日期分别代表观测时段开始时即出现沉积物输入相关颜色变化的湖泊，以及观测时段结束时仍保持变色状态的湖泊。
1984-2016年间，前四个研究区内的热融滑塌总数增长超过50倍。绝大多数热融滑塌的触发与异常暖夏相关（Lewkowicz 待刊）。超过500个湖泊因热融滑塌输入沉积物而发生颜色变化。完整数据集可用于验证当前正在开发的机器学习技术，此类技术旨在全北极范围内绘制热融滑塌及其影响的分布。