Boike, Julia; Grau, Thomas; Heim, Birgit; Günther, Frank; Langer, Moritz; Muster, Sina; Gouttevin, Isabelle; Lange, Stephan (2015): Lake maps 2001/2002 and 2009, Central Yakutia, Siberia [dataset]. PANGAEA, https://doi.org/10.1594/PANGAEA.855121,

---

In supplement to: Boike, J et al. (2016): Satellite-derived changes in the permafrost landscape of central Yakutia, 2000–2011: Wetting, drying, and fires. Global and Planetary Change, 139, 116-127, https://doi.org/10.1016/j.gloplacha.2016.01.001

Lakes were mapped based on Landsat mosaics from 2001/2002 and 2009. Maps are stored as ESRI shapefile.

Projection: UTM 52N, Datum: WGS84, Extent: South-bound Latitude: 60.40 * West-bound Longitude: 113.00 * North-bound Latitude: 65.60 * East-bound Longitude: 133.40

Methods: Waterbody mapping was carried out using two mosaics of Landsat satellite images: A Landsat TM mosaic consisting of 25 images acquired during the summer of 2009 and a mosaic of 31 Landsat ETM+ images recorded during two consecutive summers in 2001 (12 images) and 2002 (19 images). Heavy cloud cover in 2001 and 2002 made it necessary to combine remote sensing data from these two years in order to create a cloud free baseline for change detection over the entire region covered by the (cloud-free) 2009 mosaic. The average day of year (DOY) for the 12 images in our 2001 dataset was therefore 213 ±19 (1st August), and for the 19 images in the 2002 dataset 212 ±15 (31st July). Precipitation, which is likely to affect the areas of lakes, did not vary between these two years. The average DOY of our 2009 dataset was 215 ±20 (3rd August) and was therefore comparable to the 2001-2002 baseline.

Visual accuracy assessments were carried out on the water classifications (especially along overgrown lake margins), taking into account the extent of misclassifications due to shadows from clouds or mountains. We therefore did not apply any radiometric normalization during the mosaic-making process, in order to retain the original digital number (DN) values. Several classification iterations and subsequent visual inspections showed that Landsat's short wave infrared band 5, which has the same wavelength (1.55 - 1.75 µm) in both the TM and the ETM+ missions, provides the best results for surface water mapping. Characteristic for Landsat band 5 is the very high water absorption making it suitable for water detection. We therefore reclassified a narrow empirical DN value domain of high absorption (from 1 to 20, out of possible 256 DN, where higher values indicate higher reflectance) within this band as mapped water bodies. However, due to the absence of any independent control dataset for water bodies, we were unable to calculate accuracy statistics.

A filter algorithm with a 3-by-3 opening was used to smooth the contours and remove small holes and fringes in the grid mask, especially along lake margins and in areas of shallow water. This operation significantly reduced the number of small water bodies but without having any major impact on the calculated total area covered by lakes. The filtered raster data was vectorized in order to calculate the total area covered by lakes and any changes in this area over time, and also to investigate relationships between changes in lake area and changes in lake topology. Floodplain lakes located close to river meanders showed highly dynamic surface areas that changed because of river water level fluctuations, and these were therefore removed manually. Only lakes with a size greater than or equal to 4 Landsat pixels were considered for further analysis.