Dual-satellite (Sentinel-2 and Landsat 8) remote sensing of supraglacial lakes in Greenland

Abstract

Abstract. Remote sensing is commonly used to monitor supraglacial lakes on the Greenland Ice Sheet (GrIS); however, most satellite records must trade off higher spatial resolution for higher temporal resolution (e.g. MODIS) or vice versa (e.g. Landsat). Here, we overcome this issue by developing and applying a dual-sensor method that can monitor changes to lake areas and volumes at high spatial resolution (10–30 m) with a frequent revisit time (∼3 days). We achieve this by mosaicking imagery from the Landsat 8 Operational Land Imager (OLI) with imagery from the recently launched Sentinel-2 Multispectral Instrument (MSI) for a ∼12 000 km2 area of West Greenland in the 2016 melt season. First, we validate a physically based method for calculating lake depths with Sentinel-2 by comparing measurements against those derived from the available contemporaneous Landsat 8 imagery; we find close correspondence between the two sets of values (R2=0.841; RMSE = 0.555 m). This provides us with the methodological basis for automatically calculating lake areas, depths, and volumes from all available Landsat 8 and Sentinel-2 images. These automatic methods are incorporated into an algorithm for Fully Automated Supraglacial lake Tracking at Enhanced Resolution (FASTER). The FASTER algorithm produces time series showing lake evolution during the 2016 melt season, including automated rapid (≤4 day) lake-drainage identification. With the dual Sentinel-2–Landsat 8 record, we identify 184 rapidly draining lakes, many more than identified with either imagery collection alone (93 with Sentinel-2; 66 with Landsat 8), due to their inferior temporal resolution, or would be possible with MODIS, due to its omission of small lakes <0.125 km2. Finally, we estimate the water volumes drained into the GrIS during rapid-lake-drainage events and, by analysing downscaled regional climate-model (RACMO2.3p2) run-off data, the water quantity that enters the GrIS via the moulins opened by such events. We find that during the lake-drainage events alone, the water drained by small lakes (<0.125 km2) is only 5.1 % of the total water volume drained by all lakes. However, considering the total water volume entering the GrIS after lake drainage, the moulins opened by small lakes deliver 61.5 % of the total water volume delivered via the moulins opened by large and small lakes; this is because there are more small lakes, allowing more moulins to open, and because small lakes are found at lower elevations than large lakes, where run-off is higher. These findings suggest that small lakes should be included in future remote-sensing and modelling work.

Antarctic Supraglacial Lake Detection Using Landsat 8 and Sentinel-2 Imagery: Towards Continental Generation of Lake Volumes

“…Using these thresholds, we created binary (i.e., lake and non-lake) masks for each L8 scene. From the binary lake products, we removed areas that are <5 pixels (30 m resolution) in total and linear features that are narrower than 2 pixels, since these features are likely to represent areas of mixed slush or streams rather than lakes [5,41].…”

“…As with L8, the application of the physically-based model to the S2 red band TOA reflectance data produces the most accurate depth results [41]. Due to finer spatial detail of S2 compared to L8, we derived A d by averaging reflectances over a three-pixel wide ring around a lake to minimize the impact of errors associated with lake outlines.…”

“…Due to finer spatial detail of S2 compared to L8, we derived A d by averaging reflectances over a three-pixel wide ring around a lake to minimize the impact of errors associated with lake outlines. For red band data, we used g = 0.83 from Williamson et al [41]. Following a similar procedure to L8, we compile a R ∞ look-up table for S2 using a total of 12321 images acquired during 2016-2019 ( Figure 6).…”

“…Some dates with low total lake volumes are explained by low visibility due to clouds, which remain the main limiting factor in creating a high-resolution time-series record. When looking at smaller spatial scales, overlapping L8 and S2 images can improve temporal resolution to less than 5 days [41]. [41].…”

“…When looking at smaller spatial scales, overlapping L8 and S2 images can improve temporal resolution to less than 5 days [41]. [41].…”