FABIAN: A daily product of Fractional Austral-summer Blue Ice over ANtarctica during 2000–2021 based on MODIS imagery using Google Earth Engine

Where the White Continent Is Blue: Deep Learning Locates Bare Ice in Antarctica

“…(2001) and Hu et al. (2022) (both ∼0.8% to ∼1.6%). However, the estimate is considerably lower than the currently only openly available and widely‐used BIA map of Hui et al.…”

“…The uncertainty quantification of the BIA map (Figure S11 in Supporting Information ) reflects the influence of noise and biases in the individual data sets (e.g., penetration depth of radar, Figure S7 in Supporting Information ), spatial misalignment of the different data sets, absence of similar examples in the training data, under‐ or overfitting of the neural network, and physical processes. Elevated uncertainties linked to physical processes (Figure 3) include (a) more persistent temporary snow cover (Hu et al., 2022), (b) wind scouring (Das et al., 2013; Scambos et al., 2012), (c) blue slush or surface lakes on snow or on blue ice (Dell et al., 2022), (d) changing wind patterns (Takahashi et al., 1992), and (e) crevasses. Although a disentanglement of the different uncertainty sources is unfeasible, the continent‐wide uncertainty estimates allow us to quantify the total BIA‐extent uncertainty by adding/subtracting one standard deviation to/from the prediction values to obtain a lower and upper bound of 114,000 and 174,000 km 2 , respectively.…”

“…Moreover, temporary snow cover on the ice can hinder the detection of BIAs in visual imagery. Over the past decades, several efforts have been performed to detect BIAs using satellite observations in the visual and shortwave infrared spectrum (e.g., Hu et al., 2022; Hui et al., 2014; Orheim & Lucchitta, 1990; Winther et al., 2001). All of these studies use spectral properties to distinguish BIAs from other surfaces.…”

“…All of these studies use spectral properties to distinguish BIAs from other surfaces. A downside of these approaches is that the spectral properties are quantified using a limited amount of ground observations and are assumed to be applicable over a larger area, ranging from observational scenes (Jawak et al., 2023; Orheim & Lucchitta, 1990; Winther et al., 2001) to the entire continent (Hu et al., 2022; Hui et al., 2014). In these BIA mapping efforts, temporary snow cover on blue ice cannot be distinguished from accumulation areas, which results in snow cover variation being regarded as a variation of BIA extent (Hu et al., 2022; Orheim & Lucchitta, 1990).…”

Remote Sensing of Surface Melt on Antarctica: Opportunities and Challenges

“…The MODIS melt detection could potentially be refined by including additional constraints (for example based on snow grain size or blue ice presence) to the melt detection algorithm. Nevertheless, it should be noted that surface melt can occur in these low albedo areas [27], [119], and optical sensors are able to detect surface melt over these areas (albeit an overestimation), whereas blue ice regions are often underestimated or neglected by regional climate models [23] and the melt detection algorithms applied to microwave sensors.…”