Abstract:Melt ponds occupy a large fraction of the Arctic sea ice surface during spring and summer. The fraction and distribution of melt ponds have considerable impacts on Arctic climate and ecosystem by reducing the albedo. There is an urgency to obtain improved accuracy and a wider coverage of melt pond fraction (MPF) data for studying these processes. MPF information has generally been acquired from optical imagery. Conventional MPF algorithms based on high-resolution optical sensors have treated melt ponds as features with constant reflectance; however, the spectral reflectance of ponds can vary greatly, even at a local scale. Here we use Sentinel-2 imagery to demonstrate those previous algorithms assuming fixed melt pond-reflectance greatly underestimate MPF. We propose a new algorithm ("LinearPolar") based on the polar coordinate transformation that treats melt ponds as variable-reflectance features and calculates MPF across the vector between melt pond and bare ice axes. The angular coordinate theta of the polar coordinate system, which is only associated with pond fraction rather than reflectance, is used to determinate MPF. By comparing the new algorithm and previous methods with IceBridge optical imagery data, across a variety of Sentinel-2 images with melt ponds at various stages of development, we show that the RMSE value of the LinearPolar algorithm is about 30% lower than for the previous algorithms. Moreover, based on a sensitivity test, the new algorithm is also less sensitive to the subjective threshold for melt pond reflectance than previous algorithms. Plain Language Summary Melt ponds are pools of open water that form on the sea ice surface in the Arctic in summer months. Sea ice covered by melt ponds absorbs more solar heat than solid ice which speeds up the rate of sea ice melt. To study this process, we need melt pond coverage data that are generally obtained from satellite observations. There are several methods to calculate the melt pond fraction from high-resolution optical satellite imagery but these methods assume that melt ponds have constant optical properties, which is not true. In this study, we show that these existing methods significantly underestimate melt pond fraction. To address this problem, we present a new method that treats melt ponds as features having variable optical properties. Images of sea ice from aircraft are used to assess the new algorithm and two previous methods, demonstrating that the new algorithm has higher accuracy and precision. Key Points Sentinel-2 and IceBridge imagery are used to show that algorithms assuming fixed melt pond-reflectance largely underestimate melt pond fraction A new algorithm based on the polar coordinate transformation that treats melt ponds as variable-reflectance has been developed The accuracy and precision of the new Sentinel-2 algorithm, assessed with IceBridge DMS imagery, are higher than previous algorithms

Identification of Melt Pond on Sea Ice Based on Deep Learning Technology

Applying high-resolution visible imagery to satellite melt pond fraction retrieval: A neural network approach

Geo-knowledge-informed Deep Learning for Auto-identification of Supraglacial Lakes on the Greenland Ice Sheet from Satellite Imagery.

Object-based detection of Arctic sea ice and melt ponds using high spatial resolution aerial photographs

A New Algorithm for Sea Ice Melt Pond Fraction Estimation from High‐Resolution Optical Satellite Imagery

Melt pond fractions on Arctic summer sea ice retrieved from Sentinel-3 satellite data with a constrained physical forward model

Melt Pond Retrieval Based on the LinearPolar Algorithm Using Landsat Data

Detecting Melt Pond Onset on Landfast Arctic Sea Ice Using a Dual C-Band Satellite Approach

STUDY ON ARCTIC MELT POND FRACTION RETRIEVAL ALGORITHM USING MODIS DATA

Design and Implementation of an Ice-Tethered Observation System for Melt Pond Evolution with Vision and Temperature Profile Measurements

Sea Ice Image Classification Based on Heterogeneous Data Fusion and Deep Learning

Comparison of Lake Ice Extraction Methods Based on MODIS Images

Melt Pond Distribution and Geometry in High Arctic Sea Ice Derived from Aerial Investigations

Retrievals of Arctic Sea Ice Melt Pond Depth and Underlying Ice Thickness Using Optical Data

TSDCU-Net: A Deep Learning Model for Sea Ice Classification Based on SAR Images

Automated surface melt detection over the Antarctic from Sentinel-1 imagery using deep learning

Extraction of Sea Ice Concentration Based on Spectral Unmixing Method

The Identification of Ice Floes and Calculation of Sea Ice Concentration Based on a Deep Learning Method

DF-UHRNet: A Modified CNN-Based Deep Learning Method for Automatic Sea Ice Classification from Sentinel-1A/B SAR Images

Sea Ice Detection from GNSS-R Data Based on Residual Network

Deep Learning Based Sea Ice Classification with Gaofen-3 Fully Polarimetric SAR Data