Comparative evaluation of filters for smoothing the glacier surface velocity estimates

Abstract

The surface velocity of glaciers is important for the estimation of ice discharge, to understand glacier dynamics and its response to climate change, etc. However, the glacier surface velocity data collected using conventional field measurements is insufficient. Therefore, remote sensing based techniques are being used extensively in glaciological studies as it offers the comprehensive and repetitive monitoring in a cost effective manner. Recently, Co-registration of Optically Sensed Images and Correlation (COSI-Corr) tool (Leprince et al., 2007) which is based on image matching method of optical images has gained popularity for the estimation of glacier surface velocity and has proved to be robust and effective. For denoising the glacier surface velocity estimates, Non-Local Means (NLM) filter has been incorporated in COSI-Corr. However, to discard/replace erroneous values, user has to manually define the threshold to discard/replace erroneous values. Although the NLM filter in COSI-Corr gives smooth estimates, but it has been observed that the filtered values deviate from the original ones. Therefore, the main purpose of this study has been to find an optimal filter for obtaining smooth estimates while preserving the original. Various spatial domain filters such as mean filter, low-pass filter, adaptive filter and statistical filters have been applied and evaluated in this study to obtain smooth glacier surface velocity estimates. The comparative evaluation of these filters showed that the application of statistical filter with multipliers 1, 2 and 4 in sequential order followed by 3 x 3 mean filter provided close match with the original estimates.