Global snow cover estimation with Microwave Brightness Temperature measurements and one-class in situ observations

摘要

Brightness temperature (BT), which is remotely sensed by the space-borne microwave radiometer, is widely used in snow cover monitoring for its long time series imaging capabilities in all-weather conditions. Traditional linear fitting and stand-alone methods are usually uncertain with respect to the spatial distribution and temporal variation of derived snow cover, as they rarely consider local conditions and scene characteristics but fit the model with static empirical coefficients. In this paper, a novel method utilizing daily ground in situ observations is proposed and evaluated, with the purpose for accurate estimation of long-term daily snow cover. To solve the challenge that ground snow-free records are insufficient, a one-class classifier, namely the Presence and Background Learning (PBL) algorithm, is employed to identify daily global snow cover. Benefiting from daily ground in situ observations on a global scale, the proposed method is temporally and spatially dynamic such that estimation errors are globally independent during the entire study period. The proposed method is applied to the estimation of global daily snow cover from 1987 to 2010; the results are validated by ground in situ observations and compared with available optical-based and microwave-based snow cover products. Promising accuracy and model stability are achieved in daily, monthly and yearly validations as compared against ground observations (global omission error <0.13, overall accuracy >0.82 in China region, and keep stable in monthly and yearly averages). The comparison against the MODIS daily snow cover product (MOD10C1) shows good agreement under cloud-free conditions (Cohen's kappa = 0.715). The comparison against the NOAA daily Interactive Multisensor Snow and Ice Mapping System (IMS) dataset suggests promising agreement in the Northern Hemisphere. Another comparison against the AMSR-E daily SWE dataset (AE_DySno) demonstrates the efficiency of the proposed method regarding to the overestimation problem in thin snow cover region. (C) 2016 Elsevier Inc. All rights reserved.