Snow Cover Detection Over The Tibetan Plateau Using Daily Nighttime Light Data From NPP-VIIRS DNB

Traditional optical remotely sensed data have been widely used in snow-cover monitoring,but only limited to daytime detection accompanied with large data gaps caused by cloud obscuration.Since days in winter are short in mid-latitude high mountains and polar nights may last for half a year in some polar regions,it is difficult to assure spatio-temporal continuity of snow-cover simply using datasets derived from traditional optical sensors.Nighttime light remote sensing can capture the moonlight reflected by snow.Researchers have confirmed the capability of nighttime light data in snow detection by visual interpretation.However,the quantitative analysis of moonlight effect as well as the quantitative method to detect snow cover still need to be explored.Taking the Tibetan Plateau as the research area,this study quantitatively analyzed the moonlight effect on snow detection at night and proposed a NPP-VIIRS?National Polar-oribiting Partnership Visible Infrared Imaging Radiometer Suite?DNB?Day/Night Band?snow-cover detection technique based on the histogram matching and Kittler minimum error thresholding method.We successfully mapped the snow cover extent in the Tibetan Plateau from October 1,2017 to September 30,2018.To further improve the result,we improved a HMRF?Hidden Markov Random Field?-based spatio-temporal model.The model performed greatly in filling the data gaps caused by cloud obscuration and improving the quality of the original VIIRS DNB snow-cover result.The accuracy of the VIIRS DNB snow-cover result was then evaluated in comparison with in situ observations of 120 meteorological stations and MODIS snow-cover product.The major findings can be concluded as follows:1)During a lunar month,daily nighttime light data from NPP-VIIRS DNB can be used to detect snow at night in the Tibetan Plateau from the 12^th to 21^st day.Besides,moonlight intensity is the strongest when the moon is full or nearly full,which peaks the capability of daily nighttime light data from NPP-VIIRS DNB in snow detection.The farther away from this day,the worse snow detection result is shown.2)The annual average amount of cloud in VIIRS DNB snow-cover result is27.1%,which is significantly lower than MODIS snow-cover product.Besides,65%of the data vacancy caused by cloud in MODIS snow-cover product can be filled using VIIRS DNB snow-cover result as an auxiliary dataset.3)Under clear-sky condition,the overall accuracy of VIIRS DNB snow-cover result is 80.7%,which is 9.6%lower than MODIS snow-cover product.Under all-sky condition,the overall accuracy of VIIRS DNB snow-cover result is 60.5%,which is13.5%lower than MODIS snow-cover product.Data gaps,low spatial resolution and VIIRS cloud product quality are the key factors reducing the accuracy.After building the improved HMRF-based spatio-temporal model,93.4%of data gaps in the original snow-cover result have been filled and the overall accuracy under clear-sky and all-sky condition has increased by 6.3%and 19.4%respectively.The VIIRS DNB snow-cover detection technique proposed in this study can map snow cover in mountain regions globally as important supplementary data that helps fill the data gaps in MODIS snow-cover product caused by cloud.What's more,it can provide high-precision snow-cover product in polar regions when polar nights occur.