Impact of Sensor Zenith Angle on MOD10A1 Data Reliability and Modification of Snow Cover Data for the Tarim River Basin

Abstract

Snow in the mountainous watersheds of the Tarim River Basin is the primary source of water for western China. The Snow Cover Daily L3 Global 500-m Grid (MOD10A1) remote sensing dataset has proven extremely valuable for monitoring the changing snow cover patterns over large spatial areas; however, inherent uncertainty associated with large sensor zenith angles (SZAs) has called its reliability into question. Comparative analysis that utilized a paired-date difference method for parameters such as snow cover frequency, snow cover percentage, and normalized difference snow index (NDSI) has shown that overestimation of snow cover in the Tarim River Basin correlates with high values of SZA. Hence, such overestimation was associated with an increase in the NDSI, attributable to the change in reflectance between Band 4 and Band 6 imagery. A maximum threshold value of SZA of 22.37 • was used alongside a multiday refilling method to modify the MOD10A1 dataset to produce a new daily snow cover map of the Tarim River Basin, spanning a 10-year period. A comparison of benchmark results of snow cover classification produced by the HJ-1A/B satellite revealed an increase in the overall accuracy of up to 4%, confirming the usefulness of our modified MOD10A1 data.

Navigating snowscapes: scale‐dependent responses of mountain sheep to snowpack properties

“…First, in similar systems, problems with cloud cover and snow-cloud misclassification will likely necessitate excluding substantial amounts of animal location data due to the absence of an NDSI estimate for a given location in space and time. Data losses may be greater in cases where additional quality filters are employed, such as the application of sensor azimuthal thresholds to limit the use of poor quality measurements at the boundaries of a satellite's path (Xin et al 2012, Li et al 2016. At worst, data loss can lead to systematic biases associated with spatial patterns in persistent cloud cover (Parajka and Bl€ oschl 2008) and at best, substantially weaken one's inference regarding complex interactions among snow properties and other spatial covariates.…”

“…We could retain only 2.2% of sheep locations when including NDSI filtered by high-quality flags, which increased to 17.9% after performing a 10-d gap fill (Table 2). Although MOD10A1 (version 006) contains a flag for oscillating measurements akin to those described by Xin et al (2012), in the absence of an azimuthal threshold, data may contain underestimates of NDSI in areas with extensive forest cover, heterogeneous terrain, or frequent shallow sensor passes (Xin et al 2012, Li et al 2016. Data losses may be greater in cases where additional quality filters are employed, such as the application of sensor azimuthal thresholds to limit the use of poor quality measurements at the boundaries of a satellite's path (Xin et al 2012, Li et al 2016.…”

“…At worst, data loss can lead to systematic biases associated with spatial patterns in persistent cloud cover (Parajka and Bl€ oschl 2008) and at best, substantially weaken one's inference regarding complex interactions among snow properties and other spatial covariates. Data losses may be greater in cases where additional quality filters are employed, such as the application of sensor azimuthal thresholds to limit the use of poor quality measurements at the boundaries of a satellite's path (Xin et al 2012, Li et al 2016. Although MOD10A1 (version 006) contains a flag for oscillating measurements akin to those described by Xin et al (2012), in the absence of an azimuthal threshold, data may contain underestimates of NDSI in areas with extensive forest cover, heterogeneous terrain, or frequent shallow sensor passes (Xin et al 2012, Li et al 2016.…”

“…Data losses may be greater in cases where additional quality filters are employed, such as the application of sensor azimuthal thresholds to limit the use of poor quality measurements at the boundaries of a satellite's path (Xin et al 2012, Li et al 2016. Although MOD10A1 (version 006) contains a flag for oscillating measurements akin to those described by Xin et al (2012), in the absence of an azimuthal threshold, data may contain underestimates of NDSI in areas with extensive forest cover, heterogeneous terrain, or frequent shallow sensor passes (Xin et al 2012, Li et al 2016. Second, study systems with more homogenous landscapes may have insufficient heterogeneity in snow cover to inform animal resource selection (e.g., 100% snow cover throughout the study area all winter).…”

Contemporary Snow Changes in the Karakoram Region Attributed to Improved MODIS Data between 2003 and 2018

“…Clouds are a primary constraint in optical remote sensing when estimating snow cover, particularly in mountainous catchments. Additionally, a larger SZA [49] and a wide swath of MODIS cause significant overestimation of snow [50,51]. The presence of clouds in the original products could even alter the trend pattern (e.g., min and mean annual SCA trends, as shown in Table 3).…”