An improved Terra−Aqua MODIS daily cloud-free snow and Randolph Glacier Inventory 6.0 combined product (M*D10A1GL06) for high-mountain Asia between 2002 and 2019

. Snow is a dominant water resource in High Mountain Asia (HMA) and crucial for mountain communities and downstream populations. Snow cover monitoring is significant to understand regional climate change, managing meltwater, and associated hazards/disasters. The uncertainties in passive optical remote sensing snow products mainly underestimation caused by cloud-cover and overestimation associated with sensorsˈ 10 limitations hamper to understand snow dynamics. We reduced the biases in Moderate Resolution Imaging Spectroradiometer (MODIS) Terra and Aqua daily snow data and generated a combined daily snow product for High Mountain Asia between 2002 and 2019. An improved MODIS 8-day composite MOYDGL06* product was used as a training data for reducing the underestimation and overestimation of snow in daily products. The daily MODIS Terra and Aqua images were improved by implementing cloud removal algorithms followed by gap 15 filling and reduction in overestimated snow beyond the respective 8-day composite snow extent of MOYDGL06* product. The daily Terra and Aqua snow products were combined and merged with the Randolph Glacier Inventory Version 6.0 (RGI6.0) described as M*D10A1GL06 to make a more complete cryosphere product with 500 m spatial resolution. The pixel values in the daily combined product are preserved and reversible to the individual Terra and Aqua improved products. We suggest a weightage of 0.5 and 1 to snow pixels in either or 20 both of the Terra and Aqua products, respectively for deriving snow cover statistics from our final snow product. The pixels with values 200, 242, and 252 indicate snow pixels in both Terra and Aqua and has ahave 1 weight of 1, whereas pixels with snow in one of the Terra or Aqua products have a weight of 0.5.0.5 weightage. On average, the M*D10A1GL06 product reduces 39.1% of uncertainty compared to MOYDGL06* product due to cloud cover (underestimation) sensor limitations larger solar zenith angle (SZA) (overestimation) 32.9% 6.2%, observation and simulation of climate, hydro-glaciological forcings, calibration, validation, and other water-related studies. the snow underestimation due to the data gaps and overestimation of snow pixels occurring beyond the eight-day 20 maximum extent of snow in MOYDGL06* product. The daily snow M*D10A1GL06 product associated with this paper can aid a valuable input dataset for hydro-glaciological and climate modelling, snow cover dynamics, and other water-related studies.

like floods and droughts (Haq et al., 2012;Memon et al., 2015;Miyan 2015) particularly in the densely populated downstream areas (Scott et al., 2019). The vast spatial extent and topographic complexity of snow-covered mountains make field-based monitoring difficult (Immerzeel et al., 2009;Muhammad et al., 2019a), therefore remote sensing is the most appropriate tool for cryosphere observations (Muhammad and Tian, 2020).

5
Remote sensing snow products are important in hydrological and other snow-related research (Hall et al., 2002;Li et al., 2019). The temporal coverage of remote sensing snow data is sufficient for climate change studies (e.g., NOAA Advanced Very High-Resolution Radiometer (AVHRR) snow data has been available since the 1980s) (Hori et al., 2017). However, the spatial resolution before the twenty-firstthis century was relatively coarse of 4 km (Hüsler et al., 2012) which is improved since the early twenty-first century by the most popular and up-todate snow product derived from Moderate Resolution Imaging Spectroradiometer (MODIS) on-board Terra and Aqua (Hall et al., 2007). The advantage of these datasets is the daily temporal resolution and the disadvantage is the low spatial resolution and a large swath of approximately 2300 km which. These limitations causes a snow overestimation at the image edges and in the images acquired in the off-nadir view (Riggs et al., 2016). Another major constraint in these passive optical remote sensing products is the cloud cover causing the spatial and 15 temporal time-series discontinuity. The cloud contamination in the original eight-day composite MODIS snow cover products is comparatively less than the daily products (Hall et al., 2002), but remains significantlysignificant e.g., in the Karakoram 9% and 15% of the Terra and Aqua 8-day images are 9% and 15% cloud-covered on average, respectively (Thapa and Muhammad, 2020). To reduce the remaining clouds up to 99.98% in the original eight-day composite products M*D10A2, a new Terra and Aqua composite product, namely MOYDGL06* was 20 developed for HMA using a multi-step approach (Muhammad and Thapa, 2020). This product MOYDGL06* is a significant contribution to snow-related studies. However, the eight-day composite is the maximum snow for eight consecutive days, which does not detect the exact timing of snow onset and melt (Hall et al., 2006). Similar limitations are likely using the eight-day composite products for the snowmelt runoff modelling which requires daily snow information.

25
This study considers the temporal limitations in the eight-day composite data and improves the daily MODIS snow products. Various methods, including spatial and temporal filters, are used for cloud removal in MODIS data (Li et al., 2019) but less attention has been given to the removal of overestimation attributed to the large solar zenith angle (SZA) and a wide swath of each tile. In this study, a daily cloud-free product combining MODIS Terra (MOD10A1) and Aqua (MYD10A1) is generated using the eight-day composite MOYDGL06* product as 30 a reference which is not only useful for clouds removal but also reduces overestimation. Larger SZA mainly causes an overestimation which was further reduced in the daily product by combining Terra and Aqua following the MOYDGL06* product methodology with a slightly different approach. We also fill the missing data gaps, remove overestimation in the daily snow data using the respective eight-day composite snow images, and merge the improved Terra and Aqua snow assigning values reversible to the individual Terra and Aqua improved 35 products. The improved Terra and Aqua cloud-free snow composite product merged with Randolph Glacier Inventory Version 6 (RGI6.0) is developed to make a more complete daily cryosphere product covering the period between 2002 and 2019. This product will significantly improve the hydro-glaciological applications and snowrelated observations in High Mountain Asia (HMA).

Study area
MODIS Terra and Aqua combined daily snow product in this paper cover HMA similar as in Muhammad and Thapa (2020) with the geographic extent of latitude 48 o E. The ten major river basins of the Hindukush Karakoram and Himalaya (HKH) region and Tibetan Plateau are covered in this study. Snow data in this study have a daily temporal resolution and 500 m spatial resolution. The product is 5 derived from MODIS Terra (MOD10A1) and Aqua (MYD10A1), and Glacier (GL), Version 6 (06), named as M*D10A1GL06. The data in this product is available in GeoTIFF format.

Methodology
The input data for this study include collection 6 (C6) of the daily MODIS Terra (MOD10A1) and Aqua (MYD10A1) products for the period between 2002 and 2019. The snow data were downloaded from https://earthdata.nasa.gov/ (last access: 24 January 2020) of NASA's Earth Science Data Systems (ESDS) program. The algorithm in C6 has significantly reduced the errors of omission and commission in snow pixels detection mainly due to low illumination conditions and high solar zenith angle (SZA) as compared to Collection 5 (C5) (Riggs et al., 2016). The data are described as 0-100 (Normalized Difference Snow Index (NDSI) snow cover), 200 (missing data), 201 (no decision), 211 (night), 237 (inland water), 239 (ocean), 250 (cloud), 254 15 (detector saturated), and 255 (fill) (Riggs et al., 2016;Hall, 2016a, 2016b). The data for snow pixels are the NDSI values of 0−1 scaled to the range of 0−100 derived from the daily surface reflectance product (MOD09GA). We have converted the NDSI values to binary snow using the range applied in version 5 (40−100) of M*D10A1 products. The above-mentioned values were reclassified into three classes: 1) The values 40−100 as snow class and reclassified to (200), 2) value 250 is cloud and reclassified to (50), 3) the rest of the values are 20 classified as no snow (25), to make it comparable with the improved 8-day composite MOYDGL06* product (Muhammad and Thapa, 2020).
The cloudy pixels in daily Terra and Aqua snow products were replaced by snow, no snow, or remain cloudcovered using the corresponding 8-day composite improved snow (MOYDGL06*) product (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) with reduced uncertainty of underestimation and overestimation (Muhammad and Thapa, 2019; for the period 25 between 2002 and 2019. We processed the 8-day composite images for the year 2019 following the methodology of MOYDGL06* to extend the improved daily snow product to 2019. The Terra and Aqua daily products were separately processed and improved by removing clouds and overestimation. In the initial processing, the overestimation is reduced to the extent of eight-day composite images by discarding snow in daily MODIS images falling beyond the maximum extent of snow in the corresponding eight-day composites (MOYDGL06*) as shown 30 in Eqn. (1). We call the snow beyond the 8-day composite snow extent as overestimation because the 8-day composite images are the maximum extent of snow in the consecutive eight images.
The value 50 in the superscripts represents clouds and M*D10A1 represents MOD10A1 and MYD10A1. The MOD10A1 and MYD10A1 were separately processed and shown here in the same equation.
Also, the daily MODIS product contains gaps with missing data between two successive strips with an increased gap near the equator. The missing data pixels caused by such gaps in the daily Terra and Aqua products were filled using the corresponding snow or no snow pixels of the MOYDGL06* product using Eqn. (2).
The superscript NoData represents Gap in either daily MODIS Terra or Aqua data.
The improved MODIS Terra and Aqua daily snow products were combined and merged with Randolph Glacier Inventory version 6 (RGI6.0) to make an improved and combined snow and glacier product. The methodology of merging daily products is different from that of MOYDGL06* as the nature of the daily and 8-day product is 5 different to some extent. We did not replace snow pixels as no snow if it is snow either Terra or Aqua and suggest to assign 0.5 weight while using this product for snow cover analysis. The snow data in this product are also preserved to make the separated Terra or Aqua products retrievable from this product. The Terra and Aqua snow data were combined using the following Eqs. (3)-(7).
The combination of daily improved snow from Terra and Aqua with RGI was also carried out in the same way 15 except in the case of cloud in the snow data, the glacier ice is described either debris-cover or debris-free derived from RGI6.0 inventory. The glacier (debris-cover and debris-free) are described as 240 and 250 if they are exposed, otherwise given different values depending on either the glacier is covered with MODIS Terra, Aqua, or both the snow products. The description of improved daily snow combined with RGI product is described by the following values. There are thirty-six missing images in the original snow products with thirty-five in Terra snow and one in the Aqua snow equivalent to 0.29% of the total snow data which is insignificant for the time series. Missing data in the Terra snow with ordinal dates are 2003032, 2003199, 2003351−2003358, 2004050, 2004248, 2004277, 2005265, 2006172, 2006235, 2008355−2008358, 2009252, 2010065, 2010177, 2014299, 2016050−2016059, and

Results and discussion
This study improved and combined daily MODIS Terra and Aqua snow data merged with RGI6.0 separately into debris-covered and debris-free parts of the glacier (M*D10A1GL06) for the period of eighteen years between 2002 and 2019. Our methodology used the improved 8-day MOYDGL06* product as training data for improving the daily product. The eight-day data for 2019 was also improved following the algorithm described in Muhammad and Thapa (2020)  respectively. We almost completely removed cloud cover in this paper with the remaining clouds of 0.001% as shown by a straight red line in Figure 1. On average, the cloud cover in the original Terra is slightly less than Aqua data, however, the spatial distribution of clouds varies significantly with time. The couldcloud cover is significantly higher in the daily original snow product (42.7% on average) as compared to the eight -day composite product with 3.66% cloud cover. These cloud cover statistics indicate that more than 91% of the clouds were reduced in the eight-day composite M*D10A2 products available at the National Snow and Ice Data Center (Riggs et al., 2016) in HMA on average. This made our final Terra and Aqua combined daily snow product 99.99% 35 cloud-free on average. The cloud cover of the original and improved Terra and original and improved Aqua are shown in Figure 2 (a) and 2 (b). The annual average snow cover in the original Terra snow product was 6.07%, increased to 16.82% in the improved Terra snow product. Similarly, the original Aqua snow product was 5.0 5%, increased to 16.97% in the improved Aqua snow product. The original Terra and Aqua average snow was 5.56%, increased to 16.95% in the improved Terra and Aqua combined snow. An example of the original Terra and Aqua 6 images containing clouds and missing data, causing snow underestimation and the improved Terra and Aqua combined snow products is shown in Figure 3. The average annual clouds and snow statistics for original Terra MOD10A1, Aqua MYD10A1, improved Terra, MOD10A1, improved Aqua MOD10A1, and the combined Terra and Aqua productsMOYD10A1 product are shown in Table 1.  Figure 1 (a) shows cloud cover in the original MOD10A1 and improved MOD10A1 products and Figure 1 (b) shows the original MYD10A1 and improved MYD10A1 products.   Removing unmatched Terra and Aqua data in daily snow may increase the underestimation for areas where SZA is greater (Sayer et al., 2015). It is particularly challenging to detect snow when SZA exceeds 70 o (Riggs et al., 2016) which constitutes up to 8% of the data (Horváth et al., 2014). Similarly, for SZA > 60° the cloud optical thickness increases (Loeb and Davies, 1997) which is overcome by removing clouds using the eight-day composite data containing snow data overlapped by Terra and Aqua. In contrast, assigning a weight of 0.5 to such 5 data may reduce the overestimation to 50% of the data acquired from off-nadir view. To assess the variability of snow overestimation mainly due to SZA differences, we compared the minimum (sn ow overlapped by Terra and Aqua), maximum (snow in either Terra or Aqua), and mean snow (weightage of 1 to minimum snow and 0.5 to maximum snow). The maximum and minimum snow cover area showed a difference of 12.4% on average for the whole study area, whereas the mean snow differs by 6.2% on average in comparison to the minimum and 10 maximum snow. Therefore, we suggest using the mean snow for snow cover analysis using this product. Also, both the minimum and maximum snow may be analyzed for estimating a range of snow cover area. The original Terra and Aqua, minimum, maximum, and mean of the improved snow are shown in Figure 4 showing the difference explained above for the study period. There are significant variations and underestimation in the original snow mainly due to the persistence of clouds as shown in Figure 4.

15
On average, 87.6% of the individually improved Terra and Aqua snow pixels coincided in the improved Terra and Aqua combined snow products. The remaining 12.4% of the mismatched snow pixels in the individual Terra and Aqua is suggested to give 0.50 weight to be used in combination with the coincided snow for understanding snow cover dynamics, regarded as mean snow. This criterion enables to discard 50% of the mismatched snow (6.2%) in the improved Terra and Aqua composite product. The use of either minimum, maximum, and mean 20 snow data may be used with caution for small scale as the difference and mismatch may vary from region to region. Also, it is important to mention that the mismatch does not include those snow pixels in the individual Terra and Aqua snow products which fall beyond the snow extent of the respective 8 -day composite images. The mismatch of snow is mainly caused by the off-nadir view, low spatial resolution, and large swaths of the satellite (Muhammad and Thapa, 2020). The derived product is based on the improved and validated eight -day composite 25 product, therefore, we do not re-validated it. It is important to mention that the 8-day compositeMOYDGL06* product show an overestimation of 32% on average when compared with the improved daily snow dataM*D10A1GL06 product developed in this paper as shown in Figure 5. These results are quite critical for studies related to snow onset and melt timing and related hydrological simulations. The daily or 8-day snow products should be carefully selected depending on the nature of application to avoid biases and uncertainty. The daily product generated in this research is primarilymainly recommended for hydro-glaciological, water, and snow-related studies with high-temporal (daily) resolution except for very small-scale studies. An example image of the improved snow product with the description of values in the methodology and data availability sectio ns is shown in Figure 6.  improvement. The minimum Snow cover is the snow overlapped by Terra and aqua in the improved snowMOYD10A1 product, whereas, the maximum snow in the improved MOYD10A1 product is snow either in Terra or Aqua products. 5 Figure 5: Difference between daily improved dailyM*D10A1GL06 and 8-day composite snowMOYDGL06* products on monthly interval. The 8-day dataMOYDGL06* product shows overall positive bias and overestimation of ~32% on average compared to M*D10A1GL06.  clean ice under Aqua snow (249), exposed ice (250), and clean ice under Terra and Aqua snow (252). For studies using this product to analyze snow cover, we recommend to useusing 0.5 weight to snow pixels if present in either of the Terra or Aqua described by values 198, 199, 238, 239, 248, and 249 and 1 weightage to the snow pixels 15 with snow in both the Terra and Aqua with values 200, 242, and 252. The combined and improved snow product compared to the original Terra and Aqua snow products for the study period is shown in Figure 4. The combined product will serve as baseline data for hydro-glaciological and other water-related applications. The data are  (Muhammad, 2020). The source code of the algorithm for this product is available at https://doi.org/10.5281/zenodo.3862058 (Thapa, 2020). The dataset README file with the data at PANGAEA gives the information about the data and code.

Conclusion
This study results in an improved Terra and Aqua MODIS version 6 combined daily snow products merged with

10
When snow is detected in either of Terra or Aqua dataset, it is denoted as 198, 199, 238, 239, 248, and 249 where the even and odd values represent Terra and Aqua snow, respectively. The exposed debris-cover and debris-free ice are denoted as 240 and 250, similarly as in MOYDGL06* product. The average cloud persistency is 42.7 % of the original products (both Terra and Aqua) for the study region in the observed period. There is a 12.4 % mismatch between the Terra and Aqua snow in the improved snow product primarily due to thelarge SZA, wide 15 swath, and low spatial-resolution which limit accurate snow detection in the complex topography. To reduce the effect of the mismatch in snow data from 50% to 6.2% in the statistical analysis, we suggest a weight of 0.5 to the mismatched snow pixels. The clouds cause 32.9% underestimation in snow pixels, which together with a 6.2% mismatch due to larger SZA causes uncertainty of 39.1% on average. The mentioned uncertainty does not include the snow underestimation due to the data gaps and overestimation of snow pixels occurring beyond the eight -day 20 maximum extent of snow in MOYDGL06* product. The daily snow M*D10A1GL06 product associated with this paper can aid a valuable input dataset for hydro-glaciological and climate modelling, snow cover dynamics, and other water-related studies.

Author contributions.
SM designed the study and conceptualized the methodology and wrote the paper. AT developed the R code and 25 contributed to the manuscript. Both the authors contributed to the data quality control.

Competing interests.
Both the authors declare no conflict of interest.

Acknowledgements.
This work was supported by the Cryosphere Initiative of the International Centre for Integrated Mountain

15
Comments to the Author: Dear Authors thank you very much for the submission of your manuscript to ESSD, for the replies to the reviews of your paper and for the revision of your manuscript. We consider M*D10A1GL06 and the related ESSD manuscript as a distinct and useful data publication for the High 20 Mountain Asian region. We require minor revision and edits in the manuscript text and figures. Response: We thank the editor for the time to review our manuscript and suggest important changes for improved readability. We thoroughly revise the manuscript and do the necessary changes in the text and language. We do hope that the revision will satisfy the editor.

Formal Requirements
In general i) please let check the language, there are still minor issues, specifically in some of the new 30 paragraphs, and specifically in the abstract. E.g. 'The pixels with values 200, ,, indicate .. and has a …, or The data associated with this paper are available for the end-users mainly useful for observation and … or on p.5 the typo 'could cover' Response: Thank you for highlighting the minor issues. The manuscript is revised significantly for language improvement. The highlighted sentences are revised and improved.

35
The typo is corrected. ii) please spell out all abbreviations, when abbreviations first appear, e.g. NDSI Response: All the abbreviations are now given in full form including NDSI on the first appearance. iii) important: please enhance the quality in all of your figures and exchange every figure: the Response: Text and legend of figures 3-6 are enhanced. Legend names and axes labels are also improved. iv) M*D10A1GL06 is a well-chosen term for your product, please use it much more frequently, e.g., please use it in the abstract and in the conclusion, and throughout your manuscript text, please also describe in method the produced files, e.g. such as the name of 5 the product file MOYD10A1GL06_Maximum_Snow_Extent_2002289 Response: The product name M*D10A1GL06 is now used in the abstract, method, and conclusion. The product name is now also explained in the method section as suggested: "The product in this paper was named merging the names of original products e.g. combining Terra product (MOD10A1_Maximum_Snow_Extent_2002289) and Aqua product 10 (MYD10A1_Maximum_Snow_Extent_2002289) merged with RGI06 (GL06) and named as MOYD10A1GL06_Maximum_Snow_Extent_2002289 in the daily improved snow product (M*D10A1GL06).". v) please highlight in your abstract and conclusion the improvement of your product in a quantitative way, e.g. statements such as 'The effect of SZA was reduced by merging of daily 15 Terra and Aqua products with snow if the pixel is snow in both the products while giving 0.5 weight if the pixel is snow in one of the Terra or Aqua. This criterion reduces 6.2% of the overestimation in the daily composite snow product.' Response: the effect of SZA and cloud cover is now highlighted in the abstract and conclusion. In abstract, it is given as "On average, the M*D10A1GL06 product reduces 20 39.1% of uncertainty compared to MOYDGL06* product due to cloud cover (underestimation) and sensor limitations mainly larger solar zenith angle (SZA) (overestimation) of 32.9% and 6.2%, respectively." vi) please put the information on the 500 m spatial resolution in the abstract and conclusion.
Response: The information about the MODIS snow including spatial resolution is now given 25 in the abstract and conclusion as suggested. Text p.3, L36 Also, the daily MODIS contains -change to: the daily MODIS product contains p.5 The could cover Response: the daily MODIS contains is changed to the daily MODIS product contains Response: Figure 4 caption is made similar to Figure 5. The legend and text in both figures are enhanced for better readability.