Comparison of the NASA Standard MODerate-Resolution Imaging Spectroradiometer and Visible Infrared Imaging Radiometer Suite Snow-Cover Products for Creation of a Climate Data Record: A Case Study in the Great Basin of the Western United States

Abstract

A nearly continuous daily, global Environmental Science Data Record of NASA Standard MODerate-resolution Imaging Spectroradiometer (MODIS) snow-cover extent (SCE) data products has been available since 2000. When the MODIS record ends, the ‘moderate resolution’ SCE record will continue with NASA Standard Visible Infrared Imaging Radiometer Suite (VIIRS) SCE data products. The objective of this work is to evaluate and quantify the continuity between the MODIS and VIIRS SCE data products to enable the merging of the data product records. A climate data record (CDR) could be developed when 30 years of daily global moderate-resolution SCE become available if the continuity of the MODIS and VIIRS records can be established. Here, we focus on the daily cloud-gap-filled MODIS and VIIRS SCE NASA standard data products, MOD10A1F and VNP10A1F, respectively, for a case study in the Great Basin of the western United States during a period of sensor overlap. Using the methodologies described herein (daily percent of snow cover, duration of snow cover, average monthly number of days (Ndays) of snow cover, and trends in Ndays of snow cover, we show that the snow maps display excellent agreement. For example, the average monthly number of days of snow cover in the Great Basin calculated using MOD10A1F and VNP10A1F agrees with a Pearson’s correlation coefficient of r = 0.99 for our 11-year study period from WY 2013 to 2023. Additionally, the SCE derived from each data product agrees very well with meteorological station data, with a Pearson’s correlation coefficient of r = 0.91 and r = 0.92 for MOD10A1F and VNP10A1F, respectively. Our results support the eventual creation of a CDR.

1. Introduction

Snow plays a critical role in the water supply in many parts of the world, including the western United States. While snowmelt runoff currently provides ~67 percent of the inflow to the major reservoirs in the U.S. West, hydrological model forecasts show that this contribution is expected to dwindle [1]. Hence, monitoring snow cover trends is imperative, especially because snow is melting earlier in the Northern Hemisphere due to increasingly warmer late winter/spring conditions [1,2].

There exists a ~190 km resolution climate data record (CDR) of snow cover extent (SCE) in the Northern Hemisphere that has been instrumental in allowing researchers to detect changes in snow cover since the record began in 1966 [3,4]. A CDR can be developed with 30 years of satellite data which is the minimum number of years needed to reflect climate trends, establish statistical significance, and mitigate annual fluctuations [5,6].

This valuable CDR was developed from weekly Northern Hemisphere SCE data digitized from weekly snow cover maps. This is the longest satellite-based CDR of any environmental variable. However, to study global and regional changes in greater detail, a moderate-resolution time series of daily global SCE is needed.

A nearly continuous daily, global Environmental Science Data Record (ESDR) of NASA Standard MODerate-resolution Imaging Spectroradiometer (MODIS) SCE data products has been available since 24 February 2000 from the Terra satellite and 24 June 2002 from the Aqua satellite. However, the MODIS record will end when the Terra and Aqua satellites are deactivated in 2026 or 2027. At the time of writing, the Terra MODIS is operating optimally and continuing to provide well-calibrated data, resulting in reliable SCE retrievals; a decline in product reliability is not anticipated. This SCE record will continue with the Suomi-National Polar orbiting Partnership (S-NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) NASA Standard SCE data products, thus enabling a record of ‘moderate resolution’ SCE that could potentially be elevated to the status of a CDR.

The objective of this work is to quantify the agreement between the MODIS and VIIRS SCE data products in the Great Basin of the western U.S. to establish the continuity of the data records. We are not aware that this has been addressed previously. We utilize both Terra MODIS and S-NPP VIIRS satellite-derived cloud-gap filled (CGF) SCE data products to evaluate their consistency and agreement, focusing on their potential for use in constructing a moderate-resolution CDR of daily global SCE. We look at the daily percent of snow cover, duration of snow cover (including onset and melt dates), average monthly number of days (Ndays) of snow cover, and trends in Ndays of snow cover to evaluate the consistency and agreement of the snow cover products in the Great Basin. We also compare MODIS and VIIRS daily SCE results with in situ data from meteorological stations.

2. Background

2.1. Study Area

The Great Basin is a ~500,000 km² closed catchment (endorheic basin) in Nevada, Utah, Oregon, and California in the western United States (Figure 1). At least 75 percent of the water enters the Great Basin via snowfall in the winter months and leaves almost exclusively via evaporation, sublimation, and consumptive water use [7,8,9]. Snow can cover most of the Great Basin in a month having extensive or frequent snowfall (Figure 2).

Within the Great Basin, cloud cover during the snow-prone months (October through to June) is not continuous. There are many days without full cloud cover, providing ample opportunities for MODIS to capture snow cover conditions (

Table 1. Mean cloud cover frequency (CCF) in the Great Basin calculated from the cloud persistence information in the MOD10A1F data products, 2013–2023.

Month	CCF
October	0.25
November	0.41
December	0.60
January	0.57
February	0.52
March	0.49
April	0.37
May	0.36
June	0.20

). Even in December, with a mean cloud cover frequency (CCF) of 0.6 for the study period, MODIS can view the surface on 40 percent of the days on average within the Great Basin. The average number of views of the surface is even higher in the other months.

In the lower elevations of the Great Basin, desert shrublands and sagebrush are prevalent. In the mid-elevations, woodlands are common. At the highest elevations and in mountainous areas, such as in the eastern part of the Great Basin in the Wasatch and Uinta mountains, montane forests dominated by conifers are common in these cooler areas with greater precipitation. Snow cover is prevalent at higher elevations in winter months. Dense canopies can intercept snowfall, but open areas within a forest can allow more snow accumulation. Thus, there is a great deal of variability in snow and forest cover distribution at different elevations in the Great Basin.

2.2. Snow Cover Data Products

A moderate-resolution SCE CDR could begin with the MODIS Terra SCE record on 24 February 2000 and continue through the VIIRS era, which is expected to extend at least into the 2030s. The VIIRS on S-NPP was launched on 28 October 2011, followed by the NOAA-20, NOAA-21 and NOAA-22 satellites [10,11]. The Version 2 VIIRS snow-detection algorithm was designed to match the MODIS SCE mapping Version 6.1 algorithm to ensure that the MODIS ESDR of global SCE can be extended into the future with S-NPP and NOAA-20 and follow-on VIIRS snow data products.

MOD10A1F and VNP10A1F are validated data products that provide daily global SCE at spatial resolutions of 500 m and 375 m, respectively [12] (Figure 2). MODIS and VIIRS snow cover data products have been reprocessed to provide improvements for the user community [12]. MOD10A1F Version 6.1 (also known as Collection 6.1 or C6.1) and VNP10A1F Version 2 (also known as Collection 2 or C2), used in this work, are the most current versions of the NASA Standard CGF SCE products [13,14] and are available to download from the NASA Distributed Active Archive Center (DAAC) at the National Snow and Ice Data Center (NSIDC) in Boulder, Colorado.

Algorithms utilizing data from the MODIS and VIIRS sensors provide high-quality global snow cover maps under clear skies. There are many similarities and differences between the MODIS and VIIRS sensors (

Table 2. Satellite and sensor characteristics and spectral bands used for calculation of normalized-difference snow index (NDSI).

	MODIS Terra	MODIS Aqua	VIIRS S-NPP	VIIRS NOAA-20
Equator crossing time	9:30 AM	1:30 PM	1:30 PM	1:30 PM
Swath width	2330 km (±55° scan angle)	2330 km (±55° scan angle)	3060 km (±56° scan angle)	3060 km (±56° scan angle)
Resolution	500 m at nadir	500 m at nadir	375 m at nadir	375 m at nadir
Spectral bands used for calculation of NDSI	VIS B4, 0.555 µm NIR B6, 1.640 µm	VIS B4, 0.555 µm NIR B6, 1.640 µm	VIS I1, 0.640 µm NIR I3 1.610 µm	VIS I1, 0.640 µm NIR I3 1.610 µm

). The Terra MODIS has been providing daily snow maps since 2000 using a subset of the 36 MODIS channels. MODIS and VIIRS snow maps are useful when studying regional- and basin-scale SCE for hydrological applications as well as hemispheric and global snow cover. Cloud cover is the biggest issue affecting the accurate mapping of SCE using visible, near-infrared and shortwave-infrared sensors [15,16]. Cloud gap filling is used to mitigate the cloud issue [17].

There are 22 channels on the S-NPP VIIRS, many of which are very similar in wavelength range to the MODIS channels; however there should be no expectation that the MODIS and VIIRS snow maps, when acquired on the same day, are identical. Different bands are used for cloud mapping on MODIS and VIIRS, contributing to differences in cloud masking. Additionally, the Terra MODIS and the S-NPP and NOAA-20 VIIRS data are acquired at different times of the day, allowing for movement of clouds and changes in snow cover (i.e., sublimation and melt) between acquisitions [18].

3. Materials and Methods

In this work, the Terra MODIS MOD10A1F and S-NPP VIIRS VNP10A1F data products were each gridded to the same 500 m resolution latitude/longitude grid. For ease of comparison, the VIIRS products were degraded from 375 m to 500 m resolution to match the MODIS resolution using the Geospatial Data Abstraction Library nearest-neighbor reprojection. For this work, if the Normalized Difference Snow Index (NDSI) value for the grid cell was 0.1 or greater, then the grid cell is considered snow-covered. We used an NDSI threshold of 0.1 so that all possible snow, as measured using the MODIS and VIIRS SCE algorithms, would be included.

First, we measured the daily percentage of snow cover in the Great Basin and then explored the use of the MOD10A1F and VNP10A1F data products to determine the onset and melt (and duration) of snow cover in the Great Salt Lake Basin for WY 2013–2023. The date on which the daily snow cover in the basin reached 5 percent, and stayed at 5 percent or greater for five or more days, was used as the date of snow onset. The date on which the percentage of snow cover dropped and remained below 5 percent through June was used as the date of snowmelt. The onset and melt dates of the snowpack were calculated using each product, and then the snow cover duration in days was calculated for each year.

We also calculated the average monthly number of days (Ndays) of snow cover for the Great Basin for each grid cell from February 2012 to August 2023 (139 months). Ndays of snow cover at each pixel for a given month was averaged for each product. The average Ndays for all pixels in the Great Basin was then calculated.

For the Great Salt Lake Basin within the Great Basin (Figure 1), up to 102 meteorological stations [19] were selected. Using the grid cell in which each station was located, the average monthly Ndays of snow cover was derived from MOD10A1F and VNP10A1F and compared with the average monthly Ndays of snow cover derived from the meteorological stations located in that grid cell. The locations of the meteorological stations are shown in Figure 3. A station was selected if it reported for 75% of days in the snowiest part of each year (1 October–31 March) for 20 or more years. Up to 25% of the stations did not report snow conditions every year; therefore, in any given year, data from fewer than 102 stations were available for use. Then, for each year, each station’s daily snow depth data were checked, and if the value was greater than zero, then that day was counted as a day with snow. Values from all available stations were averaged to obtain the Ndays of snow for the basin as a whole, each year.

Finally, to investigate decade-scale continuity between MODIS and VIIRS data products, we calculated trends in average monthly Ndays using both MOD10A1F and VNP10A1F and compared the results.

4. Results

4.1. Daily Percent of Snow Cover in the Great Basin

The percentage of snow cover in the Great Basin was calculated for each day during the study period (Figure 4a). During the snow-prone parts of WY 2023, for example, the average daily difference between the percent of snow mapped using MOD10A1F and VNP10A1F is less than 1 percent (0.69 percent), with VIIRS mapping slightly more snow than MODIS (Figure 4b).

Over the study period, we noted lower agreement between MOD10A1F and VNP10A1F during the winter months of December, January, and February compared to the spring months of March, April, and May. The mean difference between the MODIS and VIIRS products (MODIS minus VIIRS) in the winter is −0.94 days and −0.86 days in the spring. We do not know why there is a greater difference in the winter, but we can speculate that it may be due to the greater amount of winter snow cover vs. spring snow cover. In the spring, especially later in spring, snow cover will be retained longer at the higher elevations and thus may be mapped more consistently between sensors.

4.2. Snow-Cover Duration Using MOD10A1F and VNP10A1F

As calculated from the daily percentage of snow cover in the Great Salt Lake Basin for the study period, the use of the VNP10A1F products provided ~2 days longer average daily duration of the snow season than the MOD10A1F products, as shown in Figure 5.

4.3. Average Monthly Number of Days (Ndays) of Snow Cover

We also assessed the agreement between MODIS- and VIIRS-derived average monthly Ndays of snow cover in the Great Basin using time series data from calendar year (CY) 2012 to 2023, which is slightly longer than our study period, WY 2013–2023. The average monthly Ndays of snow cover derived from the two products is very close, with a Pearson’s correlation coefficient of r = 0.99 (Figure 6a). For the study period, VNP10A1F provides 0.21 more days of snow cover on average (Figure 6a,b). In most of the 139 months, the VNP10A1F data products show slightly more days of snow cover than MOD10A1F. There is a break in the plot in the right panel (Figure 6c) because there were nine consecutive days of missing MODIS data in February of 2016.

Comparison between Satellite Data and Meteorological Station Data

When average monthly Ndays of snow cover derived from satellite data are compared with average monthly Ndays of snow cover derived from meteorological station data in the Great Salt Lake Basin (WY 2013–2023), the mean Pearson correlation coefficients for the study period are very close, with r = 0.91 for MOD10A1F and r = 0.92 for VNP10A1F (

Table 3. Statistics describing the comparison between Ndays of snow cover derived from (a) MODIS and (b) VIIRS vs. Ndays of snow cover derived from meteorological station data in the Great Salt Lake Basin, WY 2013–2023.

(a) MODIS
WY	r	r2	RMSE	Mean Bias	#Stations
2013	0.859	0.738	32.207	−9.631	84
2014	0.904	0.818	29.677	−3.932	88
2015	0.917	0.84	29.704	−6.966	88
2016	0.892	0.795	27.179	−3.58	81
2017	0.924	0.853	26	−4.155	84
2018	0.926	0.857	26.092	−8.341	88
2019	0.891	0.794	32.144	−6.035	85
2020	0.9	0.811	30.94	−8.851	87
2021	0.938	0.88	24.15	−4.258	89
2022	0.918	0.842	32.687	−14.337	86
2023	0.944	0.89	18.85	−1.056	89
(b) VIIRS
WY	r	r2	RMSE	Mean Bias	#Stations
2013	0.882	0.779	29.929	−7.071	84
2014	0.917	0.84	27.202	−1.352	88
2015	0.917	0.84	29.875	−6.318	88
2016	0.902	0.813	25.899	−3.42	81
2017	0.93	0.865	25.265	−0.131	84
2018	0.929	0.863	25.358	−6.625	88
2019	0.893	0.797	31.527	−4.647	85
2020	0.912	0.831	29.446	−6.092	87
2021	0.94	0.884	23.911	−3.843	89
2022	0.928	0.861	30.601	−11.058	86
2023	0.95	0.903	19.618	2.775	89

). This demonstrates that the agreement between the satellite-derived and station-derived snow cover is excellent, whether MODIS or VIIRS SCE data products are used to calculate the average monthly Ndays of snow cover.

4.4. Comparison of Trends in Average-Monthly Ndays of Snow Cover Using MOD10A1F and VNP10A1F

Trends in average monthly Ndays of snow cover derived from MOD10A1F and VNP10A1F were calculated for the Great Basin from February 2012 through August of 2023. Both data products show a positive but statistically insignificant trend in average monthly Ndays of snow cover, with a gain of 1.6 days over the 139-month period (Figure 6a).

5. Discussion

Establishing the continuity of NASA’s MODIS and VIIRS cloud gap-filled (CGF) snow cover products is necessary for creating a future moderate-resolution SCE climate data record (CDR) for monitoring changes and trends in climate and for validating and improving climate models using a consistent long-term dataset [5,6]. Intermittently cloudy days have a minor effect on accuracy, but extended cloudy periods can decrease the accuracy of a SCE map if snow cover conditions are changing [17,20,21,22,23]. Snow conditions can even change between the morning Terra MODIS and afternoon VIIRS overpasses, causing one sensor to detect snow while the other does not, both accurately reflecting the conditions. Despite these and other minor differences between the MODIS and VIIRS SCE daily products, the CGF data products are highly promising for use in developing a future CDR.

While we show that mean cloud cover frequency (CCF) is greatest in December during our study period (Table 1), this may not reflect the actual conditions if CCF were, for example, measured using ground-based instruments. The mean CCF varies from 20 to 60 percent in the snow-prone months (Table 1), leaving ample opportunities for the satellites to view the surface and thus update the cloud gap-filling algorithm to produce high-quality daily CGF snow products.

The average monthly Ndays of snow cover derived from MODIS and VIIRS CGF satellite-derived snow maps and average monthly Ndays of snow cover calculated from meteorological station data show excellent agreement in the Great Salt Lake Basin. Data from meteorological stations are accurate at one point in the grid cell, but the average may or may not be a meaningful representation of the actual presence or absence of snow cover in that grid cell. Even with these caveats, the satellite- and station-derived estimates of snow cover match extremely well.

The S-NPP VIIRS SCE data products map slightly more snow cover than the MODIS SCE products for daily and time-series data. This finding is consistent with earlier work [10,16]. Why does VIIRS map more days of snow cover and more snow-covered areas than MODIS? The cloud-masking algorithm used with VNP10A1F is less conservative than the cloud-masking algorithm used with MOD10A1F, allowing more snow to be mapped (if present). Also contributing to different results between MODIS and VIIRS are factors such as acquisition time, viewing geometry, and change in pixel size across a scan, as well as projecting and gridding the observations. Furthermore, the Terra MODIS, used to create MOD10A1F, has a morning overpass time, while the S-NPP VIIRS has an afternoon overpass time, possibly showing different snow conditions. Additionally, in mountainous regions, differences in snow cover are concentrated around the periphery of snow-covered regions and may be associated with changes in cloud cover or/and snow cover between data acquisition times [10].

When calculating trends from monthly averages of Ndays of snow cover in the Great Basin, the SCE derived from MOD10A1F and VNP10A1F SCE products provide nearly identical results.

6. Conclusions

The period of overlap between MODIS and VIIRS sensors provides a valuable opportunity to assess the agreement of the data products, and thus the viability of a potential CDR of moderate-resolution daily global SCE. Via the measures described in this paper: daily percent of snow cover, snow cover duration, comparison with in situ data, average monthly Ndays of snow cover, and trends in average monthly Ndays of snow cover, the NASA Standard MODIS and VIIRS SCE data products display excellent agreement and very high continuity in the Great Basin. Our results support the creation of a CDR using MOD10A1F and VNP10A1F when 30 years of data become available.