AMTAtmospheric Measurement TechniquesAMTAtmos. Meas. Tech.1867-8548Copernicus PublicationsGöttingen, Germany10.5194/amt-11-2983-2018Assessing snow extent data sets over North America to inform and improve
trace gas retrievals from solar backscatterAssessing snow extent data sets over North AmericaCooperMatthew J.cooperm2@dal.cahttps://orcid.org/0000-0002-4145-3458MartinRandall V.https://orcid.org/0000-0003-2632-8402LyapustinAlexei I.https://orcid.org/0000-0003-1105-5739McLindenChris A.https://orcid.org/0000-0001-5054-1380Department of Physics and Atmospheric Science, Dalhousie University,
Halifax, Nova Scotia, CanadaHarvard-Smithsonian Center for Astrophysics, Cambridge, Massachusetts,
USANASA Goddard Space Flight Center, Greenbelt, Maryland, USAAir Quality Research Division, Environment and Climate Change Canada,
Toronto, Ontario, CanadaMatthew J. Cooper (cooperm2@dal.ca)22May20181152983299412January201824January20181May20183May2018This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/4.0/This article is available from https://amt.copernicus.org/articles/11/2983/2018/amt-11-2983-2018.htmlThe full text article is available as a PDF file from https://amt.copernicus.org/articles/11/2983/2018/amt-11-2983-2018.pdf
Accurate representation of surface reflectivity is essential to tropospheric
trace gas retrievals from solar backscatter observations. Surface snow cover
presents a significant challenge due to its variability and thus
snow-covered scenes are often omitted from retrieval data sets; however, the
high reflectance of snow is potentially advantageous for trace gas
retrievals. We first examine the implications of surface snow on retrievals
from the upcoming TEMPO geostationary instrument for North America. We use a
radiative transfer model to examine how an increase in surface reflectivity
due to snow cover changes the sensitivity of satellite retrievals to
NO2 in the lower troposphere. We find that a substantial fraction
(> 50 %) of the TEMPO field of regard can be snow covered in
January and that the average sensitivity to the tropospheric NO2
column substantially increases (doubles) when the surface is snow covered.
We then evaluate seven existing satellite-derived or reanalysis snow extent
products against ground station observations over North America to assess
their capability of informing surface conditions for TEMPO retrievals. The
Interactive Multisensor Snow and Ice Mapping System (IMS) had the best
agreement with ground observations (accuracy of 93 %, precision of 87 %,
recall of 83 %). Multiangle Implementation of Atmospheric Correction
(MAIAC) retrievals of MODIS-observed radiances had high precision (90 %
for Aqua and Terra), but underestimated the presence of snow
(recall of 74 % for Aqua, 75 % for Terra). MAIAC generally outperforms
the standard MODIS products (precision of 51 %, recall of 43 % for Aqua;
precision of 69 %, recall of 45 % for Terra). The Near-real-time Ice and
Snow Extent (NISE) product had good precision (83 %) but missed a
significant number of snow-covered pixels (recall of 45 %). The Canadian
Meteorological Centre (CMC) Daily Snow Depth Analysis Data set had strong
performance metrics (accuracy of 91 %, precision of 79 %,
recall of 82 %). We use the Fscore, which balances precision and recall, to
determine overall product performance (F=85 %, 82 (82) %, 81 %,
58 %, 46 (54) % for IMS, MAIAC Aqua (Terra), CMC, NISE, MODIS
Aqua (Terra),
respectively) for providing snow cover information for TEMPO retrievals from
solar backscatter observations. We find that using IMS to identify snow
cover and enable inclusion of snow-covered scenes in clear-sky conditions
across North America in January can increase both the number of observations
by a factor of 2.1 and the average sensitivity to the tropospheric NO2
column by a factor of 2.7.
Introduction
Satellite observations of solar backscatter are widely used as a source of
information on atmospheric trace gases (Richter and Wagner,
2011). These observations have provided valuable information on vertical
column densities of O3, NO2, SO2, CO, HCHO, CH4, and
other important trace gases in the troposphere
(Fishman et al., 2008).
Satellite observations of trace gases have been used to assess air quality
(Duncan et al., 2014; Martin,
2008) and to gain insight into atmospheric processes including emissions
(Streets et al., 2013), lifetimes
(Beirle et
al., 2011; Fioletov et al., 2015; de Foy et al., 2015; Valin et al., 2013),
and deposition
(Geddes and
Martin, 2017; Nowlan et al., 2014). The utility of these observations is
dependent on their quality, and thus ensuring retrieval accuracy is
essential.
Previous studies have found that retrieved NO2 vertical column
densities are highly sensitive to errors in assumed surface reflectance
(Boersma et al., 2004; Lamsal
et al., 2017; Martin et al., 2002). Much of this error sensitivity results
from observation sensitivity to trace gases in the lower troposphere. The
observation sensitivity is accounted for in the air mass factor (AMF)
conversion of observed line-of-sight “slant columns” to vertical column
densities. Uncertainties in surface reflectance are a significant
contributor to AMF uncertainty.
Existing reflectivity climatologies
(e.g.
Kleipool et al., 2008; Koelemeijer et al., 2003; Liang et al., 2002; Herman
and Celarier, 1997) do not represent snow cover well, since the statistical
methods to exclude reflective clouds from the climatologies also exclude
variable snow cover. Correspondingly, surface snow may be mistaken for
cloud, leading to errors in cloud fraction and pressure estimates used in
trace gas retrievals
(Lin
et al., 2015; O'Byrne et al., 2010; Vasilkov et al., 2017). Therefore, snow
cover is particularly challenging to retrievals. Misrepresenting surface
snow cover can lead to large errors (20–50 %) in retrieved NO2
columns over broad regions with seasonal snow cover
(O'Byrne et al., 2010). For this reason, observations
over snow are often omitted or flagged as unreliable to avoid potential
errors. This limits the ability of satellite retrieved data sets to offer
adequate temporal and spatial sampling in winter months. Additionally, over
highly reflective surfaces such as snow, observation sensitivity to the lower
troposphere is larger and has less dependence on a priori NO2 profiles
(Lorente et
al., 2017; O'Byrne et al., 2010). Thus, omitting snow-covered scenes means
omitting the observations with the greatest sensitivity to the lower
troposphere. This could be remedied by using a product that would allow for
snow cover identification to be done with confidence.
Several data products provide information on snow extent using surface
station observations, satellite-observed radiances, or visible imagery.
Previous evaluations have found it difficult to determine which of these
products is definitively the best, partly due to differences in resolution.
Most products are more consistent during the winter months when persistent,
deep snow is present (Frei
et al., 2012; Frei and Lee, 2010). However, disagreements are common during
accumulation and melting seasons, over mountains, and under forest canopies.
These evaluations have largely focused on local or regional snow cover or
have
included only cloud-free observations.
The upcoming geostationary Tropospheric Emissions: Monitoring of Pollution
(TEMPO) satellite instrument will provide hourly observations of air quality
relevant trace gases over North America at an unprecedented spatial and
temporal resolution
(Zoogman
et al., 2017). As is the case for all nadir satellite retrievals, the
quality of these observations will depend on the accuracy of the surface
reflectance used in the retrieval. As a significant portion of the observed
domain experiences snow cover, an accurate representation of snow cover is
needed. Current plans to deal with snow cover for TEMPO are to rely on
external observations.
In this work, we examine the importance of accurate snow identification by
using a radiative transport model to evaluate how the vertical sensitivity
of a satellite retrieval is impacted by surface reflectance. We then assess
seven snow extent products that are expected to continue to be operational
during the TEMPO mission using in situ observations across North America with
the intent of determining which product is best suited for providing snow
cover information for TEMPO and other future satellite retrievals. Finally,
we combine radiative transfer model results with a snow extent product to
show how including snow-covered scenes improves both the quantity and
quality of information in a retrieval data set.
Data and algorithmsGridded snow productsIMS
One of the most widely used sources of snow extent data is the Interactive
Multisensor Snow and Ice Mapping System (IMS). IMS provides daily,
near-real-time maps of snow and sea ice cover in the Northern Hemisphere at
4 km resolution (Helfrich et al., 2007). The maps are
produced by a trained analyst using visible imagery from a collection of
geostationary (e.g. GOES, MeteoSat) and polar orbiting (e.g. AVHRR, MODIS,
SAR) satellite instruments, with additional information from microwave
sensors (e.g. DMSP, AMSR, AMSU), surface observations (e.g. SNOTEL), and
models (e.g. SNODAS) (Helfrich et al., 2007). By using
multiple sources of information with different spatial resolution and
temporal sampling, IMS can minimize interference from clouds.
MODIS
A second commonly used snow and ice product is derived from MODIS satellite
observations from the Terra and Aqua satellites (Hall and
Riggs, 2007). Terra and Aqua have sun-synchronous, near-polar orbits with
overpass times of 10:30 and 13:30, respectively. Snow cover is calculated
using a Normalized Difference Snow Index (NDSI), which examines the
difference between observed radiation at visible wavelengths (where snow is
highly reflective) and short infrared wavelengths (where there is little
reflection from snow). Observations are made at 500 m spatial resolution and
aggregated to produce daily snow cover fractions on a 0.05∘
resolution grid. Past evaluations of the standard MODIS snow product show
good agreement in cloud-free conditions but often snow is misidentified as
cloud (Hall and Riggs, 2007; Yang et al., 2015).
The Multiangle Implementation of Atmospheric Correction (MAIAC) algorithm is
another algorithm processing MODIS observations. MAIAC retrievals uses
radiances observed by the MODIS Aqua and Terra satellites to provide
atmospheric and surface products including snow detection on a 1 km grid
(Lyapustin et al., 2011a,
b, 2012). While the NDSI used by the standard MODIS product is also used
by MAIAC as one of the criteria, the overall snow and cloud detection in
MAIAC are different from the standard MODIS algorithm
(Lyapustin et al., 2008).
NISE
The Near-real-time Ice and Snow Extent (NISE) provides daily updated snow
cover extent information on a 25×25 km grid (Nolin et al.,
2005). NISE uses microwave measurements from the Special Sensor Microwave
Imager/Sounder (SSM/I) on a sun-synchronous, quasi-polar orbit to observe
how microwave radiation emitted by soil is scattered by snow. Products based
on microwave measurements such as NISE are known to miss wet and thin snow,
as wet snow emits microwave radiation similar to soil, and thin snow does
not provide sufficient scattering.
CMC
The Canadian Meteorological Centre (CMC) Daily Snow Depth Analysis Data is a
statistical interpolation of snow depth measurements from 8000 surface
sites across Canada and US interpolated using a snow pack model
(Brasnett, 1999). Unlike the aforementioned
satellite products that provide snow extent, CMC provides snow depths. Daily
snow maps are produced at 25 km resolution. As it a reanalysis product,
there is a time delay in availability. The CMC snow depths show good
agreement with independent observations over midlatitudes and is considered
an improvement over previous snow depth climatologies
(Brown et al., 2003).
Surface observations
These snow identification products are evaluated against surface station
observations from the Global Historical Climatology Network Daily (GHCN-D)
database, an amalgamation of daily climate records from over 80 000 surface
stations worldwide (Menne et al.,
2012a). Most observations over Canada and the United States are collected by
government organizations (Environment and Climate Change Canada and NOAA
National Climatic Data Center, respectively) with additional measurements
from smaller observation networks. While the focus of the database is
collecting temperature and precipitation measurements, many stations (1279
in Canada and 13 932 in the United States in 2015 used here) also offer snow depth
measurements.
A subset of the surface stations included in GHCN-D may also be used in the
CMC reanalysis. It is difficult to definitively know which stations are
used, as CMC does not routinely archive this information. However, we
estimate that only 5 % of the GHCN-D stations used here are located within
0.1∘ of a possible CMC station, and thus GHCN-D has sufficient
independent information sources to evaluate the CMC product.
Radiative transfer calculations
The sensitivity of satellite observations of NO2 to its vertical
distribution is calculated here using the LIDORT radiative transfer model
(Spurr, 2002). The model is used to calculate
scattering weights, which quantify the sensitivity of backscattered solar
radiation to NO2 at different altitudes (Martin et
al., 2002; Palmer et al., 2001). The observation sensitivity to lower
tropospheric NO2 is represented by the AMF. AMFs for OMI satellite observations in January 2013 are calculated as a
useful analog for future TEMPO observations as both instruments are
spectrometers observing reflected sunlight at UV to visible wavelengths.
AMFs are calculated at 440 nm, at the centre of the NO2 retrieval
window for OMI and TEMPO where NO2 has strong absorption features.
Vertical NO2 profiles, as well as other trace gas and aerosol profiles needed
for the AMF calculation shown here, are obtained from a simulation of the
GEOS-Chem chemical transport model version 11-01 (www.geos-chem.org).
Figure 1 shows maps of snow-free and snow-covered reflectances used here.
Snow-free surface reflectance at 470 nm is provided by Nadir BRDF-Adjusted Reflectances
from the MODIS CMG Gap-Filled Snow-Free Products
(Sun et al., 2017). Reflectivities at 354 nm for
snow-covered scenes are derived from OMI observations as described by
O'Byrne et al. (2010). This data set is consistent
with previous snow reflectivity
(e.g.
Moody et al., 2007; Tanskanen and Manninen, 2007) over most land types
(O'Byrne et al., 2010). Snow-covered reflectivity has
an estimated uncertainty of 10–20 % in most regions, with higher
uncertainties in regions with thin or transient snow. Although the 354 nm
wavelength is different than the 440 nm wavelength used to calculate AMFs,
snow reflectivity has weak spectral dependence in UV–visible wavelengths
(Feister and Grewe, 1995; O'Byrne et al., 2010).
Snow can increase surface reflectance by over a factor of 10 in central
North America where short vegetation is readily covered by snow.
Surface reflectivity at UV–visible wavelengths for snow-covered
and snow-free conditions for January 2013. White space in panel (a)
indicates that no snow reflectance information is available.
Methods
Here we test daily snow cover products for 2015. Snow products are regridded
from their native resolutions to a common 4 km grid (similar to the spatial
resolution of TEMPO). A grid box is considered to be snow covered if any
observations within that box are snow covered. MAIAC, NISE, and IMS give
only a yes or no flag for presence of snow. MODIS products provide a pixel snow
fraction, and we consider any pixels with nonzero snow fractions as snow
covered. Any CMC grid box with nonzero snow depth is considered snow
covered.
GHCN-D surface measurements are used as the ground “truth” for evaluating
the satellite and reanalysis snow data products tested here. If measurements
from multiple surface data networks exist in the same grid box, the most
reliable source is used per the priority order given by GHCN-D
(Menne et al., 2012b). If observations from
multiple surface stations within the most reliable network within a grid box
disagree on the presence of snow on a given day, that day is excluded from
the evaluation.
We assess the snow data sets using metrics that are commonly used for
evaluating binary data sets (Rittger
et al., 2013). These metrics are based on the possible outcomes for
identifying snow: true positive (TP), true negative (TN), false positive
(FP), and false negative (FN). Accuracy measures the likelihood that a grid
box, with snow or without, is correctly classified:
Accuracy=TP+TNTP+TN+FP+FN.
Precision is the probability that a region identified as snow covered has
snow:
Precision=TPTP+FP.
Recall is the likelihood that snow cover is detected when present:
Recall=TPTP+FN.
The F score balances recall (which accounts for false negatives) and precision
(which accounts for false positives) to measure correct classification of
snow without the influence of frequent snow-free periods, and it is therefore
the metric which is most relevant for TEMPO:
F=2⋅precision⋅recallprecision+recall.
Results
We first examine the effect of surface reflectivity on retrieval sensitivity
by using the LIDORT radiative transfer model to calculate NO2 AMFs for both snow-free and snow-covered scenarios using the
corresponding snow-free (Sun et al., 2017) or
snow-covered (O'Byrne et al., 2010) surface reflectance
over North America. We calculate AMFs over North America in
January 2013. We assume cloud-free conditions in all AMF calculations, as
the impact of surface reflectance on retrieved cloud fractions is beyond the
scope of this paper.
Figure 2 shows the sensitivity of backscattered radiation (scattering
weights) over snow-covered and snow-free surfaces for two locations: a
midlatitude location (US Midwest; 42∘ N, 99∘ W) with a
solar zenith angle of 60∘ and at a high-latitude location
(Northern Canada; 58∘ N, 76∘ W) with a solar zenith
angle of 79∘. The snow-covered scattering weights are greater
than the snow-free scattering weights throughout the troposphere, by factors
of 2.0 (2.7) below 5 km, 2.7 (3.7) below 2 km, and 2.6 (5.3) below 1 km at
the mid- (high-) latitude location. This shows that satellite-observed
backscattered radiation in clear-sky conditions is up to 5 times as
sensitive to NO2 in the boundary layer after accounting for increased
reflection by snow, due to the increased absorption by NO2 in the lower
troposphere when the surface reflects more sunlight.
Observation sensitivity to NO2. Scattering weight profiles
calculated for cloud-free OMI NO2 retrievals, with and without surface
snow cover, for January 2013 at (a) 42∘ N, 99∘ W with
a solar zenith angle (ZA) of 60∘ and (b) 58∘ N,
76∘ W with a solar zenith angle of 79∘.
Figure 3 shows the distribution of AMF values over North America with and
without reflectance from snow. The snow-free AMF distribution is unimodal
with a median of 1.2. Allowing for the presence of snow introduces a second
mode with a median of 3.2. Mean AMFs increase by a factor of 2.0 in the
presence of snow, indicating an overall doubling in the sensitivity to
tropospheric NO2 over snow-covered surfaces across North America. The
impact is larger over polluted regions, as mean AMFs increase by a factor of
2.2 in regions where NO2 columns exceed 1×1015 molec cm-2.
Maps of AMF with and without snow cover for January 2013 show that AMF
values increase over 69 % of the land surface within the TEMPO domain.
We next examine the snow datasets to identify the one most suited for the
TEMPO retrieval algorithm. Figure 4 shows the spatial distribution of false
positives and false negatives in the data sets. In all data sets, both false
positives and negatives are most frequent over mountainous regions,
particularly in the Rocky Mountain region, consistent with previous
validation studies
(Chen
et al., 2012, 2014; Frei et al., 2012; Frei and Lee, 2010). These errors are
often attributed to differences in representativeness, as snow cover in
mountain regions is often spatially inhomogeneous, and thus in situ measurements may
not be representative of the pixel. A slight increase in the number of false
positives in IMS over mid-western and prairie regions may result from crop
regions with high snow-free albedos being mistaken for snow in visible
imagery (Chen et al., 2012; Yang et
al., 2015). NISE, MODIS Aqua, and MODIS Terra have more false negatives
overall, especially in the Great Lakes and New England regions. False
positives are less frequent than false negatives in all data sets. IMS and
CMC have the lowest frequency of false negatives. NISE and MAIAC have the
lowest frequency of false positives.
(a) Distribution of air mass factors (AMFs) calculated for OMI
NO2 retrievals over North America for observation geometry of January
2013, using snow-free (Sun et al., 2017) or snow-covered
(O'Byrne et al., 2010) surface reflectance. (b)
Maps of AMF for snow-covered and snow-free conditions.
Number of false positive (FP) and false negative (FN) snow
attributions by the snow data sets in 2015. All data sets are evaluated at
4 km resolution. Total number of false snow attributions inset. White space
indicates that no ground stations present.
Figure 5 shows the metrics used to evaluate data set performance. Table 1
summarizes these results. All data sets have high accuracy numbers, owing
largely to a high number of true negatives during the summer months. MODIS
Aqua and Terra have low recall and F scores. When only observations with MODIS
cloud fractions less than 20 % are used, MODIS has better agreement with
the ground stations (F statistic increases from 0.38 to 0.49 at native
resolution for Aqua, 0.43 to 0.63 for Terra), but this reduces the
number of usable MODIS observations by up to 60 %. NISE has high precision
but low recall, indicating that, while areas classified as snow-covered by
NISE are likely correct, many snow-covered regions are missing in the data
set. This is consistent with evaluations by McLinden et al. (2014) and
O'Byrne et al. (2010). Although CMC, IMS, and MAIAC products show an
increase in frequency of false negatives over the Rocky Mountains, they
retain a high precision in this region due to frequent snow cover. While
MAIAC Aqua and Terra have high accuracy and precision, lower recall values
indicate that they are conservative in identifying the presence of snow.
This is possibly a consequence of the method used for identifying cloud,
which may incorrectly classify fresh snowfall as cloud
(Lyapustin et al., 2008). Data sets were also evaluated by
season with similar results (Appendix Table A1). All data sets have weaker
performance metrics during the spring melt season, which has been observed
in past evaluations (Frei et al., 2012). IMS has the
highest F score in winter and autumn but is slightly outperformed by MAIAC in
spring. Data sets were also evaluated at their native resolutions and at a
common 25 km resolution (Appendix Tables A2–A3). Results are similar at each
resolution with two exceptions: MODIS Aqua and Terra products perform better
when regridded from their native 0.05∘ resolution to a 4 km
resolution as it reduces the number of grid boxes missing observations due
to cloud, and MAIAC Aqua and Terra perform better at their native resolution
than at either 4 km or 25 km as degrading the spatial resolution results in
a loss of information.
Evaluation of daily snow extent data set performance for 2015.
GHCN-D surface observations are used as “truth”. All products are
regridded to a common 4 km resolution. The highest value for each metric is
shown in bold.
Statistical metrics to evaluate snow cover products. All data sets
are gridded at 4 km resolution. White space indicates that no ground stations
present.
For all data sets, recall is generally low in two regions: along the Pacific
coastline where snow depths are relatively thin and in the south when snow
is rare and generally short lived. Thin snow is likely to be less homogenous
across a pixel and more likely to be obscured by forest canopies or tall
grasses, and thus it is difficult to observe from satellite imagery. Short-lived snow in the south is likely to be missed by satellite observations,
especially since clouds are often present. However, as IMS uses multiple
observations at multiple times of day in addition to incorporating ground
station data, it is more likely to find snow in these cases than other
satellite products (Hall et al., 2010). Overall, IMS has
best agreement with in situ observations, with the highest accuracy, recall, and F
statistic and relatively high precision.
While CMC also has strong performance metrics, it is important to consider
the information source used to describe snow extent in each product.
Products based on satellite observations are advantageous when assessing how
surface reflectivity affects backscattered radiation observed from space.
For example, thin snow, or snow obscured by tree canopies, may not affect
the observed brightness from space, but would be considered snow-covered by
a product based on surface observations (e.g. CMC). Also, the reflectivity
of a snow-covered surface decreases over time as the snow ages
(Warren and Wiscombe, 1980). This effect
would not be captured by snow depth measurements. And while snow depth has
been used as an indicator of brightness (Arola et
al., 2003), it cannot account for snow aging or canopy effects. IMS is
based on visible satellite imagery and thus determines snow extent based on
brightness from space, which is more applicable to satellite retrievals.
Additionally,
while most satellite-based products rely on observations made at a single
overpass time and viewing geometry, IMS has the advantage of incorporating
observations from multiple satellites with differing measurement times and
geometries, including both geostationary and low Earth orbits. These
reasons, in addition to a strong agreement with in situ measurements and
near-real-time updates, make IMS best suited for informing TEMPO retrievals.
We next examine the effect on both spatial sampling and sensitivity to the
lower troposphere of a retrieval data set if observations with surface snow
are included rather than omitted. We use IMS to identify the presence of
snow for OMI observations over North America in January 2015. We then use
LIDORT to calculate AMFs for these observations using the corresponding
snow-free (Sun et al., 2017) or snow-covered
(O'Byrne et al., 2010) surface reflectance and examine
the results of either including or omitting snow-covered scenes. Figure 6
shows that including snow-covered scenes results in a significant (factor of
2.1) increase in observation frequency, particularly in the northern US and
Canada. Additionally, including snow-covered scenes increases the average
AMF by a factor of 2.7 in regions with occasional snow cover. The increase
in AMF demonstrates that including snow-covered scenes increases the quality
of information about the tropospheric NO2 column by increasing the
observation sensitivity to tropospheric NO2. As we assume clear-sky
conditions, these are likely upper bounds on potential increases in
observation quantity and quality. In practice, the presence of clouds and
errors in cloud retrieval algorithms will likely diminish these impacts.
OMI observation frequency (a) and average AMFs (b) over
North America in January using IMS to identify surface snow conditions.
White space indicates a lack of observations.
Conclusions
An accurate representation of snow cover is essential to ensuring satellite
retrieval accuracy, including those from TEMPO. Radiative transfer model
calculations indicate that clear-sky NO2 retrievals over reflective
snow-covered surfaces are more than twice as sensitive to NO2 in the
boundary layer than over snow-free surfaces. This makes snow an attractive
surface over which to observe tropospheric NO2. However, the lack of
confidence in snow identification has previously led many retrieval
procedures to omit observations over snow. We show that increasing this
confidence such that these observations could be included not only improves
spatial and temporal sampling but also allows the inclusion of observations
with higher-quality information on the lower troposphere.
We evaluated seven snow extent data sets to determine their usefulness for
informing satellite retrievals of trace gas from solar backscatter
observations. All products were more likely to misidentify snow over
mountains or where snow cover is thin or short lived. IMS had the best
agreement with in situ observations (F=0.85) and as a satellite-based,
operational, daily updated product, it is well suited for informing TEMPO
satellite retrievals. The low recall value (0.45) for NISE indicated that a
significant number of snow-covered pixels are missed. The standard MODIS
products showed medium precision and low recall owing to cloud
contamination. The MAIAC products had the highest precision (0.90 for both
Aqua and Terra) of those tested, but is conservative in ascribing the
presence of snow (recall of 0.74 for Aqua, 0.75 for Terra). CMC had strong
performance metrics (F=0.81), but as a reanalysis product based on ground
observations it may not appropriately represent how a surface snow
reflectivity would affect TEMPO-observed radiances.
The potential improvements in NO2 retrieval performance over
snow-covered scenes outlined here were tested for clear-sky conditions. The
accuracy of cloud retrieval schemes also impacts the quality of trace gas
retrievals. Many cloud retrieval schemes have difficulty distinguishing
between a bright surface and bright, low-altitude clouds. This may diminish
the impact that improved surface snow reflectance can have on observation
frequency and sensitivity when clouds are present. However, using accurate
surface snow cover information may also lead to corresponding improvements
in cloud retrieval accuracy.
Future work should investigate snow reflectance products that could be used
when snow is detected. This could potentially include bidirectional reflectance distribution functions that describe reflection at
different viewing angles, as this effect has been shown to have significant
impact on retrieved NO2 columns and clouds
(Lorente
et al., 2018; Vasilkov et al., 2017). Accurate knowledge of snow
reflectivity is also needed to improve retrievals over snow. A retrieval
algorithm that combines daily snow detection from IMS with a climatology of
snow reflectance has the potential to greatly improve upon current
methodologies.
IMS (10.7265/N52R3PMC, National Ice Center, 2008), NISE (10.5067/3KB2JPLFPK3R, Brodzik
and Stewart, 2016), MODIS Aqua (10.5067/MODIS/MYD10C1.006, Hall and Riggs, 2016a),
MODIS Terra (10.5067/MODIS/MOD10C1.006, Hall and Riggs, 2016b), and CMC
(Brown and Brasnett, 2010) data are available from the NASA
National Snow and Ice Data Center (http://nsidc.org, last access: 17 July 2017). MAIAC Collection 6
(Lyapustin et al., 2011a, b, 2012) re-processing of MODIS data started in September 2017 and is expected to be
completed by the end of year. This study used MAIAC Atmospheric Properties files currently available
via ftp at the NASA Center for Climate Simulations (NCCS):
ftp://maiac@dataportal.nccs.nasa.gov/DataRelease/ (last access: 15 June 2017). GHCN-D data are available
from the NOAA National Climatic Data Center
(10.7289/V5D21VHZ Menne et al., 2012b). AMF code (Spurr, 2002; Martin et al., 2002) used to calculate scattering weights and air mass
factors, as well as snow-covered Surface LER (O'Byrne et al., 2010) used here, is available at
http://fizz.phys.dal.ca/~atmos (last access: 19 June 2017) MODIS MCD43GF CMG Gap-Filled. Snow-free surface
reflectances (Sun et al., 2017) are available at ftp://rsftp.eeos.umb.edu/data02/Gapfilled/ (Sun et al., 2017).
The GEOS-Chem chemical transport model used here is available at
www.geos-chem.org (last access: 15 June 2017).
Evaluation of daily snow extent data set performance by season for
2015. GHCN-D surface observations are used as “truth”. All products are
regridded to a common 4 km resolution. The highest value for each
metric/season is shown in bold.
Evaluation of daily snow extent data set performance for 2015.
GHCN-D surface observations are used as “truth”. The highest value for
each metric is shown in bold.
Evaluation of daily snow extent data set performance for 2015.
GHCN-D surface observations are used as “truth”. All products are
regridded to a common 25 km resolution. The highest value for each metric is
shown in bold.
The authors declare that they have no conflict of
interest.
Edited by: Folkert Boersma
Reviewed by: two anonymous referees
ReferencesArola, A., Kaurola, J., Koskinen, L., Tanskanen, A., Tikkanen, T., Taalas,
P., Herman, J. R., Krotkov, N., and Fioletov, V.: A new approach to
estimating the albedo for snow-covered surfaces in the satellite UV method,
J. Geophys. Res., 108, 4531, 10.1029/2003JD003492, 2003.Beirle, S., Boersma, K. F., Platt, U., Lawrence, M. G., and Wagner, T.:
Megacity emissions and lifetimes of nitrogen oxides probed from space,
Science, 333, 1737–1739, 10.1126/science.1207824, 2011.Boersma, K. F., Eskes, H. J., and Brinksma, E. J.: Error analysis for
tropospheric NO2 retrieval from space, J. Geophys. Res.-Atmos., 109,
D04311, 10.1029/2003JD003962, 2004.Brasnett, B.: A Global Analysis of Snow Depth for Numerical Weather
Prediction, J. Appl. Meteorol., 38, 726–740,
10.1175/1520-0450(1999)038<0726:AGAOSD>2.0.CO;2,
1999.Brodzik, M. J. and Stewart, J. S.: Near-Real-Time SSM/I-SSMIS EASE-Grid
Daily Global Ice Concentration and Snow Extent, Version 5,
10.5067/3KB2JPLFPK3R, 2016.Brown, R. D. and Brasnett, B.: Canadian Meteorological Centre (CMC) Daily
Snow Depth Analysis Data, Version 1,
10.5067/W9FOYWH0EQZ3, 2010.Brown, R. D., Brasnett, B., and Robinson, D.: Gridded North American monthly
snow depth and snow water equivalent for GCM evaluation, Atmos.-Ocean,
41, 1–14, 10.3137/ao.410101, 2003.Chen, C., Lakhankar, T., Romanov, P., Helfrich, S., Powell, A., and
Khanbilvardi, R.: Validation of NOAA-Interactive Multisensor Snow and Ice
Mapping System (IMS) by Comparison with Ground-Based Measurements over
Continental United States, Remote Sens., 4, 1134–1145,
10.3390/rs4051134, 2012.Chen, X., Jiang, L., Yang, J., and Pan, J.: Validation of ice mapping system
snow cover over southern China based on Landsat Enhanced Thematic Mapper
Plus imagery, J. Appl. Remote Sens., 8, 84680,
10.1117/1.JRS.8.084680, 2014.Duncan, B. N., Prados, A. I., Lamsal, L. N., Liu, Y., Streets, D. G., Gupta,
P., Hilsenrath, E., Kahn, R. A., Nielsen, J. E., Beyersdorf, A. J., Burton,
S. P., Fiore, A. M., Fishman, J., Henze, D. K., Hostetler, C. A., Krotkov,
N. A., Lee, P., Lin, M., Pawson, S., Pfister, G., Pickering, K. E., Pierce,
R. B., Yoshida, Y., and Ziemba, L. D.: Satellite data of atmospheric
pollution for U.S. air quality applications: Examples of applications,
summary of data end-user resources, answers to FAQs, and common mistakes to
avoid, Atmos. Environ., 94, 647–662, 10.1016/j.atmosenv.2014.05.061,
2014.Feister, U. and Grewe, R.: Spectral albedo measurements in the UV and
visible region over different types of surfaces, Photochem. Photobiol.,
62, 736–744, 10.1111/j.1751-1097.1995.tb08723.x, 1995.Fioletov, V. E., McLinden, C. A., Krotkov, N., and Li, C.: Lifetimes and
emissions of SO2 from point sources estimated from OMI, Geophys. Res.
Lett., 42, 1969–1976, 10.1002/2015GL063148, 2015.Fishman, J., Al-Saadi, J. A., Creilson, J. K., Bowman, K. W., Burrows, J.
P., Richter, A., Chance, K. V., Edwards, D. P., Martin, R. V., Morris, G.
A., Pierce, R. B., Ziemke, J. R., Schaack, T. K., Thompson, A. M., Fishman,
J., Al-Saadi, J. A., Creilson, J. K., Bowman, K. W., Burrows, J. P.,
Richter, A., Chance, K. V., Edwards, D. P., Martin, R. V., Morris, G. A.,
Pierce, R. B., Ziemke, J. R., Schaack, T. K., and Thompson, A. M.: Remote
Sensing of Tropospheric Pollution from Space, B. Am. Meteorol. Soc.,
89, 805–821, 10.1175/2008BAMS2526.1, 2008.
de Foy, B., Lu, Z., Streets, D. G., Lamsal, L. N., and Duncan, B. N.:
Estimates of power plant NO x emissions and lifetimes from OMI NO 2
satellite retrievals, Atmos. Environ., 116, 1–11, 2015.Frei, A. and Lee, S.: A comparison of optical-band based snow extent
products during spring over North America, Remote Sens. Environ., 114,
1940–1948, 10.1016/j.rse.2010.03.015, 2010.Frei, A., Tedesco, M., Lee, S., Foster, J., Hall, D. K., Kelly, R., and
Robinson, D. A.: A review of global satellite-derived snow products, Adv.
Space Res., 50, 1007–1029, 10.1016/j.asr.2011.12.021, 2012.Geddes, J. A. and Martin, R. V.: Global deposition of total reactive nitrogen
oxides from 1996 to 2014 constrained with satellite observations of NO2
columns, Atmos. Chem. Phys., 17, 10071–10091,
10.5194/acp-17-10071-2017, 2017.Hall, D. K. and Riggs, G. A.: Accuracy assessment of the MODIS snow products,
Hydrol. Process., 21, 1534–1547, 10.1002/hyp.6715, 2007.Hall, D. and Riggs, G. A.: MODIS/Aqua Snow Cover Daily L3 Global 0.05Deg CMG,
Version 6, 10.5067/MODIS/MYD10C1.006, 2016a.Hall, D. K. and Riggs, G. A.: MODIS/Terra Snow Cover Daily L3 Global 0.05Deg
CMG, Version 6, 10.5067/MODIS/MOD10C1.006, 2016b.Hall, D. K., Fuhrmann, C. M., Perry, L. B., Riggs, G. A., Robinson, D.
A., and Foster, J. L.: A Comparison of Satellite-Derived Snow Maps with a
Focus on Ephemeral Snow in North Carolina, 67th Eastern Snow Conference,
Hancock MA USA, 2010.Helfrich, S. R., McNamara, D., Ramsay, B. H., Baldwin, T., and Kasheta, T.:
Enhancements to, and forthcoming developments in the Interactive Multisensor
Snow and Ice Mapping System (IMS), Hydrol. Process., 21, 1576–1586,
10.1002/hyp.6720, 2007.Herman, J. R. and Celarier, E. A.: Earth surface reflectivity climatology at
340-380 nm from TOMS data, J. Geophys. Res.-Atmos., 102, 28003–28011,
10.1029/97JD02074, 1997.Kleipool, Q. L., Dobber, M. R., de Haan, J. F., and Levelt, P. F.: Earth
surface reflectance climatology from 3 years of OMI data, J. Geophys. Res.,
113, D18308, 10.1029/2008JD010290, 2008.Koelemeijer, R. B. A., de Haan, J. F., and Stammes, P.: A database of
spectral surface reflectivity in the range 335–772 nm derived from 5.5
years of GOME observations, J. Geophys. Res., 108, 4070,
10.1029/2002JD002429, 2003.Lamsal, L. N., Janz, S. J., Krotkov, N. A., Pickering, K. E., Spurr, R. J.
D., Kowalewski, M. G., Loughner, C. P., Crawford, J. H., Swartz, W. H., and
Herman, J. R.: High-resolution NO2 observations from the Airborne Compact
Atmospheric Mapper: Retrieval and validation, J. Geophys. Res.-Atmos., 122,
1953–1970, 10.1002/2016JD025483, 2017.Liang, S., Fang, H., Chen, M., Shuey, C. J., Walthall, C., Daughtry, C.,
Morisette, J., Schaaf, C., and Strahler, A.: Validating MODIS land surface
reflectance and albedo products: methods and preliminary results, Remote
Sens. Environ., 83, 149–162, 10.1016/S0034-4257(02)00092-5, 2002.Lin, J.-T., Liu, M.-Y., Xin, J.-Y., Boersma, K. F., Spurr, R., Martin, R.,
and Zhang, Q.: Influence of aerosols and surface reflectance on satellite
NO2 retrieval: seasonal and spatial characteristics and implications for
NOx emission constraints, Atmos. Chem. Phys., 15, 11217–11241,
10.5194/acp-15-11217-2015, 2015.Lorente, A., Folkert Boersma, K., Yu, H., Dörner, S., Hilboll, A., Richter,
A., Liu, M., Lamsal, L. N., Barkley, M., De Smedt, I., Van Roozendael, M.,
Wang, Y., Wagner, T., Beirle, S., Lin, J.-T., Krotkov, N., Stammes, P., Wang,
P., Eskes, H. J., and Krol, M.: Structural uncertainty in air mass factor
calculation for NO2 and HCHO satellite retrievals, Atmos. Meas. Tech., 10,
759–782, 10.5194/amt-10-759-2017, 2017.Lorente, A., Boersma, K. F., Stammes, P., Tilstra, L. G., Richter, A., Yu,
H., Kharbouche, S., and Muller, J.-P.: The importance of surface reflectance
anisotropy for cloud and NO2 retrievals from GOME-2 and OMI, Atmos. Meas.
Tech. Discuss., 10.5194/amt-2018-32, in review, 2018.Lyapustin, A., Wang, Y., and Frey, R.: An automatic cloud mask algorithm
based on time series of MODIS measurements, J. Geophys. Res., 113, D16207,
10.1029/2007JD009641, 2008.Lyapustin, A., Martonchik, J., Wang, Y., Laszlo, I., and Korkin, S.:
Multiangle implementation of atmospheric correction (MAIAC): 1. Radiative
transfer basis and look-up tables, J. Geophys. Res., 116, D03210,
10.1029/2010JD014985, 2011a.Lyapustin, A., Wang, Y., Laszlo, I., Kahn, R., Korkin, S., Remer, L., Levy,
R., and Reid, J. S.: Multiangle implementation of atmospheric correction
(MAIAC): 2. Aerosol algorithm, J. Geophys. Res., 116, D03211,
10.1029/2010JD014986, 2011b.Lyapustin, A. I., Wang, Y., Laszlo, I., Hilker, T., G.Hall, F., Sellers, P.
J., Tucker, C. J., and Korkin, S. V.: Multi-angle implementation of
atmospheric correction for MODIS (MAIAC): 3. Atmospheric correction, Remote
Sens. Environ., 127, 385–393, 10.1016/j.rse.2012.09.002, 2012.Martin, R. V.: Satellite remote sensing of surface air quality, Atmos.
Environ., 42, 7823–7843, 10.1016/j.atmosenv.2008.07.018, 2008.Martin, R. V, Chance, K., Jacob, D. J., Kurosu, T. P., Spurr, R. J. D.,
Bucsela, E., Gleason, J. F., Palmer, P. I., Bey, I., and Fiore, A. M.: An
improved retrieval of tropospheric nitrogen dioxide from GOME, J. Geophys.
Res.-Atmos., 107, 4437, 10.1029/2001JD001027, 2002.McLinden, C. A., Fioletov, V., Boersma, K. F., Kharol, S. K., Krotkov, N.,
Lamsal, L., Makar, P. A., Martin, R. V., Veefkind, J. P., and Yang, K.:
Improved satellite retrievals of NO2 and SO2 over the Canadian oil
sands and comparisons with surface measurements, Atmos. Chem. Phys., 14,
3637–3656, 10.5194/acp-14-3637-2014, 2014.Menne, M. J., Durre, I., Vose, R. S., Gleason, B. E., Houston, T. G., Menne,
M. J., Durre, I., Vose, R. S., Gleason, B. E., and Houston, T. G.: An
Overview of the Global Historical Climatology Network-Daily Database, J.
Atmos. Ocean. Technol., 29, 897–910, 10.1175/JTECH-D-11-00103.1, 2012a.Menne, M. J., Durre, I., Korzeniewski, B., McNeal, S., Thomas, K., Yin, X.,
Anthony, S., Ray, R., Vose, R. S., Gleason, B. E., and Houston, T. G.: Global
Historical Climatology Network – Daily (GHCN-Daily), Version 3.22,
10.7289/V5D21VHZ, 2012b.Moody, E. G., King, M. D., Schaaf, C. B., Hall, D. K., and Platnick, S.:
Northern Hemisphere five-year average (2000–2004) spectral albedos of
surfaces in the presence of snow: Statistics computed from Terra MODIS land
products, Remote Sens. Environ., 111, 337–345,
10.1016/j.rse.2007.03.026, 2007.National Ice Center: IMS Daily Northern Hemisphere Snow and Ice Analysis at
1 km, 4 km, and 24 km Resolutions, Version 1,
10.7265/N52R3PMC, 2008.
Nolin, A., Armstrong, R., and Maslanik, J.: Near real-time SSM/I EASE-grid
daily global ice concentration and snow extent, Digit. Media, Natl. Snow Ice
Data Center, Boulder, CO, USA, 2005.Nowlan, C. R., Martin, R. V., Philip, S., Lamsal, L. N., Krotkov, N. A.,
Marais, E. A., Wang, S., and Zhang, Q.: Global dry deposition of nitrogen
dioxide and sulfur dioxide inferred from space-based measurements, Global
Biogeochem. Cy., 28, 1025–1043, 10.1002/2014GB004805, 2014.O'Byrne, G., Martin, R. V., van Donkelaar, A., Joiner, J., and Celarier, E.
A.: Surface reflectivity from the Ozone Monitoring Instrument using the
Moderate Resolution Imaging Spectroradiometer to eliminate clouds: Effects of
snow on ultraviolet and visible trace gas retrievals, J. Geophys. Res., 115,
D17305, 10.1029/2009JD013079, 2010.Palmer, P. I., Jacob, D. J., Chance, K., and Martin, R. V.: Air mass factor
formulation for spectroscopic measurements from satellites' Application to
formaldehyde retrievals from the Global Ozone Monitoring Experiment, J.
Geophys. Res., 106, 14539–14550, 10.1029/2000JD900772, 2001.
Richter, A. and Wagner, T.: The Use of UV, Visible and Near IR Solar Back
Scattered Radiation to Determine Trace Gases, 67–121, Springer, Berlin,
Heidelberg, 2011.Rittger, K., Painter, T. H., and Dozier, J.: Assessment of methods for
mapping snow cover from MODIS, Adv. Water Resour., 51, 367–380,
10.1016/j.advwatres.2012.03.002, 2013.
Spurr, R. J. D.: Simultaneous derivation of intensities and weighting
functions in a general pseudo-spherical discrete ordinate radiative transfer
treatment, J. Quant. Spectrosc. Ra., 75, 129–175, 2002.
Streets, D. G., Canty, T., Carmichael, G. R., de Foy, B., Dickerson, R. R.,
Duncan, B. N., Edwards, D. P., Haynes, J. A., Henze, D. K., and Houyoux, M.
R.: Emissions estimation from satellite retrievals: A review of current
capability, Atmos. Environ., 77, 1011–1042, 2013.Sun, Q., Wang, Z., Li, Z., Erb, A., and Schaaf, C. B.: Evaluation of the
global MODIS 30 arc-second spatially and temporally complete snow-free land
surface albedo and reflectance anisotropy dataset, Int. J. Appl. Earth Obs.
Geoinf., 58, 36–49, 10.1016/j.jag.2017.01.011, 2017.Tanskanen, A. and Manninen, T.: Effective UV surface albedo of seasonally
snow-covered lands, Atmos. Chem. Phys., 7, 2759–2764,
10.5194/acp-7-2759-2007, 2007.Valin, L. C., Russell, A. R., and Cohen, R. C.: Variations of OH radical in
an urban plume inferred from NO2 column measurements, Geophys. Res. Lett.,
40, 1856–1860, 2013.Vasilkov, A., Qin, W., Krotkov, N., Lamsal, L., Spurr, R., Haffner, D.,
Joiner, J., Yang, E.-S., and Marchenko, S.: Accounting for the effects of
surface BRDF on satellite cloud and trace-gas retrievals: a new approach
based on geometry-dependent Lambertian equivalent reflectivity applied to OMI
algorithms, Atmos. Meas. Tech., 10, 333–349,
10.5194/amt-10-333-2017, 2017.Warren, S. G. and Wiscombe, W. J.: A Model for the Spectral Albedo of Snow.
II: Snow Containing Atmospheric Aerosols, J. Atmos. Sci., 37, 2734–2745,
10.1175/1520-0469(1980)037<2734:AMFTSA>2.0.CO;2, 1980.Yang, J., Jiang, L., Ménard, C. B., Luojus, K., Lemmetyinen, J., and
Pulliainen, J.: Evaluation of snow products over the Tibetan Plateau, Hydrol.
Process., 29, 3247–3260, 10.1002/hyp.10427, 2015.
Zoogman, P., Liu, X., Suleiman, R. M., Pennington, W. F., Flittner, D. E.,
Al-Saadi, J. A., Hilton, B. B., Nicks, D. K., Newchurch, M. J., Carr, J. L.,
Janz, S. J., Andraschko, M. R., Arola, A., Baker, B. D., Canova, B. P., Chan
Miller, C., Cohen, R. C., Davis, J. E., Dussault, M. E., Edwards, D. P.,
Fishman, J., Ghulam, A., González Abad, G., Grutter, M., Herman, J. R.,
Houck, J., Jacob, D. J., Joiner, J., Kerridge, B. J., Kim, J., Krotkov, N.
A., Lamsal, L., Li, C., Lindfors, A., Martin, R. V., McElroy, C. T.,
McLinden, C., Natraj, V., Neil, D. O., Nowlan, C. R., O'Sullivan, E. J.,
Palmer, P. I., Pierce, R. B., Pippin, M. R., Saiz-Lopez, A., Spurr, R. J. D.,
Szykman, J. J., Torres, O., Veefkind, J. P., Veihelmann, B., Wang, H., Wang,
J., and Chance, K.: Tropospheric emissions: Monitoring of pollution (TEMPO),
J. Quant. Spectrosc. Ra., 186, 17–39, 10.1016/j.jqsrt.2016.05.008, 2017.