Cloud retrievals from the Moderate Resolution Imaging
Spectroradiometer (MODIS) instruments aboard the satellites Terra and Aqua
and the Visible Infrared Imaging Radiometer Suite (VIIRS) instrument aboard
the Suomi-NPP satellite are evaluated using a combination of ground-based
instruments providing vertical profiles of clouds. The ground-based
measurements are obtained from the Atmospheric Radiation Measurement (ARM)
programme mobile facility, which was deployed in Hyytiälä, Finland,
between February and September 2014 for the Biogenic Aerosols – Effects on
Clouds and Climate (BAECC) campaign. The satellite cloud parameters cloud
top height (CTH) and liquid water path (LWP) are compared with ground-based
CTH obtained from a cloud mask created using lidar and radar data and LWP
acquired from a multi-channel microwave radiometer. Clouds from all
altitudes in the atmosphere are investigated. The clouds are diagnosed as
single or multiple layer using the ground-based cloud mask. For single-layer
clouds, satellites overestimated CTH by 326 m (14 %) on average. When
including multilayer clouds, satellites underestimated CTH by on average
169 m (5.8 %). MODIS collection 6 overestimated LWP by on average
13 g m
Clouds are a very important component of the Earth's energy budget since they contribute to a large fraction of both the reflected shortwave radiation and absorbed longwave radiation. The magnitude and sign of the cloud impact depends on the cloud altitude, and a correct representation of the cloud distribution in the vertical is crucial to obtain a good estimate of the Earth's energy budget. One of the largest uncertainties in the global climate models used to predict the future climate is the representation of clouds, their feedbacks and their interaction with short- and longwave radiation (Dolinar et al., 2015; IPCC, 2013). Satellite cloud retrievals provide cloud distributions on a global scale, which are used to assess global climate models (e.g. Dolinar et al., 2015). It is therefore of great importance to evaluate the satellite-retrieved cloud properties and investigate whether they provide an accurate representation of the cloud fields.
The Moderate Resolution Imaging Spectroradiometer (MODIS) instrument is carried aboard the satellites Terra and Aqua, providing information on clouds (and many other terrestrial and atmospheric properties) since 2000 and 2002, respectively. Terra and Aqua are polar-orbiting with each MODIS instrument providing an image of the whole globe every 2 days (Platnick et al., 2003). A new collection (number 6) of MODIS Level 2 products was released in 2014. Several updates to the cloud product were implemented for collection 6 (hereafter C6) compared to the previous collection 5.1 (hereafter C5.1). The cloud top properties are now provided at 1 km spatial resolution along with new products, such as cloud top height (CTH) (Baum et al., 2012). Furthermore, the cloud optical properties have been modified, including updates of the radiative transfer model and look-up tables (Platnick et al., 2014) and the thermodynamic phase retrievals (Baum et al., 2012).
The Visible Infrared Imaging Radiometer Suite (VIIRS) is carried aboard the Suomi-NPP satellite, which has been in orbit since October 2011 (Cao et al., 2013). Suomi-NPP is also a polar-orbiting satellite, and the VIIRS sensor is similar to the MODIS sensor but has higher spatial resolution in the infrared bands used for cloud height retrievals. However, VIIRS has fewer bands available for CTH retrievals. Cloud products are also available from the VIIRS sensor.
In 1989 the U.S. Department of Energy initiated the Atmospheric Radiation Measurement (ARM) programme with the purpose of providing ground-based measurements of clouds, and later of aerosols and precipitation (Ackerman and Stokes, 2003). Several long-term measurement stations were implemented containing a comprehensive suite of in situ and passive and active remote sensing instruments. These long-term stations were later complemented with three mobile facilities and one aerial facility (Mather and Voyles, 2012).
ARM data have previously been used to validate satellite retrievals. Mace et al. (2005) evaluated cirrus retrievals from MODIS and Clouds and the Earth's Radiant Energy System (CERES) using ARM data from the Southern Great Plains Site (SGP). Data from this site were also used by Dong et al. (2008), who compared ARM low-level cloud properties with CERES-MODIS (CM) retrievals. One ARM mobile facility (AMF) was deployed in the Azores (Atlantic Ocean) for 18 months, and these data were compared to CM data to validate the boundary-layer cloud retrievals (Xi et al., 2014). Other researchers have also used ground-based measurements to evaluate MODIS (Liu et al., 2013) and CM (Yan et al., 2015) cloud property retrievals over China. MODIS cloud properties have also been evaluated with other satellite instruments such as Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) (Holz et al., 2008) and Multi-angle Imaging Spectroradiometer (MISR) (Naud et al., 2002).
This study uses ARM data from the AMF deployment in Hyytiälä, Finland, from February 2014 to September 2014 during the Biogenic Aerosols – Effects on Clouds and Climate, (BAECC) campaign (Petäjä, 2013; Petäjä et al., 2016). The AMF data are used to evaluate the CTH from MODIS and VIIRS. The liquid water path (LWP) from MODIS C5.1 and C6 is also assessed to quantify the improvement of the updated C6 product. The investigation is not restricted to any particular cloud type but rather includes clouds from all altitudes in the atmosphere. Because the measurement site is located at a relatively high latitude, the cloud parameters can be investigated both at high and moderately high solar zenith angles (SZAs). This is useful since satellite cloud retrievals have previously been found to be affected by SZA (Vant-Hull et al., 2007; Grosvenor and Wood, 2014). This study does not provide a complete validation of the MODIS cloud properties investigated, but rather provides insights into the performance of the satellite cloud retrievals. These insights can be used together with previous and future studies to improve satellite representation of cloud fields.
Passive instruments on orbiting satellites have a much wider field of view but lower temporal resolution than most ground-based measurements. Therefore, care must be taken when matching satellite and ground-based measurements to perform an inter-comparison at a given location. Here, 1 h averaged ground-based data centred at the satellite overpass time have been matched against satellite pixels whose centre is at maximum 15 km away from the Hyytiälä measurement station, essentially creating a circle with a diameter of 30 km around the station. Similar averaging times and areas have been used in several previous studies (Cess et al., 1996; Dong et al., 2008; Xi et al., 2014; Yan et al., 2015).
The MODIS instrument is carried aboard the polar-orbiting satellites Terra and Aqua. Terra was launched in 1999, Aqua in 2002, and both satellites are sun-synchronous with Terra in a descending orbit (equatorial crossing 10:30 local solar time) and Aqua in an ascending orbit (equatorial crossing 13:30 local solar time). MODIS is a whisk-broom scanning radiometer that scans the entire Earth every 2 days (Platnick et al., 2003). The visible and infrared spectrum is covered by 36 bands which have spatial resolution of 250 m (2 bands), 500 m (5 bands) and 1000 m (9 bands) at nadir. The MODIS data are open access and provided as calibrated data from the wavelength bands (level 1), instantaneous geophysical products (level 2) as well as spatially and temporally averaged geophysical products (level 3).
CTH is a new C6 level 2 cloud product produced from MODIS level 1 data.
Cloud top pressure (CTP) and temperature were provided in C5.1, at 5 km
spatial resolution. In C6 the spatial resolution of the cloud top properties
has been increased to 1 km, but they are also still available at 5 km resolution.
The 1 km resolution CTH will be used in this study. For high- and mid-level
clouds, CTP is retrieved from 4 spectral bands within the 15
If the CO
The other parameter compared in this study, LWP, is available in both C6 and
C5.1 cloud products at 1 km spatial resolution. LWP is derived from two
other cloud products, the cloud optical thickness (COT) and the effective
radius (
VIIRS Suomi-NPP is a scanning radiometer flying aboard Suomi-NPP, a satellite in a sun-synchronous ascending orbit crossing the Equator at 13:30 local time. VIIRS has 16 M-bands with a nadir resolution of 750 m and 6 I-bands with a resolution of 375 m at nadir. It has channels both in the infrared and the visible region of the electromagnetic spectrum.
The level 1 data used in this investigation were produced at SMHI (Swedish
Meteorological and Hydrological Institute) from local reception of VIIRS
data. The level 2 cloud products used here were produced with the Polar
Platform System (PPS) software version 2014
Level 1 data were only received by SMHI until the beginning of May 2014, and hence the VIIRS data are only available during the first 3 months of the investigation.
From February to September 2014, AMF2 was deployed at the Station for Measuring Ecosystem – Atmosphere Relations (SMEAR) II station (Hari and Kulmala, 2005) in Hyytiälä, Finland. The deployment was part of a campaign called the Biogenic Aerosols Effects on Clouds and Climate (BAECC) (Petäjä, 2013; Petäjä et al., 2016), a collaboration between University of Helsinki, Finnish Meteorological Institute, University of Eastern Finland and ARM. AMF2 contains a comprehensive suite of ground-based in situ instrumentation together with active and passive remote sensing instruments to obtain numerous atmospheric properties with very high temporal and spatial resolution.
CTH is provided by the cloud mask created from a combination of the 35 GHz Ka-band ARM Zenith-pointing cloud Radar (KAZR) and the micropulse lidar or ceilometer. Gaps in the operation of the KAZR instrument were supplemented by the 95 GHz Marine W-Band ARM Cloud Radar (MWACR). The data from these instruments were processed using the Cloudnet scheme (Illingworth et al., 2007), which diagnoses the atmospheric targets (such as aerosol, cloud, or precipitation) together with their phase if appropriate. CTH is then obtained directly as the highest cloud pixels diagnosed by this target classification. The nominal vertical and temporal resolution of CTH provided from this scheme is 30 m and 30 s.
The minimum reflectivity at 10 km is about
LWP is obtained from the Radiometrics microwave radiometer, MWR, a
vertically pointing passive instrument measuring the microwave atmospheric
BT at 23.8 and 31.4 GHz. A radiative transfer model with monthly regression
coefficients (Liljegren, 1999) is used to obtain column-integrated
water vapour and liquid water amounts. When properly calibrated, BTs are
obtained with an absolute accuracy better than 0.5 K
(Maschwitz et al., 2013), which corresponds to an LWP
uncertainty of about 20 g m
For the satellite scenes to be included in the study, the CTH retrievals had to be successful for at least 50 % of the pixels inside a 30 km circle around the measurement station. To ensure a fair comparison, homogeneity was considered and only cases where at least 90 % of all pixels were within 1000 m of the median height were included in the analysis. If these criteria were met, the geometrical average height for the clouds within the circle was calculated as this type of averaging was most suitable.
The LWP comparison was performed where satellite CTH was successfully retrieved for more than 50 % of the pixels inside the circle. Furthermore, the multilayer cloud product was used to remove pixels determined to contain several layers of clouds. Moreover, only pixels determined to contain liquid clouds by the satellite were included in the comparison since the ground-based ARM microwave radiometer measures the LWP only for liquid clouds. Only the cases where 50 % of the pixels in the circle passed every step of the screening process were included in the LWP analysis.
For ground-based data, 1 h centred on the satellite overpass time was selected for satellite overpasses that met the selection criteria above. Similar criteria for CTH homogeneity were applied: at least 90 % of the CTH values were within 1000 m of the median height, and more than 50 % of the profiles contained clouds. Again, a geometrical average CTH was calculated for cases that passed the screening. The scenes were investigated for multiple cloud layers (distinct layers separated by more than 500 m) and the fraction of multilayer clouds for each case calculated. Considering the LWP comparison, multiple cloud layers were removed from the analysis, as was performed for the satellite LWP. At least 50 % of the pixels had to contain single-layer cloud only, for a geometrical average LWP to be calculated.
Scatterplots of satellite cloud top heights (CTH) versus ARM cloud
top heights for nighttime cases for
Median and mean differences, standard deviations of the
differences, correlation coefficients and
There were 871 (Aqua) and 869 (Terra) satellite overpasses at Hyytiälä during the campaign. Of these, 322 (Aqua) and 264 (Terra) passed the selection criteria for the satellite scene CTH screening. From these scenes, 181 (Aqua) and 162 (Terra) passed the corresponding ground-based selection criteria for inclusion in the final analysis. The number of LWP cases that passed both satellite and ground-based selection is smaller since only daytime satellite data are used. There were fewer data available for the VIIRS intercomparison, with a total of 300 potential Hyytiälä overpasses, of which 127 passed the satellite CTH screening with 52 cases also meeting the ground-based selection criteria. There were not enough VIIRS cases that passed the LWP screening to enable an intercomparison because there were only data available during the winter months of the campaign; at high latitudes this precludes the use of visible/near-IR satellite retrievals since there is not enough light. Furthermore, during the period when there was sufficient light, most of the clouds in the VIIRS satellite scenes were classified as ice and no LWP was retrieved.
The CTH retrievals are compared separately for daytime and nighttime
conditions, where daytime conditions are defined as those that have a low
enough SZA for the optical properties to be retrieved. For the MODIS
retrievals, the maximum SZA for optical properties is 81.4
The CTH intercomparison during nighttime conditions is presented in Fig. 1. For both the MODIS and VIIRS datasets, different markers are used depending on the CTH retrieval that was dominant (> 50 % of the pixels) for the case. The datasets are also divided according to whether multilayer cloud fraction in the ARM data is smaller than 5 % (single-layer case) or more than 5 % (multilayered case). Since the ARM measurements do not cover the entire satellite scene there may still be multilayered clouds present in parts of the satellite scene. The statistics in Table 1 are calculated for the single-layered scenes separately and the whole dataset together. Both the median and mean differences are reported in the tables for completeness. However, only the medians will be discussed since the differences were generally not normally distributed and often contained outliers which significantly affected the means. Moreover, cases where more than 50 % of the pixels are classified as ice have a cross drawn behind them (Fig. 1).
The same as Fig. 1 except for daytime cases.
Same as in Table 1 except for daytime cases.
Most high-level cloud scenes contain multilayered clouds (Fig. 1). Between 39 and 57 % of the nighttime cases are classified as containing single-layer clouds (Table 1). For these, the median differences are positive between 309 and 407 m (10.5–16.3 %) indicating a satellite overestimate of CTH relative to the ground-based data. When multilayered cloud cases are included, the median difference decreases for all three datasets and becomes negative for Aqua and VIIRS (Table 1). There are several plausible causes for the decrease in the median differences. One is that many high cloud cases are classified as multilayer clouds, and CTHs for high-level clouds are often underestimated by satellites (Holz et al., 2008). Another explanation is that the satellite retrievals underestimate CTH when several layers of clouds are present. The satellite retrievals obtain CTH from the cloud radiating height, which corresponds to a height below the CTH, at least for optically thin clouds. This is known and corrected for, but this procedure becomes more problematic when several cloud layers are present, which may have different optical thicknesses.
Figure 1 also shows which retrieval method is selected for each cloud type.
For MODIS, the CO
Median and mean differences, standard deviations of the
differences, correlation coefficients and
Same as in Table 3 except for daytime cases.
The results for the daytime CTH comparison are displayed in Fig. 2 and
reported in Table 2. Approximately 50 % of the cases are single-layer
clouds for all three datasets and, similar to nighttime cases, very few high
clouds are defined as single-layer clouds. The Aqua (Terra) median
difference between the MODIS and ground-based CTH is 358 m (241 m) for the
single-layered clouds only and is reduced to
The daytime results show a similar percentage to nighttime in retrieval
method selection, except that now Terra and Aqua have a similar percentage
of cases where the CO
Since large errors in the CTH retrievals are a greater issue for low-level clouds than high-level clouds, we re-performed the analysis for all cases and divided them according to CTH measured from the ground with a limit between high and low clouds at 6000 m. The results are shown in Table 3 (nighttime cases) and Table 4 (daytime cases). For the low-level clouds, the CTH is overestimated by between 129 and 468 m, except for the Terra daytime cases for which the CTH is underestimated by 179 m. The CTH for the high-level clouds is underestimated for all the datasets by between 1290 and 2610 m. Thus, the differences between the satellite and ground-based measurements of CTH are lower for the low-level clouds than the high-level clouds. Moreover, the CTH is generally overestimated for low-level clouds and underestimated for high-level clouds, which may help explain the CTH results regarding single- and multilayer clouds.
There are, to our knowledge, no prior studies evaluating MODIS C6 CTH, but
previous studies have investigated the performance of the earlier
collections. Collection 4 (C4) CTP has been combined with ECMWF operational
analysis pressure profiles and compared to ground-based radar
(Naud et al., 2005) and lidar (Naud et al., 2004) data.
MODIS CTH was then found to agree with radar CTH within 1 km for mid- and
high-level clouds and within 3 km for low-level clouds (Naud
et al., 2005). The comparison with the lidar showed somewhat smaller
differences for the low-level clouds (
Both night- and daytime data were evaluated with respect to cloud fraction to determine the impact of this parameter. Cloud fraction does not appear to be associated with any specific under-/overestimates or affect the magnitude of the differences. A few of the outliers do, however, have cloud fractions close to 0.5 (minimum cloud fraction, Sect. 2.5).
Scatterplots of MODIS liquid water path (LWP) versus ARM liquid water path. The two top subfigures contain data from the C6 dataset, while the bottom subfigures contain data C5.1. The colouring of the circles is according to solar zenith angle. The lines in the figures are 1 : 1 lines.
Median and mean differences, standard deviations of the
differences, correlation coefficients and
For the satellites considered here, LWP is obtained from visible parameters and is hence only available when SZAs are below the thresholds stated in Sect. 3.1. Here, it is investigated whether LWP from the new C6 products has a better agreement with the ground-based measurements, relative to C5.1. Figure 3 and Table 5 contain the results for all satellite scenes that passed the selection criteria, while Table 6 contains the results for the satellite scenes that passed the selection criteria for both collections, i.e. can be compared directly.
Median and mean differences, standard deviations of the
differences, correlation coefficients and
As can be seen in Table 5, there were more cases selected from C5.1 than
from C6 for the Aqua data; whereas, for Terra, a similar number of cases
from both collections were selected. The satellites slightly overestimate
LWP for C6 relative to the ground-based measurements with median differences
less than 12 % (Table 5). For C5.1, Aqua marginally overestimates LWP
(4.56 %), whereas Terra shows an underestimate of 14.3 %. The Terra
LWP underestimate may be due to a drift in the reflectance bands of the
sensor which has been corrected for in C6 (Aisheng et al., 2013). The
correlation coefficients are quite high for all but the Terra C6 datasets,
and most cases are close to the 1 : 1 line when the LWP is below
200 g m
For C5.1, differences in LWP from satellite and ground-based measurements do not appear to be affected by SZA (Fig. 3). For C6, however, there does appear to be some influence with respect to SZA, with a possible bias towards a satellite overestimate at high SZA. A larger dataset is necessary to confirm if this is overestimation is systematic.
There are fewer LWP cases for the direct collection intercomparison (Table 6). For Terra, 16 cases were classified as suitable in C6 but did not pass the selection criteria in C5, indicating that the changes in the algorithms made for C6 could have a significant impact. The change in the number of cases is likely a result of modifications in the cloud phase algorithm, changing how many pixels that are classified as liquid, but adjustments to the potential multilayer cloud flag and look-up tables will also affect which pixels pass the selection criteria. In general, the performance does improve when only cases where both collections pass the selection criteria (Table 6 compared to Table 5), except for the median difference in Terra C5.1 (the standard deviation does improve).
A previous study of LWP from MODIS C5 over China found that Terra and Aqua
underestimated LWP by 43.3 and 33.6 g m
An ARM mobile facility was deployed in Hyytiälä, Finland, from February to September 2014 as part of the BAECC campaign and provided a suitable dataset for evaluating satellite cloud retrievals at high latitudes. Ground-based measurements of CTH, obtained from lidar and radar measurements, and LWP, from microwave radiometer measurements, are compared here to three satellite instruments: the MODIS instruments aboard Terra and Aqua; and the VIIRS instrument aboard the Suomi-NPP satellite.
There are no restrictions on CTH but the data are divided into single- and multiple layers according to the cloud mask derived from the ground-based measurements. For single-layer clouds, MODIS CTH is, on average, 14 % higher than ground-based measurements. For multilayer clouds, however, MODIS CTH is, on average, 5.8 % lower than ground-based measurements. Similar conclusions are made for the VIIRS intercomparisons during nighttime; during daytime there were not enough data to make any general conclusions, partly a result of the high-latitude location. The MODIS IRW method frequently overestimates CTH for high-level clouds.
Single-layer cloud situations only were selected for the LWP
intercomparison. Two different versions of MODIS products were evaluated:
collections C6 and C5.1. The LWP for C6 shows an overestimate, relative to
the ground-based measurements, of 14 % (12.5 %) for Aqua (Terra). For
C5.1, there is a slight overestimate of LWP (< 5 %) by the MODIS
instrument aboard Aqua, while Terra's exhibits an underestimate of about
14 %. The underestimation by Terra in C5.1 is most likely caused by a
known drift in the reflectance channels, which has been corrected for in C6.
Good agreement is shown between satellite and ground-based data for LWP
below 200 g m
The overall performance of the satellite retrievals show small median biases when compared to the ground-based observations. There are however some cloud scenes for which the satellite retrievals do not work well. Situations where thin cirrus clouds are present over lower clouds seem to be extra problematic. This evaluation was performed at a high-latitude location to highlight any issues with large solar zenith angles, but there seemed to be little influence on the cloud parameters investigated here. Additional evaluations of satellite cloud products performed across the globe will be necessary to draw more general conclusions regarding the performance of the investigated satellite cloud products.
The data from the MODIS sensors were provided by the US National Aeronautics
and Space Agency through the Level 1 and Atmosphere Archive Distribution
System:
The deployment of AMF2 to Hyytiälä was enabled and supported by ARM. Argonne National Laboratory's work was supported by the U.S. Department of Energy, Assistant Secretary for Environmental Management, Office of Science and Technology, under contract DE-AC02-06CH11357. The authors gratefully acknowledge the support of AMF2 (Nicki Hickmon, Michael Ritsche and others), SMEAR-II (Janne Levula and others) and the BAECC community for their support in initiating the BAECC campaign, its implementation and operation.
This work was carried out with the support of the Lund Centre for studies of Carbon Cycle and Climate Interaction, LUCCI; the European Seventh Framework Programme, ACTRIS (EU INFRA-2010-1.1.16-262254), Aerosols, Clouds, and Trace gases; Research Infra Structure Network; the Strategic Research Program MERGE, Modeling the Regional and Global Earth System; and the Swedish Research Council (diary no: 2010-4683). We are also grateful for the support by the Swedish Research Council and from the Nordic Council of Ministers for the Nordic Top-level Research initiative CRAICC: Cryosphere–atmosphere interactions in a changing Arctic climate. This work was partly supported by the Office of Science (BER), U.S. Department of Energy via BAECC (Petäjä), European Commission via projects ACTRIS-TNA, ACTRIS2, BACCHUS, PEGASOS, and Academy of Finland Centre of Excellence (project number 272041). Edited by: A. Kokhanovsky Reviewed by: three anonymous referees