Abstract

Variability and trends in total cloud cover for 1982–2009 across the contiguous United States from the International Satellite Cloud Climatology Project (ISCCP), AVHRR Pathfinder Atmospheres–Extended (PATMOS-x), and EUMETSAT Satellite Application Facility on Climate Monitoring Clouds, Albedo and Radiation from AVHRR Data Edition 1 (CLARA-A1) satellite datasets are assessed using homogeneity-adjusted weather station data. The station data, considered as “ground truth” in the evaluation, are generally well correlated with the ISCCP and PATMOS-x data and with the physically related variables diurnal temperature range, precipitation, and surface solar radiation. Among the satellite products, overall, the PATMOS-x data have the highest interannual correlations with the weather station cloud data and those other physically related variables. The CLARA-A1 daytime dataset generally shows the lowest correlations, even after trends are removed. For the U.S. mean, the station dataset shows a negative but not statistically significant trend of −0.40% decade−1, and satellite products show larger downward trends ranging from −0.55% to −5.00% decade−1 for 1984–2007. PATMOS-x 1330 local time trends for U.S. mean cloud cover are closest to those in the station data, followed by the PATMOS-x diurnally corrected dataset and ISCCP, with CLARA-A1 having a large negative trend contrasting strongly with the station data. These results tend to validate the usefulness of weather station cloud data for monitoring changes in cloud cover, and they show that the long-term stability of satellite cloud datasets can be assessed by comparison to homogeneity-adjusted station data and other physically related variables.

1. Introduction

Climate data records (CDRs) constructed from historical satellite measurements have played an increasingly important role in climate monitoring and climate change detection because of their global coverage and multidecadal length. Cloudiness is one of the essential climate variables in that it regulates Earth’s radiation budget and water cycle and exerts a strong feedback on climate change. The existence of uncertainty in satellite data calibration and intersatellite calibration, cloud detection method, and other processing involving the use of radiative transfer models and ancillary data (Dai et al. 2006; Evan et al. 2007; Heidinger et al. 2010) makes it essential to evaluate the climate quality of long-term satellite cloud products.

Past intercomparisons of satellite cloud datasets have focused mainly on climatological mean cloud amount, temperature, and height, and other physical properties as summarized by the GEWEX cloud assessment initiative (Stubenrauch et al. 2012, 2013). They compared 12 satellite products, showing diurnal and seasonal cycles of cloud amount and other cloud variables, and noting differences in long-term trends for the globe.

In this study, in contrast, we focus on interannual variations and long-term trends for the contiguous United States using independent surface weather station observations and three of the current leading satellite datasets, which have the longest period of record among the satellite products available: International Satellite Cloud Climatology Project (ISCCP; Rossow and Schiffer 1999), Pathfinder Atmospheres–Extended (PATMOS-x; Heidinger et al. 2014), and Satellite Application Facility on Climate Monitoring Clouds, Albedo and Radiation from AVHRR Data Edition 1 (CLARA-A1; Karlsson et al. 2013). This assessment is limited to the contiguous United States, where a multidecadal cloud cover dataset has been developed by adjusting discontinuities resulting from changes in reporting or data archiving procedures of synoptic weather observations from 1949 through 2010 (Free and Sun 2013, 2014). This dataset serves as “ground truth,” providing independently validated cloud information with which to assess the satellite data.

Satellite cloud cover products, including ISCCP, PATMOS-x, and CLARA-A1, are derived from visual and infrared radiance measurements and cloudy scenes are typically identified by comparing their radiance to a clear-sky threshold. Cloud cover at weather stations, on the other hand, is visually reported following the WMO synoptic code. There are differences among the satellite products in terms of sensor characteristics, retrieval method, and spatial and temporal sampling and averaging. Cloud cover definition and viewing geometry are different between satellite-derived products and surface visual observations. All of these factors can cause differences between the cloud cover amounts at a specific location and time even if the cloud is correctly detected by the respective observing systems, as discussed by Wu et al. (2014).

This assessment is therefore focused on temporal variations including seasonal and interannual time scales and especially long-term trends, but we also discuss the climatological mean characteristics of satellite products in terms of their spatial and seasonal departures from surface data in order to understand the sensitivity of satellite retrieval methods to surface type. In addition to surface cloud data, climate variables that are physically connected to cloud cover variations, including precipitation days, diurnal temperature range (DTR), and downward solar radiation, are used as proxy cloud data for an additional validation, in an attempt to narrow our uncertainty in this assessment.

2. Datasets

a. Surface cloud dataset

The U.S. surface cloud cover dataset consists of the observations from 54 National Weather Service (NWS) and Federal Aviation Administration (FAA) stations and 101 military stations. To avoid nighttime measurement bias because of poor illumination, only daytime data (1500, 1800, and 2100 UTC) were used to construct the dataset. Depending on the time zone in which the stations are located, hourly values of 0900, 1200, and 1500 central time zone; 1000, 1300, and 1600 eastern or Pacific time zone; or 1100, 1400, and 1700 mountain time zone were used, so the cloud cover dataset is primarily for afternoon. The introduction of the ASOS at NWS stations in the 1990s disrupted the record at many stations, so we restricted our dataset to use only stations that continued to make visual cloud observations. Further details of this dataset and its homogeneity adjustment are found in Free and Sun (2014). For most quantities evaluated here, the military and NWS station subsets give similar results, so we use the combined station set for our analysis except where otherwise noted.

b. ISCCP dataset

The ISCCP product was derived primarily from a succession of geostationary satellites and its cloud detection algorithm was based on one visual (0.63 μm) and one infrared (11 μm) channel. Its D2 version provides 3-hourly monthly grid-averaged data with a spatial resolution of 2.5°. The ISCCP product used in this study is accessed from the GEWEX cloud intercomparison project database as of early 2014, which provides 2.5° × 2.5° gridded monthly data for 1984–2007. For this work we examined the 0900 and 1500 LT local time (LT) datasets.

c. PATMOS-x datasets

The PATMOS-x cloud product was derived from calibrated measurements from AVHRR (Heidinger et al. 2010), an imager on board the NOAA polar-orbiting satellites that has had 0.63-, 0.86-, 3.75-, 10.8-, and 12.0-μm channels since 1981, and more recently the EUMETSAT MetOp satellites. A Bayesian cloud mask algorithm developed using collocated Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) data is used for cloud detection (Heidinger et al. 2012). PATMOS-x was processed using global area coverage (GAC) data, representing an area of approximately 3 km × 5 km, the mean of four 1.1-km AVHRR pixels. These GAC pixel data are then sampled to create a 0.1° × 0.1° product (level-2b data) by using the GAC pixel closest in distance to the grid. The level-2b product was then used to generate the 1° × 1° level-3 product by averaging all the values within the 1° × 1° grid box. The PATMOS-x AVHRR level-2b data are now hosted at the NOAA National Climatic Data Center (NCDC) as an operational climate data record (CDR). The following two level-3 datasets are used in this study:

  • The GEWEX version dataset (1982–2009) was constructed by putting together the monthly averages from individual satellites at 0130, 0730, 1330, and 1930 LT. For the months when multiple satellites were available, data from the latest satellite were used. This and the ISCCP data used are described in the GEWEX intercomparison project documentation (Stubenrauch et al. 2012). No diurnal correction was applied to this dataset. In this study, we examined 0730 and 1330 LT data.

  • For the diurnally corrected daily (DCD) dataset (1982–2012), Foster and Heidinger (2013) describe the method used to account for changes in cloud and other properties resulting from changes in observation time from satellite drifts. In summary, the “climatological” diurnal cycle of cloudiness developed from all the PATMOS-x measurements was used to correct the hourly data of individual satellites to any local time for each 1° × 1° grid box and for each month. The diurnally corrected monthly data were then constructed by averaging the daily values, which were created from the diurnally corrected values of all measurement hours averaged from values of all individual satellites available. This dataset was provided by the PATMOS-x team at CIMSS, University of Wisconsin–Madison.

d. CLARA-A1 dataset (1982–2009)

This was developed by the EUMETSAT Satellite Application Facility on Climate Monitoring (CM SAF) project by applying their cloud masking detection algorithm to the PATMOS-x calibrated AVHRR radiance dataset (Karlsson et al. 2013). Like the PATMOS-x products, the CLARA-A1 original retrievals were made at GAC resolution. After that, all pixels falling into the 0.25° grid box during a month were used in the averaging process to generate the 0.25° × 0.25° monthly datasets, consisting of nighttime and daytime data, each averaged from all satellite measurements available for each month. Daytime is defined to include times with solar zenith angle lower than 80°, including both the afternoon and morning orbits. This product was obtained from the CM SAF website (www.cmsaf.eu).

To make data comparable in spatial sampling to the 1° × 1° PATMOS-x products, we average all the 0.25° × 0.25° (“fine” product) CLARA-A1 cloud values into 1° × 1° grid boxes to generate the “coarse” CLARA-A1 product. Most results are similar between the fine and the coarse product; unless otherwise indicated, in the rest of this paper we use the coarse version.

This study focuses on the afternoon datasets, including ISCCP 1500 LT and PATMOS-x 1330 LT data, in addition to PATMOS-x DCD and CLARA-A1 daytime products, since they are closer to the in situ data in time, but we will also discuss characteristics of the datasets at other hours as necessary to gain a better understanding of the product diurnal characteristics.

e. Other datasets

DTR, precipitation days, and downward solar radiation are the variables physically related to and correlated with cloud cover that we used for comparisons with total cloud data. For NWS stations, we use DTR computed from monthly mean adjusted maximum and minimum temperature data in the Historical Climatology Network (HCN) at NCDC (Lawrimore et al. 2011). For military stations, matching HCN station data are not generally available, so we used monthly mean DTR from gridded adjusted HCN data (Lawrimore et al. 2011).

We used daily precipitation data from NWS Cooperative Observer (COOP) weather stations (Groisman et al. 2004) kindly provided by Pasha Groisman of NCDC. To minimize the effects of changes in instrumentation on measurements of very small amounts of precipitation, we computed the number of days with precipitation greater than 0.5 mm day−1 for each month and station. If data were not available for our stations, we used precipitation from the same dataset at the closest available station.

Surface radiation data from six NOAA SURFRAD (Augustine et al. 2005) sites and the Atmospheric Radiation Measurement Program (ARM) Southern Great Plains (SGP) facility (listed in Table 5) are also used for comparison. Most of the stations have data starting in 1995 with the exception of the “Desert Rock” site located at the Desert Rock Airport, Nevada, and the “Penn State” site located in Ramblewood, Pennsylvania, which start in 1998. The SURFRAD data are monthly, accessed from Global Monitoring Division, NOAA/Earth System Research Laboratory (ftp://aftp.cmdl.noaa.gov/data/radiation/surfrad/). The SGP data are the ARM Best Estimate Radiative Flux (BEFLUX) Value Added Product (VAP) 1-min data (Shi and Long 2002). In this study, these 1-min data are first averaged to hourly, then to daily and finally to monthly means for analysis.

3. Method

To extract satellite data at locations equivalent to our station locations we used the satellite grid point that was closest to the NWS or military station. We tested two alternative methods: linear interpolation between the four closest grid points, and a fractional weighting method that weighted nearby grid points according to their distances from the ground station. For the U.S. mean, the three methods gave very similar results. Differences [measured by root-mean-square (RMS) differences between results from different methods] were found mostly at locations close to oceans, where the satellite grid boxes contained a large proportion of ocean surface. Seven military stations and seven NWS stations had more than 50% ocean surface in the satellite grid boxes in which they fell. At these locations, the presence of the ocean may make it more difficult to obtain a satellite time series that is similar to that from the land station. However, the RMS differences between our station data and the collocated satellite data were not consistently worse at these coastal stations than at other stations.

We calculated Pearson correlation coefficients between station time series and collocated satellite data, and used least squares linear regression to get trends, with a correction for autocorrelation to derive confidence intervals for the trends. We consider statistics to be significant if they are at the 0.05 level or better. Data at individual locations were combined into nine regions used by NCDC for climate monitoring, shown in Fig. 1. To reduce the impact of uneven station distribution on spatially averaged U.S. mean values, U.S. mean results are created by averaging time series of anomalies in 2.5° grid boxes, then taking the mean of the grid box averages. When seasonal results are given, winter refers to December–February (DJF), spring refers to March–May (MAM), summer refers to June–August (JJA), and fall refers to September–November (SON).

Fig. 1.

Regions used for analysis of cloud cover, with daytime climatological mean cloud cover from weather station data for each region for 1984–2007.

Fig. 1.

Regions used for analysis of cloud cover, with daytime climatological mean cloud cover from weather station data for each region for 1984–2007.

Because we are primarily interested in longer-term variability and trends, most calculations are done using annual mean time series. Where monthly mean time series are used, annual cycles are removed from monthly data (by subtracting the climatological monthly means from corresponding months of each year) to minimize the impact of differences in annual cycle.

Warren et al. (2007) and Eastman and Warren (2014) observed that the daytime multidecadal trends in cloud cover over land tend to agree well with the nighttime trend in both sign and magnitude. It thus makes sense to evaluate the long-term trends in satellite products against our surface dataset (see sections 4c and 4d) even though they do not represent the same hours of the day.

4. Results

a. Climatological means

Climatological mean cloud cover amounts in the satellite products are evaluated by examining their spatial and seasonal departures relative to surface observations. Figure 1 shows the regional mean values of total cloud cover from the station data and Table 1 shows the difference from that value for the primary satellite products studied here.

Table 1.

Satellite minus NWS and military station annual mean climatological cloud cover (%) and standard deviation in parentheses for nine regions (Fig. 1) during 1984–2007. Numbers in parentheses after the region names are the number of stations in that region.

Satellite minus NWS and military station annual mean climatological cloud cover (%) and standard deviation in parentheses for nine regions (Fig. 1) during 1984–2007. Numbers in parentheses after the region names are the number of stations in that region.
Satellite minus NWS and military station annual mean climatological cloud cover (%) and standard deviation in parentheses for nine regions (Fig. 1) during 1984–2007. Numbers in parentheses after the region names are the number of stations in that region.

The ISCCP 1500 LT product has higher cloud cover than surface data across all regions, particularly over the western part of the country where ISCCP exceeds the surface data by over 9%. Similar regional variation is shown in the 0900 LT product but with much smaller differences from surface cloud cover values. Seasonally, the regional variation is stronger in warm seasons than in cold seasons (not shown). The overestimated cloud cover over the West and Southwest regions, where surface albedo is great, may result from false cloud detection, which appears to occur also in the CLARA-A1 product, as will be discussed later in this section.

The PATMOS-x 1330 LT product has lower cloud cover than surface data, particularly over the western part of the country. However, this product shows a strong seasonal variation, with summer cloud covers even higher than those in the surface data over the South and Southeast regions. PATMOS-x DCD product cloud cover is lower than surface data, particularly in the South and Central regions. Regional variations in PATMOS-x DCD are smaller than those in the 1330 LT product.

The CLARA-A1 daytime product has more cloud cover than the station data over the West, and particularly the Southwest region (by ~12%), particularly in spring and summer, but in the other regions their values are up to 4% less than the surface data. The overestimation of cloud cover in the Southwest region does not occur in the nighttime product, which underestimates cloud cover by about 6% over that region (Table 1). The misclassification of cloud-free scenes as clouds is suspected to be the cause for the daytime overestimated cloud cover over the Southwest region, where the surface is generally dry or semiarid with a greater albedo (Karlsson et al. 2013).

Standard deviations of climatological differences from weather station data computed for all sites across the contiguous United States are 5.7%, 4.1%, 3.6%, and 6.2% for ISCCP, PATMOS-x 1330 LT, PATMOS-x DCD, and CLARA-A1, respectively. These numbers indicate that, overall, PATMOS-x DCD spatial variability is in closest agreement with weather station data, followed by PATMOS-x 1330 LT, then ISCCP, with the CLARA-A1 dataset having the worst agreement. Over the Southwest region, where daytime CLARA-A1 cloud cover is highly overestimated, its station-to-station variability for both day and night is much bigger than other products, as suggested by the standard deviation values. The West region has the strongest spatial variability of cloud cover in the contiguous United States (see Eastman et al. 2014), and the large standard deviation values shown in all the products for this region indicate that it is a challenge for satellites to capture the cloud variability over an area with complex terrain.

b. Yearly and interannual variations

In this section we first examine the standard deviations of the annual anomaly time series to compare the year-to-year variability of the products, and then look in more detail at the correlations between the satellite and surface cloud time series, using both individual station locations and large-scale means. We use the detrended annual and seasonal time series to compute the standard deviation to reduce the impact of trend on the year-to-year variability. The year-to-year variability in the surface data is 1.43%, with the greatest variability in autumn and smallest in summer (Fig. 2). Relative to surface data, ISCCP tends to have a smaller year-to-year variability, particularly for winter, when the variability is 40% less. Both PATMOS-x 1330 LT and DCD products show variability higher than surface data by about 27% for winter but similar to surface data for other seasons. [This is consistent with results from the GEWEX intercomparison project (Stubenrauch et al. 2012), which showed interannual variability of ISCCP for the Northern Hemisphere midlatitudes to be less than that in PATMOS-x.] The year-to-year variability in the CLARA-A1 product is significantly higher than in the surface data, for example, by 98% for nighttime and 68% for daytime in the annual data.

Fig. 2.

Standard deviations of station and satellite cloud cover data (%) computed from time series of 1984–2007.

Fig. 2.

Standard deviations of station and satellite cloud cover data (%) computed from time series of 1984–2007.

One might expect that the correlation computed from monthly anomaly time series (after the annual cycle is removed) should be similar to the correlation computed from yearly average time series for the same time period, but we find it is not true for the cloud datasets we analyze, particularly for the CLARA-A1 dataset, which has a strong temporal trend. Correlations between individual station data and collocated satellite data are generally lower when annual mean time series are used instead of monthly mean series; Fig. 3 shows these correlations for annual means.

Fig. 3.

Pearson correlation coefficients between annual mean station cloud time series and satellite data for 1984–2007.

Fig. 3.

Pearson correlation coefficients between annual mean station cloud time series and satellite data for 1984–2007.

While most datasets show better correlations in the eastern half of the country than in the West region, the specific regional patterns vary. Across most of the country, PATMOS-x DCD has correlations higher than PATMOS-x 1330 LT which is higher than ISCCP 1500 LT and CLARA-A1 daytime. Our results for ISCCP could be affected by the lower resolution available in the GEWEX ISCCP dataset than in the other satellite datasets we used. However, we examined correlations between time series of the surface data and PATMOS-x and CLARA-A1 using data that was converted into 2° resolution to more closely resemble the resolution of the ISCCP data, and the correlations were only slightly smaller than for the higher-resolution data. We therefore doubt that spatial resolution is an important factor in this comparison.

Several stations are poorly correlated in all datasets, especially Homestead military site [WMO identifier (ID) 722026] and the Miami NWS site (WMO ID 722020) on the Florida coast, and Point Mugu (WMO ID 723910) on the California coast. For these locations, the poor correlations may be due to remaining problems in the station data, or because these coastal stations are not representative of the mostly-ocean satellite grid boxes they fall in.

Despite the systematic difference between satellite and surface total cloud cover shown in Table 1, the correlations between U.S. means of surface data and collocated satellite data computed from monthly time series of 1984–2007 are statistically significant for all the products. The correlation is 0.62 for both ISCCP 0900 and 1500 LT, 0.85 for PATMOS-x 1330 LT, 0.92 for PATMOS-x DCD, and 0.61 and 0.53 for CLARA-A1 nighttime and daytime products, respectively. If the time series are detrended first, the correlations for ISCCP and PATMOS-x remain basically the same as the ones computed without the trends removed, but the correlation increases to 0.66 for CLARA-A1 nighttime and 0.60 for daytime. Overall, based on the average time series, PATMOS-x products have higher correlations with surface data than the other two products, which is consistent with the correlations for individual stations shown in Fig. 2.

Table 2 shows correlations computed from U.S. mean time series for all four seasons and for annual mean data. As for the monthly data, correlations based on detrended annual data are similar in most cases to those for the full time series. Although correlations in this table are generally lower than those for monthly time series, correlations from annual mean time series are statistically significant for all products except CLARA-A1 daytime datasets for both full and detrended time series. As with the monthly mean time series, PATMOS-x DCD shows the highest correlation.

Table 2.

Pearson correlation coefficients between gridded U.S. mean satellite cloud fraction time series and gridded U.S. mean station data for 1984–2007. Correlations that are not statistically significant at the 0.05 level are in boldface. Numbers in parentheses are correlations in detrended time series.

Pearson correlation coefficients between gridded U.S. mean satellite cloud fraction time series and gridded U.S. mean station data for 1984–2007. Correlations that are not statistically significant at the 0.05 level are in boldface. Numbers in parentheses are correlations in detrended time series.
Pearson correlation coefficients between gridded U.S. mean satellite cloud fraction time series and gridded U.S. mean station data for 1984–2007. Correlations that are not statistically significant at the 0.05 level are in boldface. Numbers in parentheses are correlations in detrended time series.

Correlations with surface data are statistically significant for most of the seasons for ISCCP and PATMOS-x, and for all seasons for PATMOS-x DCD. The best correlations occur in fall for most datasets, but the best correlations are in winter for CLARA-A1 (daytime). The lower correlation in summer shown in most of the products is perhaps due to cumulus clouds, which are small in area but occur frequently and cannot be accurately detected by satellites. The relatively low correlation in winter seen in the ISCCP, PATMOS-x 1330 LT, and DCD datasets is very possibly caused by inaccurate cloud detection under snow cover or cold low-level atmospheric temperature. Sun (2003) noticed that the correlation of ISCCP and in situ data over the United States is improved after the data points with snow on the ground are removed.

PATMOS-x DCD has the highest correlations of any dataset for all seasons. Relative to ISCCP and PATMOS-x, CLARA-A1 shows lower correlations with surface data, particularly in summer. The pronounced negative trends in the CLARA-A1 data (Fig. 4 and Table 3), which are not shown in the surface data (and less pronounced in other satellites), contribute to the poor relationship. The CLARA-A1 correlations with surface data for most seasons become higher when they are computed from the time series with long-term trends removed, (e.g., 0.37 for summer) but are still not statistically significant.

Fig. 4.

U.S. annual mean cloud cover anomalies from satellite data collocated with U.S. weather stations for (top) military, (middle) NWS, and (bottom) all stations combined, with trends (% decade−1). Trends are for 1982–2009 except for ISCCP, which is for 1984–2007. All trends are statistically significant at the 0.05 or better level except for the NWS and combined weather station datasets.

Fig. 4.

U.S. annual mean cloud cover anomalies from satellite data collocated with U.S. weather stations for (top) military, (middle) NWS, and (bottom) all stations combined, with trends (% decade−1). Trends are for 1982–2009 except for ISCCP, which is for 1984–2007. All trends are statistically significant at the 0.05 or better level except for the NWS and combined weather station datasets.

Table 3.

Linear least squares trends (% decade−1) for four seasons in gridded U.S. monthly mean surface data and collocated satellite data for 1984–2007. Numbers in parentheses are 2 times the standard error of the trends.

Linear least squares trends (% decade−1) for four seasons in gridded U.S. monthly mean surface data and collocated satellite data for 1984–2007. Numbers in parentheses are 2 times the standard error of the trends.
Linear least squares trends (% decade−1) for four seasons in gridded U.S. monthly mean surface data and collocated satellite data for 1984–2007. Numbers in parentheses are 2 times the standard error of the trends.

c. Trends

In our previous paper (Free and Sun 2013), we found that trends in the U.S. mean military station subset were more negative by about 0.8% compared to those in the NWS subset, perhaps because of differences in the data sources and methods used to construct the homogeneity-adjusted time series. In this section, we therefore compare the trends of satellite products against NWS and military data separately as well as against the combined dataset to obtain additional insight into the degree of discrepancy between surface and satellite products. Possible reasons for the differences between trends in different products are discussed in section 5 below.

Figure 4 shows the time series of annual U.S. mean cloud cover for military, NWS, and combined locations from station and collocated satellite data, along with the trends for each time series using the full period of data availability through 2009. The ISCCP and PATMOS-x time series follow the station time series fairly well except that ISCCP does not show the dip between 1986 and 1990. CLARA-A1 shows large reductions in cloud cover at several points in time that are not seen in the other time series. All satellite datasets show negative trends for all time periods examined here. As found in our previous work (Free and Sun 2014), the U.S. mean of station data from military sites shows a negative trend (−0.67% decade−1 for 1984–2007) but the NWS station mean has a slightly positive trend of 0.28% decade−1 for 1984–2007. For CLARA-A1 and PATMOS-x 1330 LT, trends computed from the military network are more negative than those from the NWS network, but this is not true for PATMOS-x DCD or ISCCP. (One factor contributing to the differences may be the fact that the military network has more stations in coastal areas than the NWS network does).

For 1984–2007, the trend in PATMOS-x 1330 LT (−0.55% decade−1) is the closest of the satellite trends to that in the NWS station data, and PATMOS-x DCD is next closest. For the military data, PATMOS-x 1330 LT and PATMOS-x DCD trends (−0.65 and −0.79) are very close to those in the station data. For both station subsets, ISCCP trends are somewhat more negative than the PATMOS-x results and the CLARA-A1 results are almost an order of magnitude larger (e.g., −4.7% decade−1) than those in the station data.

In the NWS and military station combined dataset, the trend computed from the annual time series (shown in Fig. 4, bottom panel) for 1984–2007 is −0.40% decade−1 (Table 3). As in the results for the separate military and NWS networks, trends in PATMOS-x products are the closest to those in the surface data, the second closest is ISCCP, and CLARA-A1 products show the largest deviation from the surface data.

As shown in Table 3, the trend in PATMOS-x DCD is more negative than its 1330 LT product, ISCCP 1500 LT has a trend slightly less negative than its 0900 LT product, and the CLARA-A1 daytime product shows a more negative trend than the nighttime. Trends in the ISCCP and CLARA-A1 products are significantly different from zero, but those in PATMOS-x 1330 LT, PATMOS-x DCD, and the station data are not significant for 1984–2007, although the PATMOS-x products do have significant trends for 1982–2009. As with the separate military and NWS station subsets, extending the time series from 1984–2007 to 1982–2009 gives more negative trends in surface and PATMOS-x products and in CLARA-A1 nighttime data, but slightly less negative trends for CLARA-A1 daytime. The trends in PATMOS-x DCD become much more negative relative to PATMOS-x 1330 LT for both NWS and military networks and the combined network when the trends are computed from 1982 to 2009, and for this period the PATMOS-x 1330 LT trends are obviously much closer to the trends in the station data. As will be discussed in section 5, the diurnal data sampling used to construct PATMOS-x DCD and CLARA-A1 products changes over time. It would be worthwhile to investigate whether the increasing trend discrepancy between the two PATMOS-x products when the time period extends backward and forward is related to the diurnal sampling problem. An alternative explanation is that larger observed cloud cover for 1983 resulted from the El Chichón volcanic eruption, which would tend to increase the negative trend when the years 1982 and 1983 are included.

The seasonal patterns of trends vary widely among the datasets (Table 3). For the PATMOS-x DCD and station data, the trends are much more negative for fall than for other seasons. For all 2 ISCCP versions here, JJA and SON have more negative trends than other two seasons. In contrast to the PATMOS-x results, CLARA-A1 has more negative trends for the summer and spring than for other seasons (except for the nighttime when the biggest decline occurs in fall). Although ISCCP trends are more negative than those in PATMOS-x for annual data, PATMOS-x actually has a more negative trend than ISCCP for winter, and the two are similar for fall, with the greatest difference occurring in summer.

Figure 5 shows annual anomaly time series for nine regions and the trends for each region for 1982–2009, except for ISCCP, which covers 1984–2007. The largest declines overall are for CLARA-A1 daytime data in the Southwest region (>8%); this large negative trend is not found in CLARA-A1 for other regions (or nighttime). The time series for this region shows a larger shift in the CLARA-A1 time series in the early 1990s than is seen in other regions, a shift also not seen in the other datasets. All these suggest that the CLARA-A1 daytime cloud overestimation over the Southwest region (Table 1) may change when the diurnal sampling of the satellite data changed (see section 5). The Southwest is also the region where the negative trend in PATMOS-x 1330 LT (−2.3% decade−1) is greater than the one in PATMOS-x DCD (−1.9% decade−1), while the opposite is true for all other regions. ISCCP 1500 LT has a more moderate trend of approximately −1.5% decade−1 in this region, with its largest declines in the Southeast region, and the station data have an equally large decline in the Northwest region as in the Southwest. In the East North Central, South, Southeast, and Southwest regions, all datasets have negative trends, and most trends in the remaining regions are also less than zero, but small positive trends occur for the station data in the West and Northeast regions, ISCCP 1500 LT in West North Central region, and PATMOS-x 1330 LT in the Central and Northwest regions.

Fig. 5.

As in Fig. 4 (bottom), but for the nine individual regions in Fig. 1.

Fig. 5.

As in Fig. 4 (bottom), but for the nine individual regions in Fig. 1.

d. Comparisons with other variables

Figure 6 shows U.S. annual mean DTR and days with precipitation in comparison with surface and satellite cloud cover time series. The correlations of annual surface cloud cover with DTR and precipitation days for 1984–2007 are −0.77 and 0.82, suggesting that both of these variables are closely related to U.S. cloud cover, as shown in previous work based on different periods of data (Sun et al. 2001; Free and Sun 2013). For DTR, we find a statistically significant correlation with cloud cover for PATMOS-x products; for precipitation days, correlations with cloud cover are statistically significant for all satellite products except CLARA-A1 daytime, with the highest correlation for PATMOS-x 1330 LT and DCD.

Fig. 6.

U.S. mean time series of total cloud cover from satellites and weather stations compared to (top) diurnal temperature range and (bottom) days with precipitation, with correlation coefficients between the cloud time series and DTR or precipitation for 1984–2007.

Fig. 6.

U.S. mean time series of total cloud cover from satellites and weather stations compared to (top) diurnal temperature range and (bottom) days with precipitation, with correlation coefficients between the cloud time series and DTR or precipitation for 1984–2007.

To provide additional information about the satellite and surface cloud data, we compared the cloud datasets to surface downwelling shortwave (SW) radiation; its two components, direct and diffuse shortwave radiation; and downwelling longwave radiation at six SURFRAD sites and the SGP facility in Oklahoma [the ARM/Baseline Surface Radiation Network (BSRN) site] shown in Table 4. Table 5 shows the correlation of total cloud cover from station and satellite products with these quantities at these locations using data spanning 1995 through 2007. Note that because of the difference in starting year of radiation data and the cloud ASOS contamination in the 1990s for NWS stations and in the years starting with 2005 at military stations, the data length used for correlation computation varies among sites. Nevertheless, the correlations between cloud cover and surface radiation are basically similar across all the sites, with the lowest correlations at the Fort Peck Indian Reservation, Montana, where snow occurs more often than other sites.

Table 4.

Surface radiation measurement sites and nearby weather stations that made cloud cover measurements.

Surface radiation measurement sites and nearby weather stations that made cloud cover measurements.
Surface radiation measurement sites and nearby weather stations that made cloud cover measurements.
Table 5.

Correlations from station and satellite data over seven surface radiation measurement sites (expansions in Table 4) of downwelling shortwave radiation (first row), direct and diffuse downwelling shortwave radiation (second and third rows), and downwelling longwave radiation (fourth row) with total cloud cover. Correlations were computed using monthly data after the time period mean monthly values were removed for every month. Numbers in the parentheses after the SURFRAD site ID are the number of months of the data used for the analysis. Correlations that are not statistically significant at the 0.05 level are in boldface.

Correlations from station and satellite data over seven surface radiation measurement sites (expansions in Table 4) of downwelling shortwave radiation (first row), direct and diffuse downwelling shortwave radiation (second and third rows), and downwelling longwave radiation (fourth row) with total cloud cover. Correlations were computed using monthly data after the time period mean monthly values were removed for every month. Numbers in the parentheses after the SURFRAD site ID are the number of months of the data used for the analysis. Correlations that are not statistically significant at the 0.05 level are in boldface.
Correlations from station and satellite data over seven surface radiation measurement sites (expansions in Table 4) of downwelling shortwave radiation (first row), direct and diffuse downwelling shortwave radiation (second and third rows), and downwelling longwave radiation (fourth row) with total cloud cover. Correlations were computed using monthly data after the time period mean monthly values were removed for every month. Numbers in the parentheses after the SURFRAD site ID are the number of months of the data used for the analysis. Correlations that are not statistically significant at the 0.05 level are in boldface.

The surface SW radiation and its direct SW component are negatively correlated with total cloud cover, while the diffuse SW radiation component and downwelling longwave radiation are positively correlated with total cloud cover, as is expected from their physical relationships. For all the sites, the correlations between cloud cover and downwelling SW or direct SW radiation are much greater than the correlations between cloud cover and the diffuse SW radiation, which may be the result of differing effects of aerosols on the two quantities. The lower correlation shown for the longwave radiation is expected because the amount of longwave radiation at the surface is related more closely to cloud-base height and other cloud properties other than total cloud cover.

The station cloud data generally appear to have higher correlations with downwelling SW or direct SW radiation than the satellite cloud products do, verifying that station data can be used as a reference to evaluate the satellite products. Among the satellites, PATMOS-x 1330 LT and particularly PATMOS-x DCD have the highest correlations with downwelling shortwave radiation and its direct downwelling component, and ISCCP and CLARA-A1 are comparable to each other overall. This is basically consistent with our findings about the correlations of satellite products with station cloud data (Table 3). If we use the temporal correlation of satellite data with surface cloud or radiation data as one of the indicators to judge the quality of satellite products, one would conclude that PATMOS-x DCD has the best climate quality among the satellite products analyzed.

Trends of cloud cover for station and satellite data and surface radiation were computed for all sites. We present time series comparisons for Bondville, Illinois; the “Table Mountain” site near Boulder, Colorado; and the ARM SGP facility near Lamont, Oklahoma (the sites with the longest data records) in Fig. 7. Downwelling SW radiation at all these three sites shows increasing trends (although with different magnitudes) during 1995–2007, and continues to increase through 2013 when the data end, indicating that a solar “brightening” trend has been occurring starting in the 1990s (Wild et al. 2005; Dutton et al. 2006). Increasing tendencies are found in downwelling solar radiation (both its direct and diffuse components) and downwelling longwave radiation at these as well as the other sites we examined. Surface cloud and all satellite cloud products at the locations designated in Fig. 7 (top and middle panels) show decreasing trends, consistent with the increasing solar radiation. A similar inverse relationship between cloud and surface radiation trends was noticed by Long et al. (2009) but, unlike the data in this study, cloud cover data in Long et al. (2009) were derived from measured surface radiation data. In contrast, some other SURFRAD locations show increasing cloud in our station and/or satellite data, which is inconsistent with the solar radiation trends. Given the short time periods available and the large interannual variability in both cloud cover and solar radiation, it is difficult to draw meaningful conclusions from these trend comparisons.

Fig. 7.

Yearly time series of station and satellite cloud cover and surface downwelling shortwave radiation anomalies at the SURFRAD sites (top) TBL, (middle) BND, and (bottom) SGP. Note that the radiation measurement sites and cloud measurement stations are collocated (see Table 5 and text for detail). The cloud cover and downwelling shortwave radiation trend values, specified inside the plot, are per decade for 1995–2007.

Fig. 7.

Yearly time series of station and satellite cloud cover and surface downwelling shortwave radiation anomalies at the SURFRAD sites (top) TBL, (middle) BND, and (bottom) SGP. Note that the radiation measurement sites and cloud measurement stations are collocated (see Table 5 and text for detail). The cloud cover and downwelling shortwave radiation trend values, specified inside the plot, are per decade for 1995–2007.

5. Discussion

Cloud cover, along with other cloud properties, has a strong diurnal cycle over land areas (Bergman and Salby 1996), which can be seasonally dependent (Kondragunta and Gruber 1996; Eastman and Warren 2014). Given the presence of diurnal cloud cover variations, change in diurnal sampling over time can introduce artifacts in cloud cover time series. For AVHRR on board the NOAA and EUMETSAT polar-orbiting satellites, there are two observations per day in the beginning of the 1980s, increasing to four starting in the mid-1990s and then to six or more after 2000. Furthermore, this change shifts the focus of daily cloud observations from predominantly midday and midnight observations at the beginning of the period to mostly morning and evening observations in the end of the period. The timing of the large downward shifts in the CLARA-A1 annual mean time series (Figs. 4 and 5) appears to coincide with these changes in observation times. It is reasonable to suspect that these changes may have made the CLARA-A1 trends more negative, and since cloud amounts for times close to dawn or sunset are generally lower, this may explain the more negative trend in CLARA-A1 daytime data compared to the nighttime data shown in Table 3. An additional complication is the changes in the equator crossing time of individual AVHRR sensors because of satellite drift, which is not accounted for in generating the CLARA-A1 product. These time sampling problems are expected to affect not only the long-term cloud trend but also the cloud cover variations on interannual scales. The lower correlation between CLARA-A1 and station data (Fig. 6 and Table 2) compared to those of other satellite products could be a manifestation of those issues.

Like the CLARA-A1 product, PATMOS-x DCD was created by using all the observations available each day, but PATMOS-x DCD applies a climatologically derived diurnal correction to all available measurements for the day before creating the monthly dataset. Differences in trends between the corrected and uncorrected records are driven by changing satellite overpass times and how the cloud fractions at those times differ from daily cloudiness calculated from a climatologically derived diurnal cycle. According to Foster and Heidinger (2014), over North America centered on the contiguous United States the total cloudiness trend of −1.54% decade−1 for the uncorrected record was reduced to only −1.42% decade−1 for the corrected record for 1982–2012.

The impact of the diurnal bias suggested by Foster and Heidinger (2014) alone therefore cannot explain the strong negative trends shown in both daytime and nighttime CLARA-A1 datasets. The daytime cloud overestimation over dry/semiarid areas mentioned in section 4a, convolved with the time sampling shift from the midday in the 1980s to morning and evening at the end of record, could artificially make the daytime trend more negative, but the large negative trend is also shown in the nighttime data, suggesting that there may be other issues in generating the CLARA-A1 product.

The 1982–2009 PATMOS-x 1330 LT dataset was generated by simply merging the nondiurnally corrected records of AVHRR afternoon ascending orbits including the NOAA-7, -9, -11, -14, -16, -17, -18, and -19 polar-orbiting satellites. The later satellite records were used to form the time series when two or more satellites were flying concurrently. The fact that the 1330 LT dataset trend matches the trend in the station dataset better than the diurnally corrected dataset may suggest that the drift impact around 1330 LT is small, or there may still be issues in the diurnally corrected method, as discussed in Foster and Heidinger (2013). Another possible explanation, a possible remaining positive bias in the station data, is discussed below.

The ISCCP cloud product has had many applications on a wide range of time scales. Recent studies (Campbell 2004; Evan et al. 2007), however, point out that the long-term global trends in cloud cover from the ISCCP record are influenced by artifacts associated with changing geostationary satellite viewing geometry. This study indicates that the 1984–2007 cloud cover trend in the ISCCP dataset over the contiguous United States is 0.35% decade−1 more negative than the trend in PATMOS-x DCD and 0.77% decade−1 more negative than the trend in station data. Other work has shown much larger differences between time series of PATMOS-x 1330 LT and ISCCP for the global mean cloud amount (e.g., Fig. 3.5.1 of Stubenrauch et al. 2012). Our results suggest that the large downward shifts in ISCCP in the early part of the record that are seen in some parts of the globe (Evan et al. 2007) may not be as severe for the contiguous United States; however, further investigation is needed to assess if the discrepancies noticed in this study are due to the ISCCP satellite viewing angle problem or problems in other datasets, as discussed in this section.

The nontrivial trend difference between the NWS (−0.07% decade−1) and military (−1.00% decade−1) station datasets during 1982–2009 suggests remaining uncertainties in the surface cloud data. Given these uncertainties, we considered the possibility that trends in the station data might still be too positive even after our homogeneity adjustments, and the satellite-derived trends in some satellite products might be correct. While some individual station cloud cover time series may still contain problems (e.g., the few stations with consistently poor correlations with satellite data, mentioned in section 4b), our comparisons of U.S. mean cloud cover with other physically related variables, such as DTR, precipitation days, and surface solar radiation, indicate that station cloud data is overall more reliable in terms of interannual variability than the satellite products analyzed. Furthermore, the negative trends in DTR and positive trends in precipitation days at our station locations suggest that the negative trend in the station cloud data is overestimated to some extent. We therefore reject the assumption that there are extensive or significant errors in our station data as an explanation for the differences between U.S. mean satellite trends and station trends.

Despite the differences between the military and NWS data subsets, the primary results of our comparisons with satellite data are similar for both. For example, as shown in Fig. 4, the trends in PATMOS-x DCD and particularly 1330 LT are closer to the station data trend than those in the ISCCP product, which are far closer than those in CLARA-A1 for both data subsets and the combined station data; similar performance is obtained for those satellite datasets on interannual time scales (Figs. 2 and 6; Tables 2, 3, and 5).

6. Summary

This paper evaluates variability and trends in total cloud cover spanning the period 1982–2009 across the contiguous United States from the ISCCP, PATMOS-x, and CLARA-A1 satellite datasets using homogeneity-adjusted weather station data as the ground truth.

Climatologically, relative to weather station data, ISCCP tends to have more cloud cover, particularly in the afternoon, while PATMOS-x tends to have less cloud cover in both 1330 LT and daily average products. CLARA-A1 has lower cloud amounts, particularly for nighttime data, in most areas of the country, but anomalously higher cloud cover values over the West and Southwest regions during daytime. Relative to station data, ISCCP tends to show less year-to-year variability while PATMOS-x tends to show higher variability and significantly higher variability is shown in CLARA-A1 on this scale.

On interannual time scales, the highest correlation between U.S. mean satellite data and station cloud cover data is shown in PATMOS-x DCD (r = 0.94); the next highest is in PATMOS-x 1330 LT, followed by ISCCP, and then the CLARA-A1 daytime dataset with the lowest correlation (r = 0.20). We find similar results when analyzing the correlation of DTR and precipitation days with cloud cover, except that PATMOS-x 1330 LT tends to have a higher correlation than PATMOS-x DCD does. This result for 1984–2007 is also consistent with that shown in the correlation of surface solar radiation data with cloud cover spanning the years 1995–2007, for which period station cloud cover data tend to have a better correlation with radiation data than satellite cloud datasets do.

The station dataset shows a negative but not statistically significant trend of −0.40% decade−1 (−0.64 for 1982–2009), and satellite products show larger downward trends ranging from −0.55 to −5.00% decade−1 for the U.S. mean for 1984–2007. PATMOS-x 1330 LT trends are closest to those in the station data, followed by PATMOS-x DCD and ISCCP, with CLARA-A1 products, particularly the daytime dataset, having a large negative trend contrasting strongly with the station data.

On most measures the CLARA-A1 product is an outlier over this particular region, with lowest correlations and much more negative trends, as noted previously in Karlsson et al. (2013). Results for other regions may be different. For example, the performance over Europe was found to be climatologically superior to ISCCP and compatible with MODIS cloud data (Kotarba 2015). This product is currently being reprocessed with an improved retrieval processing system and the next version may be more suitable for regional as well as global assessment of long-term cloud cover change.

Our station data are generally well correlated with the ISCCP and PATMOS-x data and with SURFRAD surface radiation data. These results tend to validate the usefulness of weather station cloud data for monitoring climate-scale changes in cloud cover, and show that the long-term stability of satellite cloud datasets can be assessed by comparison to homogeneity-adjusted station data and other physically related variables.

Acknowledgments

We thank William Brown at NCDC for providing U.S. weather station cloud data and Pasha Groisman at NCDC for providing daily precipitation data, and Hyun Kim for GEWEX ISCCP and PATMOS-x data access and processing. Dian Seidel is acknowledged for helpful comments. The GEWEX Cloud Assessment data were obtained from the ClimServ Data Center of IPSL/CNRS, and the CLARA-A1 dataset from the Satellite Application Facility on Climate Monitoring website (www.cmsaf.eu). This work was funded in part by NOAA’s Climate Program Office. The views, opinions, and findings contained in this report are those of the author(s) and should not be construed as an official National Oceanic and Atmospheric Administration or U.S. government position, policy, or decision.

REFERENCES

REFERENCES
Augustine
,
J. A.
,
G. B.
Hodges
,
C. R.
Cornwall
,
J. J.
Michalsky
, and
C. I.
Medina
,
2005
:
An update on SURFRAD—The GCOS surface radiation budget network for the continental United States
.
J. Atmos. Oceanic Technol.
,
22
,
1460
1472
, doi:.
Bergman
,
J. W.
, and
M. L.
Salby
,
1996
:
Diurnal variations of cloud cover and their relationship to climatological conditions
.
J. Climate
,
9
,
2802
2820
, doi:.
Campbell
,
G.
,
2004
: View angle dependence of cloudiness and the trend in ISCCP cloudiness. 13th Conf. on Satellite Meteorology and Oceanography, Norfolk, VA, Amer. Meteor. Soc., P6.7. [Available online at https://ams.confex.com/ams/pdfpapers/79041.pdf.]
Dai
,
A.
,
T. R.
Karl
,
B.
Sun
, and
K. E.
Trenberth
,
2006
:
Recent trends in cloudiness over the United States: A tale of monitoring inadequacies
.
Bull. Amer. Meteor. Soc.
,
87
,
597
606
, doi:.
Dutton
,
E. G.
,
D. W.
Nelson
,
R. S.
Stone
,
D. U.
Longenecker
,
G.
Carbaugh
,
J. M.
Harris
, and
J.
Wendell
,
2006
:
Decadal variations in surface solar irradiance as observed in a globally remote network
.
J. Geophys. Res.
,
111
,
D19101
, doi:.
Eastman
,
R.
, and
S. G.
Warren
,
2014
:
Diurnal cycles of cumulus, cumulonimbus, stratus, stratocumulus, and fog from surface observations over land and ocean
.
J. Climate
,
27
,
2386
2404
, doi:.
Eastman
,
R.
,
S. G.
Warren
, and
C. J.
Hahn
,
2014
: Climatic atlas of clouds over land and ocean. [Available online at www.atmos.washington.edu/CloudMap/index.html.]
Evan
,
A. T.
,
A. K.
Heidinger
, and
D. J.
Vimont
,
2007
:
Arguments against a physical long-term trend in global ISCCP cloud amounts
.
Geophys. Res. Lett.
,
34
,
L04701
, doi:.
Foster
,
M. J.
, and
A.
Heidinger
,
2013
:
PATMOS-x: Results from a diurnally corrected 30-yr satellite cloud climatology
.
J. Climate
,
26
, 414–425, doi:.
Foster
,
M. J.
, and
A.
Heidinger
,
2014
:
Entering the era of +30-year satellite cloud climatologies: A North American case study
.
J. Climate
,
27
,
6687
6697
, doi:.
Free
,
M.
, and
B.
Sun
,
2013
:
Time-varying biases in U.S. total cloud cover
.
J. Atmos. Oceanic Technol.
,
30
,
2838
2849
, doi:.
Free
,
M.
, and
B.
Sun
,
2014
: Trends in U.S. total cloud cover from a homogeneity-adjusted dataset. J. Climate,10, 4959–4969, doi:.
Groisman
,
P. Ya.
,
R. W.
Knight
,
T. R.
Karl
,
D. R.
Easterling
,
B.
Sun
, and
J. M.
Lawrimore
,
2004
:
Contemporary changes of the hydrological cycle over the contiguous United States: Trends derived from in situ observations
.
J. Hydrometeor.
,
5
,
64
85
, doi:.
Heidinger
,
A. K.
,
W. C.
Straka
,
C. C.
Molling
,
J. T.
Sullivan
, and
X. Q.
Wu
,
2010
:
Deriving an inter-sensor consistent calibration for the AVHRR solar reflectance data record
.
Int. J. Remote Sens.
,
31
,
6493
6517
, doi:.
Heidinger
,
A. K.
,
A. T.
Evan
,
M. J.
Foster
, and
A.
Walther
,
2012
:
A naïve Bayesian cloud-detection scheme derived from CALIPSO and applied within PATMOS-x
.
J. Appl. Meteor. Climatol.
,
51
,
1129
1144
, doi:.
Heidinger
,
A. K.
,
M. J.
Foster
,
A.
Walther
, and
X.
Zhao
,
2014
:
The Pathfinder Atmospheres–Extended (PATMOS-x) AVHRR climate dataset
.
Bull. Amer. Meteor. Soc.
,
95
,
909
922
, doi:.
Karlsson
,
K.-G.
, and Coauthors
,
2013
:
CLARA-A1: A cloud, albedo, and radiation dataset from 28 yr of global AVHRR
.
Atmos. Chem. Phys.
,
13
,
5351
5367
, doi:.
Kondragunta
,
C. R.
, and
A.
Gruber
,
1996
:
Seasonal and annual variability of the diurnal cycle of clouds
.
J. Geophys. Res.
,
101
,
21 377
21 390
, doi:.
Kotarba
,
A.
,
2015
:
Evaluation of ISCCP cloud amount with MODIS observations
.
Atmos. Res.
,
153
,
310
317
, doi:.
Lawrimore
,
J. H.
,
M. J.
Menne
,
B. E.
Gleason
,
C. N.
Williams
,
D. B.
Wuertz
,
R. S.
Vose
, and
J.
Rennie
,
2011
:
An overview of the Global Historical Climatology Network monthly mean temperature dataset, version 3
.
J. Geophys. Res.
,
116
,
D19121
, doi:.
Long
,
C. N.
,
E. G.
Dutton
,
J. A.
Augustine
,
W.
Wiscombe
,
M.
Wild
,
S. A.
McFarlane
, and
C. J.
Flynn
,
2009
:
Significant decadal brightening of downwelling shortwave in the continental United States
.
J. Geophys. Res.
,
114
,
D00D06
, doi:.
Rossow
,
W. B.
, and
R. A.
Schiffer
,
1999
:
Advances in understanding clouds from ISCCP
.
Bull. Amer. Meteor. Soc.
,
80
,
2261
2287
, doi:.
Shi
,
Y.
, and
C. N.
Long
,
2002
: Best estimate radiation flux value added product: Algorithm operational details and explanations. DOE Tech Rep. DOE/SC-ARM/TR-008, 51 pp., doi:.
Stubenrauch
,
C. J.
,
W. B.
Rossow
, and
S.
Kinne
,
2012
: Assessment of global cloud datasets from satellites: A project of the World Climate Research Programme Global Energy and Water Cycle Experiment (GEWEX) Radiation Panel. WCRP Rep. 23/2012, 176 pp. [Available online at www.wcrp-climate.org/documents/GEWEX_Cloud_Assessment_2012.pdf.]
Stubenrauch
,
C. J.
, and Coauthors
,
2013
:
Assessment of global cloud datasets from satellites: Project and database initiated by the GEWEX Radiation Panel
.
Bull. Amer. Meteor. Soc.
,
94
,
1031
1049
, doi:.
Sun
,
B.
,
2003
:
Cloudiness over the contiguous United States: Contemporary changes observed using ground-based and ISCCP D2 data
.
Geophys. Res. Lett.
,
30
,
1053
, doi:.
Sun
,
B.
,
P. Ya.
Groisman
, and
I. I.
Mokhov
,
2001
:
Recent changes in cloud-type frequency and inferred increases in convection over the United States and the former USSR
.
J. Climate
,
14
,
1864
1880
, doi:.
Warren
,
S. G.
,
R.
Eastman
, and
C.
Hahn
,
2007
:
A survey of changes in cloud cover and cloud types over land from surface observations: 1971–96
.
J. Climate
,
20
,
717
738
, doi:.
Wild
,
M.
, and Coauthors
,
2005
:
From dimming to brightening: Decadal changes in solar radiation at the Earth’s surface
.
Science
,
308
,
847
850
, doi:.
Wu
,
W.
,
Y.
Liu
,
M. P.
Jensen
,
T.
Toto
,
M. J.
Foster
, and
C. N.
Long
,
2014
:
A comparison of multiscale variations of decade-long cloud fractions from six different platforms over the Southern Great Plains in the United States
.
J. Geophys. Res. Atmos.
,
119
,
3438
3459
, doi:.