GEOPHYSICAL RESEARCH LETTERS, VOL. 40, 3220–3226, doi:10.1002/grl.50569, 2013
Linear contrail and contrail cirrus properties determined from satellite data Patrick Minnis,1 Sarah T. Bedka,2 David P. Duda,2 Kristopher M. Bedka,2 Thad Chee,2 J. Kirk Ayers,2 Rabindra Palikonda,2 Douglas A. Spangenberg,2 Konstantin V. Khlopenkov,2 and Robyn Boeke2 Received 22 April 2013; revised 15 May 2013; accepted 16 May 2013; published 19 June 2013.
[1] The properties of contrail cirrus clouds are retrieved through analysis of Terra and Aqua Moderate Resolution Imaging Spectroradiometer data for 21 cases of spreading linear contrails. For these cases, contrail cirrus enhanced the linear contrail coverage by factors of 2.4–7.6 depending on the contrail mask sensitivity. In dense air traffic areas, linear contrail detection sensitivity is apparently reduced when older contrails overlap and thus is likely diminished during the afternoon. The mean optical depths and effective particle sizes of the contrail cirrus were 2–3 times and 20% greater, respectively, than the corresponding values retrieved for the adjacent linear contrails. When contrails form below, in, or above existing cirrus clouds, the column cloud optical depth is increased and particle size is decreased. Thus, even without increased cirrus coverage, contrails will affect the radiation balance. These results should be valuable for refining model characterizations of contrail cirrus needed to fully assess the climate impacts of contrails. Citation: Minnis, P., S. T. Bedka, D. P. Duda, K. M. Bedka, T. Chee, J. K. Ayers, R. Palikonda, D. A. Spangenberg, K. V. Khlopenkov, and R. Boeke (2013), Linear contrail and contrail cirrus properties determined from satellite data, Geophys. Res. Lett., 40, 3220–3226, doi:10.1002/grl.50569.
1. Introduction [2] One of the outstanding uncertainties in our understanding of aviation effects on climate is the impact of aviation on cirrus clouds over and above that due to linear contrails [Lee et al., 2009; Burkhardt et al., 2010]. This additional impact is due to a combination of spreading contrails that are not recognized as such and increased ice nuclei in the upper troposphere due to aircraft soot emissions. The former effect is likely the most influential impact and has been examined using a variety of modeling and a few observational studies. As contrails age and spread, their properties and areal coverage change [e.g., Minnis et al., 1998], and consequently, the radiative effect will likely be different from that Additional supporting information may be found in the online version of this article. 1 Climate Sciences Branch, NASA Langley Research Center, Hampton, Virginia, USA. 2 Science Systems and Applications, Inc., Hampton, Virginia, USA. Corresponding author: P. Minnis, Climate Sciences Branch, NASA Langley Research Center, MS 420, 21 Langley Blvd., Hampton, VA 23681-2199, USA. (
[email protected]) ©2013. American Geophysical Union. All Rights Reserved. 0094-8276/13/10.1002/grl.50569
of linear contrails. To reduce the uncertainty in contrail cirrus climate effects, it is necessary to quantify both the coverage of spreading contrails and their microphysical and optical properties. [3] The additional coverage caused by spreading can be quantified as the spreading factor SF, which is the ratio of total cirrus coverage from contrails (linear contrail plus contrail cirrus) to the linear contrail (Con) coverage. Contrail cirrus (CC) is defined here as the cirrus cover originating from contrails but not identified as linear contrails. Duda et al. [2004] performed a case study of spreading contrails over the Great Lakes and found SF 2. Total cirrus coverage from contrails was determined by manual inspection of Geostationary Operational Environmental Satellite (GOES) imagery, while Con coverage was obtained by an application of the contrail detection algorithm of Mannstein et al. [1999] to both GOES and Advanced Very High Resolution Radiometer imagery. Minnis et al. [2004] estimated that SF = 1.8 based on trends in cirrus clouds and an estimate of 1990 global Con coverage. Burkhardt and Kärcher [2011] simulated the life cycle of contrails over the globe using a parameterization along with flight track information in a global climate model. They found that SF 9 based on the age of young contrails rather than on linear contrails per se, but their model also produced reductions in natural cirrus clouds, partially offsetting the aircraft-induced clouds. Schumann [2012] developed a parameterization to compute cirrus and contrail coverage as well as their microphysical properties using numerical weather prediction model data and flight track information. Using that model, Schumann and Graf [2013] found that the impact of contrails, both via spreading and forming in extant cirrus, on cirrus cover and optical depth results in an order of magnitude more radiative forcing compared to linear contrails alone. [4] Using a visible reflectance method applied to lowresolution satellite data, Duda et al. [2004] also determined that the effective particle diameter D of the Con and CC together decreased with contrail age while the optical depth t initially increased, then diminished after a few hours. Burkhardt and Kärcher [2011] assumed that D and t were the same for both Con and CC to estimate CC effects. Linear contrail microphysical properties are becoming fairly well defined as retrievals based on active [e.g., Iwabuchi et al., 2012] and passive infrared observations [e.g., Bedka et al., 2013] are providing converging results. However, the visible reflectance technique often overestimates t for optically thin cirrus clouds [e.g., Minnis et al., 2011] and is very sensitive to the particle habit used in the retrieval. Thus, the sparse information currently available on CC optical properties is not very reliable and more data are
3220
MINNIS ET AL.: CONTRAIL CIRRUS PROPERTIES FROM MODIS
Figure 1. Infrared (10.8 mm) images from GOES-13 over southeastern U.S. and adjacent Atlantic, 29 December 2010. Times are given in UTC. needed to guide and validate models, such as those of Burkhardt and Kärcher [2011] and Schumann [2012]. [5] To begin addressing the paucity of information regarding the properties of contrail cirrus, this paper determines the properties of both cirrus and contrails over selected areas where it has been determined visually that only contrails produced the extant cirrus clouds. An optimal methodology applying two different techniques is used to derive D and t from Moderate Resolution Imaging Spectroradiometer (MODIS) for both the Con and CC pixels. This approach minimizes the potentially large errors associated with the reflectance methods previously used [e.g., Minnis et al., 1998]. Additionally, the contrail and cirrus coverage is determined so that a value for SF is computed for all cases.
2. Data and Methodology [6] The data sets used here include 4 km 10.8 mm images from GOES-13 and 1 km 0.64, 3.8, 10.8, and 12 mm radiances from MODIS on the Terra and Aqua satellites. Atmospheric and surface properties are the same as those used by Bedka et al. [2013]. [7] A set of 11 CC outbreaks was selected in the following manner. Sequences of GOES infrared imagery over North America were visually examined in conjunction with MODIS 10.8–12 mm brightness temperature differences (BTD) to determine areas where contrails formed and grew in otherwise clear air. Furthermore, areas were excluded from analysis if “natural” cirrus appeared to develop within or advect into the region. This necessarily subjective selection process does not always eliminate all noncontrail-induced cirrus cloudiness and bright background low-level clouds. For the selected days, an analysis region was defined for each of the available MODIS images containing the scene. The analysis regions change in each image because the areal extent of the Con and CC varies, the subject scene moves between images, the viewing zenith angles differ for each image, and ambient cloud cover may interfere with parts of the contrail-only area. The case selection process is discussed further in the supporting information of this paper.
[8] Figure 1 shows a developing CC scene in GOES-13 images taken 29 December 2010. By 13:15 UTC, a few small contrails have formed off over northern Florida and the adjacent Atlantic. The contrails continue forming and spreading at 14:15 UTC and an hour later cover parts of southern Georgia as well. At 16:15 UTC (Figure 1d), new contrails appear to have developed over both land and ocean as the older contrails spread and move eastward. This continues through at least 18:15 UTC. The area containing the Con and CC is seen clearly in the 16:30 UTC Terra 1 km BTD image, shown in Figure 2a. This image, an example of a large outbreak of a long-lived CC system, clearly shows the overlapping of new and old contrails resulting in nearly overcast skies off the coast. The system began dissipating as it moved eastward and was mostly gone by 23:00 UTC. To be more representative of the full range of CC, the selected cases include both small and large outbreaks. Table 1 lists the cases, with the latitudes and longitudes that encompass the areas used in the analyses for each satellite. The regions were selected to ensure a relatively uniform and cloud-free background. An attempt was made to track the same set of contrails for both Terra and Aqua, but because of changes in the background, it was not always possible to include all of the same contrails in each Terra and Aqua pair. The percentages of each area determined to be Con using mask C and CC (both described below) are included along with total number of pixels in the analysis domain. The nominal pixel resolution is 1 km, but increases with viewing zenith angle, which differs from image to image. [9] After defining the analysis regions, the method of Duda et al. [2013] was applied to the MODIS data to determine the linear contrail pixels at three different levels of sensitivity. The most conservative mask, A, detects the lowest number of linear contrail pixels, but has the fewest false contrail detections. Mask B detects more contrails at the expense of increasing false detections. It is assumed to be the best estimate of linear contrails, as it includes fewer very wide contrails that could be contrail cirrus. Mask C is the most sensitive mask and detects the most linear contrails and picks up some contrail cirrus coverage, but tends to misclassify
3221
MINNIS ET AL.: CONTRAIL CIRRUS PROPERTIES FROM MODIS
Figure 2. Terra MODIS (a) 10.8–12 mm brightness temperature differences and (b) contrail optical depths for mask C, 16:30 UTC, 29 December 2010. Red box indicates analysis region. many linear ambient cirrus clouds as contrails. All three masks are considered because the definitions of both linear contrails and contrail cirrus depend on the mask. [10] The contrail properties, tCon and DCon, were retrieved using the infrared bispectral (10.8 and 12 mm) technique
(IBT), which selects background radiances for each pixel by averaging the radiances for the surrounding noncontrail pixels. The IBT, described by Bedka et al. [2013], assumes a contrail temperature and iteratively solves for tCon and DCon. Figure 2b shows the retrievals of tCon for mask C applied to the Terra MODIS data taken 15 min after the case in Figure 1d. Many of the individual contrails outside the analysis region (red box) are detected, while in the region, 23.9% of the pixels are classified as contrails, yielding, on average, tCon = 0.208 0.156 and DCon = 29.7 18.1 mm. It is evident, however, that the detection algorithm misses a large amount of CC. [11] To estimate the properties and coverage of the missed CC for all cases, the following procedure is used. All pixels in a given region are classified as clear or cloudy using the approach of Minnis et al. [2008], and the cloudy pixels are then analyzed with the visible infrared shortwave-infrared split-window technique (VISST) [Minnis et al., 2011] to retrieve tCC, DCC, and cloud temperature TCC. Given the propensity of visible channel techniques to overestimate t for thin cirrus clouds, tCC and DCC are also retrieved for the CC using the IBT with the assumption that TCC is equal to the contrail temperature TCon and the background temperature is the clear-sky radiance field employed by the VISST. To further minimize the impact of the low-level clouds on the results, all pixels having tCC > 3 from either technique were eliminated from further analysis. These two retrievals provide a range of values. Contrail cirrus likely has a greater radiating temperature than the original contrails because contrails tend to spread by diffusion, precipitation, and vertical wind shear [e.g., Jensen et al., 1998], which drop the radiating center to a lower altitude. Thus, assuming that TCC = TCon could likely underestimate both tCC and DCC. Using the two methods together should constrain the ranges of t and D for the CC. [12] A best estimate of the CC properties was determined using the following procedure. The IBT result is used if DCC < 70 mm, because the differences in BTD between DCC = 70 and 130 mm are small compared to the uncertainty in the clear-sky background for small tCC. In those instances, the IBT retrieval usually tends toward the largest value and is
Table 1. Cases Used in Contrail Cirrus Analysis Case # 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Date
Time (UTC)
Latitude Range ( N)
Longitude Range ( W)
Satellite
Mask C (%)
CC (%)
Ntot
21 Jan 2006
16:45 20:00 16:50 20:05 17:45 20:55 16:45 18:20 18:25 20:00 15:50 19:05 17:15 18:55 17:05 18:40 19:15 20:50 14:45 16:30 18:05
32.0–37.0 33.2–37.7 34.3–38.0 33.1–37.1 22.9–26.0 22.6–26.5 30.2–32.6 30.5–32.7 25.7–29.3 26.2–28.5 34.8–36.5 34.5–37.5 35.9–39.9 35.0–40.7 35.5–39.0 37.8–40.5 32.5–36.5 33.4–36.2 33.0–35.1 28.3–32.5 27.9–32.2
92.0–100.0 90.0–98.0 94.0–96.5 91.4–94.1 101.6–108.6 102.4–105.6 82.4–87.0 80.9–85.2 106.6–111.1 105.5–110.9 67.8–81.6 76.5–79.9 91.7–99.3 94.5–97.5 93.6–92.0 85.0–91.7 113.3–116.1 111.5–115.1 28.0–32.0 78.3–80.3 78.6–80.2
Terra Aqua Terra Aqua Terra Aqua Terra Aqua Terra Aqua Terra Aqua Terra Aqua Terra Aqua Terra Aqua Aqua Terra Aqua
18.3 8.8 7.2 3.6 10.7 4.8 8.4 4.7 8.3 5.8 9.2 16.1 6.0 14.9 7.1 16.7 13.1 12.8 8.0 23.9 14.9
13.3 41.6 5.4 15.2 4.7 23.4 20.5 21.8 9.6 9.1 27.7 64.0 3.4 5.8 4.1 5.0 8.1 2.2 34.8 41.4 46.1
209,792 133,485 60,561 26,156 226,715 35,802 76,343 57,943 90,298 62,534 192,610 28,517 270,067 45,087 199,666 107,563 35,362 93,777 48,951 24,755 34,646
13 Feb 2006 4 Nov 2006 5 Nov 2006
6 Nov 2006 26 Dec 2006 28 Dec 2006 31 Dec 2006 7 Sep 2010 29 Dec 2010
3222
MINNIS ET AL.: CONTRAIL CIRRUS PROPERTIES FROM MODIS
likely to be highly uncertain. If DCC from the IBT exceeds 70 mm, then the VISST result is examined. In those cases, the VISST results replace the IBT results if the VISST phase is ice and |TCC TCon| < 10 K. This test is used on the basis that the VISST retrieval is likely to be accurate if the retrieved temperature (height) is close to that of the contrail. Overestimates of t by the VISST will always underestimate the cloud height and overestimate TCC. The 10 K window is selected because the actual contrail temperature could differ from the assumed TCon value and the CC radiating temperature is likely to be greater than TCon, as noted earlier. A 10 K difference is roughly equivalent to a 1.5 km altitude difference, which is a conservative constraint given the potential uncertainty in TCon [Bedka et al., 2013] and the likelihood of contrail cirrus to extend more than 1 km below the original contrail height [e.g., Heymsfield et al., 1998]. Thus, CC are all ice cloud pixels in the domain having tCC < 3 and not classified as contrails. Thus, some of the pixels defined as contrails using mask C can be considered as contrail cirrus using mask A or B. As seen in Table 1, the CC coverage is sometimes less than the contrail coverage.
3. Results and Discussion [13] Figure 3 shows an example of the analyses overlaid on the Aqua MODIS BTD image taken at 18:05 UTC for the case in Figure 1, more than 1.5 h after the Terra image (Figure 2). To minimize the uncertainties associated with the background radiance fields, the analysis region was restricted to open water east of the coast as indicated by the red box, similar to that in Figure 2. The BTD and retrievals of tCon for mask C are shown in Figures 3a and 3d, respectively, to illustrate the maximum linear contrail coverage and the contrail properties. The pixels for masks A and B are subsets of mask C and use the mask C background radiances. The retrieval results among the masks differ slightly because they use different numbers of pixels in the averaging. For this case, the contrail coverages are 4.6, 7.3, and 14.9% for masks A, B, and C, respectively. The corresponding averages for tCon are 0.165 0.128, 0.147 0.120, and 0.130 0.111 and for DCon (not shown) are 33.2, 33.5, and 33.7 mm. The tCon means are statistically different from each other at the 99% level, while the DCon values are not statistically different. For this case, TCon = 220.5 K for all three masks. The linear contrail coverage is less than that seen earlier in the Terra image (Figure 2), while tCon from the mask C data is substantially smaller and DCon is 4 mm greater than the Terra counterparts. The smaller optical depths may be due to some saturation of air traffic levels with contrails, while DCon could be larger due to increased mixing with the greater amount of extant contrail cirrus, which is characterized by larger particles than the contrails, as seen below. Some of this difference may be due to the use of slightly different analysis regions. [14] The values of DCC determined using the IBT (Figure 3b) and VISST (Figure 3c) are shown as colors for values up to 70 mm and as raw BTD values for greater values. Blocky areas in the VISST retrieval arise from errors in the background radiances. Generally, DCC(VISST) is less than DCC(IBT). In the lower part of the box, where overlapped contrails are sparse, DCC(IBT) is typically less than 30 mm, but exceeds 60 mm for many of the pixels near the top of the box. Conversely, DCC(VISST) < 45 mm for a majority
of the pixels. Remarkably, the VISST optical depths (Figure 3f) are quite similar to their IBT counterparts (Figure 3e). For this case, the average values of tCC(IBT) and DCC(IBT) are 0.387 and 56.1 mm, respectively, and tCC (VISST) and DCC(VISST) are 0.361 and 47.9 mm. After applying the filtering process described in the previous section, some of the IBT values were replaced with their VISST counterparts to obtain the best estimates of 0.388 and 46.0 mm for tCC and DCC, respectively, and CC = 48%. The best estimate of DCC for the Terra data in Figure 2 is 41.4 mm, while tCC = 0.306 and CC = 40.1%. Thus, all of the parameters except contrail coverage and tCon increased relative to the Terra results. The differences between the Terra and Aqua values of tCon, DCon, tCC, and DCC are all statistically different, having p-values < 10 8 that indicate there is essentially no probability that they are the same. All of the above averages of tCC and DCC were computed including only contrail cirrus and excluding linear contrails identified by mask C. In this case, both the CC optical depth and particle size are considerably greater than their linear contrail counterparts. [15] The results for all 21 cases were averaged to estimate the properties for the linear contrails and CC accompanying each mask. Two different approaches were used to compute the CC properties, the first only includes pixels having a best estimate DCC < 70 mm, while the other includes all pixels having a retrieval. The percentage of CC pixels having DCC < 70 mm is denoted as FD
