1. Introduction
Substantial atmospheric radiation and moisture impacts accompany jet air transportation (Grassl 1990; Brasseur et al. 1998; Wuebbles et al. 2007; Lee et al. 2009). These impacts result from the emission of gases and perturbation of clouds by particles (carbon dioxide, nitrogen oxides, water vapor, ozone, sulfur oxides, soot, and hydrocarbons) from fuel combustion (e.g., Gettelman and Chen 2013; Liou et al. 2013). The generation of linear, long-lasting ice-crystal clouds—condensation trails, or contrails—that evolve into contrail cirrus, could be the most significant atmospheric effect of commercial aviation (Burkhardt and Kärcher 2011; Burkhardt et al. 2010). Persisting contrails on time scales of 1–6 h (Minnis et al. 1998) are likely when cold ambient air is supersaturated with respect to ice (e.g., Appleman 1953; Schrader 1997; Sausen et al. 1998; Minnis et al. 2003; Schumann 2005; Immler et al. 2008; Irvine et al. 2014a). They potentially have an impact on climate (e.g., Iwabuchi et al. 2012; Spangenberg et al. 2013) upon spreading laterally and thinning, thereby extending the natural cirrus cover (Schumann and Wendling 1990; Gothe and Grassl 1993; Sassen 1997; Meerkötter et al. 1999; Kristensson et al. 2000; Schröder et al. 2000; Zerefos et al. 2003; Atlas et al. 2006; Burkhardt and Kärcher 2009; Haywood et al. 2009; Minnis et al. 2013). Moreover, persisting contrails often occur in clusters or outbreaks in otherwise clear or mostly clear skies; these episodes are referred to as “clear-sky outbreaks” (Duda et al. 2001, 2004; Carleton et al. 2008). Given a minimum threshold of higher-altitude jet-flight frequencies in a given region (Palikonda et al. 2002; Duda et al. 2005; Travis et al. 2007), clear-sky outbreaks manifest contrail-favored areas (CFAs) of upper-tropospheric (UT) meteorological conditions (e.g., temperature, humidity, vertical motion, and wind) (Pilié and Jiusto 1958; Scorer and Davenport 1970; Jensen et al. 1998; Kästner et al. 1999; Stuefer et al. 2005; Duda and Minnis 2009b) that extend over approximately 104–105 km2 (Carleton and Lamb 1986; Travis et al. 2004; Carleton et al. 2008; Carleton and Travis 2013). In terms of climate, CFAs and outbreaks occur often over northern midlatitude regions having the greatest frequencies of high-altitude flights: the contiguous United States (CONUS), Europe, and the North Atlantic Ocean (e.g., Duda et al. 2013). There, subregion-scale impacts of contrails have been implicated in trends and variations of surface temperature, cloud cover, sunshine duration, the insolation surface receipt, and outgoing longwave radiation (Changnon 1981; Lee and Johnson 1985; Liou 1992; Liou et al. 1990; Petersen et al. 1995; Liepert 1997; Boucher 1999; Nakanishi et al. 2001; Minnis et al. 2004; Stordal et al. 2005; Stubenrauch and Schumann 2005; Graf et al. 2012; Carleton et al. 2013; Schumann and Graf 2013; Magee et al. 2014; Wu et al. 2014).
The climatic impact of persisting contrails and clear-sky outbreaks derives from their associated generally positive net radiative forcing (Schumann et al. 2012; Yi et al. 2012), especially at night and over low-temperature surfaces (Bakan et al. 1994; Myhre and Stordal 2001; Stuber et al. 2006), despite their ability to reduce the incident insolation surface receipt (Kuhn 1970; Mims and Travis 1997; Meyer et al. 2002; Rädel and Shine 2008). These effects may be manifest as subregion-scale reductions in the diurnal temperature range that have occurred since the widespread adoption of jet aviation (e.g., Travis et al. 2002, 2004; Ryan et al. 2011; Bernhardt and Carleton 2015). Moreover, the positive net radiative impact of persisting contrails and contrail cirrus results in a slight warming overall (Solomon et al. 2007) and may therefore accentuate contemporary climate change due to greenhouse gas emissions (Lee et al. 2009). This possibility now makes undesirable the intentional perturbation of climate by contrails (cf. Nicodemus and McQuigg 1969; Detwiler and Pratt 1984; Minnis et al. 1999; Wuebbles et al. 2010; Muri et al. 2014).
Various technological advances and flight operational procedures—either separately or in combination—have been proposed to help to ameliorate the climatic impacts of persisting contrails and contrail cirrus (e.g., Schrader 1997; Busen and Schumann 1995; Schumann 2000; Schumann et al. 1996; Williams et al. 2003; Fichter et al. 2005; Irvine et al. 2012). A related strategy, assumed to be the most preferable for this paper, mimics that employed to avoid severe convection or clear-air turbulence: flying around a CFA, or lateral rerouting of planes (cf. Lane et al. 2012; Sheth et al. 2013). For this purpose, the ability to forecast CFAs in near–real time (1–6 h) is essential to reducing contrail incidence (Irvine et al. 2014b), especially at night when reduced flight frequencies better permit such adjustments (Newinger and Burkhardt 2012).
Contrail forecasting originated from the need to maintain aircraft stealth during military operations, and current methods comprise both statistical and physical modeling (e.g., Moss 1999; Stuefer et al. 2005; Schumann 2012). Because contrail occurrence is binary (i.e., yes or no), statistical methods to predict outbreaks for particular CONUS subregions or specific locations include logistic regression (“logit”) modeling (Newton et al. 1997; Travis et al. 1997; Jackson et al. 2001; Duda and Minnis 2009a,b) and discriminant analysis (Bjornson 1992), utilizing UT variables. Most contrail-prediction methods rely on detecting critical thresholds of UT temperature and humidity, or Schmidt–Appleman criteria (e.g., Peters 1993; Hanson and Hanson 1995, 1998; Schumann 1996). These thresholds may be transient or occur on relatively small scales (e.g., Duda et al. 2004).
The UT humidity (UTH) at 250 and 200 hPa (i.e., around jet cruising altitudes) is particularly important for determining contrail incidence and persistence; yet, reliable measurements on this variable frequently may not be available (e.g., Elliott and Gaffen 1991; Garand et al. 1992). Model-generated analyses of UTH may not always match the contrail-cirrus cloudiness (Ovarlez et al. 2000; Palikonda et al. 2005), likely because of the requirement for ice supersaturation (e.g., Gierens et al. 2000; Burkhardt et al. 2008). As an alternative, information on jet-fuel usage as a proxy for air-traffic density, engine efficiency, or flight frequency can be combined with indices of the UT cloud-generating potential—typically moisture and temperature (Sausen et al. 1998; Duda et al. 2005). This method is enhanced when contrail observations are available on daily or subdaily time scales, particularly on high-resolution satellite imagery (e.g., Mannstein et al. 1999; DeGrand et al. 2000; Minnis et al. 2005). Because most contrail-prediction methods pertain to specific locations, they may not be readily applicable to the larger subregions of the CONUS that typically experience clear-sky outbreaks, and their associated synoptic atmospheric features. Accordingly, local-scale contrail prediction in a busy operational forecasting environment could benefit from a relatively simple method that assesses the larger-scale atmospheric suitability for CFAs on a given day (i.e., “first look” approach), whereupon more objective statistical or physical models could be employed for smaller areas (e.g., expected contrail distributions relative to flight frequencies and aircraft types) or to yield a higher temporal resolution (Schumann 2012).
Prior to the possibility for application of the synoptic attributes of CFAs to their near-real-time prediction for wide areas, it is necessary that past outbreak events be accurately hindcast. Thus, detailed knowledge of CONUS subregions where clear-sky outbreaks have occurred, their timing, and their typical atmospheric conditions on subregional to regional scales (i.e., synoptic climatology) is important both for attributing the role of contrails in the local- to regional-scale surface temperature change supplementary to the array of climate forcings (e.g., Dai et al. 1999; Pielke 2003) and for determining the spatial extent and longevity of CFAs (Travis and Carleton 2005).
In pursuit of this ultimate goal, we develop a CFA visual (manual) approach for hindcasting that combines a satellite-based spatial inventory of daily contrail outbreaks for CONUS subregions in the 2000–02 midseason months (i.e., calibration period) with reanalyses of UT meteorological variables whose daily-average anomalies are shown to be related to persisting contrails and outbreaks. This method is tested on daily-level satellite images and UT reanalyses in subregions of higher-frequency outbreaks (“primary subregions”) for the 2008–09 midseason months (verification period). We emphasize primary subregions because we can assume a sufficient threshold of high-altitude flights to generate outbreaks, when UT meteorological conditions are favorable (e.g., Travis et al. 2007). For midseason months in both calibration and verification periods, hindcast success in the primary subregion—including the statistical significance of the UT variables both individually and combined—is assessed by using skill scores and by logit modeling. Moreover, we evaluate in a preliminary way the possible influences of analyst experience and academic background (an atmospheric scientist vs an earth scientist) and the influence of using daily-average versus 6-h UT analyses on the CFA hindcast success rates. This study builds upon Carleton et al. (2013), who demonstrated, for subregions of the CONUS having typically high frequencies of contrails, the utility of synoptic-map composites (i.e., geographic gridpoint means for multiple nonadjacent days) of UT and tropopause-level meteorological variables to depict the atmospheric environments—and their between-year (2008 vs 2009) variations—within which contrail outbreaks occur.
2. Data and their analysis
a. Satellite-based spatial inventories (“climatologies”) of clear-sky outbreaks
We utilize spatial inventories of outbreaks developed for the CONUS in midseason months (January, April, July, and October) of 2000–02 (Travis et al. 2007; Carleton et al. 2008) and 2008–09 (Carleton et al. 2013). These inventories comprise daily outbreak frequencies at 1° × 1° latitude/longitude resolution, derived from manual interpretation of thermal infrared (IR) high-resolution (1.1 km2) satellite images of the Advanced Very High Resolution Radiometer (AVHRR) that were available online approximately 5–8 times per day per location (http://www.class.ngdc.noaa.gov). A clear-sky outbreak (e.g., Fig. 1) is defined as a minimum of three contrails occurring simultaneously over at least three adjacent 1° grid cells (in any direction) and where any accompanying “natural” cloudiness does not exceed 50% (i.e., the contrail cloudiness composes at least 50% of the total cloud cover), determined subjectively for each image (Travis et al. 2007). The top-left and bottom-right coordinate pairs of each outbreak, in degrees of latitude and longitude, define a bounding box completely enclosing the contrail outbreak; that is, it includes contrail cloudiness, natural cloud, and clear pixels (cf. Duda et al. 2001).
AVHRR thermal IR image of a “clear-sky contrail outbreak” (see text) centered on Kentucky on 14 Oct 2002. Natural cloud—mostly cirrus and cirrostratus—is evident on the southern edge of the outbreak. A smaller outbreak is located on the KY–TN border, in the bottom-right corner of the image. Tick marks indicate the 1° × 1° latitude/longitude graticule.
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
In determining the CONUS subregions of higher versus lower outbreak frequencies for a given midseason month in 2000–02, Carleton et al. (2013) developed a GIS-based metric of the normalized frequency by which every 1° × 1° latitude/longitude CONUS grid cell (total n = 900) registered an outbreak bounding box in a given time period (a single midseason month, a given midseason month aggregated for all 3 yr of the calibration period, etc.). The resulting database of overlapping (in time and space) outbreak frequencies—or “overlaps”— is a satellite climatological metric of clear-sky outbreaks. An overlap value gives the frequency with which a given 1° × 1° latitude/longitude grid cell sees clear-sky outbreaks for a given month or other time period. It is somewhat analogous climatologically to Stuefer et al.’s (2005) mean overlap contrail forecast skill parameter, by which overlapping contrail layers from a mesoscale model are related to radiosonde data. Our overlap distributions form the basis of a semiobjective regionalization for the CONUS by midseason month, with areas of higher overlap frequency—summarized in Fig. 2—constituting distinct subregions that retain the essential latitude/longitude coordinates (i.e., four-sided shape) of the outbreak bounding boxes (Carleton et al. 2013). Most outbreak-overlap subregions occur in the eastern half of the CONUS and coincide with higher frequencies of jet flights above approximately 7-km altitude (Palikonda et al. 2005). They are coherent on horizontal length scales of at least 1 × 104 km2. The midwestern CONUS subregion is strongly represented in all midseason months except July, during which an eastern subregion (“East”) dominates (Fig. 2).
Higher-frequency subregions of contrail outbreaks, as determined using the overlap metric (see text) for the calibration period (2000–02). Line weights differ by midseason month, as follow: Midwest and South in January (both continuous thick), Midwest and Pacific South in April (both continuous thin), East in July (dashed), and Midwest–upper South in October (dotted–dashed). [Adapted from Carleton et al. (2013, their Fig. 3).]
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
b. Mapped fields of UT meteorological variables for CONUS subregions
We allied the daily AVHRR-derived outbreak-overlap data with daily-average (i.e., 4 × 6 h) data of the NCEP–NCAR reanalyses (NNR), gridded at 2.5° × 2.5° spatial resolution (Kalnay et al. 1996; Kistler et al. 2001). Midseason-month map composites of NNR UT variables were formed separately for 2000–02 and 2008/09. As in Carleton et al. (2013), we emphasize four variables—the air temperature T, relative humidity (RH), vertical motion ω (=dp/dt), and east–west wind component u—that have been shown to depart appreciably from their climatological values in the presence of contrails and outbreaks (Appleman 1953; Schumann 1996; Moss 1999; Schrader 1997; Stuefer et al. 2005; Carleton et al. 2008); here, all are at 300 hPa because this is the highest level for moisture measurements (i.e., RH) that is represented in the NNR. We use RH because of its demonstrated spatial coherence (Dee and da Silva 2003) and its dependence on temperature (Hanson and Hanson 1998). Given the absence of ice supersaturation values in the NNR dataset, higher values of RH(300) (e.g., >80%) can be assumed to increase the likelihood of the ice supersaturation (e.g., Atlas and Wang 2010) that is typically associated with persisting contrails.
For each subregion and midseason-month combination in 2000–02, two sets of NNR UT composites were formed using the Daily Mean Composites Internet page of the NOAA/Physical Sciences Division/Earth System Research Laboratory (http://www.cdc.noaa.gov/data/composites/day): one that comprises all contrail-outbreak (CON) days that were identified in the satellite-image spatial inventory and one that comprises nonoutbreak (NON) days. The NON composite excluded the day before any outbreak date, or PRE (=CON − 1 day), to reduce the possibility of including UT transitional (from NON to CON) environments on these daily-average scales. Indeed, composites formed from the PRE days for the higher-frequency outbreak subregions (not shown) mostly confirm their transitional and even mixed character (Silva 2009, his appendix A). Also, to better resolve the threshold conditions that permit outbreaks and to reduce the potential for significant temporal autocorrelation of the UT data, no two consecutive NON dates were included in the composite. To consolidate the number of output maps and to better depict visually the relationships between UT variables, the composite T(300) and RH(300) compose one map, with the composite ω(300) and u(300) composing a second map, for each subregion.
c. Outbreak-hindcast visual method
Binary (i.e., yes/no) hindcasts of CFAs on individual daily-average NNR maps [i.e., T(300), RH(300), ω(300), and u(300)] for 2000–02 involved, in the first instance, an atmospheric scientist (coauthor Silva) visually deeming that they resemble either the respective CON composite or the respective NON composite for that subregion and midseason month (Fig. 3). To highlight contrasts between CON and NON composites, difference (DIFF) maps (=CON − NON) were also available for each variable by CONUS subregion and midseason month in the calibration period (2000–02) (see Carleton et al. 2013, their Figs. 4–7). Each daily average UT map was qualitatively assessed (yes/no) for CFAs using three standard map criteria—magnitude, pattern, and gradient—for three decisions on each map per day. The magnitude criterion involves a given UT variable’s (e.g., maximum or minimum) mean value within the subregion, pattern is the arrangement of high-to-low values (e.g., location of maxima and minima and directional orientation of isolines) within the subregion, and gradient is the horizontal change in values, given by the isoline spacing.
Schematic of the decision process accompanying the visual method that determines if (Y/N = yes or no) an individual daily-average UT map (by subregion and midseason month in Fig. 2) resembles the CON composite developed for the calibration period or resembles more the NON composite (see text). The DIFF composite was consulted when there were small differences in magnitude between CON and NON composites.
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
To be deemed favorable for a contrail outbreak (i.e., yes), a daily mean NNR map had to more closely resemble the composite CON map than the NON composite for that region or midseason month. In many cases these decisions were not rigidly separate (the broken lines for the CON ellipse shown in Fig. 3). For example, a given map pattern may suggest similarity to the CON composite, but the gradient may be too weak, forcing that day’s map to be relegated to the NON group. On days for which the differences from the CON composites were small, final evaluation was assisted by comparison with the DIFF composites because these tended to highlight the areas of stronger CON-minus-NON differences (Carleton et al. 2013, their Figs. 4–7).
d. Statistical analysis of contrail-outbreak hindcasts
1) Contingency tables and skill measures
The ability of a single UT variable [T(300), RH(300), ω(300), or u(300)] to hindcast a contrail outbreak (observation category) gives four possible combinations in a contingency table, as follows (Wilks 1995, chapter 7.2; Walters et al. 2000): hit (both are yes), miss (hindcast is no but observation is yes), correct negative (both are no), and false alarm (hindcast is yes but observation is no). For the calibration period, 24 such contingency tables were created: one for each UT variable (four) for each of the six subregions (Silva 2009, 72–74). The contingency-table data were first evaluated using chi-square (χ2) analysis, which identifies the UT variable(s) that are statistically significantly associated with contrail outbreaks for a given subregion and midseason month, by comparison to an estimate of their relationship as a function of chance (Stockburger 1996). At the 0.05 significance level, the critical value of χ2 with one degree of freedom is 3.84.
Determining the success rates of daily contrail-outbreak hindcasts (by subregion and midseason month) for the calibration period utilizes variants of the skill-score technique (cf. Mason 2004; Weigel et al. 2007), applied to the contingency-table cell values (e.g., Hanson and Hanson 1998; Moss 1999; Jackson et al. 2001). Four skill measures that are typically used to determine success rates of meteorological forecasts (Wilks 1995, chapter 7.2; Walters et al. 2000) are compared with a verifying analysis—here, contrail clear-sky outbreaks as identified from the AVHRR inventory—as follows: 1) the “hit rate” H, or accuracy of predicting an outbreak occurrence or nonoccurrence from the fraction of correct hindcasts (no success = 0; maximum success = 1), 2) the critical success index (CSI), or “threat score,” which measures the success of only the correctly hindcast occurrences (hit, or both = yes) and is equal to 1 for a perfect forecast, 3) the probability-of-detection (POD) fraction, which represents the accuracy of a hindcast for instances in which the event is observed (Wilks 1995, chapter 7.2.2; POD considers only outbreak days, with the best possible value being 1 and no skill being 0), and 4) the false-alarm rate (FAR), or fraction of hindcast contrail outbreaks for which the events do not occur. Unlike the other skill measures, the best possible FAR is 0 and the worst FAR is 1.
In addition, the bias ratio B gives the correspondence between the frequency of yes predictions and the frequency of yes observations (Wilks 1995, chapter 7.2.3). An unbiased dataset has B = 1.0 because the number of predictions and observations is the same; when B > 1.0, overprediction occurs, and B < 1.0 indicates underprediction.
2) Binary logistic regression
To determine which single UT variable best hindcast CFAs in a given subregion/midseason month, a binary logistic regression was applied to the CFA dataset (i.e., that deeming the daily mean UT maps more similar to the CON composites). Logistic regression is often used to determine the relationship between a discrete binary response (yes/no)—here, contrail outbreaks—and predictor variables (Wilks 1995, chapter 6.3.2; Travis et al. 1997; Garson 2014; Duda and Minnis 2009a,b), which here are the four UT variables. Thus, for each day in the calibration dataset, five categories were input to the statistical model; 1 represents favorable for outbreaks and 0 represents unfavorable.
The output gave maximum likelihood estimates for each UT variable and gave cross-product terms for the two-way interaction model, as follows: RH × ω, RH × T, RH × u, ω × T, ω × u, and T × u. We then derived parameter estimates, standard errors, and the Wald chi-square statistic, the last being a more conservative χ2 test. To reject the null hypothesis requires probability p < 0.05.
3. Results and discussion
a. Calibration period composites (CON and NON) by subregion and midseason month
Figures 4–8 show the UT map composites for CON and NON days in the higher-frequency outbreak-overlap subregions and midseason months of 2000–02. Note that we show only those subregions/months that our statistical analyses (sections 3b and 3c; Tables 1–4) reveal as having at least one significant UT variable and map criterion; this excludes the South in January (not significant). Because we plot two variables per composite (section 2b), each map shows the significant predictor variable and one other variable—either T(300) and RH(300), or ω(300) and u(300). There are strong differences in pattern, and often also in horizontal gradients, between CON days and NON days that were shown to be particularly apparent in the DIFF composites (Carleton et al. 2013, their Figs. 4–7). For the Midwest in January (Fig. 2), the ω(300) pattern (Fig. 4) comprises a centroid of negative values (i.e., rising air) on CON days, contrasting with a more zonal pattern of sinking air on NON days, especially to the north. The co-occurrence of relatively high values of RH (not shown) and negative ω (i.e., ascendance) over the subregion is consistent with the implied synoptic-scale dynamics; rising air accompanying upper-level divergence is associated with a moistening upper troposphere.
Composite maps of ω(300) and u wind for the Midwest subregion in January of the calibration period (Fig. 2) for (a) CON days (n = 16) and (b) NON days (n = 34). Isolines of ω(300) are colored, with 0.01 Pa s−1 contour interval; isolines of u(300) are black, with 1.0 m s−1 contour interval.
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
Similar to Fig. 4, but for RH(300) and T(300) for the Midwest subregion in April for (a) CON days (n = 29) and (b) NON days (n = 25). Isolines of RH300 are black, with 1.0% contour interval; isolines of T(300) are colored, with 0.5°C contour interval.
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
As in Fig. 4, but for the Pacific South subregion in April for (a) CON (n = 8) and (b) NON (n = 40).
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
As in Fig. 5, but for the East subregion in July for (a) CON (n = 24) and (b) NON (n = 30).
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
As in Fig. 5, but for the Midwest–upper South subregion in October for (a) CON (n = 27) and (b) NON (n = 29).
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
Summary of statistically significant [p < 0.05 level (two-tailed test)] UT variable(s) and map criteria by midseason month and CONUS high-outbreak-frequency subregion, identified from the contingency table and χ2 analysis for the calibration period (2000–02).
Contingency table, along with accuracy measures (skill scores: H, CSI, POD, and FAR) and bias B values for outbreak hindcasts in the calibration period (2000–02), by significant UT variable(s) for midseason month and subregion (Table 1). For the period, the number of clear-sky outbreaks (vs no outbreaks, in parentheses) is given by subregion in the header row.
Outbreak hindcast significant results (calibration period, 2000–02) for UT variables determined using binary logistic regression (maximum likelihood estimates), by midseason month and CONUS subregion: single variable.
Outbreak hindcast significant results (calibration period, 2000–02) for UT variables determined using binary logistic regression (maximum likelihood estimates), by midseason month and CONUS subregion: two-way interactions.
For the broadly collocated Midwest subregion in April (Fig. 2), the RH(300) pattern on CON days (Fig. 5a) has increasing values to the north, in contrast to the NON composite (Fig. 5b). Also in April, only for the Pacific South subregion (Fig. 2) on CON days (Fig. 6a), the ω(300) pattern indicates strong subsidence to the east and ascendance of air to the northwest. On NON days (Fig. 6b), there is general subsidence, especially to the southwest, and relaxed spatial gradients.
For the East in July (Fig. 2), the composite RH pattern on CON days (Fig. 7a) shows moister air over the mid-South states but drier air to the north and south, in contrast to the generally meridional pattern of RH decreasing from east to west on NON days (Fig. 7b). Moreover, RH(300) gradients are considerably steeper on CON days than on NON days. Also on CON days (Fig. 7a), the upper troposphere is colder to the northeast than on NON days (Fig. 7b).
For the Midwest–upper South in October (Fig. 2), the RH(300) pattern of moister air to the west on CON days (Fig. 8a) contrasts with the generally decreasing values to the west and northwest on NON days (Fig. 8b). Thus, the map composites for CONUS subregions having UT variable(s) that are significant predictors of contrail outbreaks (Figs. 4–8; sections 3b and 3c) generally reveal large and coherent differences between CON and NON days, especially in the patterns of RH(300) and ω(300).
b. Single-variable hindcast accuracy for the calibration period
The χ2 analysis of contingency-table data (by midseason month and subregion; Table 1) shows every subregion of high outbreak frequency to have one associated statistically significant variable, except for the South in January, which has none, and the East (July) and Pacific South (April), each having two. In assessing variables individually, it is seen that no midseason month–subregion combination shows u(300) to be a significant predictor of contrail outbreaks and that the RH(300) pattern is statistically significant in four of the six subregions (section 3a). The importance of UTH in contrail prediction is consistent with previous studies.
Skill scores (i.e., H, CSI, POD, and FAR) for statistically significant variables in the calibration period (Table 1) are compared in Table 2. Despite considerable variability of hindcast success, given by the range of values on the accuracy measures among subregions and midseason months, the following general statements may be made. Overall, the hindcast skill can be characterized as modest. Frequencies in the “success” categories (i.e., yes/yes and no/no) tend to exceed the unsuccessful categories (yes/no and no/yes). The exception is T(300) for the East in July. Also, bias values tend to be lower for significant UT variables in a given subregion–month combination than for the nonsignificant variables (not shown). The exception again is T(300) for the East (July), when it is the highest in that month and subregion. “Trade offs” in skill scores are also evident. For example, in the Pacific South (April), using the ω(300) map pattern, H (=0.86) is the highest of the four UT variables across all subregions and midseason months yet the CSI (=0.28) indicates that most of these correct hindcasts were for nonoccurrences of outbreaks. In a similar way, for the East subregion (July), when the RH(300) pattern and T(300) magnitude are each significantly associated with contrail outbreaks in the calibration period, RH(300) more correctly predicted outbreak nonoccurrences than occurrences (CSI = 0.33), with fewer than one-half of all outbreaks correctly hindcast (POD = 0.46). These results (Table 2) indicate that no single UT variable or no two variables considered separately reliably predict CFAs for the calibration period.
The binary logistic regression analysis for single variables in the calibration period (Table 3) confirms the results of Tables 1 and 2 yet adds clarification. Although RH(300) pattern is the dominant variable and map criterion, its sole use for the Midwest subregion in April gives marginal results (the Wald χ2 p values). Also, the T(300) magnitude for the Pacific South in April (Tables 1 and 2) is not significant. Conversely, for the East subregion in July the logistic regression retains both UT variables and map criteria [RH(300) pattern; T(300) magnitude] as statistically significant individual predictors of contrail outbreaks for 2000–02 (cf. Tables 2 and 3), although RH(300) has a substantially higher probability of being nonrandom. No subregion studied has u(300) as a useful predictor on its own.
Although generally supporting the visual interpretation of a single UT variable and map criterion keyed to a specific subregion and midseason month (2000–02)—except the South in January—the above results suggest that it is appropriate to examine whether combinations of variables may improve confidence in the hindcasts.
c. Two-way UT variable interactions for the calibration period
Table 4 suggests that hindcasts can be improved in certain regions/midseason months by including additional variables beyond those shown in Tables 2 and 3. Thus, for the South (January)—shown to have no single UT variable as a significant outbreak predictor—the most noteworthy interaction term is the 300-hPa ω × T (p < 0.055). Although slightly above the 0.05 critical probability, this value can be considered to be statistically significant because test levels usually increase when additional terms are added to the model. Accordingly, ω(300) pattern and T(300) magnitude are both used in the verification (2008–09) for the South [section 3d(2)]. For the East in July (Table 4), RH × T is significant (p < 0.048), as are RH(300) and T(300) individually (Table 3), suggesting potential use of either RH(300) or T(300) to hindcast CFAs there on a given day. The other statistically significant interaction is T × u for the Midwest–upper South in October (Table 4). The fact that the single-variable analysis for this subregion and midseason month (Tables 2 and 3) selects RH(300) pattern as significant suggests the use of either 1) RH(300) or 2) the T(300) gradient and u(300) magnitude together for the verification study (October 2008). For the other subregions and midseason months (i.e., Midwest in January and April; Pacific South in April), outbreak hindcasting is not improved by including two-variable interactions. Accordingly, the significant single-variable predictors are used in the verification analysis for April 2009 (section 3d).
d. Verification-period outbreak hindcasts
For the verification-period midseason months (January and April of 2009 and July and October of 2008), the visual method of determining CFAs was the same as that used for the calibration period except that only the UT maps that were shown to be statistically significant—either individually or in combination—were interpreted (Tables 1–5), as follows: ω(300) pattern for the Midwest for January, RH(300) pattern for the Midwest and ω(300) pattern for the Pacific South for April, either RH(300) pattern or T(300) magnitude for the East for July, and either RH(300) pattern or both T(300) gradient and u(300) magnitude for the Midwest–upper South for October. The daily assessments were compared with an independently derived outbreak inventory, as was done for the calibration period. As shown by Carleton et al. (2013, their Fig. 9), the assumption that the outbreak high-frequency regions change little spatially between the same midseason month in different years is reasonable, although there can be substantial variations in the intensity (i.e., normalized frequency of outbreak occurrence) for a given subregion [section 3d(2), below].
Outbreak hindcasts for verification period (2008–09) by significant variable (Table 1), midseason month, and CONUS subregion (cf. with Table 2; the χ2 statistics could not be computed for the contingency tables because of the small values in some cells). The italicized values after the slash in each cell and in the header row pertain to the analysis in section 3d(3). For the period, the number of clear-sky outbreaks (vs no outbreaks, in parentheses) is given by subregion in the header row.
1) Skill scores
As with the calibration period, hindcasts of contrail outbreaks for the South for January of 2009 were unsuccessful (Table 5). By contrast, hindcast accuracy for the Midwest (January 2009)—using ω(300) pattern—was successful when judged by the H and POD, but the CSI (=0.30) and FAR (=0.63) are poor. For the broadly similar Midwest subregion in April of 2009 (Table 5), contrail outbreaks hindcast using the RH300 pattern performed better (e.g., H = 0.70 and FAR = 0.29).
For the Pacific South in April of 2009, using ω(300) pattern (Table 1), hindcast skill was mixed (Table 5); the high H (=0.8) and relatively low FAR (=0.33) contrast with low values of CSI (=0.25) and POD (=0.29). For the East (July of 2008), relying upon interpretation of either RH(300) pattern or T(300) magnitude, the combined H (=0.52) and POD (=0.57) meant that approximately one-half of all outbreaks were correctly hindcast (Table 5). Conversely, only approximately one of every three outbreaks was hindcast on those days (CSI = 0.35; FAR = 0.53). These weaker July results may reflect the reduced association of outbreaks with synoptic-scale UT fields in the summer and their closer association with the cirriform edges of deep-convection masses (e.g., DeGrand et al. 2000).
For the Midwest–upper South (October of 2008), outbreak hindcasts that were based on visual assessment of either RH300 pattern or combining T(300) gradient and u(300) magnitude gave broadly similar results to those of the East in July of 2008 (Table 5). Given these mixed results for the verification period, utilizing only the statistically significant UT map variables (usually one per midseason month), we investigated whether including nonsignificant UT variables could have successfully hindcast contrail outbreaks. Although not helpful for most subregions, in the South (January of 2009) the values of CSI, POD, and FAR all improved substantially when one used either ω(300) or T(300), as opposed to both variables in tandem (not shown).
A number of possible reasons exist for the mostly modest—albeit statistically significant—skill scores, reasons that are combinations of the data and methods used. These potentially include 1) the relatively small sample sizes of outbreaks, despite our emphasis on those CONUS subregions typically seeing the higher frequencies of contrail-outbreak overlaps, 2) the subjectivity inherent in the visual method, whereby any two analysts might differently assess the similarity of given UT maps to CON and NON composites, 3) our use of daily average NNR maps, which would, for example, obscure a UT humid area moving with the wind and entering or leaving a subregion on time scales of a few hours, and 4) the possibility that assigning the UT circulation associated with clear-sky outbreaks only to the CON category may mask the influence of different synoptic types or patterns.
In regard to limitation 1, greater confidence in the results requires the eventual development of a longer-term contrail-outbreak spatial synoptic climatology for the CONUS using the same satellite-image-based method of deriving clear-sky outbreak overlaps and better establishing their temporal variations. For limitations 2 and 3, we undertake separate preliminary assessments of their possible impacts on the skill score results, as described below. We gain insights into the potential influence of limitation 4 through a more detailed analysis of the outbreak-hindcast problem for the South subregion in January [section 3d(4)].
2) Sensitivity to analyst background and experience
To assess, in a preliminary way, the possible influence of analyst experience and academic background on CFA hindcast success using the visual method (Tables 2–4), coauthor VanderBerg—an earth scientist with training in geologic map interpretation but no previous experience in atmospheric science—subsequently and independently repeated coauthor Silva’s evaluation of outbreak favorability for the region of highest outbreak frequency in each midseason month of the verification period (i.e., Midwest in January and April of 2009, East in July of 2008, Midwest–upper South in October of 2008, and the South in January of 2009). The results (italicized in Table 5) show broad agreement between both scientists: averaged over the five subregions and months the skill scores balance out, with the atmospheric scientist having a slightly higher H and slightly lower FAR than the earth scientist but with almost identical CSI and POD. Therefore, we infer that the relatively modest skill scores we obtain are not primarily a function of the subjectivity inherent in the visual hindcast-map method.
3) Sensitivity to temporal resolution of UT reanalyses
To evaluate the possibility that the temporal resolution of the reanalyses—NNR daily average versus the 6-h maps nearest in time (≤3 h) to the satellite observation of each outbreak—affects the hindcast success rates, we analyzed two additional months in the verification period (South for January of 2008 and Midwest for April of 2008). Table 6 suggests that use of 6-h maps tends to improve the H value but also increases FAR. The CSI and POD values show no consistent change. Thus, we conclude that increasing the temporal resolution of the visual retrodiction method to the nearest 6-h UT map likely would not appreciably improve hindcast skill using the visual map method.
Comparison of accuracy measures and bias for contrail outbreak hindcasts using daily-average (first cell value) vs nearest 6-h (italicized cell value after the slash) NNR maps: Midwest (April 2009 and April 2008) and South (January 2008).
4) The problem of outbreak prediction for the South
As shown, the visual hindcast method struggled particularly for the South subregion in January, in both the calibration and verification periods. Plausible reasons for this weak performance include 1) the importance of (an)other UT variable not considered in the analysis and/or 2) the use of a single pressure level (300 hPa) rather than a UT layer, including the possibility that 300 hPa represents too low of an altitude given the more equatorward location of this subregion; also, 3) the UT map visual interpretation method may not distinguish important meteorological features associated with outbreaks in the South in January, especially if 4) the UT variations occur on scales that are smaller than those captured in the NNR synoptic-scale data. Observational evidence supporting one or more of these hypotheses includes the following: 1) January outbreaks in the South are associated frequently with the subtropical jet stream (STJ) rather than the baroclinic waves that typically accompany outbreaks in the Midwest (DeGrand et al. 2000; Carleton et al. 2008), 2) the climatological wind profile associated with the STJ in this subregion and month peaks between 250 and 200 hPa rather than at 300 hPa (Carleton et al. 2013, their Table 4), 3) at least for longer-lived (i.e., >4-h duration) outbreaks occurring in this subregion in January of 2009 (n = 9), the vertical shear of the wind in the 300–100-hPa layer and the horizontal gradient of the specific humidity at 300 hPa were both statistically significantly greater than their respective long-term (1979–2010) synoptic climatologies, determined using the higher-spatiotemporal-resolution North American Regional Reanalysis (NARR) dataset (Mesinger et al. 2006), and 4) these longer-lived outbreaks were spread across three distinct circulation categories at 500 hPa: within a trough axis, east of a ridge, and east of a trough. Thus, the ω and RH composite maps may each represent a mix of synoptic patterns.
To try to improve hindcasts of clear-sky outbreaks in the South (January), it is appropriate to more fully characterize their UT synoptic climatology, going beyond that disclosed by mapped composites at a single atmospheric level. Accordingly, we examine outbreak occurrences in that subregion for an additional winter midmonth (January of 2008). The South exhibits considerable interannual variability in outbreak-overlap frequencies, as demonstrated by maps of their normalized departures for January of 2008 and January of 2009 (Figs. 9a,b). Given that we are comparing individual months with a very short period “climatology” (2000–02), it is not appropriate to generate statistical significances of the difference patterns, but grid cells containing the largest differences of either sign are likely important. The identification of typical variations for outbreaks, and the statistical significance of individual-month departures for the South, in particular would benefit from a longer-term synoptic climatology of clear-sky outbreak overlaps.
Maps of contrail-outbreak overlaps for the wider southern CONUS region in (a) January 2008 and (b) January 2009. Grid-cell values depict normalized departures from the January 2000–02 means. (c) Difference map of normalized grid-cell values of contrail-outbreak overlaps for January 2008 − January 2009.
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
The daily synoptic analysis of UT variables (NARR dataset) for all longer-duration (i.e., >4 h) outbreaks in the South in January of 2008 (n = 15) and 2009 (n = 9) includes the vertical shear of u wind for the 300–100-hPa layer. The results (Table 7) show that three of the eight 500-hPa synoptic types identified by DeGrand et al. (2000) were sufficient to categorize the outbreaks in both January months: a trough axis (category 3), east of a ridge (category 5), and east of a trough (category 6). These were represented in equal proportions in January of 2009, but category 6 was 3 times as frequent in January of 2008. Accordingly, while warm (W) and cold (C) advection in the 300–100-hPa layer occurred about equally in January of 2009, the frequency of W was 2 times that of C in January of 2008. Moreover, in both January months the enhanced horizontal gradient anomaly of 300-hPa specific humidity [SH(300)] was statistically significant, as was the stronger UT u vertical shear (Table 7), suggesting their possible use in improving prediction of contrail outbreaks for the South in winter. The horizontal gradient anomaly of T(300) was not significant in either January. Because a UT vertical wind shear anomaly accompanies an anomaly of the layer-averaged temperature (i.e., thickness), we expect a steepened lapse rate in the upper troposphere, particularly between 300 and 200 hPa. We derived composite anomaly (from 1979–2010 means) vertical profiles of u and T at standard levels in the mid- and upper troposphere for all longer-duration outbreaks (i.e., January 2008 plus 2009; n = 24), using the spatially nearest NARR 3-h sounding. Those composites (Figs. 10a,b) support this association: the greater UT horizontal and vertical wind shear associated with the STJ increases the likelihood that persisting contrails will spread laterally (e.g., Travis 1996; Jensen et al. 1998; Kästner et al. 1999). Thus, reliable hindcasting of clear-sky contrail outbreaks in the South for midwinter using the visual method should consider the vertical shear anomaly as well as the map pattern RH(300) (Table 5), and possibly also the UT thickness gradient.
Synoptic analysis of longer-duration (>4 h) contrail outbreaks for the South subregion (Fig. 2) in January 2008 (n = 15) and January 2009 (n = 9). Subtotals of n are given in boldface, and averages are given in lightface. The low and high ends of the range are given in italics. Synoptic types at 500 hPa follow DeGrand et al. (2000); i.e., type 3 = trough axis; type 5 = east of ridge; and type 6 = east of trough. The departure of the absolute value of vertical shear from the long-term (1979–2010) normal (Δ|SHEAR| = S; m s−1) and the sign of the temperature advection (C = cold; W = warm) pertain to the 300–100-hPa layer. Departures of the horizontal gradients of temperature (ΔT-grad) and specific humidity (ΔSH-grad) from the long-term normals at 300 hPa are computed across the extent of each outbreak. A single asterisk indicates statistical significance at the p < 0.05 level; two asterisks indicate significance at the p < 0.01 level.
Composite (a) u-wind (m s−1) and (b) air temperature (°C) anomaly vertical profiles (500–100 hPa) of the spatially nearest NARR 3-h sounding accompanying each of 24 longer-duration (>4 h) contrail outbreaks in the South subregion (Fig. 2) for January 2008 and 2009. Anomalies are with respect to the 1979–2010 means.
Citation: Journal of Applied Meteorology and Climatology 54, 8; 10.1175/JAMC-D-14-0186.1
4. Summary and concluding remarks
This study addressed the question of the extent to which a synoptic-climatological approach can contribute to the prediction of persisting contrails and clear-sky outbreaks on subregion scales for the CONUS. Reliable prediction of contrails is an objective for sustainable aviation given their radiative—and surface climate—impacts. Before subregion-scale operational forecasts of clear-sky contrail outbreaks for the CONUS can reasonably be attempted, as a basis for potentially rerouting planes around CFAs, it is important to show skill in hindcasting outbreak occurrences. To this end, we presented a subjective visual map-based method (pattern, magnitude, and gradient) to hindcast CFAs on a daily basis by utilizing the midseason UT composites (2000–02) for CONUS subregions having frequent outbreaks and testing it on corresponding months in the 2008–09 period. The frequencies of outbreaks hindcast and observed in satellite imagery in the study periods were evaluated statistically by subregion and midseason month/year to determine which UT variable(s) performed better than chance, using contingency-table and χ2 analysis, skill scores, and binary logistic regression (logit modeling). The last approach was applied separately to each UT variable and also considered two-variable interactions. The χ2 tests and binary logit model identified similar UT synoptic variables associated with contrail outbreaks in each higher-frequency subregion and midseason month, supporting their evaluation for the verification period.
The dominant outbreak-related variables in both the calibration and verification periods are the map patterns of RH(300) and ω(300). This result makes physical sense because increased UT humidity [here, higher composite RH(300) values] is required for persisting contrails and outbreaks and is also enhanced where air is rising (i.e., −ω). In addition, interseasonal and subregional differences are associated with the relative influences of UT variables on outbreak occurrence, such as the apparently additional importance of T(300) (magnitude) for the East in July and the combination of T(300) and u(300) for the Midwest–upper South in October. In considering individual and two-variable interactions, more skillful hindcasts were obtained for the Midwest (January and April), and the Pacific South (April); these skill levels can still be considered “modest,”however. In other subregions, the skill was poor. Outbreak hindcasts consistently underperformed in the South (January), although including the vertical shear of u wind for 300–100 hPa captured the unique influence of the STJ there.
Preliminary evaluations of the possible impacts of analyst experience and academic training and of using the daily-average versus 6-h UT analyses in the hindcasts suggests that these influences may be relatively minor, at least for the midseason months/years analyzed here. Accordingly, we conclude that improvements in the visual map method to accurately hindcast and predict CFAs—identified not just as statistically significant but also as reliable predictors—will likely require the increased sample sizes of clear-sky outbreaks resulting from eventual development of a longer-term satellite-based synoptic climatology. Moreover, as we demonstrated for the South subregion, improved outbreak predictability for other CONUS subregions might also benefit from composites that are formed on the basis of synoptic situations or types, beyond the binary CON versus NON options.
The subjective hindcast method ultimately could provide a useful larger-scale first look at CFAs, whereupon objective statistical or physically based models of contrail prediction could then be applied (e.g., Duda and Minnis 2009a,b; Schumann 2012). Moreover, its extension to include all other months of the year obviously is desirable. The eventual advance identification of CFAs may allow flight-plan alterations to reduce persisting contrail incidence, thus limiting the climatic impacts of future potential outbreaks.
This research was supported by NSF Grants 0099011 and 0099014 (Travis et al. 2007; Carleton et al. 2008) as well as 0819396 and 0819416. Armand Silva was partially supported by a Pennsylvania State University (PSU) Bunton–Waller Graduate Fellowship. The assistance of PSU graduate students Matt Aghazarian and Adam Naito is appreciated. Images derived from the NNR data (Kalnay et al. 1996) were provided online by the NOAA/ESRL/Physical Sciences Division, Boulder, Colorado (http://www.esrl.noaa.gov/psd). We are grateful to the PSU Department of Meteorology for making available the NARR dataset. The Dean of PSU’s College of Earth and Mineral Sciences (Bill Easterling) provided partial support for this article’s page charges. The revised manuscript benefited greatly from detailed and insightful comments from the four reviewers.
REFERENCES
Appleman, H., 1953: The formation of exhaust condensation trails by jet aircraft. Bull. Amer. Meteor. Soc., 34, 14–20.
Atlas, D., , and Z. Wang, 2010: Contrails of small and very large optical depth. J. Atmos. Sci., 67, 3065–3073, doi:10.1175/2010JAS3403.1.
Atlas, D., , Z. Wang, , and D. P. Duda, 2006: Contrails to cirrus—Morphology, microphysics, and radiative properties. J. Appl. Meteor. Climatol., 45, 5–19, doi:10.1175/JAM2325.1.
Bakan, S., , M. Betancor, , V. Gayler, , and H. Grassl, 1994: Contrail frequency over Europe from NOAA-satellite images. Ann. Geophys., 12, 962–968, doi:10.1007/s00585-994-0962-y.
Bernhardt, J., , and A. M. Carleton, 2015: The impacts of long-lived jet contrail ‘outbreaks’ on surface station diurnal temperature range. Int. J. Climatol., doi:10.1002/joc.4303, in press.
Bjornson, B. M., 1992: SAC contrail formation study. U.S. Air Force Environmental Technical Applications Center Rep. USAFETAC/PR-92/003, 40 pp. [Available from USAF Environmental Technical Applications Center, Scott AFB, IL 62225.]
Boucher, O., 1999: Air traffic may increase cirrus cloudiness. Nature, 397, 30–31, doi:10.1038/16169.
Brasseur, G. P., and Coauthors, 1998: European scientific assessment of the atmospheric effects of aircraft emissions. Atmos. Environ., 32, 2329–2418, doi:10.1016/S1352-2310(97)00486-X.
Burkhardt, U., , and B. Kärcher, 2009: Process-based simulation of contrail cirrus in a global climate model. J. Geophys. Res., 114, D16201, doi:10.1029/2008JD011491.
Burkhardt, U., , and B. Kärcher, 2011: Global radiative forcing from contrail cirrus. Nat. Climate Change, 1, 54–58, doi:10.1038/nclimate1068.
Burkhardt, U., , B. Kärcher, , M. Ponater, , K. Gierens, , and A. Gettelman, 2008: Contrail cirrus supporting areas in model and observations. Geophys. Res. Lett., 35, L16808, doi:10.1029/2008GL034056.
Burkhardt, U., , B. Kärcher, , and U. Schumann, 2010: Global modeling of the contrail and contrail cirrus climate impact. Bull. Amer. Meteor. Soc., 91, 479–483, doi:10.1175/2009BAMS2656.1.
Busen, R., , and U. Schumann, 1995: Visible contrail formation from fuels with different sulfur contents. Geophys. Res. Lett., 22, 1357–1360, doi:10.1029/95GL01312.
Carleton, A. M., , and P. J. Lamb, 1986: Jet contrails and cirrus cloud: A feasibility study employing high-resolution satellite imagery. Bull. Amer. Meteor. Soc., 67, 301–309, doi:10.1175/1520-0477(1986)067<0301:JCACCA>2.0.CO;2.
Carleton, A. M., , and D. J. Travis, 2013: Aviation-contrail impacts on climate and climate change: A ready-to-wear research mantle for geographers. Prof. Geogr., 65, 421–432, doi:10.1080/00330124.2012.697795.
Carleton, A. M., , D. J. Travis, , K. Master, , and S. Vezhapparambu, 2008: Composite atmospheric environments of jet contrail outbreaks for the United States. J. Appl. Meteor. Climatol., 47, 641–667, doi:10.1175/2007JAMC1481.1.
Carleton, A. M., , A. D. Silva, , M. S. Aghazarian, , J. Bernhardt, , D. J. Travis, , and J. Allard, 2013: Mid-season climate diagnostics of jet contrail ‘outbreaks’ and implications for eastern U.S. sky-cover trends. Climate Res., 56, 209–230, doi:10.3354/cr01148.
Changnon, S. A., 1981: Midwestern cloud, sunshine and temperature trends since 1901: Possible evidence of jet contrail effects. J. Appl. Meteor., 20, 496–508, doi:10.1175/1520-0450(1981)020<0496:MCSATT>2.0.CO;2.
Dai, A., , K. E. Trenberth, , and T. R. Karl, 1999: Effects of clouds, soil moisture, precipitation, and water vapor on diurnal temperature range. J. Climate, 12, 2451–2473, doi:10.1175/1520-0442(1999)012<2451:EOCSMP>2.0.CO;2.
Dee, D. P., , and A. M. Da Silva, 2003: The choice of variable for atmospheric moisture analysis. Mon. Wea. Rev., 131, 155–171, doi:10.1175/1520-0493(2003)131<0155:TCOVFA>2.0.CO;2.
DeGrand, J. Q., , A. M. Carleton, , D. J. Travis, , and P. J. Lamb, 2000: A satellite-based climatic description of jet aircraft contrails and associations with atmospheric conditions, 1977–79. J. Appl. Meteor., 39, 1434–1459, doi:10.1175/1520-0450(2000)039<1434:ASBCDO>2.0.CO;2.
Detwiler, A., , and R. Pratt, 1984: Clear-air seeding: Opportunities and strategies. J. Wea. Modif., 16, 46–60.
Duda, D. P., , and P. Minnis, 2009a: Basic diagnosis and prediction of persistent contrail occurrence using high-resolution numerical weather analyses/forecasts and logistic regression. Part I: Effects of random error. J. Appl. Meteor. Climatol., 48, 1780–1789, doi:10.1175/2009JAMC2056.1.
Duda, D. P., , and P. Minnis, 2009b: Basic diagnosis and prediction of persistent contrail occurrence using high-resolution numerical weather analyses/forecasts and logistic regression. Part II: Evaluation of sample models. J. Appl. Meteor. Climatol., 48, 1790–1802, doi:10.1175/2009JAMC2057.1.
Duda, D. P., , P. Minnis, , and L. Nguyen, 2001: Estimates of cloud radiative forcing in contrail clusters using GOES imagery. J. Geophys. Res., 106, 4927–4937, doi:10.1029/2000JD900393.
Duda, D. P., , P. Minnis, , L. Nguyen, , and R. Palikonda, 2004: A case study of the development of contrail clusters over the Great Lakes. J. Atmos. Sci., 61, 1132–1146, doi:10.1175/1520-0469(2004)061<1132:ACSOTD>2.0.CO;2.
Duda, D. P., , P. Minnis, , and R. Palikonda, 2005: Estimated contrail frequency and coverage over the contiguous United States from numerical weather prediction analyses and flight track data. Meteor. Z., 14, 537–548, doi:10.1127/0941-2948/2005/0050.
Duda, D. P., , P. Minnis, , K. Khlopenkov, , T. L. Chee, , and R. Boeke, 2013: Estimation of 2006 Northern Hemisphere contrail coverage using MODIS data. Geophys. Res. Lett., 40, 612–617, doi:10.1002/grl.50097.
Elliott, W. P., , and D. J. Gaffen, 1991: On the utility of radiosonde humidity archives for climate studies. Bull. Amer. Meteor. Soc., 72, 1507–1520, doi:10.1175/1520-0477(1991)072<1507:OTUORH>2.0.CO;2.
Fichter, C., , S. Marquart, , R. Sausen, , and D. S. Lee, 2005: The impact of cruise altitude on contrails and related radiative forcing. Meteor. Z., 14, 563–572, doi:10.1127/0941-2948/2005/0048.
Garand, L., , C. Grassotti, , J. Halle, , and G. L. Klein, 1992: On differences in radiosonde humidity-reporting practices and their implications for numerical weather prediction and remote sensing. Bull. Amer. Meteor. Soc., 73, 1417–1423, doi:10.1175/1520-0477(1992)073<1417:ODIRHR>2.0.CO;2.
Garson, D. G., 2014: Logistic Regression: Binary & Multinomial. Statistical Associates Publishers, 224 pp. [Available online at http://www.statisticalassociates.com/logistic.htm.]
Gettelman, A., , and C. Chen, 2013: The climate impact of aviation aerosols. Geophys. Res. Lett., 40, 2785–2789, doi:10.1002/grl.50520.
Gierens, K., , U. Schumann, , M. Helten, , H. Smit, , and P.-H. Wang, 2000: Ice-supersaturated regions and subvisual cirrus in the northern midlatitude upper troposphere. J. Geophys. Res., 105, 22 743–22 753, doi:10.1029/2000JD900341.
Gothe, M. B., , and H. Grassl, 1993: Satellite remote sensing of the optical depth and mean crystal size of thin cirrus and contrails. Theor. Appl. Climatol., 48, 101–113, doi:10.1007/BF00864917.
Graf, K., , U. Schumann, , H. Mannstein, , and B. Mayer, 2012: Aviation induced diurnal North Atlantic cirrus cover cycle. Geophys. Res. Lett., 39, L16804, doi:10.1029/2012GL052590.
Grassl, H., 1990: Possible climatic effects of contrails and additional water vapour. Air Traffic and the Environment—Background, Tendencies and Potential Global Atmospheric Effects, U. Schumann, Ed., Springer-Verlag, 124–137.
Hanson, H. M., , and D. M. Hanson, 1995: A reexamination of the formation of exhaust condensation trails by jet aircraft. J. Appl. Meteor., 34, 2400–2405, doi:10.1175/1520-0450(1995)034<2400:AROTFO>2.0.CO;2.
Hanson, H. M., , and D. M. Hanson, 1998: The application of the revised algorithm for the prediction of the formation of exhaust condensation trails by jet aircraft. J. Appl. Meteor., 37, 436–440, doi:10.1175/1520-0450(1998)037<0436:TAOTRA>2.0.CO;2.
Haywood, J. M., and Coauthors, 2009: A case study of the radiative forcing of persistent contrails evolving into contrail-induced cirrus. J. Geophys. Res., 114, D24201, doi:10.1029/2009JD012650.
Immler, F., , R. Treffeisen, , D. Engelbart, , K. Krüger, , and O. Schrems, 2008: Cirrus, contrails, and ice supersaturated regions in high pressure systems at northern mid latitudes. Atmos. Chem. Phys., 8, 1689–1699, doi:10.5194/acp-8-1689-2008.
Irvine, E. A., , B. J. Hoskins, , and K. P. Shine, 2012: The dependence of contrail formation on the weather pattern and altitude in the North Atlantic. Geophys. Res. Lett., 39, L12802, doi:10.1029/2012GL051909.
Irvine, E. A., , B. J. Hoskins, , and K. P. Shine, 2014a: A Lagrangian analysis of ice-supersaturated air over the North Atlantic. J. Geophys. Res. Atmos., 119, 90–100, doi:10.1002/2013JD020251.
Irvine, E. A., , B. J. Hoskins, , and K. P. Shine, 2014b: A simple framework for assessing the trade-off between the climate impact of aviation carbon dioxide emissions and contrails for a single flight. Environ. Res. Lett., 9, doi:10.1088/1748-9326/9/6/064021.
Iwabuchi, H., , P. Yang, , K. N. Liou, , and P. Minnis, 2012: Physical and optical properties of persistent contrails: Climatology and interpretation. J. Geophys. Res., 117, D06215, doi:10.1029/2011JD017020.
Jackson, A., , B. Newton, , D. Hahn, , and A. Bussey, 2001: Statistical contrail forecasting. J. Appl. Meteor., 40, 269–279, doi:10.1175/1520-0450(2001)040<0269:SCF>2.0.CO;2.
Jensen, E. J., , A. S. Ackerman, , D. E. Stevens, , O. B. Toon, , and P. Minnis, 1998: Spreading and growth of contrails in a sheared environment. J. Geophys. Res., 103, 31 557–31 567, doi:10.1029/98JD02594.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kästner, M., , R. Meyer, , and P. Wendling, 1999: Influence of weather conditions on the distribution of persistent contrails. Meteor. Appl., 6, 261–271, doi:10.1017/S1350482799001231.
Kistler, R., and Coauthors, 2001: The NCEP/NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82, 247–267, doi:10.1175/1520-0477(2001)082<0247:TNNYRM>2.3.CO;2.
Kristensson, A., , J.-F. Gayet, , J. Strom, , and F. Auriol, 2000: In situ observations of a reduction in effective crystal diameter in cirrus clouds near flight corridors. Geophys. Res. Lett., 27, 681–684, doi:10.1029/1999GL010934.
Kuhn, P. M., 1970: Airborne observations of contrail effects on the thermal radiation budget. J. Atmos. Sci., 27, 937–942, doi:10.1175/1520-0469(1970)027<0937:AOOCEO>2.0.CO;2.
Lane, T. P., , R. D. Sharman, , S. B. Trier, , R. G. Fovell, , and J. K. Williams, 2012: Recent advances in the understanding of near-cloud turbulence. Bull. Amer. Meteor. Soc., 93, 499–515, doi:10.1175/BAMS-D-11-00062.1.
Lee, D. S., , D. W. Fahey, , P. M. Forster, , P. J. Newton, , R. C. N. Wit, , L. L. Lim, , B. Owen, , and R. Sausen, 2009: Aviation and global climate change in the 21st century. Atmos. Environ., 43, 3520–3537, doi:10.1016/j.atmosenv.2009.04.024.
Lee, J. E., , and S. D. Johnson, 1985: Expectancy of cloudless photographic days in the contiguous United States. Photogramm. Eng. Remote Sens., 51, 1883–1891.
Liepert, B. G., 1997: Recent changes in solar radiation under cloudy conditions in Germany. Int. J. Climatol., 17, 1581–1593, doi:10.1002/(SICI)1097-0088(19971130)17:14<1581::AID-JOC214>3.0.CO;2-H.
Liou, K.-N., 1992: Radiation and Cloud Processes in the Atmosphere: Theory, Observation, and Modeling. Oxford University Press, 487 pp.
Liou, K.-N., , S. C. Ou, , and G. Koenig, 1990: An investigation on the climatic effect of contrail cirrus. Air Traffic and the Environment—Background, Tendencies and Potential Global Atmospheric Effects, U. Schumann, Ed., Springer Verlag, 154–169.
Liou, K.-N., , Y. Takano, , Q. Yue, , and P. Yang, 2013: On the radiative forcing of contrail cirrus contaminated by black carbon. Geophys. Res. Lett., 40, 778–784, doi:10.1002/grl.50110.
Magee, N. B., , E. Melaas, , P. M. Finocchio, , M. Jardel, , A. Noonan, , and M. J. Iacono, 2014: Blue Hill Observatory sunshine—Assessment of climate signals in the longest continuous meteorological record in North America. Bull. Amer. Meteor. Soc., 95, 1741–1751, doi:10.1175/BAMS-D-12-00206.1.
Mannstein, H., , R. Meyer, , and P. Wendling, 1999: Operational detection of contrails from NOAA-AVHRR data. Int. J. Remote Sens., 20, 1641–1660, doi:10.1080/014311699212650.
Mason, S. J., 2004: On using “climatology” as a reference strategy in the Brier and ranked probability skill scores. Mon. Wea. Rev., 132, 1891–1895, doi:10.1175/1520-0493(2004)132<1891:OUCAAR>2.0.CO;2.
Meerkötter, R., , U. Schumann, , D. R. Doelling, , P. Minnis, , T. Nakajima, , and Y. Tsushima, 1999: Radiative forcing by contrails. Ann. Geophys., 17, 1080–1094, doi:10.1007/s00585-999-1080-7.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360, doi:10.1175/BAMS-87-3-343.
Meyer, R., , H. Mannstein, , R. Meerkötter, , U. Shumann, , and P. Wendling, 2002: Regional radiative forcing by line-shaped contrails derived from satellite data. J. Geophys. Res., 107, doi:10.1029/2001JD00426.
Mims, F. M., , and D. J. Travis, 1997: Aircraft contrails reduce solar irradiance. Eos, Trans. Amer. Geophys. Union, 78, 448, doi:10.1029/97EO00281.
Minnis, P. J., , D. F. Young, , D. P. Garber, , L. Nguyen, , W. L. Smith Jr., , and R. Palikonda, 1998: Transformation of contrails into cirrus during SUCCESS. Geophys. Res. Lett., 25, 1157–1160, doi:10.1029/97GL03314.
Minnis, P. J., , U. Schumann, , D. R. Doelling, , K. M. Gierens, , and D. W. Fahey, 1999: Global distribution of contrail radiative forcing. Geophys. Res. Lett., 26, 1853–1856, doi:10.1029/1999GL900358.
Minnis, P. J., , J. K. Ayers, , M. L. Nordeen, , and S. P. Weaver, 2003: Contrail frequency over the United States from surface observations. J. Climate, 16, 3447–3462, doi:10.1175/1520-0442(2003)016<3447:CFOTUS>2.0.CO;2.
Minnis, P. J., , J. K. Ayers, , M. L. Nordeen, , R. Palikonda, , and D. Phan, 2004: Contrails, cirrus trends, and climate. J. Climate, 17, 1671–1685, doi:10.1175/1520-0442(2004)017<1671:CCTAC>2.0.CO;2.
Minnis, P. J., , R. Palikonda, , B. J. Walter, , J. K. Ayers, , and H. Mannstein, 2005: Contrail properties over the eastern North Pacific from AVHRR data. Meteor. Z., 14, 515–523, doi:10.1127/0941-2948/2005/0056.
Minnis, P. J., and Coauthors, 2013: Linear contrail and contrail cirrus properties determined from satellite data. Geophys. Res. Lett., 40, 3220–3226, doi:10.1002/grl.50569.
Moss, S. J., 1999: The testing and verification of contrail forecasts using pilot observations from aircraft. Meteor. Appl., 6, 193–200, doi:10.1017/S1350482799001115.
Muri, H., , J. E. Kristjánsson, , T. Storelvmo, , and M. A. Pfeffer, 2014: The climatic effects of modifying cirrus clouds in a climate engineering framework. J. Geophys. Res., 119, 4174–4191, doi:10.1002/2013JD021063.
Myhre, G., , and F. Stordal, 2001: On the tradeoff of the solar and thermal infrared radiative impact of contrails. Geophys. Res. Lett., 28, 3119–3122, doi:10.1029/2001GL013193.
Nakanishi, S., , J. Curtis, , and G. Wendler, 2001: The influence of increased jet airline traffic on the amount of high level cloudiness in Alaska. Theor. Appl. Climatol., 68, 197–205, doi:10.1007/s007040170045.
Neter, J., , W. Wasserman, , and M. H. Kutner, 1985: Applied Linear Statistical Models: Regression, Analysis of Variance, and Experimental Designs. 2nd ed. Richard D. Irwin, 1127 pp.
Newinger, C., , and U. Burkhardt, 2012: Sensitivity of contrail cirrus radiative forcing to air traffic scheduling. J. Geophys. Res., 117, D10205, doi:10.1029/2011JD016736.
Newton, B., , A. Jackson, , D. Hahn, , and A. Bussey, 1997: Statistical contrail forecasting. Preprints, Seventh Conf. on Aviation, Range, and Aerospace Meteorology, Long Beach, CA, Amer. Meteor. Soc., 463–468.
Nicodemus, M. L., , and J. D. McQuigg, 1969: A simulation model for studying possible modification of surface temperature. J. Appl. Meteor., 8, 199–204, doi:10.1175/1520-0450(1969)008<0199:ASMFSP>2.0.CO;2.
Ovarlez, J., , P. Van Velthoven, , G. Sachse, , S. Vay, , H. Schlager, , and H. Ovarlez, 2000: Comparison of water vapor measurements from POLINAT 2 with ECMWF analyses in high-humidity conditions. J. Geophys. Res., 105, 3737–3744, doi:10.1029/1999JD900954.
Palikonda, R., , P. Minnis, , P. K. Costulis, , and D. P. Duda, 2002: Contrail climatology over the USA from MODIS and AVHRR data. Preprints, 10th Conf. on Aviation, Range, and Aerospace Meteorology, Portland, OR, Amer. Meteor. Soc., J1.3.
Palikonda, R., , P. Minnis, , D. P. Duda, , and H. Mannstein, 2005: Contrail coverage derived from 2001 AVHRR data over the continental United States of America and surrounding areas. Meteor. Z., 14, 525–536, doi:10.1127/0941-2948/2005/0051.
Peters, J. L., 1993: New techniques for contrail forecasting. Air Weather Service Scott Air Force Base Tech. Rep. AWS/TR-93/001, 31 pp. [NTIS AD-A269686.]
Petersen, M. S., , P. J. Lamb, , and K. E. Kunkel, 1995: Implementation of a semiphysical model for examining solar radiation in the Midwest. J. Appl. Meteor., 34, 1905–1915, doi:10.1175/1520-0450(1995)034<1905:IOASMF>2.0.CO;2.
Pielke, R. A., Sr., 2003: Heat storage within the Earth system. Bull. Amer. Meteor. Soc., 84, 331–335, doi:10.1175/BAMS-84-3-331.
Pilié, R. J., , and J. E. Jiusto, 1958: A laboratory study of contrails. J. Meteor., 15, 149–154, doi:10.1175/1520-0469(1958)015<0149:ALSOC>2.0.CO;2.
Rädel, G., , and K. P. Shine, 2008: Radiative forcing by persistent contrails and its dependence on cruise altitudes. J. Geophys. Res., 113, D07105, doi:10.1029/2007JD009117.
Ryan, A. C., , A. R. MacKenzie, , S. Watkins, , and R. Timmis, 2011: World War II contrails: A case study of aviation-induced cloudiness. Int. J. Climatol., 32, 1745–1753, doi:10.1002/joc.2392.
Sassen, K., 1997: Contrail-cirrus and their potential for regional climate change. Bull. Amer. Meteor. Soc., 78, 1885–1903, doi:10.1175/1520-0477(1997)078<1885:CCATPF>2.0.CO;2.
Sausen, R., , K. Gierens, , M. Ponater, , and U. Schumann, 1998: A diagnostic study of the global distribution of contrails. Part I: Present day climate. Theor. Appl. Climatol., 61, 127–141, doi:10.1007/s007040050058.
Schrader, M. L., 1997: Calculations of aircraft contrail formation critical temperatures. J. Appl. Meteor., 36, 1725–1728, doi:10.1175/1520-0450(1997)036<1725:COACFC>2.0.CO;2.