We compare, for the first time, the performance of a simplified atmospheric radiative transfer algorithm package, the Corti–Peter (CP) model, versus the more complex Fu–Liou–Gu (FLG) model, for resolving top-of-the-atmosphere radiative forcing characteristics from single-layer cirrus clouds obtained from the NASA Micro-Pulse Lidar Network database in 2010 and 2011 at Singapore and in Greenbelt, Maryland, USA, in 2012. Specifically, CP simplifies calculation of both clear-sky longwave and shortwave radiation through regression analysis applied to radiative calculations, which contributes significantly to differences between the two. The results of the intercomparison show that differences in annual net top-of-the-atmosphere (TOA) cloud radiative forcing can reach 65 %. This is particularly true when land surface temperatures are warmer than 288 K, where the CP regression analysis becomes less accurate. CP proves useful for first-order estimates of TOA cirrus cloud forcing, but may not be suitable for quantitative accuracy, including the absolute sign of cirrus cloud daytime TOA forcing that can readily oscillate around zero globally.
Cirrus clouds play a fundamental role in atmospheric radiation balance and their net radiative effect remains unclear (IPCC, 2014; Berry and Mace, 2014; Campbell et al., 2016; Lolli et al., 2017). Feedbacks between cirrus dynamic, microphysical and radiative processes are poorly understood, with ramifications across a host of modeling interests and temporal/spatial scales (Kuo-Nan, 1986; Liou, 1986, Khvorostyanov and Sassen, 1998). Simply put, different models parameterize ice formation in varied, yet relatively simplified, ways that impact how cirrus are resolved, and how their macro/microphysical and radiative properties are coupled with other atmospheric processes (e.g., Comstock et al., 2001; Immler et al., 2008). Consequently, models are very sensitive to small changes in cirrus parameterization (Soden and Donner 1994; Min et al., 2010; Dionisi et al., 2013).
Cirrus clouds constitute the only tropospheric cloud genus that exerts either a positive or negative top-of-the-atmosphere (TOA) cloud radiative forcing effect (CRE) during daytime. All other clouds exert a negative daytime TOA CRE. Cirrus clouds exerting negative net TOA CRE cool the earth–atmosphere system and surface below them. This occurs as the solar albedo term is greater than the infrared absorption and re-emission term. Positive forcing occurs when the two are reversed and infrared warming and re-emission exceed scattering back to space. In contrast, all clouds cause a positive nighttime TOA value, with an infrared term alone and no compensating solar albedo term. This dual property makes cirrus distinct, and why it is crucial to understand how well radiative transfer models are resolving their TOA CRE properties.
The burgeoning satellite and ground-based era of atmospheric monitoring
(Sassen and Campbell, 2001; Campbell et al., 2002; Welton et al., 2002;
Nazaryan et al., 2008; Sassen et al., 2008) has led to a wealth of new data
for looking at global cirrus cloud properties. In particular, TOA CREs are evaluated by means of radiative
transfer modeling, designed with different degrees of complexity. What is not
yet known is how the relative simplicity of some models translates to a
relative retrieval uncertainty, given that the CRE effect of cirrus clouds,
at both the ground and TOA, is typically on the order of 1 W m
Corti and Peter (2009; CP) describe a simplified radiative transfer model that relies upon a constrained number of input parameters, including surface temperature, cloud top temperature, surface albedo, layer cloud optical depth, and the solar zenith angle. CP simplifies drastically the framework of the Fu–Liou–Gu radiative transfer model (Fu and Liou, 1992; Gu et al., 2003, 2011; FLG), for instance, through a parameterization of the longwave and shortwave fluxes derived from the FLG model calculations for realistic atmospheric conditions. Moreover, CP does not directly consider gaseous absorption. The model has increasingly been used to assess cirrus cloud radiative effects (Kothe et al., 2011; Kienast-Sjögren et al., 2016; Burgeois et al., 2016) from lidar measurements, owing to its relative simplicity and lower computational burden compared with a model like FLG.
To date, CP model performance vs. FLG model has been evaluated for sensitivities only to simulated synthetic clouds and never on real measurements, especially those collected over long periods (Corti and Peter, 2009). Such evaluation, however, can readily be conducted using the unique NASA Micro-Pulse Lidar Network (MPLNET; Welton et al., 2002; Campbell et al., 2002; Lolli et al., 2013, 2014), established in 1999 to continuously monitor cloud and aerosol physical properties (Wang et al., 2012; Pani et al., 2016).
The objective of this technical note is to then assess differences between CP and FLG in terms of net annual daytime TOA CRE. CP and FLG model performance are evaluated using MPLNET datasets collected from Singapore in 2010 and 2011, a permanent tropical MPLNET observational site, and at Greenbelt, Maryland, in 2012, a midlatitude site. Our goal is to more appropriately characterize the sensitivities of CP relative to what is generally considered a more complex, and presumably more accurate, model, with the hopes of better understanding relative uncertainties and thus interpreting whether such uncertainties are appropriate for long-term global cirrus cloud analysis.
FLG is a combination of the delta four-stream approximation for solar flux
calculations (Liou, 1986) and a delta-two–four-stream approximation for IR
flux calculations (Fu et al., 1997), divided into 6 and 12 bands,
respectively. It has been extensively used to assess net cirrus cloud daytime
radiative effects, most recently for daytime TOA forcing characteristics
within MPLNET datasets at both Greenbelt, Maryland, and Singapore,
respectively (Campbell et al., 2016; Lolli et al., 2017). The results from
these studies have led to the hypothesis of a meridional gradient in cirrus
cloud daytime TOA radiative forcing existing, with daytime cirrus clouds
producing a positive daytime TOA CRE at lower latitudes that reverses to a
net negative daytime TOA CRE approaching the non-snow and ice-covered polar
regions. They estimate absolute net cirrus daytime TOA forcing term between
0.03 and 0.27 W m
To calculate daytime cirrus cloud radiative effects from MPLNET datasets, the
lidar-retrieved single-layer cirrus cloud extinction profile (Campbell et
al., 2016; Lewis et al., 2016; Lolli et al., 2016, 2017) is transformed into
crystal size diameter (using the atmospheric temperature profile) and ice
water content (IWC) profiles using the parameterization proposed by
Heymsfield et al. (2014). Those parameters, at each range bin, are input into
FLG. The thermodynamic atmospheric profiles, together with ozone
concentrations are obtained with a temporal resolution of
Calculations here are performed for the same MPLNET observational sites, Singapore and Greenbelt, Maryland (i.e., NASA Goddard Space Flight Center; GSFC). For the former site, two different values of the surface albedo, which is a common input parameter in both models, are fixed at 0.12 and 0.05, respectively, as Singapore is a metropolitan area completely surrounded by sea. This allows us to more reasonably characterize forcing over the broader archipelago of Southeast Asia, and follows the experiments described by Lolli et al. (2017). At NASA GSFC, only a single over-land albedo is used, though one that varies monthly between 0.12 and 0.15 based on climatology.
Total NET, SW, and LW fluxes
(W m
Here, we reconsider these results by first intercomparing those solved with FLG and CP for net daytime TOA CRE over a practical range of cloud optical depth (COD). As described in both Campbell et al. (2016) and Lolli et al. (2017), daytime is specifically defined in these experiments as those hours where incoming net solar energy exceeds that outgoing. Only under such circumstances can the net TOA CRE term become negative. Otherwise, it is effectively nighttime, as the term is positive and all clouds induce a warming TOA term. Those nighttime results presented within the analysis below will instead be considered as context for understanding net diurnally averaged differences between the models specifically for the GSFC dataset.
The daytime cirrus net TOA CRE, normalized by corresponding occurrence
frequency, in this case as a function of COD, was evaluated at Singapore
(1.3
Analysis over land (albedo
Analysis over land (albedo
An initial sensitivity study was carried out to evaluate how the input parameters, and eventually their uncertainties, influence the net TOA CRE calculations. Results are summarized in Table 1. Model input parameter sensitivities were investigated for surface albedo, COD, land/ocean surface temperature, and cloud top temperature. Table 1 shows how much net, SW, and LW fluxes change by varying each individual parameter alone. For instance, changing the surface albedo from 0.12 to 0.14 and keeping the other three parameters fixed produces similar changes in both models (26 % for CP model and 25 % for FLG model). Changing COD from 1 to 1.1 produces a change of 16 % for CP and 21 % for FLG. Changing surface temperature and cloud top temperature of 1 K produces respective changes of 10 and 7 % for CP and 7 and 6 % for FLG. Though subtle, the models exhibit some differences in variance relative to the input parameters required to initialize them.
FLG and CP were compared over a total of 33 072 total daytime single-layer cirrus clouds at Singapore from 2010 to 2011. Figures 1, 2, 3, and 4 reflect histograms of cirrus cloud relative frequency and net annual daytime TOA CRE normalized by corresponding frequency, for surface albedo values of both 0.05 (Figs. 3 and 4; i.e., over sea) and 0.12 (Figs. 1 and 2; i.e., over land) at 0.03 COD resolution from 0 to 3. This latter COD range was chosen to distinguish cirrus clouds in a phenomenological manner consistent with Sassen and Cho (1992). Note that, since a common cloud sample is used, the 20 sr samples vary in COD between only 0 and approximately 1 in contrast to the 30 sr sample topping out at 3. The observed differences in net radiative effect can be ascribed to the different lidar ratio. Overall, the results here complement the work of Berry and Mace (2014), who first recognized the significance of optically thin cirrus influencing the net normalized term so significantly.
Intercomparison of net daytime TOA CRE vs. COD over the ocean at 30 sr, we
obtain
Same as Fig. 1, but over the ocean (albedo
Same as Fig. 2, but over the ocean (albedo
To better understand the different outputs between the two models, a scatter
plot between from FLG barplot entries is shown in Figs. 2 and 4, and the
corresponding CP barplot values are plotted, over land and over ocean, in
Figs. 5 and 6. The blue line represents the actual linear data regression,
while the red line represents an ideal case (i.e., slope
From Figs. 5 and 6, the FLG-derived net daytime CP TOA CRE values are
systematically greater in absolute value than the corresponding FLG values
by 60 %. More in detail, CP TOA CRE of 1 Wm
For the sake of completeness, and to cover all the variability related to
the chosen LR, we performed the same analysis though excluding the 20 sr
solution. Over the ocean, we derive an overall forcing of 1.34
for CP and 0.48 W m
Scatter plot and linear regression for 30 sr solution for FLG and CP
CRE in 2010–2011 over land
Analysis on 2010 dataset from MPLNET GSFC observational site for
30 sr solution daytime
Summary of principal CRE (Wm
To limit potential assessment ambiguity based on a single-site analysis, we
performed a second model comparison using the 2012 NASA GSFC dataset. A
summary of this dataset and net daytime TOA CRE results can be found in
Campbell et al. (2016). As this site is land-locked, only the single albedo
was, again, used, though varied monthly based on climatological passive
satellite estimates. A total of 21 107 daytime cirrus cloud profiles were considered.
Shown in Fig. 6 (upper panel) are the total net TOA CREs vs. COD at 30 sr,
for CP (
With this NASA GSFC dataset, we further consider an additional 32 185
nighttime cirrus cloud cases within the analysis (Fig. 6, lower panel).
Relative to prior estimates of CP uncertainty compared with more complex
models, a diurnal average would be likely to produce a different, and
plausibly closer, relative agreement consistent with prior studies. That is,
since during for most of the period we define here as night there is no
solar input, a simplification of the infrared forcing terms and
parameterizations alone would potentially yield a closer comparison between
the two models. For the NASA GSFC dataset, we solved a relative net
nighttime TOA CRE of 29.1 Wm
Same as Fig. 6, taking out those measurements with a land surface
temperature
It is useful at this point to discuss some of the potential elements driving
these differences. The larger discrepancies between the two models are
likeliest ascribed to the parameterization of three specific parameters in
the CP model: the first two,
The 20 % relative model accuracy claimed in Corti and Peter (2009) may be
verified for special conditions in tropical latitudes, where the three
parameters discussed above are well optimized. However, that is clearly not
found from our study. Corti and Peters (2009) expressly stated that they
used fixed values for those three parameters (i.e.,
We advise that those looking to apply CP to long-term climate/cirrus cloud study should carefully analyze the relevance of these settings to their given experiment before directly applying the model, especially when land surface temperatures are warmer than 288 K.
Annual single-layer cirrus cloud top-of-the-atmosphere (TOA) radiative effects (CREs) calculated from the Corti and Peter (2009) radiative transfer model (CP) are compared with similar results from the more complex, and presumably more accurate, Fu–Liou–Gu (FLG) radiative transfer model. The CP model calculates CREs using a parameterization of longwave and shortwave fluxes that are derived from real measurements optimized for a tropical environment through a regression analysis to simplify the radiative calculations. Values for these parameterizations, as suggested in Corti and Peter (2009), lead to relative differences in TOA CRE that far exceed the stated 20 % in the original paper. This includes parsing results out for daytime, nighttime, or diurnal averages. It is believed that specific parameterizations with the simplified model cannot be considered global constants, as originally defined for CP, but that they should be carefully evaluated on single case basis for each experiment. Moreover we find that the land surface temperature is responsible for significant discrepancies when larger than 288 K, because the original CP regression analysis is less accurate for larger temperatures. However, CP uses less input parameters compared with FLG, making it practically and computationally more efficient, particularly for large climate datasets. This is the first time, however, that the two models are compared using long-term cirrus clouds datasets, as opposed to synthetic datasets, with experiments conducted using NASA Micro-Pulse Lidar datasets collected at Singapore in 2010 and 2011 (Lolli et al., 2017) and Greenbelt, Maryland, in 2012.
Net daytime TOA CRE was evaluated versus cloud optical depth (COD) for steps
of 0.03 (COD range: 0–1) at 20 sr and for steps of 0.1 at 30 sr (COD range:
0–3) for Singapore datasets, while at 30 sr for Greenbelt, Maryland. Our
findings suggest that the difference in annual net TOA CRE between the two
models approaches 65 % in one experiment at Singapore. At Greenbelt,
Maryland, the sign of the net annual daytime TOA CRE term differs, and the
absolute difference varies between by nearly 2.5 Wm
In spite of this comparison, even if we reasonably speculate that FLG is the more accurate model overall, because of its relative complexity compared with CP, we are still missing regular comparisons of FLG with real observational data. Thus, the practical gains to long-term application of a simplified model like CP cannot be overstated, given lower computational demands. However, we believe that the results from this study are noteworthy because they show that the differences between the two models are significant. With respect to cirrus annual net daytime TOA CRE, and given the perspective on their global distribution described by Campbell et al. (2016) and Lolli et al. (2017), these sensitivities can lead to completely different conclusions about global cirrus TOA forcing effects. Therefore, in future work, it is imperative for the community to continue understanding and refining the global parameterizations used in all radiative transfer models regarding cirrus. Continued intercomparisons between models with real observation at the ground (using flux measurements), in situ (aircraft measurements) and at TOA (using satellite-based measurements) remain critical interests.
Data used for analysis are publicly available at
The authors declare that they have no conflict of interest.
This study and the NASA Micro-Pulse Lidar Network (MPLNET) are supported by the NASA Radiation Sciences Program (H. Maring). James R. Campbell acknowledges the Naval Research Laboratory Base Program (BE033-03-45-T008-17) and support of NASA Interagency Agreement NNG15JA17P on behalf of MPLNET. Edited by: T. von Clarmann Reviewed by: two anonymous referees