Geophysical Research Letters

Investigating cloud radar sensitivity to optically thin cirrus using collocated Raman lidar observations

Authors


Abstract

[1] The sensitivity of the millimeter cloud radar (MMCR) to optically thin single-layer cirrus at the Atmospheric Radiation Measurement (ARM) Southern Great Plains (SGP) site is investigated using collocated Raman lidar observations. The sensitivity is characterized in terms of cloud optical depth (OD) and infrared (IR) radiative flux using over three years of coincident Raman lidar and MMCR observations. For cases when the Raman lidar is not fully attenuated (OD < 2.0) the MMCR detects approximately 70% of the total cloud OD with the majority of missed cloud OD occurring near cloud top. If only MMCR observations are used for computing cloudy top-of-the-atmosphere (TOA) IR flux, the missed cloud OD results in TOA flux biases from 0 to over 100 W/m2; however, the most frequently occurring bias is approximately 16 W/m2. This result highlights the importance of combining Raman lidar, or other sensitive cloud lidars that are able to measure cloud extinction directly, with the MMCR in order to accurately characterize the cloud radiative forcing for thin cirrus cases.

1. Introduction

[2] Clouds are one of the primary agents involved in the regulation of the planetary radiative balance, impacting both the incoming solar and outgoing IR radiation. Cloud radiative feedbacks have been identified as one of the largest sources of uncertainty in understanding climate sensitivity [Intergovernmental Panel on Climate Change, 2007]. The reason for this large uncertainty results from both challenges in measuring cloud radiative properties [Ohring et al., 2005] and the complex interactions between the cloud forcing, global circulation and hydrologic cycle [Stephens, 2005; Mace et al., 2006]. Cirrus clouds have proven very difficult to both accurately measure and to model, and as a result, represent a significant source of uncertainty in both cloud climatologies and global climate models. To rectify this problem accurate and long-term measurements of cirrus cloud properties are needed.

[3] The ARM SGP Central Facility in Lamont, Oklahoma is an atmospheric and surface observation facility that is heavily instrumented, including both Raman lidar [Goldsmith et al., 1998] and MMCR [Moran et al., 1998] systems. The Raman lidar at SGP, which was originally developed to retrieve atmospheric water vapor profiles, was upgraded in September 2004 to improve the system signal-to-noise performance thereby providing better cloud sensitivity [Ferrare et al., 2006]. As will be discussed, the upgraded Raman lidar system provides the capability of direct measurements of the vertically resolved cloud OD (extinction); however, at present, ARM has not implemented a cloud extinction retrieval for the SGP system. As part of this study an algorithm to derive cirrus extinction profiles was implemented and processed for over three years of observations at the SGP site. These extinction profiles made it possible to quantify the improvements to the characterization of cirrus heating rate profiles, as well as the surface and TOA IR fluxes, for optically thin cirrus when the Raman lidar observations are merged with those traditionally used from the MMCR [e.g., Mace and Benson, 2008].

2. Data Sets and Data Analysis

[4] The Raman lidar system is a multi-channel lidar that measures the energy backscattered elastically at the laser wavelength (i.e., from both particulates such as cirrus ice crystals and molecules) and inelastically from only molecules (nitrogen Raman backscatter) separately. The molecular (nitrogen Raman scattered) signal combined with an atmospheric temperature and pressure profile provides a direct measurement of the particulate extinction profile. This is possible because the Raman backscatter cross-section is dependent only on the molecular density and temperature and independent of the particulate scattering medium providing a known range resolved backscatter target. The difference between the calculated clear sky Raman return and the measured return is only a function of the particulate extinction. Furthermore, by combining the direct measurements of extinction with the observations from the elastic channel, the cirrus backscatter coefficient and backscatter phase function can also be derived [Ansmann et al., 1990].

[5] A significant source of uncertainty in the Raman OD measurements is from multiple scattering (MS) within the receiver field-of-view (FOV). The Raman system has a relatively large FOV (300 μrad) and short wavelength (355 nm) resulting in a significant component of the forward diffraction peak remaining within the receiver FOV. Using a multiple scatter model [Eloranta, 1998] MS uncertainties for a range of ice cloud optical depths, particle sizes, and cloud geometrical characteristics were investigated. The uncertainty in OD due to MS was found to range from 20–40% for this lidar, depending on the crystal effective radius and cloud optical depth. Correcting each Raman lidar optical depth profile for MS is challenging because of the strong dependencies on particle size, cloud optical depth, and geometrical inhomogeneities in the vertical (which is not well constrained). For this reason, the results presented here do not have a MS correction applied and must be considered when interpreting the results.

[6] The derived extinction profiles from the Raman lidar were combined with reflectivity data from the MMCR to produce a merged dataset on a 5-minute interval, with a 300 m vertical resolution between 6–16 km, from mid September 2004 (when the Raman lidar system was upgraded) to the end of 2007. Note that the MMCR operates by cycling through several different operating modes that have different sensitivity to atmospheric hydrometeors [Kollias et al., 2007]; in this analysis the most sensitive “cirrus” mode was used. This 3-year record of single-layer cirrus cloud events at SGP combines the best of both the MMCR and Raman lidar systems; the ability of the MMCR to characterize optically thick cirrus clouds that could fully attenuate the lidar and the ability of the Raman lidar to detect optically thin cirrus that the MMCR can miss. Furthermore, when both instruments sense the cirrus in the same volume, the combination of measurements from the two systems allows an estimate of particle size to be derived.

[7] This merged dataset was then used to derive cirrus cloud microphysical properties for a case study. When only the MMCR data were available, reflectivity was related to ice water content through the Liu-Illingworth power law relationship [Liu and Illingworth, 2000] and effective radius was assumed to be a simple function of temperature [Ivanova et al., 2001], which is the methodology currently used by ARM in its broadband heating rate and flux products [Mlawer et al., 2002]. When only lidar data were available, the lidar extinction profile was used with an assumed small effective radius (13.0 μm). And for those cases when both radar and lidar data were available simultaneously, direct measurements of extinction and retrieved effective particle size were used following the work by Donovan and van Lammeren [2001] with an assumed ice habit of hexagonal columns to be consistent with the methodology used in the radiative transfer model.

[8] The derived cirrus cloud microphysical properties were combined with additional atmospheric state profile and surface information and used to compute IR TOA fluxes and heating rates using the fast radiative transfer model RRTM [Mlawer et al., 1997], using Ebert and Curry's [1992] parameterization of ice cloud optical properties. Temperature, pressure, and humidity profile information used in the RRTM calculations came from temporally interpolated radiosonde profiles with the US standard atmosphere appended above to an altitude of 68 km. The humidity profiles in this interpolated dataset were scaled to yield the same precipitable water vapor as retrieved from the collocated microwave radiometer (MWR) [Turner et al., 2007]. Surface temperatures were calculated from downward-looking surface IR radiometer measurements. Carbon dioxide was assumed to be evenly mixed throughout the atmospheric column at 380 ppmv and the US standard profile was used for O3, N2O, and CO molecular species. Heating rate profiles and fluxes were computed for two situations: (a) using only the MMCR data to specify the extent and microphysical properties of the cirrus, and (b) using a combination of MMCR and Raman lidar data.

3. Results

3.1. Case Study

[9] To illustrate the utility of the combined radar and lidar measurements a case study is presented that was characterized by a single layer high altitude cirrus for 2.5 days from 8 to 11 November 2005 (Figure 1). Reflectivity measurements from the MMCR and cloud boundaries from the Raman lidar are overlaid in Figure 1a, which reveal regions where each instrument was sensing the cirrus cloud. In general, a significant portion of this cirrus cloud case had coincident radar and lidar measurements. Two brief periods of optically thick cirrus occurred where the MMCR observed the cloud higher in the atmosphere than the lidar because the lidar was fully attenuated by the cloud: just after 09:00 UTC on 09 November 2005 and 00:00 UTC on 10 November 2005. However, the lidar shows that a cirrus layer was over the site almost continuously during this period, whereas the MMCR missed a significant portion of the uppermost part of the cirrus cloud and in some regions did not detect any cloud overhead at all.

Figure 1.

An example of a single layer cirrus event from Nov at 1800 UTC to 11 Nov at 1200 UTC 2005: (a) MMCR reflectivity (in color) with Raman lidar derived cloud boundaries (black); (b) IR heating rate profile computed using MMCR data only to derive the cloud properties; (c) IR heating rate profile computed using a combination of MMCR and Raman lidar data to derive the cloud properties; (d) IR TOA fluxes for clear sky conditions (black), using MMCR-only observations (red), using MMCR and Raman lidar observations (green), and as derived by GOES (blue dots); and (e) flux difference between the (MMCR and lidar) minus the MMCR-only methods.

[10] Figure 1b shows the IR heating rates computed when the cloud extent and microphysical property information were derived using only MMCR data. In those regions where there was significant radar reflectivity there was also significant radiative heating and cooling, with heating towards the cloud base and cooling towards the cloud top. Adding the Raman lidar information and using the combined radar-lidar logic to derive the cloud extent and microphysical properties resulted in more pronounced heating and cooling throughout the case study, and is particularly more evident during the second half of the study period, as shown in Figure 1c.

[11] IR TOA fluxes are shown in Figure 1d for four cases: clear sky conditions (black), using only MMCR data (red), using MMCR and Raman lidar data (green), and as derived from the 10-km GOES-WEST observations over the SGP site using the VISST algorithm as a reference [Minnis et al., 1995] (blue dots). Significant differences can be seen in the TOA fluxes computed using the cloud properties derived from the combination of MMCR and Raman lidar observations versus those fluxes computed using the cloud properties determined from only the MMCR observations. These differences in TOA Flux, between the combined MMCR-lidar method (green) and the MMCR-only method (red), are shown in Figure 1e. Very large flux differences, greater than −100 W/m2, can be seen around 00:00 UTC on 09 November 2005 and greater than −50 W/m2 on 11 November 2005. These large flux differences are due to the lack of sensitivity of the MMCR, as the cloud radar responds primarily to the maximum dimension of the scattering medium and not the OD. The MMCR is strongly sensitive to particle size which results in limited sensitivity to cirrus composed of small particles, even when these clouds have significant radiative impact.

3.2. Multi-year Statistics

[12] The previous example provides a perspective on the possible improvements to the IR TOA flux when MMCR data is supplemented with Raman lidar data for optically thin cirrus cases. However, an analysis based on a longer data set is needed in order to better characterize what fraction of the cloud OD is not detected by the MMCR and its radiative impact. In the following analysis the OD missed by the MMCR is quantified over a three-year period (17 September 2004–31 December 2007); periods where the MMCR was known to be operating at degraded sensitivity relative to its nominal state have been excluded from this analysis. Since the MMCR does not directly measure cloud OD, we have inferred this from the Raman lidar cloud extinction profile data. This is done simply by integrating the lidar extinction profile from the base to the cloud top as determined by the cloud radar, which can only be done for those cirrus cases when the lidar is not attenuated.

[13] Figure 2 shows the fraction of the OD sensed by the MMCR versus the lidar OD with the color-scale depicting the number of occurrences. Since Figure 2 includes only the cases where the lidar beam was not fully attenuated by the cloud, the lidar OD is the total cirrus cloud OD. Furthermore, if the MMCR detected the cloud over the same vertical extent as the lidar, then the ratio of the OD fraction sensed by the radar would be unity. However, Figure 2 shows that for a large fraction of cases when the lidar cloud OD < 0.4 the fraction of the OD sensed by the radar can vary widely. Even for lidar ODs greater than 1, the MMCR does not detect the entire cloud vertical extent, typically only showing sensitivity to about 70% of the total OD of the cirrus. In addition there are a significant number of optically thin cases where the Raman lidar detects a cloud (approximately 40% of the time that single-layer cirrus occur and does not attenuate the lidar) that was not sensed by the radar; this is shown by the high concentration of points with a zero OD fraction sensed by the radar at the bottom of Figure 2. This implies that there are many optically thin clouds at the SGP site, and to best characterize these clouds a combination of both radar and lidar data is needed.

Figure 2.

Fraction of total cloud OD sensed by the MMCR versus the lidar OD with the color scale indicating the number of occurrences in the multi-year data set. Median values with interquartile ranges overlaid in black do not include samples where the cloud was undetected by the MMCR.

[14] The impact of the fraction of the cloud OD that was undetected by the MMCR on IR TOA fluxes is shown in Figure 3. Figure 3a shows the flux differences between the MMCR-only and MMCR and Raman lidar detection methods versus the lidar OD, with the color scale representing the number of occurrences; Figure 3b provides a two dimensional histogram of the same flux differences. However, it should be noted that the fluxes that contributed to these differences were not calculated in the same manner as those in the case study shown in Figure 1d. In order to minimize flux differences induced by the assumptions used to derive the cloud microphysical properties, the ODs were used to directly compute the IR TOA fluxes using a gray approximation. As a result any differences in fluxes can be assumed to be from measurement differences in the sensitivity to cloud OD, and not to differences in particle size. Adding the Raman lidar information resulted in a reduction in the IR TOA flux, because the lidar was able to detect the cloud extent higher in the atmosphere than the radar. The flux difference distribution peaks around −16 W/m2 at a lidar OD of approximately 0.1, which is primarily a result of those clouds that were undetected by the MMCR. However, the flux biases can be much larger than this value and are seen to be greater than −100 W/m2 for a smaller number of cases (Figures 3 and 1e).

Figure 3.

IR TOA flux difference between calculations using both MMCR and lidar data to characterize the properties of the cloud minus calculations that use only the MMCR over the multi-year data set: (a) versus lidar OD with the color scale indicating the number of occurrences and (b) as a two-dimensional histogram.

[15] These large IR flux differences are the result of a combination of effects: incorrect temperature assigned to the cloud due to the inadequate sampling of the vertical extent of the entire cloud and an underestimate of the cloud OD. While both impact the accuracy of the IR TOA fluxes, only the latter will impact the shortwave radiative fluxes at both the TOA and at the surface. Therefore, this analysis provides a first-order estimate of the error of IR TOA flux when only the cloud radar is used to determine the properties of the cirrus cloud used in the radiative transfer calculation.

4. Conclusions

[16] Cloud extinction profiles in single-layer cirrus clouds were derived from the Raman lidar system at SGP, and a 39-month long merged dataset of coincident MMCR and Raman lidar data was produced. This dataset was used to investigate single-layer optically thin cirrus cloud (OD < 2), where it was found that in order to accurately characterize thin cirrus a combination of MMCR and Raman lidar data should be used. The MMCR frequently fails to detect optically thin cirrus (OD < 0.2) and is often not sensitive to the full vertical extent of the cloud. Using coincident Raman extinction profiles over the multi-year record, the cloud missed by the MMCR was quantified in terms of OD and found that the MMCR missed on average 30% of the total cloud optical depth focused primarily near the cloud top, and nearly 40% of the optically thinnest cirrus clouds altogether. These uncertainties can result in large errors in calculated LW TOA flux, with the most common bias being −16 W/m2 and with biases as large as −100 W/m2. These findings demonstrate the importance of leveraging Raman lidar observations, or equivalent observations from other systems able to directly measure cloud OD profiles such as high-spectral-resolution lidars (HSRL) [Eloranta, 2005; Hair et al., 2008], to supplement the MMCR when developing ARM cloud products at SGP for radiative closure and climate model parameterization studies.

Acknowledgments

[17] The data used in this analysis were obtained from the Atmospheric Radiation Measurement (ARM) Program sponsored by the U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research, Climate and Environmental Sciences Division. This work was supported by DOE grants DE-FG02-90ER61057 and DE-FG02-06ER64167 as part of the ARM program. We would like to thank Mandana Khaiyer and Patrick Minnis for providing the GOES IR fluxes, and Ed Eloranta for discussions on profiling cloud extinction with Raman lidars and HSRLs.

Ancillary