Closure Experiments

Closure experiments aim to measure the consistency of the atmospheric state parameters with respect to models of the underlying physical processes (Ogren

1995; Quinn et al. 1996). The methodology consists of measuring input parameters to initialize a model and derive output parameters; concomitantly, the control parameters are measured for comparison with model predictions. To illustrate this, we will describe the observational strategy of the ACE-2 Cloudy-Column experiment (Brenguier, Chuang et al. 2000), which was the first field study dedicated entirely to the aerosol indirect effects in extended boundary-layer cloud systems. The hypotheses to be tested can be summarized in the form of three key questions:

1. For specified cloud fields, is the droplet concentration consistent with the predictions of aerosol activation models?

2. Do cloud radiative properties vary with droplet concentration, as anticipated by Twomey (1977)?

3. For a particular value of LWP, is the precipitation rate modulated by the droplet concentration?

Temporal and Spatial Scales

Sampling a single convective cell during its vertical ascent while measuring aerosol properties, vertical velocity, and cloud droplet concentration to examine CCN activation is not feasible using the existing airborne platforms. In addition, radiative transfer raises serious difficulties for single cells because it is, in essence, three dimensional and thus measurements of irradiance performed from above a single convective cell will necessarily be affected by radiation from neighboring cells. Finally, the cycle of precipitation formation in a single convective cell is short (a few tens of minutes), resulting in very heterogeneous drizzle patches below cloud base.

An alternative strategy is to examine the phenomenon at a larger scale at which aerosol properties, turbulence, cloud microphysics, and precipitation are statistically homogeneous (ergodic), and the three-dimensional heterogeneities of the radiation and precipitation fields are smoothed over a large number of cells. Such conditions are often satisfied in boundary-layer marine stratocumulus clouds at a scale of a few tens of kilometers.

Aerosol Activation Closure

This experiment aims to evaluate 0-D kinetic models of CCN activation to predict the cloud droplet concentration (control parameter) from the vertical velocity at cloud base and the physicochemical properties of the aerosol (input parameters) (Guibert et al. 2003; Snider et al. 2003).

Since we are unable to perform a closure experiment on individual CCN activation events, a statistical approach to the problem must be adopted, which necessarily must encompass the spatial variability of the system being studied. Aerosol properties can, far from the aerosol sources, be reasonably assumed to be uniform over the area and the duration of the experiment. Vertical velocity, on the other hand, varies from a few cm s-1 up to more than 1 m s-1 in the most active cells. Thus, comparison involves the probability distribution function (PDF) of measured droplet concentration and its comparison with the predictions of a CCN activation model initialized with the full spectrum of measured vertical velocities. Figure 21.1 shows the comparison of the deciles of the measured droplet concentration PDF with the predictions of the model initialized successively with the deciles of the measured vertical velocity distribution. This figure demonstrates that the range of concentration variability resulting from vertical velocity fluctuations is broader than the difference between the mean values of a pristine and a polluted case.

Therefore, a closure experiment on CCN activation will not be conclusive if the vertical velocity is not fully constrained by observations. A consistent definition of the cloud droplet concentration used here as a control parameter is also crucial. In fact, the droplet concentration measured in a cloud system is different from the one resulting from CCN activation, even though both are tightly related. After CCN activation is completed, additional processes (e.g., mixing with the environmental dry air and scavenging by precipitation) dilute droplet concentration significantly. Thus, for comparison with a CCN activation model prediction, it is sensible to select only those droplet concentration samples that are not affected by either mixing or precipitation scavenging. In ACE-2, for instance, the droplet concentration after selection was

2 180

2 180

/ -



— #21

— #30


90 180 270 360 Concentration predicted (mg-1)

90 180 270 360 Concentration predicted (mg-1)

Figure 21.1 Deciles of the observed droplet concentration probability distribution function (PDF) versus predictions of the droplet concentration with a CCN activation model initialized successively with the deciles of the observed vertical velocity PDF for flight 21 (pristine case) and flight 30 (polluted case).

typically 30% higher than the average over all the samples (Pawlowska and Brenguier 2000).

The main limitation of this closure experiment was the incomplete characterization of the size-segregated chemical composition of the particles, which introduces uncertainties in the prediction of their hygroscopic properties. Since the development of airborne mass spectrometers and improved chemical analysis, activation closure has been significantly refined (Conant et al. 2004; Fountoukis et al. 2007). Further work remains to be conducted to characterize the mass accommodation coefficient for small growing particles, and instruments need to be designed to characterize their state of chemical mixture.

Column Closure on Radiative Transfer

This type of experiment aims to corroborate the Twomey hypothesis (i.e., that aerosol-induced microphysical changes are reflected by changes of cloud radiative properties). The input parameters measured in situ are the vertical distribution and horizontal variability of the cloud droplet size distribution, and the output parameter is the optical thickness of the cloud layer, derived as the vertical integral of extinction. The control parameter is the optical thickness derived independently from multispectral radiances measured with a second aircraft flying above the cloud layer.

In ACE-2, the statistics of all input parameters were very robust because of the long duration of sampling with two collocated aircraft: one was dedicated to the microphysical fields while the other focused on radiation. Thus, ACE-2 provided the fi rst observational evidence of the Twomey effect in extended cloud systems, e.g., the scaling of the optical thickness with LWP and droplet concentration as anticipated by Twomey (see Fig. 6 in Brenguier, Pawlowska et al. 2000). One serious limitation, however, was that in-situ airborne measurements do not provide information on how the microphysical fields sampled at various levels are distributed vertically. The optical thickness used as the output parameter was derived by assuming either random or maximum overlap of the microphysical fields measured in situ. The accuracy of the prediction was thus significantly degraded by the overlap uncertainty. Since ACE-2, other closure experiments have been performed using remote-sensing systems that better constrain the vertical organization of the microphysical fields (Feingold et al. 2003).

An alternative approach to radiative closure is to validate the same radiative transfer model but in the inverse mode. Indeed, inverse models are currently used to derive cloud properties from space measurements of multispectral radiances (Nakajima and King 1990). In this approach, radiance measurements are used to derive cloud geometrical thickness, or LWP, and droplet concentration, which are then compared to the ones measured in situ (Schüller et al. 2003).

Column Closure on Precipitation

This experiment contributes to the improvement of precipitation formation parameterization in global climate models. Recent field studies suggest that the precipitation rate of stratocumulus clouds, averaged over a large domain containing numerous cloud cells, scales with the mean cloud thickness or LWP and the typical droplet concentration: ACE-2 (Pawlowska and Brenguier 2003), EPIC (Comstock et al. 2004; Wood 2005), and DYCOMS-II (Van Zanten et al. 2005). The nature of such a relationship is a major determinant of the magnitude of the aerosol impact on cloud extent thickness and lifetime. Large eddy simulations (LES) are therefore used to corroborate these observations and quantify the empirical relationship better.

The observations are summarized in Figure 21.2a. For each field campaign, the precipitation rate at cloud base scales well with the cloud thickness and cloud droplet concentration. Each dataset appears, however, to have offsets that mainly refl ect measurement biases and differences in the methodology: precipitation rate averaged over the cloud layer (ACE-2) or at cloud base only (EPIC and DYCOMS-2), droplet concentration measured in cloud samples that are not affected by mixing or precipitation scavenging (ACE-2), averaged over the cloud layer (DYCOMS-II), or extrapolated from remote sensing (EPIC), cloud thickness derived from detection of cloud base and top (ACE-2 and DYCOMS-II) or derived from remote sensing of the LWP (EPIC). These discrepancies reveal how sensitive the results can be to the definition of the physical parameters derived from diverse measurement and data processing techniques. In addition, the results demonstrate just how sensitive the precipitation rate is to cloud macrophysical properties, with a doubling of precipitation requiring only a change of ~100 m in cloud thickness. This emphasizes the importance of controlling for meteorological variability when examining microphysical impacts.

Numerical simulations of similar cloud systems were performed with an LES model over a broad range of LWP and CCN concentration values to explore the parameter space of the measurements (Geoffroy et al. 2008). Figure 21.2b-d show the comparison of the model results with the measurements, using the same parameters and scaling laws as in each fi eld campaign, respectively. The similarity between observations from three different areas and the results of numerical simulation suggests that the large-scale relationship between LWP, droplet concentration, and the precipitation rate at cloud base is robust.

Summary and Recommendations

These three examples illustrate different types of closure experiments. Each type has clearly testable hypotheses. When the model is straightforward, such as the 0-D model of CCN activation, the closure experiment follows closely the basic methodology: measured input parameters, numerical simulations, comparison of model predictions with the control parameter. In the second example, the distinction between input and control parameters is less obvious, depending on whether the model is used in the direct or in the inverse mode, such as for satellite 1-D retrieval techniques. The third example, with its 3-D model of stratocumulus clouds, suggests how the technique can be extrapolated to compare relationships between specified physical parameters that have been observed and further simulated with a model over the same parameter space. The three approaches share in common the following methodological rules, which are generally not given sufficient attention in most of the closure studies:

1. Models need to be fully constrained: All parameters, which might impact the prediction of the model to be tested, must be documented according to a level of accuracy consistent with their impact (e.g., as for vertical velocity).

2. Ensure consistency in the definition of the measured and model parameters: The measured and model values of a parameter must be defined over the same spatial and temporal scales. For example, droplet concentration will exhibit significant differences, depending upon whether it is defined as the mean value over a cloud system or the value specifically measured in regions of CCN activation.

3. Redundancy in measurements is highly desirable: Single validation experiments often succeed, whereas redundant controls are more difficult to reconcile, but allow for a higher degree of confidence. For closure to be robust, attempts must be made to combine redundant closures of the same process, such as combining a CCN activation spectrum and a droplet activation closure on the same data set (Snider et al. 2003) or radiation closures on both transmitted and reflected light in the same cloud system (Platnick 2000).

In general, dynamic, thermodynamic, and microphysical properties exhibit important variability and covariability on the kilometer scale (i.e., scales smaller than a typical climate model grid box), and this variability has marked impacts on how aerosol-cloud interaction affects the large-scale properties of clouds. It is important to design a sampling strategy that allows us to characterize further these important subgrid statistical connections between variables (see, e.g., Illingworth and Bony, this volume; Larson et al. 2001, 2002). For instance, airborne measurements in clouds are often optimized by targeting cloud cells along the flight track; however, such an approach introduces bias into the data base (overestimated cloud fraction). Thus, it is crucial to adopt unbiased sampling or provide additional information to reduce potential biases in the data base.

Column closure experiments are useful to validate models of physical processes and their parameterizations for general circulation models, as long as the

0 0

Post a comment