loading

Logout succeed

Logout succeed. See you again!

ebook img

NASA Technical Reports Server (NTRS) 20070030218: Evaluation of a Cloud Resolving Model Using TRMM Observations for Multiscale Modeling Applications PDF

file size2 MB
languageEnglish

Preview NASA Technical Reports Server (NTRS) 20070030218: Evaluation of a Cloud Resolving Model Using TRMM Observations for Multiscale Modeling Applications

Source of Acquisition NASA Goddard Space Flight Center Evaluation of a Cloud Resolving Model Using TWIM Observations for NIuItiscale Modeling Applications Derek J. Posselt, Tristan S. L'Ecuyer, Wei-Kuo Tao, Arthur Y. Hou, and Graeme L. Stephens Submitted to J: Climate Popular Summary As an intermediate step between conventional general circulation models (GCMs) and global cloud resolving models (CRMs), many climate prediction centers are embedding a CRM in each grid cell of a conventional GCM. These Multiscale Modeling Frameworks (MMFs) represent a theoretical advance over the use of conventional GCM cloud and convection parameterizations in that they directly simulate a distribution of convective elements in a given GCM grid box. Though they represent a theoretical advance over conventional GCMs, MMFs have been shown to exhibit an overproduction of precipitation in the tropics during the northern hemisphere summer. This precipitation bias is particularly pronounced over the Tropical Western Pacific and South China Sea during the Asian Monsoon. In this study, the CRM component of the NASA Goddard MMF is evaluated against retrievals derived from multiple instruments aboard the Tropical Rainfall Measuring Mission (TRMM) satellite platform. To examine the CRM in isolation from the parent GCM, the NASA Goddard Cumulus Ensemble (GCE) model is driven with observed large-scale forcing derived from soundings taken during the South China Sea Monsoon Experiment. Simulations of clouds, precipitation, and radiation over the South China Sea are then compared with TRMM retrieved clouds, precipitation, and radiative fluxes. It is found that the GCE configuration used in the NASA Goddard MMF responds too vigorously to the imposed large-scale forcing, accumulating too much moisture and producing too much cloud cover during convective phases, and overdrying the atmosphere and suppressing clouds during monsoon break periods. Sensitivity experiments reveal that changes to microphysical parameters that determine the precipitating (snow and graupel) ice particle size distribution have a relatively large effect on simulated clouds, precipitation, and radiation, while changes to grid spacing and domain length have little effect on simulation results. The results motivate a more detailed and quantitative exploration of the sources and magnitude of the uncertainty associated with specified cloud microphysical parameters in the CRM components of MMFs. Data assimilation experiments, in which this uncertainty is quantified for each microphysical parameter, as well as comparisons between microphysical schen~esw ith differing levels of complexity, are underway. Evaluation of a Cloud Resolving Model Using TRMM 0 bservations for Multiscale Modeling Applications Derek J. Posselt , Tristan L'Ecuyer, Wei-Kuo Tao, Arthur Y. Hou, and Graeme L. Stephens Submitted to Journal of Climate 1 February 2007 Abstract The climate change simulation community is moving toward use of global cloud resolving models (CRMs), however, current computational resources are not sufficient to run global CRMs over the hundreds of years necessary to produce climate change estimates. As an intermediate step between conventional general circulation models (GCMs) and global CRMs, many climate analysis centers are embedding a CRM in each grid cell of a conventional GCM. These Multiscale Modeling Frameworks (MMFs) represent a theoretical advance over the use of conventional GCM cloud and convection parameterizations, but have been shown to exhibit an overproduction of precipitation in the tropics during the northern hemisphere summer. In this study, simulations of clouds, precipitation, and radiation over the South China Sea using the CRM component of the NASA Goddard MMF are evaluated using retrievals derived from the instruments aboard the Tropical Rainfall Measuring Mission (TRMM) satellite platform for a 46-day time period that spans 5 May - 20 June 1998. The NASA Goddard Cumulus Ensemble (GCE) model is forced with observed large- scale forcing derived from soundings taken during the intensive observing period of the South China Sea Monsoon Experiment. It is found that the GCE configuration used in the NASA Goddard MMF responds too vigorously to the imposed large-scale forcing, accumulating too much moisture and producing too much cloud cover during convective phases, and overdrying the atmosphere and suppressing clouds during monsoon break periods. Sensitivity experiments reveal that changes to ice cloud microphysical param- eters have a relatively large effect on simulated clouds, precipitation, and radiation, while changes to grid spacing and domain length have little effect on simulation results. The results motivate a more detailed and quantitative exploration of the sources and magnitude of the uncertainty associated with specified cloud microphysical parameters in the CRM components of XIINIFs. One of the greatest sources of uncertainty in current estimates of climate system response to surface warming is the influence of clouds and precipitation. Specifically, it is uilknocvn whether, given a globally averaged rise in surface temperature, clouds will act to enhance or mitigate the climate system response. Of particular interest is deep convection in the tropics, which plays an important role as it links the fluxes of short and longwave radiation with the large scale circulation and provides an important mid-tropospheric energy source through the release of latent heat. Though the effects of deep convection and the associated cloud field are felt on scales greater than 1000 km, the processes that critically determine the amount and intensity of precipitation, as well as the properties and extent of upper-level cirrus operate on the scales of a few km. It is in part due to this intrinsic separation in scales that it has been traditionally difficult for General Circulation Models (GCMs) to accurately simulate the observed distribution of clouds and precipitation in the tropics (Bony et al. 2006, Soden and Held 2006). Clouds and precipitation have typically been parameterized in climate models by as- suming a distribution of cloud heights and convective cores within a single GCM grid cell (Arakawa and Schubert 1974, Zhang and McFarlane 1995, Sud and Walker 1999). This type of representation suffers from uncertainty in the inherent assumptions about the character- istics of clouds and precipitation on the sub-GCM grid scale, leading to a wide range of uncertainty in the interaction between clouds and radiation, as well as in the vertical distri- bution of latent heat release. After three decades of concentrated efforts designed to develop, evaluate, and improve conventional GCM cloud and convective parameterizations, it is gen- erally acltnowledged that clouds, precipitation, and their interaction with radiation can be most realistically simulated using models that are run on the scales of the cloud processes themselves (grid lengths of at most four ltilometers). Though conlputational resources are continually increasing, it is still not possible to perform global simulations of climate change on cloud-resolving scales. As an intermediate step between GCMs and global cloud resolv- 3 ing models, selected centers have implemented so-called "Multi-scale Modeling Franieworks" (MMFs). These models employ a Cloud Resolving Convective Parameterization (CRCP, Grabowski 2001) or "Super-Parameterization" (Khairoutdinov and Randall 2001, Randall et al. 2003), and use a cloud resolving model to represent cloud-scale processes by embedding a small two-dimensional CRM in each GCM grid cell. Though certain features of the observed climate system (e.g. the Madden Julian Oscillation) are more realistically simu- lated in MMFs, there are problems in MMFs that are not observed in conventional GCMs. The most prominent example is the existence of the so-called "Great Red Spot"; a region of anomalously large precipitation centered over the South China Sea and Bay of Bengal during the northern hemisphere summer (Khairoutdinov et al. 2005, Tao et al. 2007). Sensitivity tests have indicated that the use of a small three-dimensional domain, and the adjustment of 2D CRM orientation within each GCM grid cell may help to reduce this anomaly, and Luo and Stephens (2006) have hypothesized that the problem arises from a convection-wind- evaporation feedback operating on a small cyclic CRM domain. In general, the problem illustrates the need for a more systematic evaluation of the CRM component of a multiscale modeling framework. In particular, it remains to be demonstrated that the CRM simu- lated distribution of clouds and precipitation and the associated interaction with visible and infrared radiative fluxes is consistent with 'observations. This paper addresses two fundamental questions regarding use of a CRM as a replace- ment for the conventional convective parameterization in a GCM. First, given the limited domain and two-dimensional nature of CRM simulations in a MMF, can the CRM correctly reproduce the observed statistics of clouds, precipitation, and radiation? Second, what are the dominant sources of uncertainty in the CRM, and can observations be brought to bear to reduce this uncertainty? The first question is addressed by comparing statistics of the clouds, precipitation, and radiation produced by the NASA Goddard Cumulus Ensemble (GCE) cloud resolving model used in the NASA Go.ddard Space Flight Center MMF to re- trievals of clot~dsp, recipitation, and radiation from the Tropical Rainfall Measuring Mission (TRhlIM) satellite platform. A partial answer to the second question is provided through a set of sensitivity experiments, in which changes to domain size, grid spacing, and cloud niicrophysical assumptions are applied to the model and the resulting fields are compared to a control simulation. A multiscale modeling framework typically consists of a two-way interaction between CRM and GCM, in which GCM "large-scale" tendencies of temperature and water vapor are first applied over the a CRM simulation time equal to a GCM timestep, after tvhich the CRM tendencies of temperature and water vapor are applied on the GCM grid. The resulting CRM fields are therefore affected by errors in both CRM and GCM. To evaluate clouds, precipitation, and radiation produced by the CRM apart from uncertainties in the parent GCM, CRM simulations are performed in the presence of observed large-scale forcing computed from a sounding network deployed during the South China Sea Monsoon Experi- ment (SCSMEX) that ran from 5 May through 20 June, 1998. Large-scale forcing includes the effects of advection of temperature and water vapor, as well as large-scale tendencies of the three-dimensional wind. A thorough description of the methodology used to generate the forcing dataset is contained in (Johnson and Ciesielski 2000). Forcing the GCE with observed large scale fields is analogous to embedding the CRNf within a GCM that perfectly simulates the large-scale flow, with the primary difference being that this is a purely one-way interaction with no feedback allowed from the CRM back to the large scales. When forced with observed large-scale advective tendencies of wind, temperature, and moisture, the CRM should produce clouds and precipitation that are consistent with observed conditions, and ideally, consistent with the observed distribution of clouds and precipitation. The comparison of model with observations is done in a statistical manner since it is the temporal and spatial distribution of clouds and cloud properties that have an effect on key climate variables. As such, it is not necessary for the model to reproduce each observed convective element or system at the exact time and place it occurred in reality. Instead, it is sufficient that the model produce the appropriate distribution of clouds and precipitation and that the effect of clouds on radiative fluxes and heating rates is consistent with observations. To this end, statistics of GCE-simulated clouds, precipitation, and radiation, accumulated over the SCSMEX field experiment are compared to TRNlM observations via two quantitative metrics. The first is a measure of the center of mass of the distribution, which can be defined as the mean, median, or mode, depending on the specifics of the comparison. The second is an integrated measure of the difference between each histogram over the combined range of values found in observations and model. This measure is effectively the sum of differences between histograms in each bin, and can be computed as the absolute value (integrated absolute difference, IAD) or alternatively as the root mean square (integrated RMS). The IAD carries the added benefit of an intuitive interpretation, as a 50% difference in PDF mass is computed as an IAD of 1.0, while a perfect mismatch (no shared mass between histograms) is computed as an IAD of 2.0. An illustration of the utility of both measures can be found in figure 1, in which a comparison of two idealized histogram is presented. While the mean of the two PDFs is identical in the first case, the IAD reveals a difference in structure. The second case has the identical IAD as the first case, but the means are shifted, indicating that the difference in PDF may be due more to the fact that there is a bias in the solution than to a structural difference in PDF. Two time periods are examined in this study: a 46-day interval, equal to one full TRMM precession cycle, and a shorter 9-day period during which the SCSMEX domain was charac- terized by repeated development of convective squall lines evolving in the presence of strong vertical wind shear. This second case is of particular interest for two reasons (1) it allows the evaluation of CRM simulations of convection in the region of the Great Red Spot and (2) in past numerical studies of convection it has been demonstrated that strongly sheared convective squall lines are more likely to be realistically represented in the two-dimensional framework than is convection that develops in weak shear or under suppressed conditions (Grabowski et al. 1998, Tornpkiiis 2000, Petcli and Gray 2001). Comparisons between model data and observations will be used to demonstrate that the GCE develops a moist bias over the course of the 46-day integration, and that this bias exists independent of changes to grid length or spacing. The steady accumulation of tropospheric water vapor leads to overpre- diction of cloud fraction, more frequent and intense precipitation, liquid and ice cloud that is thicker than observed, and cloud radiative forcing that is too strong. It will also be shown that modification of cloud microphysical parameters can lead to significant changes to the stati stics of all simulated fields, and potentially to an improved agreement with observations. The results imply that the details of the cloud microphysical parameterization play a key role in determining the statistics of clouds and precipitation simulated by the CRM, and that changes in grid spacing or geometry cannot eliminate problems associated with uncertainties in the cloud microphysical scheme. It is clear from this study that prior to use of a CRM in an MMF, uncertainties in the cloud microphysical representation must be quantified and mitigated. The remainder of this paper is organized as follows. A brief overview of the SCSMEX field campaign and computed large-scale forcing fields, along with a description of the TRMM multisensor retrieval algorithm, retrieved products, and estimated errors is presented in section 2. The details of the GCE model and of the specific configuration used in the GSFC MMF are described in section 3. Results of the GCE comparison with TRMM are presented in section 4, while sensitivity to grid and cloud microphysical parameters is explored in section 5. Summary, conclusions, and suggestions for future work are offered in section 6. 2 TRMM Retrievals and Description of SCSMEX 2.1 South China Sea Monsoon Experiment The South China Sea Monsoon Experiment (SCSMEX, Lau et al. 2000) was condtlcted during May and June 1998 to examine the mechanisms associated with the onset of mon- soon convection over the northern and southern South China Sea (SCS). It also served as a validation campaign for the newly-launched TRMM satellite (Kumn~erawe t al. 2000). During SCSMEX, radiosondes were regularly launched from several sites on and around the SCS (Ding and Lau 2001) with frequencies of four tinies daily over each of two Enhanced Sounding Arrays; one centered around the TOGA-BMRC dual-doppler radar network in the northern SCS, and the other centered around the Kexue #I GPS sounding site. Gridded fields of temperature, specific humidity, geopotential height, and the horizontal components of the wind were produced over the South China Sea at one-degree resolution using a mul- tiquadric interpolation scheme (Nuss and Titley 1994, Johnson and Ciesielski 2000). These fields were subsequently used to compute the pressure vertical velocity (w) and large-scale advective forcing of temperature and water vapor at six-hourly intervals averaged over both the Northern Enhanced Sounding Array (NESA) and Southern Enhanced Sounding Array (SESA) regions. These forcing fields have been used to drive 2D GCE simulations in previous studies of the Asian monsoon (Tao et al. 2003a, Tao et al. 2004). Depiction of the SCS with NESA and SESA regions is provided in figure 2; only results from the NESA will be used in this paper. Timeseries of TRMM observed daily domain-averaged precipitation rate and outgoing longwave radiation (OLR) over the NESA region (Fig. 3) reveals convection occurring in two distinct phases; a pre- and during-monsoon onset period that lasted from 18-26 May, and a post-onset episode that lasted from 2-11 June. Examination of the mean vertical shear for each of these periods (not shown) reveals weak unidirectional shear (approximately 10 m s-l) that extends through the depth of the troposphere for the May time period, with stronger shear (approximately 15 m s-l) during 2-11 June that reverses sign in the mid-troposphere. Consistent with the expectation of more strongly organized propagating convection under conditions of strong low-level vertical shear (Fovell and Ogura 1989), convection was observed to be more vigorous during the June time period with nearly double the mean precipitation rate observed during the pre-onset period (Johnson and Ciesielski 2002). OLR is generally anti-correlated with the precipitation rate, with relatively large values of OLR observed during periods in which precipitation was light, and small OLR observed during periods of heavy precipitation. Consistent with the production of stratiform precipitation regions and the persistence of clouds following a period of deep convection, the tinling of maxima in OLR lag the peak precipitation slightly. 2.2 TRMM Multisensor Retrieval Algorithm Observations used in this study derive from a multisensor retrieval algorithm that combines ice cloud information from the Visible and Infrared Scanner (VIRS) (Cooper et al. 2003), liquid cloud information from the TRMM Microwave Imager (TMI) (Greenwald et al. 1993), and precipitation information from the TMI-based Goddard Profiling algorithm (GPROF) (Kumrnerotv et al. 2000) to produce an analysis of the precipitable water vapor (PWV), precipitation rate, liquid water path (LWP), ice water path (IWP), and liquid and ice cloud fraction. Visible and infrared top of atmosphere fluxes and short-cvave and longwave heating and cooling rates computed in three discrete layers; low (0.5 - 2.5 ltm), middle (2.5 - 5.0 km), and high (5.0 - 17.0 km) (L'Ecuyer and Stephens 2003) are obtained from broadband radiative transfer calculations in each pixel. The estimated uncertainties in all retrieved parameters (Table 1) are based on a combination of rigorous sensitivity studies and product intercomparisons (L'Ecuyer and Stephens 2002, Cooper et al. 2003, L'Ecuyer and Stephens 2003). The time period that spans 0000 UTC 5 May - 0000 UTC 20 June 1998 was specifically chosen to match a single 46-day TRMM precession period,l the time interval over which TRMM repeats an observation of the identical position on the Earth's surface at the identical local time. Use of the full precession period helps avoid a bias toward observing at a particular local time, and consequently avoids biases in observations of the diurnal cycle of clouds and precipitation. . 'See http : //tsdis .gsfc .nasa. gov/overfl i~ht/PredictLocalSolar html for more details.

See more

The list of books you might like