The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite provides robust and global direct measurements of the cloud vertical structure. The GCM-Oriented CALIPSO Cloud Product is used to evaluate the simulated clouds in five climate models using a lidar simulator. The total cloud cover is underestimated in all models (51% to 62% vs. 64% in observations) except in the Arctic. Continental cloud covers (at low, mid, high altitudes) are highly variable depending on the model. In the tropics, the top of deep convective clouds varies between 14 and 18 km in the models versus 16 km in the observations, and all models underestimate the low cloud amount (16% to 25%) compared to observations (29%). In the Arctic, the modeled low cloud amounts (37% to 57%) are slightly biased compared to observations (44%), and the models do not reproduce the observed seasonal variation.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Clouds are the primary modulators of the Earth's radiation budget and still constitute the main source of uncertainty in model estimates of climate sensitivity [Randall et al., 2007], and a major limitation to the reliability of climate change projections [e.g., Dufresne and Bony, 2008]. To improve the reliability of climate change projections, it is therefore imperative to improve the representation of cloud processes in models. This first requires evaluating cloud descriptions in climate models. Until recently, this evaluation has been largely based on satellite data of the Earth's radiation budget (the Earth Radiation Budget Experiment, ERBE [Kandel et al., 1994], the Scanner for Radiation Budget, ScaRaB [Barkstrom and Smith, 1986], and the Clouds and the Earth's Radiant Energy System, CERES [Wielicki et al., 1996]). Basic aspects of cloudiness, as fundamental as the vertical distribution of the cloud cover were crucially lacking. Direct robust observations of the cloud amount over highly reflective surface (ice-sheets, deserts, continents) or in sparse shallow cumulus clouds near the surface, and of cloud vertical structure were not available at global scale. This lack of knowledge is particularly critical in the tropics [e.g.,Bony and Dufresne, 2005] and the polar regions [e.g., Winton, 2006; Kay and Gettelman, 2009; Kay et al., 2012], where model-based estimates of future climate have shown to be significantly sensitive to the description of clouds in the models.
 The A-Train constellation includes active remote sensing satellites (CALIPSO, launched in 2006 [Winker et al., 2009] and CloudSat [Stephens et al., 2002]) that can observe directly some of the key missing cloud properties like the cloud vertical distribution at high spatial resolution (30 m to 480 m), and to detect clouds over reflective surfaces. The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP [Winker et al., 2009]) lidar is specifically well adapted for observing i) clouds with optical depth lower than 3, ii) sparse clouds like shallow cumulus, iii) occurrence of clouds within the two first km above the surface (continent and ocean). Its main limitations regarding clouds, are i) the heliosynchroneous orbit of the satellite does not give access to the cloud diurnal cycle (not studied in this paper), and ii) its laser cannot penetrate an optical thickness larger than 3. This precludes the observations of clouds within deep convection along the Intertropical Convergence Zone (ITCZ) and within optically thick storm tracks in mid-latitudes.
 This study aims at evaluating the description of the cloud cover and the vertical structure within the Coupled Model Intercomparison Project Phase 5 (CMIP5) [Taylor et al., 2012] climate models using the GCM-Oriented CALIPSO Cloud Product (CALIPSO-GOCCP [Chepfer et al., 2010]). The methodology used for comparing climate models and CALIPSO-GOCCP observations is shortly described inSection 2. Results are primarily shown as averages and over the globe, over oceans and over continents (Section 3). We then focus our study on two regions where predictions of future climate have been shown to be significantly sensitive to clouds: the tropics (Section 4) and the Arctic region (Section 5).
2. CALIPSO-GOCCP Observations and Climate Model Outputs
 CALIPSO-GOCCP was developed from the CALIPSO level 1 attenuated backscatter measurements for evaluating clouds in climate models [Chepfer et al., 2010]. Here we used six years (2006–2011) of monthly CALIPSO-GOCCP observations including the layered low-level (z < 3.36 km), mid-level (3.36 km < z < 6.72 km), high-level (z > 6.72 km), and total cloud cover, as well as the cloud fraction profile at a vertical resolution of 480 m, averaged over a 2° × 2° horizontal grid. These observations are based on CALIOP lidar footprints which have a diameter of about 70 meters with a center-to-center spacing of 1/3 km. This small footprint size reduces concerns about partially filled pixels [Pincus et al., 2012]. The CALIPSO-GOCCP cloud climatology has been compared with that from CALIPSO Science Team (CALIPSO-ST) and found significant differences in zonal cloud profiles, because different cloud detection thresholds and horizontal averaging are used in the two algorithms (H. Chepfer et al., Comparison of two different cloud climatologies derived from CALIOP Level 1 observations: the CALIPSO-ST and the CALIPSO-GOCCP, submitted toJournal of Atmospheric and Oceanic Technology, 2012).
 We evaluated model output (AMIP experiment, atmosphere only, monthly time frequency and r1i1p1 ensemble) from years 1979 through 2008) of the CMIP5 experiment [Taylor et al., 2012] by the following climate models: IPSL-CM5B [Hourdin et al., 2012], CNRM-CM5 [Voldoire et al., 2011], HadGEM2 [Jones et al., 2011], CanAM4 [Cole et al., 2011] and MPI-ESM [Jungclaus et al., 2010]. As the definition of a cloud is not the same in models and observations nor between models, the lidar simulator [Chepfer et al., 2008] that is integrated into COSP (Cloud Feedback Model Intercomparison Project, CFMIP Observational Simulator Package [Bodas-Salcedo et al., 2011]) has been used in each model to simulate the cloud amount that would be observed by CALIPSO above the atmosphere predicted by the model. Use of a simulator reduces instrument biases and ensures that the cloud covers are defined consistently across the models and with the observations. Nevertheless, even a perfect agreement between CALIPSO-GOCCP and simulator outputs would not guaranty that the model reproduces perfectly clouds, because CALIPSO-GOCCP and COSP do not detect the optically thinnest clouds [Chepfer et al., 2010; also submitted manuscript, 2012].
3. Global Scale Analysis
3.1. Zonal Mean Cloud Cover
Figure 1shows the zonal mean cloud cover observed by CALIPSO-GOCCP and as simulated by the climate models through the lidar simulator. The total, mid and low cloud covers (Figures 1a, 1c, and 1d) are underestimated by most models at all latitudes except north of 50°N, where inter-model spread is significant around the observations (+/− 20% at 80°N). The global underestimation of total cloud cover (51% to 62% in models vs. 64% in observations) was already pointed out [Zhang et al., 2005] in former climate models (CMIP3 [Meehl et al., 2007]) compared to International Satellite Cloud Climatology Project (ISCCP) observations [Rossow and Schiffer, 1999]. The present evaluation (Figure 1a) suggests that this systematic model defect remains in CMIP5 models. The models do not produce enough clouds, and compensate by making clouds optically too thick, in order to get correct fluxes at the top of the atmosphere [Kay et al., 2012; D. Konsta et al., Evaluation of clouds simulated by the LMDZ5 GCM using A-train satellite observations (CALIPSO-PARASOL-CERES), submitted toClimate Dynamics, 2012]. In polar regions, where the observed cloud cover was highly uncertain before active space-based remote sensing, the inter-model spread is significant around the observations (20% to 50%), and all models underestimate the Antarctic cloud cover. The modeled cover of low-level clouds is too low, especially in the tropics (Figure 1d). Mid-level clouds vary a lot between models in the polar regions. Modeled high clouds do not show a systematic bias. Compared to passive remotes sensing evaluation used byZhang et al. , the CALIPSO evaluation suggests that the inter-model spread in low, mid, high cloud cover is reduced, the underestimate of mid-level clouds by all models is confirmed, and the high latitude clouds are significantly different than the ones seen by passive remote sensing.
 Near 60°S (Figures 1a–1d) where most of the climate models exhibit substantial bias against the flux observed at top of the atmosphere (TOA) [e.g., Cole et al., 2011], the total cloud cover (Figure 1a) is not particularly biased compared to others latitudes. Four of the 5 models reproduce surprisingly well the observed low and total cloud cover. This suggests a model's ability to produce the right cloud cover may not explain why the modeled TOA flux is too large here. The amount of condensed water (liquid and ice, or the cloud optical depth) is more likely the reason of this discrepancy.
 Over continents (Figures 1i–1l), the biases in modeled cloud cover are similar to the previous section. In deep convection along the ITCZ, the observed high altitude clouds cover is higher there (70% Figure 1j) than above ocean (50%, Figure 1f) because the continent is warmer and produces stronger convective motions. The models reproduce roughly this continent/ocean contrast at high altitude, but the simulated mid- and low-level continental clouds (Figures 1k and 1l) spread significantly more than the global/ocean ones (Figures 1g and 1h), especially in the tropics and polar regions. It confirms that the representation of continental clouds remains a challenge for climate models.
3.2. Zonal Mean Cloud Profiles
 The zonal mean cloud profiles observed by CALIPSO-GOCCP (Figure 2a) illustrate that clouds follow the atmospheric circulation: deep convection along the equator leads to high clouds, the subsidence branches of the Hadley cells around 25° preclude cloud formation in the free troposphere, the mid latitude storm tracks cover the entire troposphere, and boundary layer clouds (z < 4 km) produced by small local convection occur at all latitudes.
 Although the models roughly reproduce this structure (Figures 2b–2f), the quantitative comparison with observations (Figure 2a) shows systematic model bias. The highest simulated clouds are between 14 km (CNRM-CM5 and HadGEM2-A) and 18 km (CanAM4) but are near 16 km in CALIPSO-GOCCP. All models generate too many high-level clouds compared to observations, at all latitudes, including in the subsidence branch of the Hadley cells. Most models produce too many boundary layer clouds in mid and high latitudes, and too few in the tropics. This underestimation may, in some cases, be caused by an overestimation of the high clouds, which tends to mask the low clouds in the lidar simulator.
 Exchanges of energy in the tropics influence the climate of the entire Earth, and tropical clouds play a key role in its redistribution. All cloud types influence the tropical climate, but inter-model studies [Bony and Dufresne, 2005] suggest that the representation of low level clouds in subsidence regions (nearly 65% of the tropics) impacts substantially the tropical clouds climate sensitivity.
 As describe in Section 1, CALIOP can detect the fractionated and small shallow cumulus in subsidence regions close to the surface [e.g., Konsta et al., 2012]. The observed boundary layer cloud cover is larger than 15% almost everywhere (Figure 3a), with a maximum of 100% in stratocumulus over the East part of oceans, where the subsidence is strong. Figures 3b–3f exhibit the model cloud covers together with the regions of subsidence, identified by a positive mean air mass vertical speed at 500 hPa (w500). Stratocumuli are reproduced by most models but their horizontal extent is underestimated, in particular along the Californian and Australian coasts. The shallow cumulus cloud cover is significantly underestimated (10% instead of 25%) by half of the models. Further analysis (not shown) indicates the high cloud cover in subsidence regions (w500 > 0) is small in both models and observations, which suggest that the model underestimation of low clouds is not due to masking by higher clouds (Section 3.2).
 Tropospheric polar clouds modulate the radiation reaching the surface and regulate the Arctic climate [e.g., Morrison et al., 2011]. Most reliable observations of Arctic clouds were collected by ground based sites [Shupe et al., 2006; de Boer et al., 2009] that do not provide a complete view of the region. Thanks to its capability to observe clouds above reflective surface, CALIPSO provides relevant information over the polar region equatorward of 82° latitude.
 The annual mean low-level cloud cover (z < 3.36 km) observed by CALIPSO-GOCCP in the Arctic (Figure 4a) shows that the dry atmosphere above continents contains a smaller but significant amount of low clouds (30% to 45%), except over Greenland and high altitude regions where it is lower (<30%). Above ocean, the moister atmosphere produces a larger low-level cloud cover (typically > 60% up to 80%). To first order, their significantly asymmetric distribution is related to the sea surface temperature: cloud covers are largest above the warmest Barents and Greenland seas, and smaller (50–60%) above the cold Beaufort sea. All models (Figures 4b–4f) except MPI-ESM, reproduce this asymmetry, but not the correct low cloud cover, and the inter-model spread is large especially in the Arctic sea (between 40% and 70%). Similarly, the models mimic the ocean-continent contrast but the cloud cover is often substantially different from observations.
 The annual cycle of the monthly mean low cloud cover in the Arctic is presented in Figure 4g. Observations show that it is maximum in May and October (>50%) and minimum in winter between December and March (35%) with a secondary minimum in July (42%). Most of models reproduce the winter minimum cloud cover, but none reproduce the July minimum, except CNRM-CM5 for which it is the only minimum. No model simulated the maximum in fall season. The minimum model cloud cover in winter is strongly variable: between 20% and 50%, vs. 35% in the satellite data.
 In this paper, active remote sensing satellite observations are used to evaluate cloud cover and cloud vertical structure simulated by five climate models. To ensure that differences between model and observations can be attributed to model defects, we used CMIP5 climate models including the COSP/lidar simulator, which mimics the lidar profile that would be observed by CALIPSO over the modeled atmosphere. We compared the “model + simulator” outputs with CALIPSO-GOCCP observations that are consistent with the simulator algorithm.
 Results show that all models underestimate the total cloud cover (51% to 62%) against observations (64%) at all latitudes, except in the Arctic. Low- and mid-level altitude clouds are underestimated by all the models (except in the Arctic), while high altitude cloud cover is overestimated by some models. The discrepancy between models and observations, and the inter-model spread is more pronounced over continents than over ocean. The zonal cloud fraction profiles (every 480 m in the vertical) indicate that some models shift the altitude of the clouds along the ITCZ by 2 km (higher or lower) compared to observations. The models hardly reproduce the cloud free subsidence branch of the Hadley cells, and the high-level cloud cover is often too large. In the tropics, the low-level cloud cover (29% in CALIPSO-GOCCP) is underestimated by all models in subsidence regions (16% to 25%). In the Arctic, the simulated winter low-level cloud cover varies between 20% and 55%. Despite the significant discrepancy between modeled and observed cloud covers, most models roughly reproduce the observed spatial distribution of low clouds over open ocean. The pronounced seasonal cycle observed in low-level Arctic clouds is hardly simulated by some models.
 This article shows how CALIPSO-GOCCP observations and COSP/lidar simulator can provide simple and robust benchmark for identifying systematic multi-model deficiencies in the description of the cloud vertical structure at global scale, and in the cloud cover in regions typically hard to observe (e.g. poles, tropical oceans, continents). Future work will include more advanced evaluation using complementary observations from others A-train instruments.
 We would like to thank NASA, the ICARE and ClimServ centers for access to the CALIPSO level 1 data, and CNES for supporting the development of CALIPSO-GOCCP. We also acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (Canadian Centre for Climate Modelling and Analysis, Centre National de Recherches Meteorologiques, Institut Pierre-Simon Laplace, Max Planck Institute for Meteorology and Met Office Hadley Centre) for producing and making available their model output. Thanks are due to Vincent Noel for his internal review and to the anonymous reviewers for their useful comments that helped to improve the manuscript.
 The editor thanks the two anonymous reviewers.