Characterization of vertical cloud variability over Europe using spatial lidar observations and regional simulation

Abstract

In this paper we characterize the seasonal and inter-annual variabilities of cloud fraction profiles in both observations and simulation since they are critical to better assess the impact of clouds on climate variability. The spaceborne lidar onboard CALIPSO, providing cloud vertical profiles since 2006, is used together with a 23-year WRF simulation at 20 km resolution. A lidar simulator helps to compare consistently model with observations. The bias in observations due to the satellite under-sampling is first estimated. Then we examine the vertical variability of both occurrence and properties of clouds. It results that observations indicate a similar occurrence of low and high clouds over continent, and more high than low clouds over the sea except in summer. The simulation shows an overestimate (underestimate) of high (low) clouds comparing to observations, especially in summer. However the seasonal variability of cloud vertical profiles is well captured by WRF. Concerning inter-annual variability, observations show that in winter, those of high clouds is twice the low clouds one, an order of magnitude that is is well simulated. In summer, the observed inter-annual variability is vertically more homogeneous while the model still simulates more variability for high clouds than for low clouds. The good behavior of the simulation in winter allows us to use the 23 years of simulation and 8 years of observations to estimate the time period required to characterize the natural variability of the cloud fraction profile in winter, i.e. the time period required to detect significant anomalies and trends.

Appendices

Appendix 1: Lidar equation

The scattering ratio SR is given by (1):

$$SR\left( z \right) = \frac{{ATB_{tot} \left( z \right)}}{{ATB_{mol} \left( z \right)}}$$

(1)

where $ATB_{tot}$ and $ATB_{mol}$ are respectively the attenuated backscattered signals for particles and molecules and for molecules only and are given by (2) and (3):

$$ATB_{tot} (z) = \left( {\beta_{sca,part} (z) + \beta_{sca,mol} (z)} \right).e^{{ - 2\eta \int_{{z_{TOA} }}^{z} {\alpha_{sca,part} \left( z \right) + \alpha_{sca,mol} \left( z \right).dz} }}$$

(2)

$$ATB_{mol} (z) = \beta_{sca,mol} (z).e^{{ - 2\eta \int_{{z_{TOA} }}^{z} {\alpha_{sca,mol} \left( z \right).dz} }}$$

(3)

ATB_mol and ATB_tot products are averaged vertically to obtain SR over 40 layers (Chepfer et al. 2008, 2010).

β_sca,part, β_sca,mol are lidar backscatter coefficients (m⁻¹ sr⁻¹) and α_sca,part and α_sca,mol attenuation coefficients (m⁻¹) for particles (clouds, aerosols) and molecules. η is a multiple scattering coefficient that depends both on lidar characteristics and size, shape and density of particles. It is about 0.7 for CALIPSO (Winker et al. 2003; Chepfer et al. 2008).

Figure 10 illustrates two instantaneous SR profiles to help understand what a lidar signal looks like and how cloud detection is computed in this study. Above 10 km, SR(z) is around 1, indicating clear sky for both profiles. High clouds are detected in both profiles between 8 and 10 km: SR(z) of the blue profile reaches the value of 8 and while SR(z) of the red one goes up to 22. The magnitude of SR(z) depends on the cloud optical thickness from the Top Of Atmosphere (TOA) to the level z and the cloud microphysical properties such as the size of the particle or its shape. While the signal is fully attenuated for the red profile below 8 km (SR(z) is almost zero), the blue profile still detects low clouds around 2 km.

Fig. 10

Two instantaneous observed SR vertical profiles (blue around [5°E; 47°N] and red around [5°E; 43°N]) in 2009/01/19 at night, and vertical black line represents the SR = 5 threshold for cloud detection. Red box in (a) represents the Mediterranean Sea area while the blue box is for Europe area

Appendix 2: Simulated cloud fraction maps

Figure 11 shows that: for high clouds, a north–south gradient exists in winter with about 10 % of clouds over North Africa and more than 50 % above continental Europe, while in summer, this gradient is north–west/south–east, with almost no high clouds over Turkish and eastern part of Mediterranean basin. In winter, most mid and low clouds occur above the north-eastern part of Europe. In summer, very few mid and low clouds are simulated and they are mostly induced by orography.

Fig. 11

Winter cloud fraction maps (CF ^T_WRF : cloud fraction computed from the model without lidar simulator) averaged from 2006 to 2011 for simulation low clouds (a), mid-clouds (b) and high clouds (c). d–f are the same but for summer

Appendix 3: Simulated winter 2010 high clouds anomaly

See Fig. 12.

Fig. 12

Winter 2010 high clouds anomaly computed with CF ^T_WRF+sim relative to the average high cloud map of winters from 1990 to 2011

Appendix 4

The CALIPSO undersampling error estimation from observed cloud fraction profiles is defined as:

$\varepsilon \, \left( z \right) = \left| {CF_{GOCCP} \left( z \right) - CF_{GOCCP}^{T} \left( z \right)} \right|with\,CF_{GOCCP}^{T} \left( z \right)$ a theoretical cloud fraction that we would have with a complete sampling (observations over all the grid-boxes every 00 UTC).

We define α(z) as the relative model bias, so $\alpha \left( z \right) = \frac{{ CF_{WRF + sim} \left( z \right)}}{{CF_{GOCCP} \left( z \right)}}$

We used a set of different samplings to test if α(z) can be considered as constant, i.e. independent of the number of profiles in the sampling. To do that, since we need both observations and simulation to test this hypothesis, we reduced the CALIPSO sampling using only 1 profile over 2 (test 7), 1 over 3 (test 6), and so on down to one profile over 20. Table 4 presents the results of these tests and indicates the α values for low, mid and high clouds. This shows that if the number of profiles become greater than 1/15 of the CALIPSO sampling, α(z) can be considered as nearly constant.

Table 4 Computing model biases over continent ($\alpha = \frac{{CF_{WRF + sim} \left( z \right)}}{{ CF_{GOCCP} \left( z \right)}}$) for low clouds (1st row), mid clouds (2nd row) and high clouds (3rd row) by testing different samplings (test 1 means we extract 1 profile over 20 and test 8 means we extract all the profiles)

We deduce that: $\alpha \left( z \right) = \frac{{ CF_{WRF + sim}^{T} \left( z \right)}}{{CF_{GOCCP}^{T} \left( z \right)}}$

and ε (z) can be written as $\varepsilon = \frac{{\left| {CF_{WRF + sim} \left( z \right) - CF_{WRF + sim}^{T} \left( z \right)} \right| }}{\alpha \left( z \right)} = \frac{\beta \left( z \right)}{ \alpha \left( z \right)}$ with β(z) defined as the error of undersampling estimated by the simulation.