Mesoscale cellular convection Boundary layer clouds exhibit a tremendous range of structures in the mesoscale, from relatively featureless and homogeneous stratus clouds to intricate convective patterns on with dominant scales of several tens of kilometres. Often the clouds form mesoscale convective cells, either open (sharp updraughts with clouds around the edges of the cells and clear slow subsidence in the centre) or closed (slow ascent and clouds at the centre of the cells, with sharp downdraughts at the edges) form.

The nature of mesoscale cellular convection (MCC) has been the focus of considerable attention since the 1960s. The similarity to Benard-Rayleigh (B-R) convection are often commented upon. However, the aspect ratios (ratio of cell width to height) can be very much larger for atmospheric MCC. Early laboratory and numerical model studies by Krishnamurti (J. Atmos. Sci., 32 1353-1363, 1364-1372 and 1373-1383, 1975) suggested that the preferred form of MCC (open or closed) depends upon whether there is large-scale ascent (closed cells) or descent (open cells). Determining the distribution of open and closed mesoscale cellular convection over the oceans Diagnosis of cellular convection type using MODIS. From examination of many MODIS 1km resolution images, a suitable diagnosis of open celled convection can be obtained by examining the power contained in the high frequency (0.2-0.5 km^-1) end of the liquid water path LWP power spectrum. Open cells generally contain considerably more power. In addition, the skewness of LWP (cloud+clear) also tends to be larger in the open celled convection. The following criteria are chosen to determine whether each 256x256 km scene has open or closed mesoscale cellular convection:

OPEN MCC: LWP POWER (0.2-0.5 km^-1) > 250 g² m^-4

and LWP SKEWNESS > 1.8

and CLOUD FRACTION 0.1-0.75

The reason that cloud fraction alone is not a great predictior is that many scenes contain edges between e.g. open celled convection and clear air, or open celled convection and closed cell convection.

CLOSED MCC: Open MCC not diagnosed

and Characteristic length scale > 10 km

The characteristic length scale is derived using the LWP power spectrum. An example here (Fig. 1) shows three 256x256 km LWP images and their respective power spectra below. First, an "average" wavenumber is calculated using the LWP power spectrum as a weighting function (see e.g. Jonker et al., J. Atmos. Sci., 56, 801-808, 1999). The inverse of this value is a characteristic lengthscale (shown by the filled triangles on the abscissae) but because it is sensitive to the power spectrum at low freqencies, is not a good measure of the size of the mesoscale convective cells. However, it is useful for separating the power spectrum into a MCC component (scales larger than the inverse average wavenumber) and a large-scale component. We then calculate the exponent of the power spectrum in the "intertial subrange" defined here as being wavenumbers larger than 1.5 times the average wavenumber. The straight dotted line is the resulting power law fit to the LWP spectrum in the "intertial subrange". The solid circles and the dashed curves above and below the spectra represent the smoothed power spectrum and the 95% confidence limits on the spectrum. The point at which the upper 95% confidence interval curve falls below the power law fit (b and c below) is designated as the integral scale lambda (vertical dashed lines in Fig. 1 below). We found that this matched well with the characteristic length scale of the convective cells estimated by eye. In (a) below, the upper 95% curve never passes below the power law fit. We found this only occurred when the eye-estimated length scale was very small, close to the resolution of the data. We designate the lenthscale undetermined in these cases, but note that it is small.

Figure 1. Three examples showing stratocumulus clouds with markedly different convective length scales. The two-dimensional LWP power spectra are shown below the respective images. For case (a) the integral scale cannot be determined because it is too small. For cases (b) and (c) the integral scales are 10.3 and 48.2 km respectively.

The characterisation process is described graphically below in Fig. 2. Here we see a 1168x1023 km image of LWP at 1km resolution from MODIS. The two boxed regions (both 256x256km) are selected because they represent open and closed cellular convection. Their power spectra are shown below (left upper panel: note that the wavenumber has been multiplied by the power in this plot). The open celled convection has much more power at high freqencies than the closed cellular convection. The probability density functions for the two cases are shown (upper right panel). The increased skewness in the open cellular convective case is apparent. The lower left panel shows a plot of the power in the 0.2-0.5 km^-1 range. against skewness. The dotted lines distinguish the open celled convection (to left and top) from non open celled convection. The lower right panel shows the classification corresponding to the image above into closed (red) and open (blue) cellular convection, derived using 256x256 km subscenes overlapped by 32 km. The diagnosis matches quite well with where the eye perceives there to be open and closed cellular convection.

Diagnosis of open and closed cellular convection using MODIS. Open cells are characterised by larger power at high wavenumbers and by higher skewnesses than are found in closed cellular convection. The method leads to a reasonably satisfactory, although by no means perfect, diagnosis of the locations of open and closed cellular convection.