Smith and Rutan (2003) have performed the analysis on the ERBE OLR fluxes
data from land and ocean separately, they found that the first and second PCs contributes 72%-78% and 8%-10% to the total variance respectively. Their first PCs for all seasons over land are extremely similar as they describe the response of the surface to solar radiation heating. It was found that the curves peak at about noon and are symmetric about the peak with the minima positioned at night time.
Their corresponding first EOF dominated the diurnal cycle over land, implying that the daily (non-decomposed) signals for each region would look largely similar to the first PC. They found the strongest signal to be mostly over deserts where there is little moisture to permit solar heating to be converted into latent heat [Smith and
Rutan, 2003].
Figure 2.5.: Map of second EOF of OLR for land in boreal summer time, taken from [Smith and Rutan, 2003]
They have observed the second PCs for all seasons again to be quite similar to each other, taking the shape of a sinusoidal curve, peaking at approximately 1600 local solar time (LST) with a minimum at around 0800 LST which makes it about 4 hours out of phase with PC1. Their map of EOF2 (see fig.2.5) for boreal summer over land shows maxima over northern Africa, the Kalahari Desert and the Middle Eastern deserts. This signal is due to a combination of cloud development, surface and orography. Deep convective systems over regions of Brazil and Africa have
shown negative values of EOF2. This shifts the maxima of the diurnal cycle to the morning due to the effect of high cloud development during the afternoon. The PC3 for land showed two maxima during the day and the EOF3 showed that the regions with the maxima overlap with regions of deep convective activities.
In contrast to [Smith and Rutan, 2003], Comer et al. (2007) have found their first PC contributes 82.3% of the total variance. They found that the first PC carries an asymmetric pattern, which shifts the peak from noon [Smith and Rutan, 2003] to around 1300 LST, thought to be primarily due to the finite heat capacity of the surface but also influenced by the atmosphere and surface feedbacks to the heat- ing. This suggests that the analysis results may be highly sensitive to the dataset temporal resolution or region observed. They have suggested that the main OLR effect in the EOF2 is mainly due to convective cloud over land triggered by sur- face heating and that the cloud optical depth maximises at about 1700-1800 LST. Thus, the corresponding OLR PC2 signal minimises at this time with a maximum at about 0800 LST. They also analysed the EOF2 signal with respect to the topo- graphical features. They have found in the map of EOF2 that the strongest signals for convective clouds are over mountains (see Fig. 2.6).
The topographical features 1, 2, 4, 6 and 7 in Fig. 2.6 expressed strong signals in the EOF2 component, indicating a relationship between topography forced uplift and convective clouds. However, features 4 and 5 are more subtle in terms of size of spatial distribution, which suggests topographical induced uplift also requires the appropriate atmospheric conditions for the convective clouds to develop.
There is also a big region of strong signal just below feature 6 which overlaps with the Kilimanjaro mountain and Victoria lake. This will be investigated in chapter 5 where we discuss our diurnal analysis results.
In their results, the combined variance from the PC1 and PC2 adds up to over 95% of the variance. The PCA technique is powerful in decomposing the diurnal cycle as the physical signals of surface heating and cloud response in EOFs 1 and 2 are clearly identified. However, it would be important to observe the location at which the residual signal concentrates.
Allan et al. (2007) have used averaged GERB data for July 2006 over equatorial
Africa (7◦45◦E, 10◦S10◦N), they have found the cycle of OLR to peak sometime near noon, a lower temporal resolution sample (3 hours) of the data (compared with
Figure 2.6.: Map of second EOF with topographical features in North Africa, taken from [Comer et al., 2007]
GERB data resolution at 17 minutes) was employed for the purpose of comparison with climate models, hence the exact timing of the peak is uncertain. The diurnal cycle of OLR produced by the model peaks 3 hours earlier than observational data, where the resolution of the model output could have partially contributed to the large discrepancy.
Allan et al. (2007) have observed differences in their map of model and GERB
Sahara and a correspondent underestimate in RSR over the sub-Saharan regions in the clear-sky comparison. The clear-sky RSR differences are likely to be due to an unrealistic spatial distribution of modelled surface albedo, whilst the OLR differences are likely to arise from the high mineral-dust optical depths which are not represented by the models. Most of these significant clear-sky differences in both RSR and OLR are also highlighted on the total OLR and RSR plots. The total plots also indicate modelling errors in higher-altitude cloud properties and the cloud fraction over the Ethiopian highlands. The net radiation comparison of GERB and the model exhibits a substantial overestimate in the model over the marine stratocumulus regions and the ITCZ.
Although, the studies reviewed in this section have analysed the diurnal cycle of OLR, some of them specifically analyse the OLR cycle over land. However, they have all treated land as a whole domain, which may not necessarily be the best selection of data to analyse, as different surface types will show various surface heating cycles, and could possibly lead to a different diurnal variation of the convective cloud signal. Thus, a separation of the data according to surface types may be appropriate for analysing both OLR and cloud cycles.