Text S3.
Understanding the “Faster than FAT” Response in the Tropics of IPSL
To determine if the “Faster than FAT” response in IPSL may be an effect of the MISR discretization we investigate how a “thin” distribution of high clouds appears to move upward given the MISR grid discretization. Specifically, we assume a gaussian distribution for the vertical profile of high clouds (Figure S5 ). We find that if the width (standard deviation) of the underlying CTH distribution is 700 m or larger, then the WCTH captures the increase in CTH well. However, if the width is less than 700 m (400 m is used in the example inFigure S5 ), the MISR discretization poorly captures the slow rise in the peak position of the distribution (left panel) and WCTH can appear to rise faster or slower than the true peak depending on whether the true peak is above or below the bin mid-point; and in the limit where there is no width in the distribution (meaning all CTH occurs at a single fixed height), the WCTH would appear as a flat except for a single step of 2 km (the size of the MISR vertical bins) as the CTH moves from one MISR bin to the next. In Figure S5 a gaussian distribution of CTH with a width of 400 m is slowly lifted with the position of the peak moving from 12 to 14 km (top right panel) and the resulting MISR CTH distribution and WCTH are shown in the top left panel and bottom left panels, respectively. The bottom right plot is designed to mimic the results produced by tropical cloud in IPSL by adding white noise to the position of the Gaussian peak. In summary, IPSL has a narrow high cloud distribution in the tropics, the rapid rise in WCTH seen in IPSL after 2040, and slow down after 2060, as well as the increased variability between 2040 and 2060 (as seen in Figure 7 of the main text) are consistent with how a narrow cloud distribution would be seen by the MISR simulator. It is noted that if the distribution of CTH in the real world were as narrow as suggested by the IPSL model, the WCTH based on the MISR observations would have this same undesirable property. However, that does not appear to be the case (as can be seen in Figure S1 , as observed high cloud is spread across several MISR height bins). Of course, it is worth noting that the MISR retrievals have some random error associated with them (largely due to uncertainty in wind correction, see Marchand et al. 2010), which will cause the observed distribution to be broader than the true distribution. Given the potential for artifacts, it might be wise to increase the resolution of the MISR CTH-OD product and simulated output, in the future.