Cancelling the Tropical Cancellation

Guest Post by Willis Eschenbach

There’s a much-cited paper (129 citations) from 1994 called “On the Observed Near Cancellation between Longwave and Shortwave Cloud Forcing in Tropical Regions” by J. T. Kiehl (hereinafter Kiehl1994), available here. The paper makes the following claim (emphasis mine):

ABSTRACT

Observations based on Earth Radiation Budget Experiment (ERBE) satellite data indicate that there is a near cancellation between tropical longwave and shortwave cloud forcing in regions of deep convective activity. Cloud forcing depends on both cloud macrophysical properties (e.g., cloud amount, cloud height, etc.) and on microphysical properties (e.g., cloud particle size, particle shape, etc.). Hence, the near cancellation in the tropics could be due to either the macrophysical or the microphysical properties of these clouds, or a combination of these effects. 

Now, to me that’s a pretty curious and surprising claim. The paper says that both in the Indonesian Region, as well as over the entire tropical Pacific, deep convective tropical clouds have no net effect on radiation, ostensibly because the longwave (positive) and reflected shortwave radiation (negative) cancel each other out. Let me call this the “cancellation hypothesis”. Kiehl supports the cancellation hypothesis inter alia with his Figure 1:

Figure 1. The first figure in Kiehl1994. The vertical axis shows shortwave cloud forcing, which is the amount of sunlight reflected by clouds in watts per square metre (W m-2). The horizontal axis shows the longwave cloud forcing, which is the change in top-of-atmosphere longwave forcing from the clouds. By convention, the reflected solar radiation is shown as negative, presumably because it is cooling the earth. Data is from April 1985.

And that looks pretty convincing … but, despite the 129 citations of the paper, I’m a suspicious fellow who believes firmly in the famous fallibility of experts. So I thought I’d use the CERES data and see if the cancellation hypothesis held up. Figure 2 shows that result, for the same Indonesian Region used by Kiehl.

Figure 2. Replication of the Kiehl study, using the ten-year April averages for the specified region to remove annual variations. Each square symbol represents a 1°x1° gridcell (N = 1500).

Instead of one single month’s data, I’ve used the averages of the ten Aprils in the CERES dataset.

Now, as you can see from Figure 2, the Kiehl hypothesis of cancellation has a big problem. The CERES results do not bear out Kiehl’s claims in the slightest. Instead, they support my hypothesis that increased tropical clouds cool the surface. As you can see, on average the loss from the reflected sunlight is about 20% or so greater than the gain from increased IR. This means that there is no cancellation. Instead, the clouds have a net cooling effect.

The paper goes on to say:

One feature of the cloud radiative forcing obtained from the Earth Radiation Budget Experiment ( ERBE) is the near cancellation between the longwave cloud forcing and the shortwave cloud forcing in tropical deep convective regions. This result was clearly shown by Kiehl and Ramanathan ( 1990) for the Indonesian convective region (here reproduced in Fig. 1 ). Further analysis of tropical deep convective regions of net cloud radiative forcing indicates that this is a ubiquitous feature that occurs over either ocean or land regions.

This is a more expansive claim than the one in the Abstract. Here the paper says that the cancellation happens over both land and ocean. So I thought I’d divide the Indonesian Region shown above into land and ocean regions. In addition, rather than average the months as in Figure 2, Figures 3 & 4 show all available April data for the entire time period. First, Figure 3 shows the land. I have colored the points by surface temperature of the gridcell.

Figure 3. As in Figure 2, but showing only the land. N=1,320.

Well, this shows that whatever might be happening in the Indonesian Region in the way of a linear relationship between reflected shortwave and longwave, it is definitely NOT happening over the land. The land shows little in the way of any relationship between shortwave and longwave.

Figure 4 shows the ocean data for the same Indonesian region.

Figure 4. As in Figure 2, but showing only the ocean. Note that the scale is slightly larger, to include all of the individual data. N=17,736

Well, this clarifies matters somewhat. First, about 40% of the land gridcells, but only about 10% of the ocean gridcells, have longwave cloud forcing greater than the shortwave forcing. Next, the land is pretty tightly clustered, with no apparent pattern. The ocean is different. Over the ocean the longwave is proportional to the shortwave … but it is a long ways from cancelling out. Instead, there’s a net cooling of -13.7 watts per square meter over the region. And the amount of cooling increases as the forcings increase. By the time the cloud reflections are up to 100 W m-2, the longwave is only up to 75 W m-2. That is a cooling from the clouds of about 25 W/m2, and not a cancellation under any meaning of the word.

In addition, there are a number of the warmest gridcells (red) which are on the left of the group (lots of reflection, little longwave).

Kiehl goes on to show his Figure 2, which compares the entire tropical Pacific region, from 10N 140E to 10S 90W. He shows a different kind of graph for this region, viz:

Figure 5. Kiehl Figure 2, showing the longwave and shortwave cloud forcing separately as functions of temperature.

Based on this graph, he makes the even more expansive claim that the cancellation of long-and shortwave radiation occurs across the Pacific, viz:

This figure illustrates that the cancellation between these two forcings occurs not only in the western tropical Pacific region of Fig. 1 but also across the entire tropical Pacific Ocean region. Even in regions of colder SSTs (298 K) the cancellation is apparent.

I found his style of graph in Figure 2 to be notably uninformative regarding the purported Pacific-wide cancellation. So I repeated his Figure 1 style of graph using the Pacific data, to see whether the forcings actually cancelled all across the Pacific. Figures 6 and 7 shows that result, for land and ocean, and reveals that there are large problems with this second claim as well.

Figure 6. As in Figure 2, but covering a larger area, the entire tropical Pacific Ocean as specified in the title. This figure shows land only. N=348.

Again, there is no clear pattern over the land, merely a cluster of data.

Figure 7. As in Figure 6, but covering ocean only. Note the slightly larger scale than in Figure 4. N=63,132.

This actually is pretty interesting. First off, again the ocean and land are different, with the land results clustered as in the Indonesian Region. Regarding the ocean, in only 5% of the gridcells does the longwave ever exceed the shortwave (area above the central diagonal line). In order for Kiehl’s cancellation hypothesis to be true, the average of a number of gridcells over an area would have to fall on the central diagonal line … which is clearly impossible with only 5% of them above the line.

Nor do things get better when we look at the entire dataset, and not just the April data. Figure 8 shows all of the data for the Pacific-wide area shown in Figure 7 (ocean only).

Figure 8. All months of data for the Pacific-wide area as in Figure 7

Next, rather than cancellation, there are a whole lot of gridcells where the longwave is smaller, and often much smaller, than the reflected shortwave. When there is solar reflection of a hundred watts per square metre, the longwave is only around sixty watts per square metre, for a full 40 W/m2 of cooling. And even on average the cooling is nearly 20W/m2 … not what we call cancellation on my planet.

Finally, his claim that “Even in regions of colder SSTs (298 K) the cancellation is apparent” is not upheld by the data. The temperature of 298 K [25°C] is shown in blue in Figure 7, and it is the farthest from cancellation of any of the data.

CONCLUSIONS

The main result is that the CERES data clearly and emphatically falsifies the cancellation hypothesis. In general, the longwave and shortwave are far from cancelling each other in the tropical deep convective areas.

A secondary result is that this clearly shows how the politicization of the field has affected the scientific process. Kiehl’s claims were very tempting to the theorists and modelers, because cancellation meant that they didn’t have to concern themselves with the deep convective processes, aka thunderstorms—the could simply repeat Kiehl’s claim that the shortwave and longwave cancelled each other out. And as a result, when more detailed data became available, the original claims of Kiehl1994 were never questioned.

Now, all we need is some automated method to notify the 129 people who cited the cancellation hypothesis in other scientific papers that the rumored cancellation has been cancelled for the duration …

All the best,

w.