Booker, North, and Willis on the IPCC Amazongate affair

Firstly, I would like to thank Dr Nepstad for being willing to engage on scientific topics. However, I respectfully disagree on one point in particular.
From Dr. Nepstad’s comments:

5. Is there any evidence that climate change will provoke drying in the Amazon?
Yes. See Mali et al. 2007. Science. Most global circulation models predict a decline in rainfall in the eastern Amazon, which is where the forests are already most stressed.
Note: These climate model runs do not include the inhibitory effect of dense smoke on rainfall, and only a few include the inhibitory effect of deforestation itself on rainfall.

Professor Koutsoyiannis from the NTUA demonstrated that predictions of local changes in precipitation from GCMs are inconsistent with reality, even in hindcast; and model predictions are less accurate on larger time scales than shorter ones. See peer-reviewed reference below. This is from 2008 – like most of Dr Nepstad’s references, well after the AR4 document cut-off – but Dr Nepstad’s arguments hinge on this. Due to the Hurst phenomenon, precipitation levels are prone to substantial large scale changes without any human interaction. Add a further unpredictable element of human interaction and it is anyone’s guess what might happen. Indeed, I would go as far to say, based on the knowledge that we have, that human induced climate change is as likely to save the rainforests as it is to damage them.
It shows that using the peer-reviewed literature, any conclusion can be justified; the science is too uncertain. The IPCC should be presenting an objective, balanced view and is failing in this. Of course, Dr Nepstad is entitled to put forward his views on the science. But the current IPCC structure allows advocates to narrowly present one viewpoint when there are many valid viewpoints underpinned by scientific research.
On the credibility of climate predictions
Koutsoyiannis, D., A. Efstratiadis, N. Mamassis, and A. Christofides, On the credibility of climate predictions, Hydrological Sciences Journal, 53 (4), 671–684, 2008.

@ur momisugly Spence_UK says:
June 29, 2010 at 6:45 am
“Firstly, I would like to thank Dr Nepstad for being willing to engage on scientific topics.”
Well, so far, I wouldn’t characterize his posting as an engagement, but rather something akin to a “drive by.” I saw very few questions on this thread which he answered. From my interpretation of his posting, I read ‘I’m the expert,(he even referenced his work as proof that his work was correct), myself and a computer model says I’m right, so just believe it and move on’. He couldn’t even be bothered to clear up the “slight reduction in rainfall” vs. “drought” question. Compared to other published scientists that have posted here, his explanations for his thoughts are lacking. It would seem to me, most would defend their work a little more fervently than this effort. With that in consideration, I’m wondering if it was the good doctor or not. Right now, I’m writing it off as a pretender in a lame attempt to close the conversation.

Show me that rising CO2 increases the severity of droughts.
===========================

Spence … thanks for that. I was unaware of the followup study… it appears not to have gained much traction outside of Climate Audit where I go but rarely. If anything, the increased number of stations increases the mismatch between individual model runs and observations.
However, it is not possible from a quick perusal of the poster presentation to determine whether the expanded study (which surely is as worthy of journal publication as the original – even as a comment) addressed the flaws identified by Gavin Schmidt… viz
They are using single realisations of model runs, and so they are not testing the forced component of the response (which can only be determined using ensembles or very long simulations). By correlating at the annual and other short term periods they are effectively comparing the weather in the real world with that in a model. Even without looking at their results, it is obvious that this is not going to match (since weather is uncorrelated in one realisation to another, let alone in the real world). Furthermore, by using only one to four grid boxes for their comparisons, even the longer term (30 year) forced trends are not going to come out of the noise.
Remember that the magnitude of annual, interannual and decadal variability increases substantially as spatial scales go from global, hemispheric, continental, regional to local. The IPCC report for instance is very clear in stating that the detection and attribution of climate changes is only clearly possible at continental scales and above. Note also that K et al compare absolute temperatures rather than anomalies. This isn’t a terrible idea, but single grid points have offsets to a co-located station for any number of reasons – mean altitude, un-resolved micro-climate effects, systematic but stable biases in planetary wave patterns etc. – and anomaly comparison are generally preferred since they can correct for these oft-times irrelevant effects. Finally (and surprisingly given the attention being paid to it in various circles), K et al do not consider whether any of their selected stations might have any artifacts within them that might effect their statistical properties.
Therefore, it comes as no surprise at all that K and colleagues find poor matches in their comparisons. The answer to their effective question – are very local single realisations of weather coherent across observations and models? – is no, as anyone would have concluded from reading the IPCC report or the existing literature. This is why no one uses (or should be using) single grid points from single models in any kind of future impact study.
The last point is germane – Nepstad was referring to ensemble projections for the East Amazon region, so a study showing that single model runs correlate poorly with small clusters of grid cells is apples and oranges – and does not invalidate his conclusions, seems to me.

Nepstad’s confident assertion that my comment reveals “an important misinterpretation of the IPCC statement” actually reveals more about what could charitably be called the IPCC’s failure to communicate.
This has been compounded by the supporters of the IPCC, including Nepstad, who have equally lacking in this department. They are in a poor position to lay down ex cathedra assertions on what might or might not be true, especially in offering a novel interpretation of the IPCC statement which cannot be adduced from any facts that it or its supporters have so far offered.
Nepstad might reflect, therefore, that it is the duty of those who seek to communicate information to make themselves clear – it is no part of ours to puzzle out what they mean to say, in the absence of any coherent elucidation.
In that context, if he wishes to posit an entirely new argument to the effect that that the “current” cycle of drought is one which has been super-imposed on the historical climate pattern, he might start by declaring that he is introducing a new argument, ex post facto, rather than complain that we have misinterpreted the previously opaque communications.
If Nepstad then wishes to argue about variations in historical climate patterns, he might start by identifying those patterns. For instance, we see Coe et al 2002 who tells us that there is considerable climatic variability in the Amazon basin”, with “short (∼3–4 years) and long (∼28 years) modes of precipitation variability. Others write of different and longer cycles.
Thus, in order to accept Nepstad’s assertion, we would need evidence (rather than speculation and the more typical ex cathedra pronouncements) that the droughts experienced over the relatively short timeframe of 1983-2005, to which he refers, differs significantly from climate patterns in the past. That information, so far, seems to have been notably lacking in his published work.
Further, while he refers to droughts in 1983 and 2001 (with a reference to other short-lived episodes) he must also be aware of reports of “major flooding” in Amazon basin in the 1984-2001 period.
And, although as far as his published record goes, history seems to stop in 2005, he will undoubtedly be aware episodes of major flooding in the region for every year since 2005, recorded here and here and currently in the drier northeast.
Some might think that commentators should temper predictions of “severe drought” with at least a hat-tip to the reality, which at the moment is extremely soggy in parts.

Phil,
Whilst Gavin might be correct, his views are moot.
Firstly, with regard to a single grid cell – what people experience is local climate, not global climate. What the Amazon rainforest will experience will be local climate, not global climate. If the results of four grid cells (which in a typical model run is an area of 500km x 500km or so – not trivial!) bears no relation to what is felt on the ground, then models are likely to be useless for assessing what happens on this kind of scale.
OK, the Amazon rainforest is somewhat larger than 250000 sq km – but if you read the poster session carefully, you will see they expanded the study to the contiguous 48 states of the US, an area of (approx) 8 million sq km – some 32 times larger, or around 128 grid cells, and very similar in scale to the Amazon rainforest (at 5.5 million sq km, from wiki). The results got worse, not better – once again confirming that increasing scale and averaging makes GCM output less accurate, not more accurate.
Asking whether the stations have problems is a valid question, but are you really suggesting that by selecting 63 stations at random, all 63 are likely to be severely affected by artefacts that will render their data so far incorrect? If you or Gavin really believe that, then climate science has some even more severe problems than the Koutsoyiannis paper! However, in the 48 contiguous states case, the 70 selected stations are a spot on match to the NOAA average temperature for the same area. Station quality may still be an issue, but if it is, the Koutsoyiannis paper is the last of the problems to worry about in climate science – the whole underpinnings of the present consensus would be flawed.
With regard to failure to account for the fact that tests are performed against single model runs rather than ensembles, this shows Gavin does not understand Professor Koutsoyiannis’ appoach. In his paper, Koutsoyiannis outlines (and references papers in support) of his approach (stochastic and probabilistic modelling) against Gavin’s preference (ensemble modelling) and the paper includes references to explain these choices. Koutsoyiannis’ approach is based on a valid statistical approach using stochastic modelling techniques.
Ensemble modelling is a popular alternative to stochastic modelling though, and it is important to underline some issues around this choice – if it were to be made instead.
There are two substantial issues with Gavin’s claims regarding ensemble approaches. The first – and most obvious – is that Koutsoyiannis does include multiple model runs, and it is evident – particularly for precipitation – that there is a bias in the results (i.e., results from most model runs are in error on one side of the correct value). Ensembles do nothing to remove bias, and would certainly not solve the problem for rainfall estimates.
The second problem is that what happens in the Amazon rainforest WILL be a single realisation of climate. If a single realisation of climate from a model is so far removed from an ensemble to turn Professor Koutsoyiannis’ work upside down, then an ensemble output is so far removed from a single realisation of the actual climate to be worthless.
In summary, Gavin’s criticisms raise interesting points, some of which have since been addressed, and some misunderstandings about the approach. Ironically though, the most valid criticisms Gavin makes actually underline how irrelevant the models are for assessing (say) precipitation over a small part of the globe (say of the order of millions of sq km).

So the follow-on study did not address the basic category error – using grid-cells from a global model to try and reproduce local climate. As Gavin wrote, he could tell that the results would not correlate before he even looked at them. An analogy would be tossing my Road Atlas because it cannot navigate me from my living room to the kitchen. Wrong tool, wrong scale.
The reasons are many but simply: Climate models are sensitive to boundary conditions, and it is not possible to initialise these at grid cell level. Downscaling to continental and smaller scales has been done successfully, notably Chou et al., 2000; Nobre et al., 2001; Druyan et al., 2002 attempted to downscale to Brazil (with limited success) but downscaling to an individual station is just absurd. Global climate models cannot possibly reproduce local microsite features such as elevation, proximity to topographical features, [e.g. a mountain, a lake, an airport!]. This does not make the model useless, any more than it means the station is inaccurate.
There are many examples of regional climate being successfully predicted, see for example Hurrell et at 2006 on CLIVAR, or the IPCC on regional projections, where you can see observed vs modelled hindcasts for continental-scale regions, including South America. Miraculously, these results are not contradicted by a study expecting to find exact correlation between a global climate model output (why not the regional climate models or the empirical-statistical downsizing studies? these might have been slightly more valid.) at the grid-cell scale.

Nepstad can offer a “closing comment” but the issue is far from closed. Nor is it appropriate or accurate to dismiss the controversy over the IPCC statement as a “misinterpretation”. That is the height of arrogance.
There is no more evidence to support his novel (and equally unsupported) claim that the IPCC is “talking about future reductions in precipitation below the current variable rainfall regime” than there is to support the original assertion that 40% of the forest is at risk from a slight reduction in precipitation.
Interestingly, in 2005, writing in an IPAM pamphlet headed: “Tropical Deforestation and Climate Change”, Nepstad and others (including one of his co-authors from the 1999 pamphlet, Ane Alencar) declare:

Although the occurrence of logging or fire are perhaps the major determinants of human-induced forest biomass reduction, other variables influence the magnitude of these effects. The influence of logging on forest biomass and forest flammability, for example, depends on the intensity of the logging – the wood volume harvested per area and the type of damage reduction measures that were employed (Holdsworth and Uhl, 1997; Gerwing, 2002). Rainfall history and natural characteristics of the forest site as soil and vegetation type also influence the occurrence of fire on forests in the Amazon (Cochrane and Schulze, 1999; Cochrane et al., 1999; Barbosa and Fearnside, 1999; Haugaasen et al., 2003).
All these studies demonstrate that fire provokes significant reductions in the total biomass (alive and dead) of Amazon forests – from 15% to 40% of mature forest – and that this reduction is directly related to the intensity of logging, the intensity of drought, and the occurrence of previous fire between an unburned forest (undisturbed) and a logged and burned or just burned forest.

The suggestion by Nepstad in a 2005 IPAM document, stating that up to 40% of mature forest is at risk, is uncanny. Although, this time, fire is the proximate cause.
Which is it to be? Nepstad can’t have it both ways. Either it is a “slight reduction in precipitation” or it is fire. His 40% figure points to the latter.
Then, as to current climate patterns, Nepstad is working within the span of less than one 28-year cycle (yet there may be other longer, overlying cycles), on top of which different cycles prevail in different climatic zones, which do not necessarily relate. Thus, his use of the conditional “may” and “if” reflects the considerable uncertainty, which may also include an increase in precipitation in some or all of the zones – as suggested by other workers.
That is where I am going with my comment on flooding. We are discussing periods of intense rainfall that have been occurring in the recent past and currently and could continue into the future. If those future flooding episodes are more severe than in the past, then lots of damage from flooding could ensure, although you might have a little difficulty keeping your matches dry.

A summary of the state of play.

Dr Nepstad:
Many thanks for your comments
You say in (1) “This exchange has helped me to understand this fundamental misinterpretation.” with respect to precipitation regimes.
I am afraid it is not possible to say “slight reduction in precipitation” and mean “an even more severe drought”. Any defence of the “When the IPCC says X, they actually mean Y” kind, cannot be sustained for long.
In any case, if that was indeed the case, (that severe droughts alone could radically flip the Amazon), there is enough available peer-reviewed literature that supports the opposite conclusion. Indeed Nepstad 2007 is cited as a reference in this regard.
With your comments, a clearer view of what the experts such as yourself think about the Amazon system has no doubt emerged. But it is inescapable that this clear meaning is not what the IPCC statement conveys.
Your point (3) says: There seems to be a great deal of concern about the word “slight” and “severe” in the IPCC statement. These are not quantitative words.
This comment is misleading, at the least. All quantities do not have to be of a continious nature, to be measured by numericals, they can be ordinal too. In fact, it is surprising that such a question should be raised at all, because most climate change impacts are measured or quantified in such categories.
The IPCC itself places great stress on its authors to be careful when choosing adjectives and terms to quantify impact and offers guidelines to this effect, and any end-user looking at the final report would assume that IPCC statements satisfy its criteria.

Phil Clarke, your comment below:

So the follow-on study did not address the basic category error – using grid-cells from a global model to try and reproduce local climate.

… is just astonishing. There are so many mistakes in that short sentence it is difficult to know how to answer them all without writing an essay. Firstly, you should try looking up the definition of a category error. Local and global climate are not different categories, they are the same categories on different scales. If you want to know what a category error really is – as opposed to your incorrect usage – wikipedia has an article describing it here. In short, a category error would only apply if GCM output bore no relation to climate whatsoever – a position which some sceptics may adhere to, but I would assume you do not.
You then claim that the follow up study did not address the issue of spatial scale, which it clearly does when it creates a spatially averaged area of over 100 grid cells and assesses the performance of this region. Notably, this performs worse than the individual stations. This is an important point – spatial averaging makes the GCM output worse, not better. The example used by Koutsoyiannis is around 8 million sq km, larger than the rainforest example.
Koutsoyiannis et al corrects for issues such as elevation in the analysis, contrary to your statement above.
So your claims are – unfortunately – glaring false with even a brief perusal of the Koutsoyiannis work. And having spent several posts attempting to pick holes in the paper, you then shift tactic and claim the result is blindingly obvious (which is a bit bizarre given that just a couple of sentences earlier you claim it to be a category error). You seem to argue that the AR4 GCM runs are inappropriate for this kind of grid cell scale analysis. You even try to justify this with a strange statement about initialisation of boundary conditions (tip: boundary conditions don’t need initialisation – that is why they are boundary conditions, and not initial conditions).
The problem with this – the claim that the IPCC AR4 runs are completely inappropriate for this kind of grid cell analysis – is that this is exactly what Mahli et al do in the 2007 Science article that Dr Nepstad refers to. They look at grid cell results from IPCC AR4 model runs – the same models, the same set up as Koutsoyiannis et al looked at, just run forward into prediction rather than hindcast. They look at data from grid cell levels up to 5.5 million sq km – right in the type of scales Koutsoyiannis et al looked at.
So your statement – “Wrong tool, wrong scale” – completely undermines Dr Nepstad’s claims regarding Mahli et al 2007. As they have used the IPCC AR4 GCM runs at the same scales as the Koutsoyiannis work to make predictions that you note do not correlate with reality.
At the end of your post, you ask why we don’t look at regional climate models which would be more appropriate for this type of activity. Clearly, your question should be directed at Dr Nepstad and Dr Mahli. They are the ones who have used IPCC AR4 GCM runs to analyse local climatic changes; not the rest of us. The rest of us just observed (much like you did) that this is the not the correct thing to do.