Wednesday, December 14, 2011

Extended peer review: Assumptions in Arnell's article

Addition, 25.12.2011

Now, the three known participants of the effort spelling out the assumptions in Arnell's study have published their results as comments further down. For some reason, a number of readers could not resist the temptation to please the rest of the crowd with their opinions on other issues related to Arnell - either that any assertions that the paper would not be fine (which have not been made), are baseless, or that one should discuss more if Meichsner article about AfricaGate was right, or that I would have a biased attitude favouring skeptics - , but somehow we have now have ended the "listing" phases of merely listing the assumptions made, without judging their plausibility or legitimacy.

The three commentators, Reiner Grundmann, Günter Heß and myself, have not come to the same list - and I invite now everybody to comment whether our three listings are in order, and what you think about these assumptions. To make this easier, all comments apart of the three listing comments as well a few procedural explanations are deleted now. Please keep in mind that the purpose of this effort is to find out if we can agree on certain issues, independently of what we think about the severity of anthropogenic climate warming. Parallel to this, I am opening a new thread, where all kind of comments on this process, on Arnell, Meichsner, on my biases, on the unfairness of deleting comments, are welcome. But please stay disciplined with the present thread - it is only on talking about the assumptions employed by Arnell, not on other publications and also not on the conclusions drawn by Arnell.

 ---------------- original posting -------------------------------------

During a previous discussion on "Africagate" many comments were dealing with Arnell's analysis of possible future drought conditions or water stress in Africa. The participants of the discussion could not really agree how robust the analysis would be, and whether it was wise by the IPCC to base its perspectives mainly or even exclusively on this paper. (Correct presentation so far?)

I understand that we talk about this paper

N. W. Arnell, 2004: Climate change and global water resources: SRES emissions and socio-economic scenarios, Global Environmental Change 14, 31-52

(The link was provided by Vinny Burgoo on 5 December 2011).

This is the right paper, correct?

A few days earlier I had suggested that participants of the discussion here on the Klimazwiebel would screen this article with respect to the key assumptions made by the author. So far two plus myself have volunteered (Günter Heß and Reiner Grundmann); if others want to join, you are welcome. None of us is a real expert in the field, so the exercise is among educated lay people - and the suggested process reminds on postnormal's "extended peer review".

What I suggest is
  1. Readers comment until Saturday, 17 December 2011, on the two questions I have raised above, namely on the significance of the Arnell paper for the IPCC assessment, and whether the link above leads to the right article.
  2. The three volunteers, and hopefully more, will come up with a list of assumptions and publish them as comments on this thread - in this phase without assessing the legitimacy or plausibility of the assumptions. We should limit ourselves to key assumptions, which are significant for the outcome of the analysis. This "listing"-phase will be about 1 weeks beginning Sunday, 18 December 2011.
  3. After one weeks, I will announce that the "listing" phases is over, and invite comments on the completeness of the list, asa well as on the legitimacy and plausibility of the key assumptions.
Please keep in mind that we are not discussing the conclusions of the article, but the implicit basis for the analysis. We will see if we are able to agree what the key assumptions are. I expect that we will find disagreement among the participants concerning plausibility and legitimacy of these assumptions.

If you want to comment on this process, you can do that until coming Saturday (17 Dec.) on this thread or by mailing me (

Do not use the present thread for issues unrelated to the question of assumptions made in the Arnell paper. I will delete comments which are off-topic. 


Hans von Storch said...

Got a confirmation by mail that it is really the right article.

Hans von Storch said...

No opposition, so far - to my short summary of state of affairs, and to the key role of the paper with an URL given above. Thus, we are entering now the phase that readers should try to analyze the ASSUMPTIONS made in the Arnell paper.

The fact that hardly any reactions, could also mean that nobody is interested in this effort :-(

Günter Heß said...

Dear Prof. Storch,

I am still interested, but my spare time is also limited. However, I think as a first step we should agree on the problem description Arnell chooses for himself.

Arnell writes in his abstract:
“This paper describes an assessment of the relative effect of climate change and population growth on future global and regional water resources stresses, using SRES socio-economic scenarios and climate projections made using six climate models driven by SRES emissions scenarios.”

Therefore, I think the purpose of the paper is to describe a method to assess this relative effect using the tools Arnell mentions in his abstract. It might seem trivial, but in the heated debate about climate change, we should limit our discussion here to the purpose of the paper.

Arnell writes:
„The paper compares the relative effect of emissions scenario and population growth rate on the effects of climate change.“

„The study adopted the conventional approach to climate change impact assessment, following a change in climate through to change in runoff, and then calculating the implications for the number of people at risk of increased water resources pressures.“
“The primary innovation of the study lies in the use of a consistent set of emissions and socio-economic scenarios.”

I think we have at least two assumptions here that we need to discuss:

The first one being that the climate projections simulate the change in runoff reasonably well.

The second one being that the emissions and socio-economic scenarios are consistent.

„The key assumption is that population changes everywhere within a country (and after 2050 a region) at the same rate. A more sophisticated approach would allow for differential growth rates between urban and rural areas, but this would not give substantially different results when populations are summed back up to the watershed scale.“

Arnell assumes that aggregating the effects on the country level is sufficient for the purpose of the a paper.

I started this now, fairly neutral and hope that we can start the discussion here. Please comment also, if this is what you expected for Rainer and me and others. Feedback is welcome, positive and negative feedback if you allow me the pun.

Best regards

@ReinerGrundmann said...

I can see three basic assumptions made in the paper:

1 - the problem is water scarcity. I assume this assumption was made on the basis of model simulations. This assumption could introduce a (confirmation) bias into the analysis.

2 - no adaptation will occur. Otherwise the projections do not make sense.

3 - the urban/rural divide is neglected. It is assumed that this will not make a difference on aggregate level.

However, 2 combined with 3 suggests that the level and quality of infrastructures would make a big difference. The author addresses this problem only in the conclusion.

Hans von Storch said...

Comments on assumptions made by Arnell in his 2004-paper (see also Section 8.1 Caveats, on p. 49): For me, the most significant assumptions are:

1. Possibly significant changes of global water resources are only due to population change and climate change. Adaptation, including efficiency of water use, water management, re-use of water or changing irrigation technology, are secondary and therefore not considered.

2. The spatial distribution of population change within a country (and after 2050 a region) is irrelevant for assessing the number of people affected by changing water resources.

3. Monthly statistics generated by global models are sufficient to describe relevant rainfall (and evaporation) statistics on a 0.5 x 0.5 degree grid so that the water balance in each grid cell is given with a daily resolution (p. 34)

4. The chosen indicators of water resources stress describe the primary effects and impacts, but not "issues such as access to safe drinking water" (p. 38).

and the less significant assumptions are

5. The range of plausible possible change in terms of population and climate are described by the SRES (IPCC) scenarios. While population change is provided explicitly by SRES, climate change is derived through a series of global model simulations.

6. The effects of population change and climate change on water resources is conceptualized as additive and independent in the paper (even if population growth has a first-order effect on climate).

7. The so-called Delta-approach of specifying the future as the sum of present statistics plus the change of statistics in a pair of control/scenario simulation is adequate for describing possible future run-off.

8. Run-off, and its change, associated with upstream glaciers are only of secondary relevance and are therefore disregarded.

9. "The current study assumes that areas with less than 1000 m**3/capita/year are water stressed". (p.37)

10. "The current study assumes a ‘‘significant’’ change in runoff, and hence water stress, occurs when the percentage change in mean annual runoff is more than the standard deviation of 30-year mean annual runoff due to natural multi-decadal climatic variability" (p. 38) derived from a long term control run. Thus, in about 1/3 of time unaffected by climate change induced hydrology, changes would be "significant".

11. The effect of the hazard of drought can be summarized by the 10-year return period minimum annual run-off (p. 45).

12. Instead of observed annual run-offs, simulated estimates of current run-off are used (p. 49)

The author is open and explicit with his assumptions. His evaluation of the significance of his assumptions (key assumptions are summarized in 8.1 Caveats on p. 49) differs from my ranking.

Hans von Storch said...

All comments, which were unrelated to the issue of this thread, namely to orderly discuss the assumptions employed in the Arnell (2004) study, have been deleted.

For comments on the effort of this thread and related issues (including the deletion of quite a few comments), a new thread has been opened, namely:

Günter Heß said...

I started my discussion of the assumptions with the assumption 1 of Prof von Storch.

1. “Possibly significant changes of global water resources are only due to population change and climate change. Adaptation, including efficiency of water use, water management, re-use of water or changing irrigation technology, are secondary and therefore not considered.”

I did not find this assumption explicitly spelled out like this in the text.
But it seems to be implicitly supported by the following two statements of Arnell in the text:

„The study adopted the conventional approach to climate change impact assessment, following a change in climate through to change in runoff, and then calculating the implications for the number of people at risk of increased water resources pressures.“

„Finally, and perhaps most importantly, the actual impact of the effects of climate change on water scarcity will depend on how water resources are managed in the future. This will vary between the SRES storylines, depending not only on economic prosperity but also attitudes towards environmental management and protection.“

I think these statements of Arnell qualify and mark the study clearly as a projection within influence parameter space for water stress on the influence parameter plane with the two axes “population” and “runoff”.

Furthermore, Arnell himself brings his study in perspective with the following statement:

“The net effect of all these caveats is that the numerical estimates of the implications of climate change on future water resources stresses are not to be taken too literally. Rather, the numbers in the tables can be used to indicate the relative effects of different emissions, climate and population scenarios.”

Best regards
Günter Heß

MH said...

A bit late, but I had to catch up.

Part I

1) Some methodological considerations

An assumption is something that is taken as a fact or as true without proof. There are different kinds of facts and it may be usefull to distinguish between different kinds of assumptions:
a) statements (The sun keeps shining, ...)
b) form of a function (linear, quadratic, ...)
c) arguments of a function
d) choice of statistical model (linear/non-linear, normal distribution, ...)
e) ...

Assumptions are used in arguments (computations) to draw conclusions (calculate results). An assumption is important if it is likely to be false and if its falseness has a considerable impact on the conclusion. If the assumption is guaranteed to be true then it impact does not matter and the assumption is irrelevant (Arnell assumes that the sun will keep shining). If the assumption is evidently false but has no impact on the conclusion then it is also irrelevant.

The likelihood of an assumption to be false can be estimated independently of the argument. To judge the impact of an assumption the whole argument must be considered, not just some part of it. For example, if f(x,y) varies little with x and g(s,t) is very sensitive to s then in g(f(x,y),t) x may have a large impact.

Disagreement on the interpretation of the argument leads to disagreement on the assumptions. A clear statement of the argument should precede a listing of the assumptions. A clear statement of the argument will also facilitate finding the assumptions. For example, if we compute a certain quantity as f(x,y) then for each z we have the assumption that f varies little with z. The choice of the particular form of f is also an assumption.

Assumption are used to draw conclusions. These conclusions together with new assumptions lead to new conclusions. So there is a hierarchy (or network?) of assumptions.

The falseness of an assumption may imply the irrelevance of other assumptions. For example, if we conclude that an indicator is not adequate for measuring something then all assumptions used for computing the indicator become irrelevant. If the falseness of A1 implies the irrelevance of A2, does this mean that A2 is a less important assumption than A1?

When an indicator is used to measure something there is an assumption that the indicator measures the something adequately. This is the assumption of 'adequacy of definition'. When the value of the indicator is computed from data there an the assumption the computation gives the correct value for the indicator. This is the assumption of 'adequacy of computation'. For example, we can measure welfare by measuring income. The assumption then is that welfare depends only on income. If we accept that then we must still compute the income. This may be non-trivial if we have to take into account things like tax evasion and may imply the use of statistical techniques.

2) Arnells argument

Indicators in Arnell

'people-stress' = measured via watershed stress
'watershed-stress' = a) average annual runoff, b) 10-year return period minimal annual runoff = drought runoff (p.45)

These indicators are estimated for a given SRES scenario.

MH said...

Part II

3) Comments on the listed assumptions

Hess@A-1 = "The first one being that the climate projections simulate the change in runoff reasonably well."
Adequacy of compution of 'runoff'

Hess@A-2 = "The second one being that the emissions and socio-economic scenarios are consistent."
I don't understand this. What does 'consistent' mean. What happens when this assumption is false?

Hess@A-3 = "The key assumption is that population changes everywhere within a country (and after 2050 a region) at the same rate. ...urban...rural..."
= Storch@A-2 = Grundmann@A-3 = urban/rural divide

Grundmann@A-1 = problem is water scarcity
I don't see how this is an assumption. Where in Arnells argument is it used? What happens when this assumptions is false?

Grundmann@A-2 = Storch@A-1 = no adaptation

Grundmann@A-3 = Storch@A-2 = Hess@A-3 = urban/rural divide

Storch@A-1 = Grundmann@A-2 = no adaptation
Adequacy of computation of 'people-stress'. Assumption about the arguments of f.
The assumption is formulated as D-people-stress = f(D-population,D-climate) where D represents a change of some quantity. It may be better to formulate this assumption as people-stress = f(population,climate).

Storch@A-2 = Grundmann@A-3 = Hess@A-3 = urban/rural divide
Adequacy of definition/computation of 'people-stress'
Is this the same assumption as Arnetll p. 37: 'assuming implicitely that resources in a watershed are equally available throughout that watershed'?

Storch@A-3 = monthly statistics (p.34)
Adequacy of computation of 'runoff'.
Is this about monthly versus daily? Or just about the adequacy of the hydrological model?

Storch@A-4 = indicators of water resources stress describe primary effects (p.38)
Adequacy of definition of 'watershed-stress'

Storch@A-5 = adequacy SRES scenarios
This assumption is somewhat different from the others. The other assumptions are about the adequacy of definition/computation of an indicator. The indicator is computed for a given SRES scenario. The result is a range of values for the indicator. The assumption is that the future corresponds to one of these scenarios. There is a tradeoff: the more scenarios there are => the more likely it is that the assumption is true => the larger the range for the indicator => the less usefull the indicator becomes. And vice-versa.

Storch@A-6 = additivity of runoff
Adequacy of computation of 'runoff'. Assumption about form of the function f.
Where does this assumption appear in the paper?

Storch@A-7 = Delta-approach
Adequacy of computation of 'runoff'?
I'm not familiar with this approach.

Storch@A-8 = runoff and upstream glaciers (p.35)
Adequacy of computation of 'runoff'. Assumption about the arguments of the function f.
Arnell also mentions transmission loss along the river channel and evaporation of water (p.34).

Storch@A-9 = water stress threshold of 1000 (p.37)
Adequacy of definition of 'watershed-stress'

Storch@A-10 = 'significant change' in runoff (p.38)
Adequacy of definition of 'significant change' of runoff

Storch@A-11 = hazard of drought effect (p.45)
Adequacy of definition of 'drought runoff'

Storch@A-12 = simulated/observed annual runoffs (p.49)
Adequacy of computation of 'runoff'

4) Conclusion

I think that a further formalization of Arnells argument will reveal many more (and possibly less important) assumptions. But such a formalization is far from trivial.