Dear Kenneth crop model require daily weather data, which can be obtained from the Regional climate models. I have been advised that there is no problem to use climate simulation from CORDEX regional climate models forced by GCMs for analysis the future inter-annual variability of maize yields, thank you very much

yes, it is always good to use the ensemble mean to get a more robust estimate of the inter-annual variability.

But I think this is true regardless you study the inter-annual climate variability or the long term mean/variability. It is just the belief that the ensemble mean is more robust, single GCM can be quite far away from the mean in the worst case. 

Dear all thank you very much for good answers. But if I use the ensemble average as input into crop model, It will smooth the yields and will not allow the analysis of uncertainity from the projections. The best way I have been advised is to use individual model as input into crop model and simulate the yields after that I will find the ensemble average of the yields.

Kenneth the task is to find inter-annual variability of maize yields from 2010-2099 using RCP 4.5 and 8.5.

Dear Kenneth thank you very much for this paper. I can see they simulated for historical climate variability. I will simulate for historical and future yields. I plan to simulate for period 1989-2008 using climate data from RCMs forced by ERA -Interim reanalysis data and then I will simulate for present (2010-2039) mid (2040-2069) and end (2070-2099) centuries. I need to know the relationship in inter-annual variation of climate and maize yields.

Dear Kenneth, I entirely depend on the RCMs derived by GCMs. Any fault is not me to blame we will blame the climate modellor

1. Have a look at chaos theory as it relates to weather and climate. Peter Carl on Researchgate has published on chaos and monsoons.

A General Circulation Model en Route to Intraseasonal Monsoon Chaos

Peter Carl, Full-text available · Chapter · Jan 2013

2. Re chaos theory as applied to climate, see Edward Norton Lorenz, http://tinyurl.com/h4r8fd3

3. The 3rd IPCC report that made the bold statement that prediction of climate was not possible because the Earth's climate system is non-linear and chaotic.

"Improve methods to quantify uncertainties of climate projections and scenarios, including development and exploration of long-term ensemble simulations using complex models. The climate system is a coupled non-linear chaotic system, and therefore the long-term prediction of future climate states is not possible. Rather the focus must be upon the prediction of the probability distribution of the system's future possible states by the generation of ensembles of model solutions. Addressing adequately the statistical nature of climate is computationally intensive and requires the application of new methods of model diagnosis, but such statistical information is essential. ."

Executive Summary of Chapter 14 Working Group 1, 2001, Co-ordinating Lead Author, B. Moore III, Lead Authors, W.L. Gates, L.J. Mata, A. Underdal

Subsequent IPPC reports have accepted this advice and focused on ensembles of model solutions. Most of the variance in model states is in the treatment of clouds and water vapor feedbacks, the estimates of which have not converged for a couple of decades. 

Alternative approaches:

4. The paper by Belda et Al (2014) shows the climate regions of the world (except Antarctica) for two periods, 1901-1931 and 1975-2005. Between the two periods separated by 75 years.

I analyzed their results using an Excel spreadsheet. Belda found about 8% of climate zones had change category, mainly steppe and savanna had become wetter. Also frozen land had become tundra. The CRU (UK) has revised the climate data Belda used to remove wet bias.

 For reasons Belda explains the Koppen-Trewartha system is preferred over Koppen-Geiger used by research teams.

Climate classification revisited from Köppen to Trewartha, Belda, M. et al, Climate Research, 2014

http://www.int-res.com/articles/cr_oa/c059p001.pdf

http://tinyurl.com/hqh2bwe

5. Lecture by Richard Lindzen discussing Budyko and Izrael at 22:14 URL: http://tinyurl.com/hd6o3rt

Budyko, MI, Izrael, Y. Anthropogenic Climate Change. Tuscon: The University of Arizona Press; 1987, pp.277-318.

Lindzen also discusses a paper by Stanley Grotch, Lawrence Livermore Lab, Monthly Weather Review, Volume 115 No. 7, July 1987, American Met Society

Details here: http://tinyurl.com/h3kjeew

Dear Frederick,

I would really wonder if this statement is in the 3rd IPCC report, but I will check it.

Dear Patrick Laux 

I have edited my original post to include references. Unfortunately I was unwell when I posted it. I'm well now but still ancient.

Dear Aleš Kralj ·

Interesting. Nicola Scafetta has done a lot of work on this,

Nir Shaviv is worth checking out for astrophysical aspects.

Thank you very much for the fruitful discussion in prediction of climate. My question Can I use climate simulations from the Regional Climate Models for studying the future climate variability (inter annual climate variability). Is there any published work on this??.. I was thinking myself that this study will be conducted in Tropics, where the interannual variation of rainfall depends on ENSO events which I doubt if is properly represented in any regional climate models. May be we can say that the interannual variation of rainfall in tropics (east africa) are not always shaped by ENSO events. But also the ITCZ which play role on distribution of rainfall in Africa I doubt if is represented well in regional climate models. Anyway my question need to be resolved before I waste lot of my energy to produce projection which can not be justified scientifically.

Dear Philbert Modest Luhunga ·

Have a look at the Budyco-Izrael graph that Richard Lindzen discusses in his video lecture. See my comment above.

I have seen your comments and heard some talk by Richard Lindzen that prediction of future states of climate is not possible. If that is the case why do we have large centers running climate predictions???.

Philbert Modest Luhunga

If youy reread my comment you will see that the IPCC is following the advice that climate is non-linear and chaotic. These models are scenarios. The IPCC claims that the ensemble of the models define the probability distribution of future states. 

The Fifth Assessment Report was unique in not attempting to define the most probable future state. And if you look carefully at the whole series of assessment reports you will see that the most probable states were selected based on subjective judgment.

Problem is nobody seems to read the science reports. Nobody seems to analyze the reports to determine just what the statements mean. Nobody seems to ask what are the error bars how the errors are propagate when the models are iterated forward.

Pat Frank has a lot to say about this subject.

Since I read the term :"climate modeler", I have to remark:

• that kind of modelling is so successful just like economic collapse predictions: only after the crash 'a posteriori prophets' claim that they had predicted it...

There is not such a thing like weather prediction, sorry.

As for the initial task,my only advice is to study correlations only.

Regards.

Kenneth M Towe sais "There is a very strong correlation between people (population) and CO2"

Dear Kenneth

Time series correlations don't count for much. 

A well-known example described by Yule in 1926 is the correlation between the fall in the proportion of all marriages in the Church of England and the fall in the mortality rate between 1866 and 1911.

Spurious correlation,is so common in econometrics. In 2003, Clive Granger and Robert Engle were awarded the Nobel Memorial Prize in Economic Sciences for their work on cointegration, a technique that identifies spurious correlation.

Engle and Granger (1987) is the seminal paper on cointegration and perhaps the most cited paper in the history of econometrics, treating specification, representation, estimation and testing.

http://en.wikipedia.org/wiki/Granger_causality

Based on the econometric technique called polynomial cointegration analysis an Israeli group concluded, "We have shown that anthropogenic forcings do not polynomially cointegrate with global temperature and solar irradiance. Therefore, data for 1880–2007 do not support the anthropogenic interpretation of global warming during this period."

Beenstock, Reingewertz, and Paldor, Polynomial cointegration tests of anthropogenic impact on global warming, Earth Syst. Dynam. Discuss., 3, 561–596, 2012

http://www.earth-syst-dynam.net/3/173/2012/esd-3-173-2012-discussion.html

Kenneth

Norman Newell was an eminent paleontologist. However, we do not know much about his proficiency in statistical analysis. He does not describe his methodology in the 1987 paper.

I am not saying that Newell's claim to have established a causal relationship is false. Only that his statistical methodology was probably not robust.when applied to time series. Neither of the variables that Newell used were stationary and non-stationarity often leads to spurious correlation. The paper shows he might have applied first differences to the population data to make one variable stationary. But he does not say so.

Newell's paper (1987) was written before Granger and Engle did their work on cointegration. But already a lot of work had been done on autocorrelation of time-series data. No discussion in the paper of auto-correlation. No discussion of the methodology of the preparation of the CO2 data. A lot of work has been done showing how to remove the annual cycle while keeping the inter-annual changes.

I reviewed methods for pre-processing the data here and provided a few references: http://tinyurl.com/js33bg7

Beenstock, Reingewertz, and Paldor,claim that it is not the increase of CO2 in the atmosphere that is correlated with climate factors, but the acceleration of the increase of CO2 in the atmosphere. Further they argue that the acceleration of CO2 is linked to the acceleration of economic growth that is occurring in some regions and the rate of economic growth will will level off sometime during this century. As the rate of growth in GDP declines the rate of increase in CO2 in the atmosphere will level off and cause climate to stabilize.

The Beenstock paper seems to be a reasonable effort, best of the bunch of papers on the same subject using cointegration, but I don't think it's the last word.

The main problems seem to be the coupled GCMs and the short time period over which good quality data has been collected on a global basis.

1. model treatment of clouds and

2. water vapor feedbacks.

3. model treatment of albedo including the data (short time series of good measurements)

4. potential for confounding effect of cloud albedo with other cloud effects.

5. reliability of data used to model the world ocean as a calorimeter. The entire atmosphere has the heat capacity of the world ocean to a depth of 10 meters. (However most studies model the ocean down to the thermocline.) 

6. short time period of reliable ocean and bulk atmosphere data (satellites and buoys)

UAH satellite model is auto-calibrated. RSS relies in part on sea surface temperature for calibration. The UAH and RSS are consistent and both are consistent with balloon data in providing information about the bulk atmosphere.

7. contamination of surface data with urban heat island effect. (Great reduction in rural stations globally during the last 50 years or so.)

8. elevation of the Stephenson screens between 1.24 and 2 meters from the ground (d

Does not appear to me to be a suitable configuration to yield information about atmospheric physics, whatever the benefits to agriculture and to weather forecasting might be. More likely merely tells us how much heat is near the ground as a result of direct radiative effects and the local environment (Heat from pavements, asphalt roads, air-conditioner exhaust heat and other heat from buildings and vehicles.)

I confess I am agnostic when it comes to anthropocentric global warming. I have studied climate for over 50 years and still believe Hubert Lamb was right. (Newell too.) What we need is watchful waiting and continued study not connected to policy prescriptions.

It's too early yet for dramatic intervention.

Dear all thanks for the advice that I should use the ensemble average to drive my crop model to simulate maize yields. I used ensemble of 8-RCM-GCM combination to drive the crop model. The results I am starting to get shows that maximum temperature has negative effect on maize yields in present century. The minimum temperature has indirect effect on maize yields since it correlate negatively with the length of growing season which has positive correlation with the yields. This means that when the length of growing season increases the yields also increase and vice versa. Surprisingly the yields are weakly correlated with rainfall in the present century with a non significant correlation coefficient of 0.227.  

Attached is the table with the correlation coefficient. My analysis is limited to find the correlation coefficient

Kenneth

I agree with almost all of your comments. Just not the statistical methodology.

Cheers

Many of  the western forests are fire climax. Suppression of fire allows undergrowth to build up (read fuel). Then when they burn, they burn right down to the subsoil and if the forest is on a hillside erosion that means the forest may not regenerate as it might have had the undergrowth been allowed to burn naturally and more frequently.

In my opinion, to understand this data one would have to do a lot of detailed work to locate the forests and to determine to what extent forest management has changed over time to allow bigger fires to increase in number. We don't know whether we have bias in the data merely because more recent fires are bigger than before.

When we have done that could we begin to explore other factors. 

Michael Crichton discussed some of the issues of park management in one of his lectures still available on Youtube focusing on Yellowstone National Park.

I did some work on modern human impact, mainly the effect of road building in fragmenting forests, easily identified in satellite images.

Philbert, the advice you got is quite good: 

"But if I use the ensemble average as input into crop model, It will smooth the yields and will not allow the analysis of uncertainity from the projections. The best way I have been advised is to use individual model as input into crop model and simulate the yields after that I will find the ensemble average of the yields."

Use every single RCM data and run your crop model. Then you may calculate the ensemble mean at the last step for your crop data together with the ensemble spread. You should be aware however, that the ensemble spread for future simulations is not the uncertainty. Every RCM and hence crop simulation for the future is just one physically plausible solution representing one trajectory of the climate system assuming certain forcing conditions (scenarios). The uncertainty for the future cannot be assessed. This is only possible for the present/past where the "real" state has already been observed.

Philbert, don't do that!

"Dear all thanks for the advice that I should use the ensemble average to drive my crop model to simulate maize yields."

If you average over N ensemble members, the inter-annual variability goes towards zero with increasing N. The only thing left is the mean warming trend. Your crop will never experience such a mean trend. Extreme warm/cold or wet/dry periods are more important and you would average them out in an ensemble mean.

E.g. assume you would have more extreme periods of drought and wet spells in the future with negative impacts on crops. If you are lucky, some of the different GCM/RCMs may even simulate them, but the different simulations would not be correlated in time. The ensemble average would smooth out the things you actually want to look at.

Dear Frederik thank you very much. I have done a hard work to run the crop model with the individual RCMs and the ensemble average. What I will do is to analyse the results from the crop model forced by the ensemble average of RCMs and the results from crop model forced by individual RCM. I think the ensemble of the crop yields will not be different from the yields obtained when the crop model is forced by ensemble of the RCMs

It is possible that there is no big difference. It may depend on the complexity of your crop model as well. How different are the most extreme members from your simulations? Maybe you could plot the frequency distributions of each run and compare them with the others whether there are changes at the tails or e.g. a divergence of mean and median etc.

Thank you very much Frederik. I also thought that there might be no difference since the crop model is kept as a constant, only climate data varies. I will need to analyse the results to see the spread of the yields from individual models.

I don't get that: "the crop model is kept as a constant, only climate data varies"

Dear Frederik the spread between the ensemble member is mixed, in some years the spread is too much but in other years the spread is small. I am starting to believe the simulation of these models (RCMs) may reproduce the seasonal climate and the effect can be seen in the simulated yields. For example all the model simulations indicate decline in maize yields in year 2015-2016 this was due to increased maximum temperature and decreased rainfall. This actually happened in Tanzania. We now face food shortage (maize yields) in the study region. However, my simulation shows that if management practices remain constant this year2017-2018 we will have increased maize yields. I think we should believe in these climate models. They can be used to guide though not accurately

I think we are talking about two things here in parallel. One question is how well the RCMs perform for the present and past when they are driven by reanalysis data. Another question is how to setup and run your crop model for the future.

It is nice to hear that the RCMs seem to work fine for 2015-16 once they are driven by reanalysis data. But it does not tell you much about how well the RCMs perform once they are driven by GCM data for the future. If the GCM has a bad skill for the mean climatology over your region, the RCM will have a bad skill as well. This explains why some people say you should use the ensemble mean for the future (it is assumed to be more robust than a single potentially bad GCM/RCM run). But the ensemble mean is an artificial climate state and the physically most unrealistic trajectory of the climate evolution. Any model driven with such an ensemble mean is much more stable than the climate would ever be and you miss most if not all extreme excursions.

The performance of the RCMs driven by ERA-reanalysis is good, the RCMs captured the annual and interannual cycles of rainfall and temperatures. However the all the RCMs underestimate the maximum temperature climatology. The most important is that the RCMs reproduce the inter-annual variation of climate variables. The setup for future I am using RCMs driven by GCMs as you advised. What am doing is to change the weather file in the crop model replace historical climate with the future climate. 

OK, once you have some plots and results, it would be interesting to have a look at it. Sounds like an interesting work.

andu, 1)the best way is to predict the yield by forcing with each ensemble member then you can average your ensemble yield

another way to do is, if you have large ensemble members (~> 25 ) look for the likelihood(the most likely occurrence of a member at a given particular day) and take for instance 80% rather than the ensemble average

@Robel: There is no "most likely occurrence" for the future. Every run is equally likely. You may limit the runs based on your confidence in how well GCM/RCMs performed in the past. But it cannot be ruled out that these models might not be the best for the future (as models are to a large extent tuned to today's climate but not to the future).

Frederik I have produced attached graphs. Would you suggest any types of graphs that can suite my analysis. ?. I tried to plot the distribution of the simulated yields all are under normal distributed

The ensemble mean is the ensemble over the Maize yields? I guess it would be important here to understand why the light blue (RCA4-CNRM) behaves very differently from e.g. the green (RACMO22T-ICHEC). The ensemble mean seems to suggest that there will be no change.

Yes Frederik the ensemble mean is over the maize yields. I guess the deviation is coming from the GCM (CNRM). For RACMO22T forced by ICHEC deviate from other RCMs (HIRHAM and RCA due to the different RCMs formulation I guess.

''The ensemble mean seems to suggest that there will be no change''

Yes, I remember we published a paper about the impacts of climate change over wami ruvu basin in Tanzania and we found that climate change will cause small decrease in maize yields. In the first century (2010-2039) climate change will cause small increase in maize yields. I think the issue to fear is not climate change but its variability. As you can see in some years decrease in yields in some year increase in yields up/down. This need careful response for developing nations where agriculture is dominated by poor resources and unskilled farmers. Most farmers tend to cultivate small farms to help get yields that can  help him in one season.

Do you know any paleoclimate reconstructions for Tanzania for the last 100 to 20.000 years? How stable has the climate been during that period? This kind of information might help you to interpret whether the model runs for the future are plausible or not.

In the linked publication, it looks like internal variability dominates but models may not be doing a good job over the region.

Article Comparison of simulated and reconstructed variations in East...

Dear Frederik I dont have data for paleoclimate. Where can I get this kind of data?. About the climate of Tanzania it has been changed, temperatures have increased in many areas, rainfall has decreased in many areas. The performance of the RCMs in the region is not bad see the attached paper where we validated the models using observed data from meteorological stations