Discussions:

Hi Folks,

Regarding the ensemble:

Thanks for clarifying this, Yongkang - sorry we had not noticed

the crucial sentence that only one run is required!

Could you also clarify for everyone (by email and on the website)

which ensemble member of the UKMO-GCM we should begin with - I

guess the first experiment listed would be the obvious choice,

which is 'afoif'?

Regarding soil moisture (SM) initialisation:

I've also now chatted to Richard Jones about this, who knows far

more about regional climate modelling than I. He too is concerned

about model spin-up from an initial SM field that is not in balance

with the model.

We (Richard, Will and I) have decided that the way we will

proceed at the Hadley Centre is as follows. We will first do a

continuous 8-year integration forced by NCEP reanalysis at the

lateral boundaries for 1998-2005. From this we will take the SM

fields for 1 April for the 4 WAMME years. In fact these are

themselves essentially a reanalysis of the real SM that occurred

in those years, but having been produced with our RCM (PRECIS)

instead of the NCEP GCM. Thus they are equally valid as a

reanalysis of the real SM, but with the benefit of being

consistent with our RCM. We would then use these to initialise

the 7-month runs for the 4 WAMME years, along with NCEP

reanalysis data to initialise the atmosphere. (Note that the

1998 start date of the first run [PRECIS 'SM Reanalysis']

allows a 2-year period for the SM to spin-up from whatever Will

starts it with [I guess compatible GCM data].)

For the other RCM groups, our view remains that your RCMs should

also be initialised with a SM field produced by your RCM, as

detailed in the experimental protocol on the website, and for

the reasons I described in the last email. Ideally, it would

be great if everyone could follow the above approach. However,

we appreciate everyone may not have the resources for an 8-year

'SM reanalysis' with their own RCM, so it may be that some/

all of you prefer to use some other readily available SM field

from your model. This may necessitate usinge the same SM

initialisation in all 4 years. If so, we are willing at the

Hadley Centre to run an additional sensitivity experiment to

test the response to using the same versus different SM

initialisation in a given pair of years. (Note this is not the

same as testing the sensitivity to using NCEP SM versus own-RCM

SM - this sensitivity will likely vary between models and

according the the way in which the NCEP SM is imported into

the RCM's land surface scheme.)

I look forward to hearing what you think of all this, Kerry,

Len and Ben.

Finally, just to say that I am away from now until 16 April,

so discussion from me will go quiet! Also, Will is away when I

get back, so in order to meet the 15 June deadline, Will will

be starting our integrations tomorrow as described above -

since it's based on the original protocol and our best

scientific judgement, we're hoping this will be more acceptable

than missing the deadline!

Regards, Dave

To: dave.rowell@metoffice.gov.uk

From: Yongkang Xue <yxue@geog.ucla.edu>

Subject: Re: WAMME Initial Soil Moisture

Cc: Edward Kalman Vizy...snip... lau@climate.gsfc.nasa.gov"

Dear Dave,

Thanks for your questions and suggestions. First I have to say that there is vagueness in the model experiment design.  The experimental design says "RCM will conduct one run for each LBC.  This could be adjusted should we find ensemble runs necessary."  For the reanalysis data, there is only one LBC for each year.  For GCM LBC,   there could be multi-LBCs.  But I think we need to discuss whether ensemble runs are necessary.

The issue of RCM downscaling ability is quite different from that of GCM's predictability. Now there is an ongoing serious argument on whether RCM has any ability to do the down scaling.   I believe our work will contribute to this basic understanding.

To provide an idea about this issue and possible controversy, I attach our recent paper for N. American climate downscaling.  It will be published in J. Climate in next a couple of months.  This study using Eta RCM indicates that ensemble runs are necessary only when the domain size is large enough to cause large internal variability (see figures 2b, 2c, and 2d).  As to initial soil moisture, it shows marginal impact (see section 4.4).  Although this study is for N. America, I believe other domain will have similar features.

Of course, our model design depends on our group's perception.  Since Len and Ned have done some African runs, if they believe our model domain warrant ensemble approach, we could produce 16 ensemble runs for GCM LBC.  As to initial soil moisture, we will test its impact in next stage experiment.  At this stage, we just need to set a rule, for example, using the same source for initial atmospheric and soil conditions is one choice.   I would like to have other opinions/suggestions.

Yongkang

All:  Sorry to be so late in responding to these issues.The WAM brings ever increasing precip, gradually advancing northward, to West Africa during the spring. Doesn't that scenario tend to minimize  the influence of the November through March climate regime on the April soil moisture?

Yongkang wrote that the US climate was not too sensitive to initial  SM. Although I have no evidence, I would not expect much variability of rainy season precip to be caused by the very small range of SM during the dry season. Also, reanalysis SM, while perhaps not compatible with each RCM, may provide enough of an ic on Apr 1 to allow each RCM to spin-up the SM by May. I would be surprised if the SM could be still "spinning up" into June  when the  heavy rains are already moving northward.  Admittedly, this is only my hunch.

Len

Hi,

I agree that the models are unlikely to be sensitive to *realistic*

variations of the initial SM. But if the NCEP Reanalysis SM is

significantly different from that normally produced by the model

when left to run on it's own for a few years (say), then the impact

would be larger. But how much larger? Yongkang shows that for his

model over the US there's not a problem - this may or may not also

apply to the WAM with other models.

I suppose, although your hunch may well turn out to be right Len

(and your argument concerning the impact of the climatological

annual cycle of rainfall is well taken), I'm questioning whether

we should risk *possible* oddities in our experimental data based

on this. I guess a key question is whether in practice you have an

existing 1 April SM field from your model, and if not, whether you

have the resources to produce one?

Regards, Dave

Hi Dave - For all of our previous work, we began simulations on May 15. Theoretically, we could make a special run from, say, the previous Oct. to spin up the SM to April 1. This is not the same as your proposed 8  yr downscaling of reanalysis, and I doubt  whether the adjustments to SM that would occur during one dry season would make a difference.

As you have surmised, it would be a strain and anyway we have already begun with reanalysis SM on April 1. This has been our practice also for the May 15th starts. We once looked at the trend of SM in the rain belt at the beginning of a simulation, and found that it reached what looked like a quasi steady state after only several days.  (But, it was not a rigorous study.) In any case,  even according to your suggestion all participant models will be using a different SM initial condition, so different initial SM will be one of the factors explaining different results in the intercomparison. In our case, we will have to accept the risk that results could be "contaminated" by SM spin-up. This doesn't concern me too much because we were getting time-space correlations with TRMM near 0.9 after only 15 days of downscaling reanalysis with reanalysis initial SM. Yongkang previously wrote that his design also included using reanalysis initial SM to go along with the reanalysis initial atmospheric state on April 1, so I guess we are in good company. In any case, it has been an enlightening discussion. Thanks.

Len

Dear all,

Thanks for your contributions to the discussion of initial soil moisture issue.  It is very enlightening.  I would like to clarify my opinions on this issue.

(1) Initial soil moisture has impact on the model simulation even in the RCM simulation. In our North American experiment, we tested the initial soil moistures from reanalysis 1 and North American regional reanalysis.  One initial soil moisture condition produced better correlation between simulated and observed daily precipitations in 3-month simulations than another, by about 5%.  However, these two runs produced similar monthly means of large scale circulation and precipitation.  As to West Africa, we have not done similar tests.

(2). I think David and Will's approach is certainly an ideal one.  However, it requires a substantial amount of resources.  The good news is that AMMA ALMIP should be able to produce a set of "best" soil moisture data.   Since Aaron and other ALMIP organizer participate in our project, I wish this data set will be available for us soon.  I will propose to have some tests in our second stage experiment for initial soil moisture.

(3). Comparing to Reanalysis I, the Reanalysis II soil moisture seems relatively dry in most part of northern hemisphere, but it was wetter over Sahel.   The dry initial condition in Reanalysis II causes problem in our N. American simulation.  Len's experiment indicates it is not a problem in West Africa.  For short term simulation, I normally like initial soil moisture a little bit wet.  It is easy for model to spin down rather than spin up.  Since WAMME does not require the same initials soil moisture, each group certainly could test and select best one for their use.

Yongkang