GROUP and Objectives

• Persons: Frederique, Agnes, Jan, Catherine, Fuxing, Tao, Philippe P., Jean Louis, ?
• Main responsibles: Frederique ?
• Objectives / roadmap : Possible roadmap (to be refined and detailed)
  • Integrate a « unique soil discretisation for W and E » (travail de Fuxing) => date? Current version of Fuxing is 8 Meters for both E and W, with same discretization. Questions: AD and JP wonder about the impact on the runoff and thus river discharges and possibly the time required for the spin up (such time is likely to increase with a 8 m soil) ?
  • Modify the coupling of the new multi-layer snow with soil to be fully implicit. => date ? The current version in MICT works (couple and force mode) but is not fully implicit and the energy conservation at each time step should be checked! The work to make it fully implicit needs to be quantified with Tao (Anticipated to be relatively straightforward)
  • Eventually implement a soil freezing for the "thin layers of the new vertical discretization" => ?? A solution to avoid large oscillations needs to be proposed and tested.
  • Improve the albedo of the vegetation and bare soil: Work that can be pushed by a Post-doc from the CMUG project (Patricia is responsible of the project; person to be recruited); => Aout 2015; propositions: Make some tests with the We can try to re-activate the dependance of bare soil albedo to sol humidy with the 11 layers scheme.
• Remarks : The agreement and the details of such roadmap should be done; Most likely we need to separate the different TASKs with a different group of people for each?

Discussions

Jan Polcher, 13/01/15 (mail)

après notre discussion cet après-midi j'ai un peu réfléchi à comment "merger" les 4 points de Philippe dans le trunk, tout en permettant à tout le monde de suivre ce qui se passe et donc de construire une confiance commune dans ces changements.

Je propose un plan en 4 étapes ou nous laissons quelques semaines entre chaque étape afin que tout le monde puisse faire des runs et des tests avec cette nouvelle version. Ceux ceux qui ont piloté les changement pourront écrire une documentation technique pendant ce temps. J'ai essayé aussi de séparer les changements structuraux des améliorations des paramétrisations actuelles. Donc voici mes 4 étapes :
1) implémentation de la discrétisation verticale commune entre T et W. L'important est de garder la profondeur en T et W comme des paramètres indépendants afin qu'on soit libre de choisir plus tard ce qui convient.
2) Enlever la neige de Thermosoil et la mettre en sandwich entre enerbil et thermosoil. Ceci est donc la partie thermodynamique de la neige que Thao vient d’implémenter dans MICT mais avec la l'interface neige/sol finalisée.
Après ces deux changements on aura une version qui a la nouvelle structure et les nouvelles approches numériques. Les paramètre auront peu changés et donc à priori le comportement du modèle sera similaire au trunk. Un environnement idéal pour voir si tout marche bien techniquement.
3) On rajoute les nouvelles capacité et diffusion thermiques de Fuxing avec le transport de la chaleur par l'eau.
4) L'albedo de la neige et les autres composantes du modèle de neige de MF sont rajouté afin d'arriver à la même configuration que dans MICT aujourd'hui.

Si on veut on peut rajouter une 5ème étape qui est l'activation du gel du sol et la vérification des processus dans la première couche.

Je pense qu'après ce processus on sera au stade ou voulait en venir Philippe pour CMIP6. Si à chaque étape on échange sur ce qui a été fait, que tout le monde fait les tests qu'il pense important on aura tous la même compréhension des évolutions faites.

Pour réaliser ce plan je suis disposé à travailler avec Fuxing et Thao sur les étapes 1 et 2. Je ne peux pas quantifier le travail pour chaque étape, mais une semaine me semble un bon first-guess. Je pense que pour la neige ce sera un peu plus long. Mais comme cela vient en second on peut préparer les choses en amont.

Philipe Peylin, 17/01/15 (mail)

Concernant la proposition de Jan:

• De mon coté je vois un interêt au point 1) et à le traiter tout de suite en discussion avec ce qui va devoir être fait pour inclure le permafrost (déja inclus dans MICT (version de Charlie Koven)) car dans ce cas la discretisation pour la Temperature devra aller beaucoup plus profond.
• Ensuite pour le point 2 et 4, personnellement j'aurais traiter cela ensemble (mais je ne suis pas le plus compétent pour juger de l'intéret de séparer)

Agnès Ducharne, 23/01/15 (mail)

My opinion is that the main challenges are :
1) to get the new snow scheme, as it improves the albedo, but in a way that fully respects the surface energy budget and the implicit coupling with LMDZ
2) to define the best vertical discretizations for T and W, in a way that keeps them as close as possible for convenience. The new soil termal properties are an improvement that can be added even if we do not change the vertical discretizations compared to the present Trunk, and heat convection might probably be overlooked in CMIP6.v1 if we cannot achieve identical discretizations, as it has a small effect based on Fuxing's results.
3) define "MODIS-like" reference albedos for the PFTs

I focus today on the second point: define the best vertical discretizations for T and W. Again, we can identify several issues, and from my point of view, the main ones are:
a) the deep soil required for T may be to deep to reasonable runoff values compared to observed river discharge
b) the very thin top layer required for W may be too thin for T and induce numerical instabilities of Ts and the surface energy fluxes
c) this behavior might be exacerbated in case of freezing

Attached is a series of plots by Fuxing illustrating point b ​https://forge.ipsl.jussieu.fr/orchidee/attachment/wiki/Meetings/CMIP6/Physic/fuxing_np_point_numerical.pdf.

Explanations of the plots: If we look at the temperature at each time step, we can find oscillations over the upper 1.4cm soil depth, but the oscillations are not too large. Three simulations were compared (only 5-day simulation, at point: 25E,0N) (see attachment): CTL: the standard vertical discretization is used (2-meter for moisture, 5-meter for temperature), 7.5-min time step (black line) EXP: the revised vertical discretization (8-meter for both moisture and temperature), 7.5-min time step (red line) EXP2: same as EXP, but 15-min time step (green line) Main conclusions include: (1) in the revised soil layer (8-meter), the oscillations appear at upper 4 layers (0.5mm, 2mm, 6mm, 14mm). the oscillations disappear at 5th layer (29mm) (see Figs. c-g). But there is no very large or very small strange temperature in the sub-surface (Figs. a-b). (2) There are also oscillations in the FLUXLAT, FLUXSENS, SOILFLX, RAIN, QAIR (Figs. h-l), but the variations are in normal range. Besides, there are also oscillations by using the standard discretization.

The good thing is that there seems to be solutions to the above issues:
a) if pb a is proven (we are still waiting for an answer about this), we could do as in SURFEX-ISBA and only consider the same discretization over 2-m, below which T is solved under the assumption of a uniform W profile (consistent with teh gravitational drainage assumption). After talks with Jan, Frédérique and Philippe, I feel it's a solution anyone could agree on if cannot make deep W discretization work.
b) we know we have a pb here (Fuxing grahs), but Jan mentioned several times that numerical solutions exist. It's about filtering the variability, but I personnaly have no idea what is recover in practice, and it woud be nice if Jan could give us more information on this. Another solution might be to lump the 4 top layers of the W discretization to define the first layer for T It's seems rather easy, especially if we agree on eth proposition made by Aurelien during his PhD: stick to the present W discretization in the top 25cm, even we allow to change it below.
c) this still needs to be assessed, and we need to do it for CMIP6.v1 if we want to activate soil freezing.

Summary of a talk between Fuxing Wang and Aaron Boone at the AMA2015 in Toulouse

(1) Potential problem with the too deep soil depth: A. BOONE said that not only the annual mean discharge changes by using very deep soil layers, the seasonal cycles of discharge (phase and amplitude) are changed a lot by using very deep soil layers in ISBA model. And the second situation may be more serious that the first problem. It seems the hydrology model in MPI also has this problem mentioned by A. BOONE. Currently, the soil depth is 2-3m (changing for different regions) for hydrology,and it is 12m for thermal (constant) in ISBA model. The soil moistures below 2-3m are computed by Darcy's law.

Decharme, B., A. Boone, C. Delire, and J. Noilhan (2011), Local evaluation of the interaction between soil biosphere atmosphere soil multilayer diffusion scheme using four pedotransfer functions, J. Geophys. Res., 116, D20126, doi:10.1029/2011JD016002.

Decharme, B., E. Martin, and S. Faroux (2013), Reconciling soil thermal and hydrological lower boundary conditions in land surface models, J. Geophys. Res. Atmos., 118, 7819–7834, doi:10.1002/jgrd.50631.

(2) Spinup: For the spin up in the ISBA model, it is about 10 years (or more) for dry conditions, and it is much shorter for humid regions (2 year). And it is confirmed that the spinup length for temperature is shorter than that for moisture. For our revised model, A. BOONE commented that it maybe take several decades to get soil moisture equilibrium (but A. BOONE is also not sure exactly how many years are necessary).

(3) 1st layer depth It is 1cm for the 1st soil layer in ISBA. A. BOONE's opinion is that the 1mm soil depth is too thin, and A. BOONE's suggested it better to use at least 1 cm for the 1st layer.

Jan Polcher, 26/01/15 (mail) to Agnès 23/01/15

thank you for this analysis ... which I totally share. I will give you my feeling on a few of the points you raise :

The challenges :

• I have put the snow second because I find the interaction with Tao more difficult and thus I wanted to give us more time for it.

The depth : This should be configurable by the user as different applications will have different needs. We should code things so that each one can test the configuration which suits him/her best. The limiting conditions are :

• The top layers need to have the fine vertical resolution of the soil moisture.
• The total depth for soil moisture should be equal or smaller than for temperature (The total depth of soil moisture and temperature need to be parameters in the run.def).
• After that we can debate if some upper and lower bounds for the depth should be introduced.

As for the technicalities :

• soil moisture below its total depth will be take as constant and equal to that of the deepest layer.
• The smoothing of the soil temperature profile at depth of less than 5cm needs to be done in thermosoil once the new profile is calculated. That should be enough to avoid numerical instabilities.

For the snow, I am still waiting to see the equations to be used to coupled the bottom temperature of snow pack with the one of the soil. Once we have all the equations of the temperature diffusion (from the skin temperature to the bottom of the soil and through the snow pack) in implicit form, things should flow smoothly.

Fuxing Wang, 26/01/15 (mail)

(1) The depth. This should be configurable by the user as different applications will have different needs. We should code things so that each one can test the configuration which suits him/her best.

In the new discretization, the HYDROL_SOIL_DEPTH (=dpu_max), THERMOSOIL_NBLEV (=ngrnd) can be defined in run.def. I also added a flag (VERDIS_FLAG, can be defined in run.def ) for different discretizations: verdis_flag = 1, the standard discretization; verdis_flag = 2, the revised discretization. But 'nbdl' (Number of diagnostic levels) and 'nslm' (number of CWRR levels) in 'constantes_soil_var' (=11 in standard discretization) was not defined in run.def. Is this kind of configuration OK ? Do the variables 'nbdl' and 'nslm' also need to be defined in run.def ? Because they also change when using the revised discretization.

(2) The technicalities : The smoothing of the soil temperature profile at depth of less than 5cm needs to be done in thermosoil once the new profile is calculated.

In this case, we smooth the top 5 soil layers for soil temperature ? The bottom positions from top 4st to 6th layers are : 2.151cm, 4.497cm, 9.189cm.

Frédérique Cheruy, 27/01/15 (mail)

About the optimal vertical discretization for W and T
1- We still need to verify that the 8m deep soil, is a problem for the runoff and the drainage.
2-Afterwards, as already discussed last week, and reported in Agnes mail, we might want to have different depths for moisture and temperature but we want to have the dependance of the soil thermal properties with the soil texture (3 or 12?), and the same vertical discretization for moisture and temperature to avoid interpolations.
I am not sure we want to externalise nslm or ngrnd, and let the user play with it. If it is an advanced user he can go into the code and change the value of the parameter, if not may be it's better he does not play with it at all.
But we need to test upper and lower bound that we want to adopt for CMIP6 at least.
Once we get the answer to point 1, Is the 8m T, 2m W with same vertical discretization for the upper 2 m and moisture constant and equal to the one of the deepest layer a good point to start with? Can we agree on that?
3- Is someone able to explain to me (sorry ), why a smoothing is a good way to handle with numerical instabilities, to me it looks like an artificial way of masking a potential problem.
4- Do we want to consider the problem of the soil freezing now?

About the albedo
Whoever will work on that subject, it would be good to know if the bare soil albedo maps prepared by Sebastian and team can be directly pluged into Orchidee-trunk or if we need to see with ISBA people, or to redo the work ourselves.

Jan Polcher, 27/01/15 (mail) to F. Chéruy, 27/01/15

Je suis d'accord avec toi qu'il faudra définir la "configurabilité" des niveaux verticaux en fonction des différentes contraintes. Le nombre de nivaux sera sans doute un résultat plutôt qu'un paramètre.

Sinon pour le filtre. Tu as une onde thermique qui a une vitesse de propagation dans le sol. Il faut avec la discrétisation satisfaire le critère de Courrant-Levy : ne pas sauter plus d'une maille par pas de temps. Cela n'est pas possible avec les niveau de l'hydrologie. Au lieu de filtrer les ondes rapides (On ne veut pas de cela !) il faudra filtrer le résultat et aboutir à une discrétisation effective (au plan numérique) plus lâche que celle de l'hydrologie.

C'est la même chose qui est fait dans LMDZ au pôle !

Jan Polcher, 06/02/15 (mail)

I have given it some thought on how to specify the parameter in order to choose the common levels between hydrol and thermosoil.

In order to see what works best from a user point of view, I have written a simple python script to select the properties of the vertical discretisation and plot the result. The 4 parameters I propose for the moment are :
Depth of top hydrology layer [9.77517107e-04 m] >
Depth of hydrology [2.0m] >
Depth at which linear hydrology layers start [0.75 m] >
Depth of thermodynamics [8.0m] >
in [] are the default values you can if you do not enter anything to the script and which produce a result very similar to what we have today. Attached is also the plot produced by the script ​http://forge.ipsl.jussieu.fr/orchidee/attachment/wiki/Meetings/CMIP6/Physic/VerticalLayers.pdf.

There is still the open question if we need a vertical filter for the top layer of thermosoil, and if so we need to specify the filtering depth.

So please play with this and see if these are the parameters which are best suited for ORCHIDEE or if we should change some.

The way I have coded the calculation of the vertical levels and number of layer in Python, are very FORTRAN like. So once we agree it should be easy to put that into a module of ORCHIDEE and test again if we need or not a numerical filter.

So once we agree on the choice it can be implemented in the trunk.

Agnès Ducharne, Fuxing Wang, Fréderique Chéruy, 06/02/15 (meeting)

We had a skype meeting today about the soil discretization for water and T. We looked at three discretizations (D1: as in Trunk; D2: 8m for water and T, with the same 17 nodes, and the same location of the top 11 nodes as in D1 for water; D3: 2m for water as in D1 + assumption of uniform profile below, 8m for T as in D2, so that the two diffusion schemes use the same nodes in the top 2m)

• Spin-up: Fuxing tested this in off-line mode, by repeating the same year, globally (forcing dataset to be specified). Almost all land points reach equilibrium in 20 years for D1, and you need more than 40 years with D2. Fuxing still needs to check how long it takes to reach thermal equilibrium in D3, then to do the same tests online.
• The goal of the spin-up phase is to compare the different simulations in a comparable state: the next steps will be to compare the water budget + seasonal cycles of T, top SM, surface fluxes and river discharge, between D1 and D2, and the thermal state between D1, D2 and D3. To be tested off-line then on-line. This should allow us to decide which kind of vertical discretization is preferable for CMIP6.v1. No conflict with the above suggestion by Jan, which makes the vertical discretization more flexible.
• Then, we'll have to check how the chosen discretization performs with the developments of Fuxing regarding the soil termal properties (dependence on soil texture) and heat convection (from which we don't expect a large effect).
• We shall not forget the issue of the large T variations in the top soil layer in D2 and D3, with a thin top soil layer for heat diffusion.
• If the group agrees on using the Reynolds map of texture for CMIP6, with 12 texture classes, this will impose to have a soil depth of at least 9 m for T (then why not 10 m?) + basic checks on the water budgets and T behavior given the new textures.
• Shall we include permafrost for CMIP6.v1 ? with which soil depth and discretization for T ?

Agreed actions and Agenda