Hydraulic architecture improvements

Meeting the 07 October 2021: comparison of Julien and Yitong schemes

The purpose of the meeting was to compare Yitong Yao’s model with Julien Alléon’s one in order to i) Understand the differences between both models and ii) prepare a strategy in order to implement only one “mixed” model in ORCHIDEE_trunk.

The meeting was organised around the presentation of the two models and the discussion on the advantages/drawbacks of each model:

Summary of the differences:

  • Starting point:
    • Yitong: Emilie Joetzjer’s previous version of the hydraulic architecture in the module hydraulic_arch.f90. This previous version was more a proof about the necessity of introducing and hydraulic architecture than a clean implemented model as ORCHIDEE would need.
    • Julien: New model based on ORCHIDEE_trunk version of May 2020. New subroutines introduced in hydrol module.
  • Overall scheme: Both scheme rely on the description of the water transport from soil/root interface towards leaves thanks to a resistance/capacitance scheme.
  • Root absorption:
    • Yitong: Rely on a development by Emily which, in addition to a soil/root resistance based on soil conductivity and root biomass, adds a weighting of the absorption according to the amount of water present in each layer. The idea is to weight more by the amount of water than by the root biomass.
    • Julien: Two ways to model root absorption:
      • Radial resolution of Richards’ equation around a fictive root which have the length of the total amount of roots in each layer;
      • Classic root absorption scheme relying on a dynamic resistance
  • Transport through canopy
    • Resistances:
      • Yitong: Dynamic resistances linked to the water potential at each stage
      • Julien: fixed resistances
    • Capacitances:
      • Yitong: fixed values
      • Julien: Dynamic values linked to the water potential of each storage pool.
  • Stomatal conductance:
    • Yitong: Two decoupled stomatal conductances for C assimilation and water control. Only the one that controls water is linked to leaf water potential
    • Julien: One stomatal conductance calculated thanks to the leaf water potential. The conductance controls assimilation and water supply.
  • Resolution method:
    • Yitong; Global method: the scheme calculates transpiration demand and supply. If the supply does not reach the demand, energy budget is recalculated (enerbil or mleb has to iterate to converge towards the supply). Inside the model: Iteration process in order to converge towards the fluxes at each stages. Uses minpack package for the resolution
    • Julien: Resolution without iterations based on prediction/correction schemes for differential equations and estimation/corrections for non differential equations. ==> Causes instabilities.


Both models seem to have good results on the site studied: Yitong's model works well at Hesse and Julien simulate the temporal variations of LE at Caxana but with too low transpiration because he is using 2m of soil instead of 4m in Yitong. Several options are possible in ORCHIDEE, one would be a flag to activate one scheme or the other (but hard to maintain). The best option would be to implement a mixed version of both models keeping a resolution without iterations and introducing the dynamic resistances:

  • Root absorption: Best option would be to activate a flag to choose between the three models;
  • Transport: Best option would be to have both dynamic resistances and capacitances ;
  • Stomatal conductance: Best option would be to have only one stomatal conductance (this option leads to instabilities as mentioned in Julien’s model;
  • Resolution method: Best option would be to avoid iterations as much as possible. Moreover, recalculating the energy budget has to be avoid in order to remove iterations in mleb.
  • To avoid instability, several options were given:
    • Look more precisely at the storage pools that should act as a buffer;
    • Improve the "predictor / corrector" approach

Note that dynamic resistances are probably crucial for the mortality calculation (by hydraulic failure).

Next steps:

Yitong is entering the final stretch of her PHD and will not have much time to allocate to the project before March 2022. Suggestion: She could possibly do a simulation with a precise configuration in order to look at the extra-time calculation added by the iterations and the minpack solver.

Julien will look closer on how to implement some features of Yitong in his scheme or just the two versions in the same code in order to compare them more easily.

Last modified 8 months ago Last modified on 10/09/21 16:41:15

Attachments (2)