220 likes | 439 Views
Modeling dynamic vegetation with ELM-FATES: v2 status update. E3SM All-Hands Meeting March 19-21, 2019. Jennifer Holm, Ryan Knox, Khachik Sargsyan, Daniel Ricciuto, Bill Riley FATES Modeling Team!. Big leaf model (ELM). Individual based model. FATES = cohort model with demography, •
E N D
Modeling dynamic vegetation with ELM-FATES: v2 status update E3SM All-Hands Meeting March 19-21, 2019 Jennifer Holm, Ryan Knox, Khachik Sargsyan, Daniel Ricciuto, Bill Riley FATES Modeling Team!
Big leaf model (ELM) Individual based model FATES = cohort model with demography, • Patches of different forest age, • Succession that moves through time since disturbance, • Fast fluxes & slow dynamic vegetation growth and mortality • No climate envelopes Cohort model (FATES) FATES Developed in NGEE-Tropics and Partnering with E3SM
FATES in E3SMv2 ---- Current and New Features • Features already existing in ELM-FATES – Heterogeneity in patches and cohorts – Multiple allometry schemes (user specified) – Options in the canopy architecture ‘Perfect Plasticity Approximation’ (user specified) – Logging & fire disturbances – Leaf ages; variable SLA profile; variable soil depth – Plant hydraulics; • Previous challenge - needs to occur with dynamic successional development of plants. Done! • Features still be developed and tested for E3SM v2 – Heterogeneity in patches and cohorts • Needs improvement and improved PFT co-existence – Plant hydraulics • Testing and parameterization – Nutrient competition • Finalizing testing nitrogen & phosphorus, then couple with belowground BGC code. – Phenology improvements • Water stressed deciduous plants – Global PFT parameterization and UQ (boreal forests underway) – Higher resolution wind disturbance (NGEE-Tropics only) ✓ ✓
Plant Hydraulics in FATES • Continuous plant hydrology while plants continuously grow; i.e., “dynamic vegetation” – in ELM-FATES – Hydraulics updated as a function of height and biomass – Tracks water in newly recruited plants; Water balance conserve during mortality of plants – Updates to plant mortality where death is based on fractional losses of conductivity and conductivity threshold – Work by: Chonggang Xu, Brad Christoffersen, Ryan Knox, others. • Still needed: – Updating hydraulic parameters for all global PFTs; 4 plant organs ✓ ✓ Parameters = Pressure-volume curves (4 organs) Xylem conductivity curves (3 organs) Stomatal closure Root architecture stomata leaf storage stem xylem stem storage stem xylem SOIL SURFACE Absorbing root Transport root root xylem … PLANT RHIZOSPHERE
E3SM v2; 3 options E3SM v3; 4 options Big-leaf ELM Big-leaf ELM ELM-FATES ELM-FATES BradChong BradChong + Simple PHS VSFM_PHS VSFM soil + VSFM soil + VSFM soil Simple PHS BradChong 4 potential Plant Hydraulic schemes in total = • Simple PHS = based on the CLM5 plant hydraulics • VSFM_PHS = variably saturated flow model with plant hydraulics • BradChong = Brad and Chonggang’s FATES plant hydraulics • BradChong + VSFM = Brad and Chonggang’s plant hydraulics and VSFM in soil Red arrows = Default setting
Enabling Plant Nutrient Dynamics in ELM-FATES Module Design Will Allow For: Plant Biophysics Mortality Disturbance Lots of Stuff... ELM- FATES 1) Multi-hypothesis intercomparison, including future approaches. } 1) Could hypothetically work in any vegetation model (e.g. FATES). ✓ ✓ New Stand- alone Module Plant 1) Has its own lightweight coupler. Nutrient Allocation + Transport 1) Advanced software techniques give scientists a protected sand-box to work in. PARTEH = Plant Allocation and Reactive Transport Extensible Hypothesis (work by Ryan Knox, and others)
Nutrient Competition in ELM-FATES Catm Steps = 1. Lots of defining, meetings, coordinating. 2. Develop single-plant testing module that isolated plant allocation. 3. Convert current carbon only scheme into new module. 4. Add C+N+P to module, test in single tree simulator. 5. Add module into FATES, test C+N+P with simplified boundary conditions. 6. Couple new reactive transport (PARTEH) with below-ground BGC code. 7. Utilize new interface (EMI) development. ✓ ✓ Cleaf GPP Nleaf ✓ ✓ ✓ ✓ ✓ ✓ Csap Nsa p Ndead Cdea d Nlitter Clitter Nfr Cfr Nmin
ELM-FATES One grid-cell Plant functional types: Needleaf Evergreens Broadleaf Evergreen Shrubs
NPP Difference compared to MODIS Mean 3.4 (gC /m2/d) High NPP bias in FATES compared to MODIS Mean +2.1 (gC /m2/d)
Towards improving global application of FATES • Moving outside the tropics. Successfully simulating boreal forests. more water vapor Popular and Birch more productive biosphere 1) more vegetation ice melt warming darker surface Swann et al. 2010 White and Black Spruce How are water and carbon cycling impacted with shifts to more evergreen (under warming) or deciduous (under drying, fires, changes in nutrient availability)?
FATES in Boreal Forests • Updating allometric traits for wood and leaf biomass; very important for demography • Alaska biomass testing against field data Leaf Biomass Allometry
Parameter Sensitivity Testing and UQ Carbon stress mortality (%) Leaf allometry coefficient Vcmax Stem allometry coefficient Growth respiration fraction 2) Leaf Nitrogen Stoichiometry Carbon storage target Leaf Biomass 1) Height Allometry Vcmax 3) Specific Leaf Area Leaf vs. storage C priority Max. diameter to crown Maximum DBH Min. diameter to crown 4) Target for Carbon Storage 5) Leaf Allometry
Generic UQ Workflow • git clone git@github.com:ACME-Climate/Uncertainty-Quantification.git • Python scripts utilizing UQTk Toolkit (www.sandia.gov/uqtoolkit) a lightweight C/C++ UQ toolkit from SNL-CA, part of tools within FASTMath SciDAC Institute • Several demos available (uncertainty propagation, surrogate, sensitivity) • Plotting scripts for quick automated analysis • See confluence • Bottom line
Global Sensitivity Analysis (GSA) and Forward UQ • Also called Sobol indices or variance-based decomposition • Fraction of output variance explained by the given parameter • … i.e., how much output variance would reduce if we had to fix a given parameter to its nominal value • Essentially, forward propagation of uncertainties • Major UQ challenge: large number of uncertain input parameters • We developed and employ weighted iterative Bayesian compressed sensing for high-dimensional polynomial surrogate construction • Approach: • Generate ensemble of model simulations with inputs sampled over given ranges • Build surrogate approximation of input-output maps over the prior range of parameter variability • Perform GSA screening, pick a subset of impactful parameters • Repeat the study with fewer parameters to obtain a more accurate surrogate
Boreal forest run over 300 years, influence on LAI Leaf Area Index Sensitivity Index (0-1)
Global Sensitivity Analysis across many QoIs Needleleaf Evergreen Boreal PFT in FATES
Next up: Bayesian calibration of the model Employ the surrogate for Bayesian calibration of model inputs given observational data
Thanks! Questions? Jennifer Holm (jaholm@lbl.gov) Khachik Sargsyan (ksargsy@sandia.gov)