towards a diagnostic terrestrial c - h2o model-data …data & hypothesis driven model...

Towards a diagnostic terrestrial C-H2O model-data fusion analysis and

prediction framework

A. Anthony Bloom, ESM workshop, Mar 2018

A. Anthony Bloom*1,ESM workshop Mar 27 2017

*[email protected]: Junjie Liu1, Kevin Bowman1, Sassan Saatchi1, Michael Keller1, Alexandra Konings2, John Worden1, Meemong Lee1, David Schimel1, Mathew Williams3,

Nicholas Parazoo1, John Reager1,others.1Jet Propulsion Laboratory, California Institute of Technology, 2Stanford University, 3University of Edinburgh, UK, 4California

Institute of Technology

Improving predictions depends on quantitatively understanding of present-day terrestrial C cycle dynamics:- Climate sensitivity of terrestrial C fluxes- Ecosystem memory (lags, initial conditions, residence times)- C cycle feedbacks and interactions with water, energy and nutrient cycles


?

Le Quéré et al. (2015)

Presenter

Presentation Notes

Here’s the evolution of the land sink (50+ years)? Predictive ability hinges on understanding of present-day dynamics I willl show how global EO allow us toplace a join constrainy on these Emphasis on (a), (b), (c) I am going to argue that joint knowledge of these is central to achieving an unbiased prediction of the terrestial C balance!

Terrestrial carbon-climate feedbacks


ΔCO2 AtmosphereΔclimate

BiosphereΔCO2

global

regionalForcing

Presenter

Presentation Notes

CO2 effect of atmosphere is global, stastical shifts in climate variability are regional Need spatially explict understanding of regional responses of biospheric C fluxes to regional climate variability to understand the sign and magnitude of the terrestrial C response

Liu et al., 2017

A. Anthony Bloom, AGU Fall Meeting, Dec 2017

OCO-2 & GOSAT CO2

MOPITT CO GOSAT SIF

Presenter

Presentation Notes

Why do we need spatially explicit results Localized results in terms of climate extremes, need to use these results to constrain space-time behaviour of terrestrial C cycle Responses are localized, suggested process controls vary across continents, to the same climate event

Terrestrial C cycle: towards reducing process uncertainty.

MODIS LAI

Tropical Biomass

HWSD Soil Carbon

GOSAT SIF

Atmospheric CO2 (GOSAT)

MechanisticLinks?

Atmospheric CO(MOPITT)

Le Quéré et al., 2016

Friedlingstein et al., 2014

Land sink

Land sink projections


Presenter

Presentation Notes

This motivates us to understand the mechanistic links between Variety of datasets = constrain components How do the quantities for qwhcih we have a spatially explicit understanding link to each other? What do these tell us about the evolution of the terrestrial land sink? How do we build on this?


Williams et al. (2005), Bloom & Williams (2015), Bloom et al. (2016)

Spatially-explicit optimization of terrestrial C cycle parameters

Carbon data-model framework (CARDAMOM)

Bloom et al. 2016, PNAS

CARDAMOM: Terrestrial carbon cycle state and process variables


GOSAT and OCO-2 NBE – assimilation & prediction


Black = GOSAT-derived NBE (assimilated into CARDAMOM)Blue = OCO-2 derived NBE (withheld for validation)

Red = CARDAMOM meanPink = CARDAMOM 1σ

Do memory effects contribute to NBE IAV?Lagged effect: attributable to ecosystem memory effects

Direct effect: attributable to instantaneous met. forcings

Net biosphericexchange

Year

Lagged (memory) effect

Direct + Indirect

NBE

(Pg

C)

Direct only

ΔNBEtotal = ΔNBEdirect + ΔNBElagged


Direct & lagged NBE Attribution approach: meteorological forcing sensitivity of optimized CARDAMOM NBE


Ecosystem memory in tropical ecosystems

Blue = totalGreen = direct

Orange = lagged

Bloom et al. (2018, in prep.)

Why?

IAV and trajectories of plant-available H2O, canopy C and soil C stocks influence NBE responses to climate variability.


Bloom et al. (2018, in prep.)

Blue = Plant-available H2OGreen = Foliar COrange = Soil C

Memory effects on tropical NBE IAV

• Memory effects account for ~ 50% of tropical NBE IAV

• Quantitative knowledge of NBE legacy effects is necessary to predict the evolution of the terrestrial C balance.

• Process-oriented diagnostic model central to advancing mechanistic understanding of terrestrial C cycling.


Direct vs total Indirect vs total Direct vs indirect

Presenter

Presentation Notes

Quantitative knowledge of NBE direct vs memory contributions is necessary, in part because we need these to back out the sensitvity of the biosphgere to climate variability.

Example: data-oriented evolution of a diagnostic carbon-water model structure


Presenter

Presentation Notes

I’m going to show how we can advance our diagnostic model structures to match the availability of Earth Observation data

Prediction error vs model complexity

Lever et al., 2016

Insufficient process insight

Overconfidence in modeled processes

What level of complexity do we need given:(a) The current Earth-

Observing system(b) The quantities we’re

trying to predict?


Presenter

Presentation Notes

Deciding what we’re trying to predict will immediately define the required model complexity. More measurements = higher level of justified complexity

CARDAMOM modular structure


1. ModelDiagnostic C & H2O

models (DALEC), etc.

2. Drivers & Observations

Meteorology, LS datasets, plant traits, etc.

4. Data assimilation

MCMC, EnKF, etc.

3. Cost functionObservation operator, prior

parameter constraints, ecological & dynamic

constraints.

Science, prediction.

Presenter

Presentation Notes

Modular structuure allows us to swap models in and out. This allows us to rapidly investigate the role of specific process representations on C cycle prediction

DALEC genealogy: data & hypothesis driven evolution

DALEC_CDEA

DALEC2

DALEC evergreen

DALEC deciduous

DALEC GSI

Data

& h

ypot

hesis

driv

en m

odel

dev

elop

men

t

DALEC_FIREBUCKET

Spadaveccia et al. 2011

DALEC-FGRowland et al. 2014

Smallman et al. 2016, Exbrayat et al.(2018, in prep)

Bloom & Williams 2015

Bloom et al. 2016

Fox et al. 2009

DALEC & C-H2O

Bloom et al., 2018 (in prep.)

DALEC2’ (structural sensitivity tests)

DALECcRevill et al., 2014

Williams et al. 2005

DALEC-GSI-HBV


DALECN-GSI-BUCKET

Presenter

Presentation Notes

Model of optimal complexity to (a) represent assimilated datasets and (b) test hypotheses on the mechanistic links of the terrestrial C cycle. This is reflected in the evolution and process diversity of Mat’s DALEC model, and the process advances give rise to testable hypotheses.

Carbon-water feedbacks

dW/dt = P – ET – R

dC/dt = GPP – Reco – D

Change in terrestrial water balance

Change in terrestrial carbon balance

Precip.Evaporation + plant transpiration

Runoff

DisturbancesEcosystem respiration

Gross primary production


Gravity recovery and climate experiment (GRACE)


Presenter

Presentation Notes

Motivates us to look at terrestrial water storage as a constraint on the land-surface carbon-water balance.


GRACE total water storage anomaly (ΔTWS)

Example: data-driven model structure evolution

Presenter

Presentation Notes

JOINT ESTIMATION of parameters

Red = GRACE (withheld)Δterrestria

lwat

er

stor

age

[mm

]Δterrestria

lwat

er

stor

age

[mm

]

Black = GRACE (assimilated)

Blue = CARDAMOM

Blue

= C

ARDA

MO

M


CARDAMOM: C and GRACE constraints


CARDAMOM: C constrains only

Conclusions & questions• Joint understanding of climate sensitivity and

ecosystem memory effects on present-day NBE IAV necessary for prediction.

• Observational constraints on both carbon and water cycle dynamics are central to building a land-surface predictive ability.

• How do we bring and test process-based model knowledge in a diagnostic C-H2O-energy model-data framework?

• What is the optimum balance between process fidelity and data consistency for decadal-to-centennial predictions?


Presenter

Presentation Notes

What is the optimum balance between process fidelity and data consistency?

Do ecosystem memory effects matter for NBE IAV prediction?


|δ20

13 N

BE|

[nor

mal

ized]

|δ Lagged effect contribution| [normalized]

Optimized CARDAMOM parameters and state variables.

Random perturbations (δx)on optimized CARDAMOM parameters (x)

CARDAMOM (x + δx)

r=0.39 r = 0.56 r = 0.41

r = 0.40 r = 0.50 r = 0.42

Example: 2013 NBE mismatch

1:1

Presenter

Presentation Notes

0.3912 0.5595 0.4110 0.4027 0.5019 0.4190

towards a diagnostic terrestrial c - h2o model-data …data & hypothesis driven model...

Documents