assimilating earth observation data into vegetation models tristan quaife darc seminar 11 th july...

39
ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Upload: sierra-snyder

Post on 28-Mar-2015

221 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELSTristan QuaifeDARC seminar 11th July 2012

Page 2: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Some context – the residual sink

http://www.whrc.org/global/carbon/residual.html

PgC

yr-1

Page 3: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Some context – Cox et al. 2000

The red lines represent the fully coupled climate/carbon-cycle simulation, and the blue lines are from the 'offline' simulation which neglects direct CO2-induced climate change. The figure shows simulated changes in vegetation carbon (a) and soil carbon (b) for the global land area (continuous lines) and South America alone (dashed lines).

Cox P et al. (2000) Acceleration of global warming due to carbon-cycle feedbacks in a coupled climate model. Nature. 408, 184-187

Ch

an

ge

in

ve

ge

tati

on

ca

rbo

n (

GtC

)

0 0

Page 4: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

The Land Surface DA problem At first glance similar to NWP DA

problem. Of the form:

xt+1=M(xt, p, dt) But… Observation time scales tend to be much

shorter than many of the key process In general M is not fully understood and

typical for many parameters to be determined empirically

Page 5: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Assimilating products

Data Assimilation Scheme(KF, EnKF, 4DVAR, etc)

MODEL

Assumptions

Observations

Observations

Assumptions

Assumptions

For example: soil moisture from SMOS or photosynthesis (GPP) from MODIS

Page 6: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

MODIS GPP/PSN

http://www.ntsg.umt.edu/remote_sensing/netprimary/

MODIS data

Climate data

Look up table

Page 7: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Data Assimilation Scheme(KF, EnKF, 4DVAR, etc)

Observations

Observations

MODEL

Assumptions

Observation Operator

Assumptions

Quaife T, Lewis P, De Kauwe M, Williams M, Law BE, Disney MI and Bowyer P (2008) Assimilating canopy reflectance data into an ecosystem model with an Ensemble Kalman Filter. Remote Sensing of Environment. 112(4):1347-1364

e.g. reflectance, backscatter, etc…

Assimilating low level data

Page 8: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Vegetation

Foliage

Humus

LitterRoots

Wood

GPP

Af

Ar

Aw

Ra

Lf

Lr

Lw

Rh

DMet Data

Soil

DALEC

Page 9: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Ensemble Kalman Filter

Aa = A + A′A′THT(HA′A′THT + Re)-1(D - HA)

H = observation operatorA = state vector ensembleA′ = state vector ensemble – mean state vectorD = observation ensembleRe = observation error covariance matrix

Page 10: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

EnKF – augmented analysis

Aa = A + A′Â′TĤT(ĤÂ′Â′TĤT + Re)-1(D - ĤÂ)

Ĥ = augmented observation operator = augmented state vector ensemble

 = h(A) A

h is a canopy reflectance model

Page 11: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Simple observation operator

Page 12: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Observation operator

Source: N Gobron, JRC

Page 13: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Shaded crownIlluminated

crown

Illuminated soil

Shaded soil

Geometric Observation Operator

Page 14: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Modelled vs. observed reflectance

MODIS Band 1 (red) MODIS Band 2 (NIR)

Quaife T, Lewis P, De Kauwe M, Williams M, Law BE, Disney MI and Bowyer P (2008) Assimilating canopy reflectance data into an ecosystem model with an Ensemble Kalman Filter. Remote Sensing of Environment. 112(4):1347-1364

Page 15: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Assimilating reflectance into DALEC

No assimilation

Assimilating MODIS surface reflectance bands 1 and 2

Page 16: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Carbon balance for 2000-2002

15 65

gC/m2/year

4.5 km

Flux Tower

Spatial average = 50.9

Std. dev. = 9.7

(gC/m2/year)

Page 17: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Parameter sensitivity

Problem using optical EO data is most vegetation model parameters are not sensitive to it

Broadly this is true for all EO data May change with advent of CO2 observations

Have taken a different approach for some problems: Use models driven by satellite data Assimilate available ground data

Page 18: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Mountain pine beetle

Page 19: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Mountain pine beetles

Science question: what is the impact of MPB on carbon balance of ecosystem?

Problem: most veg models are not adequately parameterised for mountain forests tend to exhibit quite different photosynthetic

responses to temperature than other forests Use simple photosynthesis model driven by

EO data Assimilate ground observations using

standard MCMC-MH Bayesian parameter estimation

Page 20: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Posterior PDF

Page 21: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Mountain pine beetle

Moore DJP, Trahan NA, Wilkes P, Quaife T, Desai AR, Negron JF, Stephens BB, Elder K & Monson RK (submitted 2012) Changes in carbon balance after insect disturbance in Western U.S. Forests.

Page 22: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

A re-think…

Started to think about how we could approach the land surface problem a little differently

First, most land surface models do not have RT physics that is consistent with EO observations Make this a design goal of vegetation models A good place to start given volume of EO data

Second, there may be additional constraints that are applicable specifically to the land surface Generally does not undergo rapid change

Page 23: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

f = kernel weightK = kernel valuen = number of kernels

λ = wavelengthρ = BRFΩ = view geometryΩ' = illumination geometry

Kernel driven BRDF model

Page 24: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

f = (KTC-1K)-1KTC-1ρ

• Formulation used for the NASA MODIS BRDF/albedo product (MCD43)

• Requires an 16 day window (Terra + Aqua) that is moved every 8 days

Standard Least Squares

Page 25: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

MODIS data product (MOD43)

Page 26: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

MODIS data product (MOD43)

Page 27: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

f = (KTC-1K + γ2BTB)-1KTC-1ρ

B is the required constraint. It imposes:

Bf = 0

and the scalar γ is a weighting on that constraint.

Constrained formulation

Page 28: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Constraint matrix

Page 29: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Constrained result

Quaife T and Lewis P (2010) Temporal Constraints on Linear BRDF Model Parameters. IEEE Transactions on Geoscience and Remote Sensing, 48 (5). pp. 2445-2450.

Page 30: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

EOLDAS

European Space Agency Project to improve data retrievals and inter-sensor calibration

Variational scheme using the following cost function:

Page 31: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

EOLDAS variational assimilation

Lewis P, Gomez-Dans J, Kaminski T, Settle J, Quaife T, Gobron N, Styles J & Berger M (2012), An Earth Observation Land Data Assimilation System (EOLDAS), Remote Sensing of Environment.

Leaf Area Index

Chlorophyll

Time

Page 32: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Spatial DA example – Synthetic Truth

NDVISource: P Lewis & J Gomez-Dans, UCL

Page 33: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Spatial DA example – Observations

NDVISource: P Lewis & J Gomez-Dans, UCL

Page 34: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Spatial DA example – Posterior

NDVISource: P Lewis & J Gomez-Dans, UCL

Page 35: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Multi-scale DA using a particle filter

Page 36: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Multi-scale DA using a particle filter

Hill TC, Quaife T & Williams M (2011) A data assimilation method for using low-resolution Earth observation data in heterogeneous ecosystems, J. Geophys. Res., 116, D08117.

Leaf Area Index:

Page 37: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Current ESA project

Project with Reading and UCL Builds on the existing EOLDAS framework Constructing a land surface scheme that

includes trace gas and energy fluxes Key aim is to have the broadest possible

range of EO observations available for DA Design goal to invest most complexity in

the physics required for the observation operator

Page 38: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Routes to collaboration inside DARC DALEC code setup in flexible framework

Already has EnKF & PF – easy to add more Easy to add non-linear observation operators Lots of test data available

EOLDAS code available Official public release very soon Very general, but also very slow

Lots of data for vegetation type problems available… ask me…

Page 39: ASSIMILATING EARTH OBSERVATION DATA INTO VEGETATION MODELS Tristan Quaife DARC seminar 11 th July 2012

Any Questions?