Today’s Big Picture Question Regarding Predicting and Quantifying the ‘Breathing of the Biosphere’:

• Can We Produce Flux Information that is ‘Everywhere All the Time’ with a Mechanistic Biophysical Model?

Dennis BaldocchiYoungryel Ryu & Hideki KobayashiUniversity of California, Berkeley

Stomata: 10-5 m

Leaf: 0.01-0.1 m

Plant: 1-10 m

Canopy: 100-1000 m

Landscape: 1-100 km

Continent: 1000 km (106 m)

Globe: 10,000 km (107 m)

Bacteria/Chloroplast: 10-6 m

How Do We Transcend Flux Information from the Scales of the Stomata to the Leaf, Plant, Lanscape and Globe?

A Challenge for Leaf to Landscape Upscaling:

Transform Weather Conditions from a Weather Station to that of the Leaves in a Canopy with Their Assortment of Angles and Layers Relative to the Sun and Sky

And use that information to drive a variety of Non-Linear Functions (photosynthesis, energy balance, stomatal

conductance)

Hierarchy of Canopy Abstractions

Cornelus T deWit (1970)

‘.. To build a model we have to consider and join two levels of knowledge. The level with the sort of relaxation times is then the level which provides the explanation or the explanatory level and the one with the long relaxation times, the level which is to be explained or the explainable level…’

The Perils of Upscaling Leaf-Scale Fluxes

Upscaling from Landscapes to the Globe

‘Space: The final frontier … To boldly go where no man has gone before’

Captain James Kirk, Starship Enterprise

Piers Sellers, Biometeorologist (and Astronaut), broke the ‘deWit’ barrier by attempting to incorporate Soil-Vegetation-Atmosphere Transfer models (SVATS) into Global Circulation Climate Models, but at coarse

spatial resolution

Global-Scale SVAT Modeling is Possible Today

Challenge for Landscape to Global Upscaling

Converting Virtual ‘Cubism’ back to Virtual ‘Reality’

Realistic Spatialization of Flux DataRequires the Merging Numerous Data Layers with

varying Time Stamps (hourly, daily, weekly), Spatial Resolution (1 km to 0.5 degree) and Data Sources

(Satellites, Flux Networks, Climate Stations)

To Develop a Scientifically Defensible Virtual World‘You Must get your boots dirty’, too

Collecting Real Data Gives you Insights on What is Important & Data to Parameterize and Validate Models

-180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180

Longitude

-90

-75

-60

-45

-30

-15

0

15

30

45

60

75

90

La

titu

de

FLUXNET 2007

• Motivation for a High-Resolution, Space-Driven, and Mechanistic Trace Gas Exchange Model– Current Global-Scale Remote Sensing Products tend to rely on

• Highly-Tuned Light Use Efficiency Approach– GPP=PAR*fPAR*LUE (since Monteith 1960’s)

• Empirical, Data-Driven Approach (machine learning technique)• Some Forcings come from Satellite Remote Sensing Snap Shots, at fine

Spatial scale ( < 1 km)• Other Forcings come from coarse reanalysis data (several tens to

hundreds of km resolution)– Hypothesis, We can do Better by:

• Applying the Principles taught in Biometeorology 129 and Ecosystem Ecology 111 which Reflect Intellectual Advances in these Fields over the past Decade and Emerging Scaling Rules

• Merging Vast Environmental Databases at same resolution• Utilizing Microsoft Cloud Computational Resources

Lessons Learned from the John Norman, Experience with the CanOak Model, and Reading the Literature

We Must:• Couple Carbon and Water Fluxes• Assess Non-Linear Biophysical Functions with Leaf-Level

Microclimate Conditions• Consider Sun and Shade fractions separately• Consider effects of Clumped Vegetation on Light Transfer• Consider Seasonal Variations in Physiological Capacity of

Leaves and Structure of the Canopy

Atmosphericradiativetransfer

Canopy photosynthesis,Evaporation, Radiative transfer

Soil evaporation

Beam PAR NIR

Diffuse PAR NIR

Albdeo->Nitrogen -> Vcmax, Jmax

LAI, Clumping-> canopy radiative transfer

dePury & Farquhar two leaf Photosynthesis model

Rnet

Surface conductance

Penman-Monteithevaporation model

Radiation at understory

Soil evaporation

shade sunlit

BESS, Breathing-Earth Science Simulator

Necessary Attributes of Global Biophysical ET Model: Applying Lessons from the Berkeley Biomet Class and CANOAK

• Treat Canopy as Dual Source (Sun/Shade), Two-Layer (Vegetation/Soil) system– Treat Non-Linear Processes with Statistical Rigor (Norman, 1980s)

• Requires Information on Direct and Diffuse Portions of Sunlight– Monte Carlo Atmospheric Radiative Transfer model (Kobayashi + Iwabuchi,, 2008)

• Couple Carbon-Water Fluxes for Constrained Stomatal Conductance Simulations– Photosynthesis and Transpiration on Sun/Shade Leaf Fractions (dePury and Farquhar,

1996)– Compute Leaf Energy Balance to compute Leaf Saturation Vapor Pressure and Respiration

Correctly– Photosynthesis of C3 and C4 vegetation Must be considered Separately

• Light transfer through canopies MUST consider Leaf Clumping to Compute Photosynthesis/Stomatal Conductance correctly (Baldocchi and Harley, 1995)

– Apply New Global Clumping Maps of Chen et al./Pisek et al.• Use Emerging Ecosystem Scaling Rules to parameterize models, based on remote sensing

spatio-temporal inputs– Vcmax=f(N)=f(albedo) (Ollinger et al; Hollinger et al; Wright et al.)– Seasonality in Vcmax is considered (Wang et al., 2008)– Vcmax scales with Jmax (Wullschleger, 1993 )

Role of Proper Model Abstraction

But, We Need Big Iron to Play with the Big Guys and Gals

MOD04

MOD05

MOD06

MOD07

aerosol

Precipitable water

cloud

Temperature, ozone

MCD43 albedo

MOD11 Skin temperatureAtm

ospheric radiative transfer

Net radiation

MOD15 LAI

POLDER Foliage clumping

Canopy radiativetransfer

Challenge for a Computationally-Challenged Biometeorology Lab:Extracting Data Drivers from Global Remote Sensing to Run the Model

Youngryel was lonely with 1 PC

Barriers to Global Remote Sensing by the Berkeley Biometeorology Lab

• Data processing– Global 1-year source data: 2.4 TB (10 yr: 24 TB)– 150,000+ source files– Global 1-year calculation: 9000 CPU hours– That is, 375 days.– 1-year calculation takes 1 year!

Help from ModisAzure -Azure Service for Remote Sensing Geoscience

Scientific Results Download

Reduction #1 Queue

Source Metadata

AzureMODIS Service Web Role Portal

Request Queue

Analysis Reduction Stage

Data Collection Stage

Source Imagery Download Sites

. .

.

Reprojection Queue

Derivation Reduction Stage Reprojection Stage

Reduction #2 Queue

DownloadQueue

Scientists

Science results

AZURE Cloud with 200 CPUs cuts 1 Year of Processing to <2 days

Photosynthetic Capacity Leaf Area Index

Solar RadiationHumidity Deficits

Ryu et al (Accepted) Global Biogeochemical Cycles

118±26 PgC yr-1

BESS vs Machine Learning Upscaling Method

Ryu et al (Accepted) Global Biogeochemical Cycles

Global Evaporation at 1 to 5 km scale

An Independent, Bottom-Up Alternative to Residuals based on the Global Water Balance, ET = Precipitation - Runoff

<ET> = 503 mm/y == 6.5 1013 m3/y

BESS vs Machine Learning Upscaling Method

Ryu et al (Accepted) Global Biogeochemical Cycles

Big Picture Question Regarding Predicting and Quantifying the ‘Breathing of the Biosphere’:

• Can We Produce Flux Information with a Mechanistic Model that is ‘Everywhere, All the Time?’...Yes

Gross Photosynthesis, GPP, Across the US

Lessons for Biofuel Production

Indicates Less GPP in the Corn Belt, than the Adjacent Temperate Forests

Key point: 4. Temporal upscaling of fluxes from snap-shots to 8-day mean daily sum estimates

d gPOT

gPOT

d

d dttR

tRs

dttEtEstSF

)(

)(1800

)()(1800)(

8

18 )(

)(180081d dd

dday tSF

tEsE

Ryu et al (2011) Agricultural and Forest Meteorology Accepted

RgPOT =f(latitude, longitude, time)

Instantaneous LE

Rg at TOA

Day (1-8)

30 minSatellite overpass time

Ryu et al (Accepted) Agricultural and Forest Meteorology

Tested the scheme using 33 flux tower data from the Arctic to the Tropics

Ryu et al (Accepted) Agricultural and Forest Meteorology

Conclusion• Three-Dimensional Radiative Transfer models should be used

to compute Mass and energy exchanges of Heterogeneous canopies– Models can be implemented with new generation of LIDAR data and

powerful clusters of computers• Advances in Theory, Data Availability, Data Sharing and

Computational Systems Enable us to Produce the Next-Generation of Globally-Integrated Products on the ‘Breathing of the Earth’

• Data-Mining these Products has Much Potential for Regional and Locale Decision making on Environmental and Agricultural Management

• Data standardization

MODIS Land products: standardized tiles (sinusoidal projection)

Barriers for global RS study

• 2. Data standardization

MODIS Atmospheric products: swath=> Should be gridded to overlay with the land products

Current status

• The Cloud includes– 10-year MODIS Terra and Aqua data over the US (1

km resolution)– 3-year MODIS Terra for the global land (5 km

resolution)• Quota:

– 200 CPUs– 100TB storage

Help from MODIS-AZURE

Necessary Attributes of the Next-Generation Global Biophysical Model, BESS

• Direct and Diffuse Sunlight– Monte Carlo Atmospheric Radiative Transfer model (Kobayashi, xxxx)– Light transfer through canopies consider leaf clumping

• Coupled Carbon-Water for Better Stomatal Conductance Simlulations– Photosynthesis and Transpiration on Sun/Shade Leaf Fractions (dePury and

Farquhar, 1996)– Photosynthesis of C3 and C4 vegetation considered

• Ecosystem Scaling Relations to parameterize models, based on remote sensing spatio-temporal inputs– Vcmax=f(N)=f(albedo) (Ollinger et al; Hollinger et al;Schulze et al.; Wright et al.– Seasonality in Vcmax is considered

• Model Predictions should Match Fluxes Measured at Ecosystem Scale hourly and seasonally.

Seasonal pattern of Vmax@25 follows the seasonal pattern of LAI(modified version of Houborg et al 2009 AFM)

Size and Number of Candidate Data Sets is Enormous

US: 15 tilesFluxTower: 32 tilesGlobal: 193 tiles

1. Global 1-year source data: 2.4 TB (10 yr: 24 TB)2. How to know which source files are missed among >0.1 million files

• Automation– Downloads thousands of files of MODIS data from NASA ftp

• Reprojection– Converts one geo-spatial representation to another. – Example: latitude-longitude swaths converted to sinusoidal cells

to merge MODIS Land and Atmosphere Products• Spatial resampling

– Converts one spatial resolution to another. – Example is converting from 1 km to 5 km pixels.

• Temporal resampling – Converts one temporal resolution to another.– Converts daily observation to 8 day averages.

• Gap filling – Assigns values to pixels without data either due to inherent

data issues such as clouds or missing pixels.• Masking

– Eliminates uninteresting or unneeded pixels.– Examples are eliminating pixels over the ocean when

computing a land product or outside a spatial feature such as a watershed.

Reprojected Data (Sinusoidal format - equal land area pixel)

Tasked Performed with MODIS-AZURE

h12v04h13v04h11v04h10v04h09v04h08v04

h12v05h11v05h10v05h09v05h08v05

h11v06h10v06h09v06h08v06

remote sensingof CO2

Tem

pora

l sca

le

Spatial scale [km]

hour

day

week

month

year

decade

century

local 0.1 1 10 100 1000 10 000 global

forestinventory

plot

Countries EUplot/site

talltowerobser-

vatories

Forest/soil inventories

Eddycovariance

towers

Landsurface remote sensing

Components of an Integrated Earth System EXIST, but are Multi-Faceted

From: Markus Reichstein, MPI

Computing Carbon Dioxide and Water Vapor Fluxes Everywhere, All of the Time

Dennis BaldocchiYoungryel Ryu & Hideki KobayashiUniversity of California, Berkeley

AGU, Fall 2011