Integrating Remote Sensing, Flux Measurements and Ecosystem Models Faith Ann Heinsch Numerical Terradynamic Simulation Group (NTSG) University of Montana NCAR ASP 2007 Colloquium Regional Biogeochemistry June 12, 2007
Spatial scale (m) Time scale (s) hour day year decade Tree Rings Chambers Inventories Historical observations Remote Sensing Stream Flow Method Hopping Sap Flow Eddy Flux Climate gradients Manipulations
Spatial scale (m) Time scale (s) hour day year decade Remote Sensing Method Hopping Eddy Flux
Spatial scale (m) Time scale (s) hour day year decade Eddy Flux Eddy Covariance
Spatial scale (m) Time scale (s) hour day year decade Remote Sensing Remote sensing
MODIS GPP (MOD17)
ε = ε max * m(T min ) * m(VPD) Stress Scalars for Light Use Efficiency VPDTemperature Light Use Efficiency
Land Cover (MOD12Q1) –Biome Type –Annual, 1-km 8-Day FPAR/LAI (MOD15A2) –FPAR and living biomass –8-day, 1-km Daily Meteorological Data (DAO) –Environmental conditions –Driving forces –Daily, 1.00 x 1.25 GPP/NPP (MOD17A2/A3) Inputs to the MOD17 GPP/NPP Algorithm
MOD17 BPLUT – v. 4.8
MODIS GPP
Comparison of GPP from Terra-MODIS and AmeriFlux Network Towers Biome types used in comparison: forests (evergreen needleleaf, deciduous broadleaf, and mixed species), oak savanna, grassland, tundra, and chaparral.
Calibration / Validation Tests VEGETATION: Forests, Grass, Shrubs, and Crops CLIMATE: Cold-Dry, Cold-Wet, Warm-Dry, and Warm- Wet GEOGRAPHIC PATTERNS GROWING SEASON (Start and End) STRESS (Mid-Summer Water Stress, Cold Temperatures, High Vapor Pressure Deficits) SEASONAL PATTERNS FLUX MAGNITUDE
Location of the AmeriFlux network sites AmeriFlux: Fluxnet:
Uncertainties in the MOD17 (GPP/NPP) Algorithm 1. Meteorological DAO IPAR, Temperature, VPD 2. Radiometric MODIS FPAR and LAI 3. Ecological MOD17 representation of plant physiology (BPLUT) Accurate mapping of landcover type Each of these requires a different mode of validation.
Uncertainties in the MOD17 (GPP/NPP) Algorithm 1. Meteorological DAO IPAR, Temperature, VPD 2. Radiometric MODIS FPAR and LAI 3. Ecological MOD17 representation of plant physiology (BPLUT) Accurate mapping of landcover type Each of these requires a different mode of validation.
Non-linear interpolation of DAO AB
Methods of DAO Smoothing The non-linear distances The weighted values The interpolated DAO variables
Climate – Niwot Ridge, CO Heinsch et al. IEEE 44: , 2006
Climate – Tonzi Ranch, CA Heinsch et al. IEEE 44: , 2006
Global Daily Surface Meteorology vs Fluxtowers across 9 biomes From D.P.Turner et al. Remote Sensing of Env. 102:
Uncertainties in the MOD17 (GPP/NPP) Algorithm 1. Meteorological DAO IPAR, Temperature, VPD 2. Radiometric MODIS FPAR and LAI 3. Ecological Accurate mapping of landcover type MOD17 representation of plant physiology (BPLUT) Each of these requires a different mode of validation.
MODIS LAI vs. Tower GPP for 15 Ameriflux Sites Heinsch et al. IEEE 44: , 2006
Uncertainties in the MOD17 (GPP/NPP) Algorithm 1. Meteorological DAO IPAR, Temperature, VPD 2. Radiometric MODIS FPAR and LAI 3. Ecological Accurate mapping of landcover type MOD17 representation of plant physiology (BPLUT) Each of these requires a different mode of validation.
Blodgett Forest, CA 1 = ENF 5 = Mixed Forest
Gainesville, FL (Austin-Carey) 1 = ENF 2 = EBF 5 = MF 8 = Woody Savanna 12 = Cropland
Uncertainties from Land Cover (MOD12Q1) 4 = Deciduous Broadleaf Forest (DBF) 5 = Mixed Forest 8 = Woody Savannas WLEF Tall Tower, Wisconsin
Land Cover Heinsch et al. IEEE 44: , 2006
Daily GPP by Biome Type, July 20~27, 2001 Credit: Sinkyu Kang, NTSG
r = % Error = 19% MODIS GPP vs. Tower GPP (DAO met.)
r = % Error = -2% MODIS GPP vs. Tower GPP (Tower met.)
Metolius (P. pine) Sylvania (dbf) Tonzi Ranch (oak savanna) Niwot Ridge (subalpine fir)
MODIS GPP/NPP vs. Flux Towers across 9 Biomes From D.P. Turner et al. Remote Sensing of Env 102:
Summary of Results MODIS GPP follows the general trend, capturing onset of leaf growth, and in many cases, leaf senescence, while tending to over-estimate total tower GPP. The MODIS GPP algorithm effectively captures the effects of stress events, such as late-summer dry-down, on canopies. Substituting tower meteorological data in the MODIS algorithm leads to GPP values which are very similar to tower GPP, suggesting the algorithm adequately estimates site GPP. If DAO meteorology and tower meteorology are similar, MODIS GPP is comparable to tower GPP. But, if the coarse-resolution DAO data is not representative of the site, MODIS GPP can differ greatly from tower GPP. Comparisons of site data that have been received are weighted heavily towards forest biomes. Other sites need to be studied to determine if results are similar in other ecosystems.
Integrating Ecosystem Process Models (e.g., Biome-BGC)
Integrating Ecosystem Process Models Does the MODIS GPP contain enough information regarding water stress? –VPD is sole water stress scalar –Soil water stress?? Test by comparing with Biome-BGC –U.S.A. –China Mu et al., JGR, 2007
Integrating Ecosystem Process Models
Water Stress Scalars (Growing Season) Mu et al., JGR, 2007
Correlation between water stress scalars in Biome-BGC and MOD17 (Growing Season) Mu et al., JGR, 2007
Correlation between water stress scalars in Biome-BGC and MOD17 (Monthly) Mu et al., JGR, 2007
Correlation between Biome-BGC and MOD17 GPP Estimates (Monthly)
Does MOD17 Capture Water Stress? Water not strongly limiting for most of the wetter areas of China and the conterminous USA – –m(VPD) reflects full water stress from air & soil as determined by Biome-BGC Using only VPD underestimates the water stress in dry regions & in areas with strong monsoons – –Western China, the northeast China plain, the Shandong peninsula, and the central and western United States – –MOD17 overestimates GPP; add soil water stress? – –Need for improved precipitation data to include soil moisture VPD alone reflects interannual variability in most areas, – –Current MOD17 adequate for global studies.