University of Oklahoma Oct. 22-24, 2007 Assimilating terrestrial remote sensing data into carbon models: Some issues Shunlin Liang Department of Geography University of Maryland at College Park, USA Sliang@geog.umd.edu, http://www.glue.umd.edu/~sliang
Development of Carbon Cycle Data assimilation systems Georeferenced emissions inventories Atmospheric measurements Remote sensing of Atmospheric CO2 Data assimilation link Climate and weather fields Atmospheric Transport model optimized Fluxes optimized model parameters Ocean time series Biogeochemical pco2 Surface observation pco2 nutrients Water column inventories Ocean carbon model Coastal studies Ocean remote sensing Ocean colour Altimetry Winds SST SSS Terrestrial rivers carbon model Lateral fluxes Ciais et al. 2003 IGOS-P Integrated Global Carbon Observing Strategy Remote sensing of Vegetation properties Growth Cycle Fires Biomass Radiation Land cover /use Eddy-covariance flux towers Biomass soil carbon inventories Ecological studies
Remote Sensing Technologies Multispectral Hyperspectral RADAR / SAR Thermal Atmospheric LIDAR Surface LIDAR Passive Microwave RADAR Altimetry Limb Sounding Microwave Ranging Irradiance/Photometry Scatterometry
Terrestrial remote sensing products for C studies Land cover & land use Leaf area index (LAI) & FPAR Vegetation indices (e.g., NDVI) & phenology Incident PAR Canopy height Foliage nitrogen Forest age class Biomass Fires, disturbances
MODIS Land Cover Map LC (0) Water (1) Grasses/cereal crops (2) Shrubs (3) Broadleaf crops (4) Savannah (5) Broadleaf forest (6) Needle leaf forest Collection 5 Collection 4 LC (0) Water (1) Grasses/cereal crops (2) Shrubs (3) Broadleaf crops (4) Savannah (5) Evergreen Broadleaf forest (6) Deciduous Broadleaf forest (7) Evergreen Needle leaf forest (8) Deciduous Needle leaf forest M. Fridel, Boston University
MODIS Anisotropy and Albedo White sky albedo April 2004 C. Schaaf, Boston University
MODIS Terra 8-day LAI (185-192.2000) Collection 5 LAI 0.0 Collection 4 0.1 0.5 0.8 1.0 1.5 2.5 3.0 4.0 5.0 6.0 7.0 R. Myneni, Boston University
MODIS Terra 8-day FPAR (185-192.2000) Collection 5 FPAR Collection 4 0.00 0.10 0.20 0.25 0.35 0.40 0.50 0.60 0.65 0.75 0.85 1.00 R. Myneni, Boston University
Leaf type % broadleaf % needleleaf Leaf longevity % deciduous % evergreen MODIS product, John Townshend at University of Maryland
Production Efficiency Principles: GPP = ε fpar PAR g NPP = ε fpar PAR n Grid cell level regression of net primary production (NPP) (kg C yr -1 m-2) against absorbed photosynthetically active radiation (APAR) (GJ yr -1 m-2) Cramer, et al., 1995, Net primary productivity model inter-comparison activity, IGBP/GAIM report series 5
Incident direst & diffuse PAR from MODIS (Liang, S. T. Zheng, H. Fang, R. Liu, S. Running, and S. Tsay, 2006. Mapping incident Photosynthetically ActiveRadiation (PAR) from MODIS Data. J. Geophys. Res. Atmosphere, 111, D15208)
The increasing availability of spatial data and the growing interest in quantifying terrestrial carbon flux have driven rapid progress in the integration of modeling and remote sensing Most studies use remote sensing data to drive models Few studies integrate terrestrial remote sensing and carbon models through data assimilation at the regional scales
Many C studies have used remote sensing products as inputs (reflectance/radiance or Vegetation indices) Inversion model (e.g., LAI) Carbon process model
Few studies are assimilating terrestrial remotely sensed products into carbon models spatially From http://www.geog.ucl.ac.uk/~mdisney/ctcd/misc/www_new/science/eo/eo_overview.html
Two data assimilation schemes Remote sensing data Remote sensing data Estimation Algorithms Minimization LAI Minimization Time Carbon Model Adjust model parameters Adjust model parameters Observation operator: Radiative transfer models Carbon model method (A) method (B)
Surface energy balance estimation Qin, J., S. Liang, R. Liu, H. Zhang, and B. Hu, (2007), A Weak-Constraint Based Data Assimilation Scheme for Estimating Surface Turbulent Fluxes, IEEE Geoscience and Remote Sensing Letters, in press. The objective is to assimilate land surface temperature (LST) from remote sensing to a land surface model to estimate latent heat and sensible heat. surface temperature ( K ) sensible heat ( wm -2 ) sensible heat ( wm -2 ) 320 310 300 290 280 270 600 500 400 300 200 100 0-100 600 500 400 300 200 100 0-100 235 235 240 240 245 Julian day from 231 to 260 in 1999 245 Julian day from 231 to 260 in 1999 250 250 255 255 measured estimated measured estimated measured estimated (a) 260 (b) 260 (c) 235 240 245 Julian day from 231 to 260 in 1999 250 255 260
Coupled model Canopy radiative transfer model: Kuusk Markov reflectance model Leaf optical model: PROSPECT DSSAT crop growth model produces LAI, leaf nutrition ->leaf chrolophyll -> leaf optics Fang, H., S. Liang, G. Hoogenboom, J. Teasdale, and M. Cavigelli, (2007), Crop yield estimation through assimilation of remotely sensed data into DSSAT-CERES, International Journal of Remote Sensing, DOI: 10.1080/01431160701408386
Fig. 3. (a) A sample time series of MODIS EVI data and the estimated phenological transition dates for a corn pixel in Indiana, 2000. (b) The LAI products
Fig.4. Comparison of the simulated corn yield and USDA NASS yield data for selected counties (43) in Indiana, 2000. (a) S1 (LAI); (b) S2 (EVI); (c) S3 (NDVI); (d) S4 (EVI+LAI); (e) S5 (NDVI+LAI).
Data Issues Since the number of unknowns >> number of observations, remote sensing inversion is always an ill-posed problem. Although extensive validation efforts have been made, product uncertainties have not well characterized. It is difficult to define the error matrix in the cost function Validation of MODIS leaf area index (LAI) collection-4 product
Data Issues Remotely sensed products are not continuous in space and time due to cloud contamination and other factors (A) LAI distribution around August 12, 2000: MODIS collection4 product (A) and processed product (B) (B) Fang, H., S., Liang, J. Townshend, R. Dickinson, (2007), Spatially and temporally continuous LAI data sets based on an new filtering method: Examples from North America, Remote Sensing of Environment, in press.
Data Issues Multiple satellite sensors are producing the same product with variable accuracies, spatial and temporal characteristics The instrument teams continuously update the versions of their products It makes the user difficult to select
Fraction of the absorbed PAR by the green vegetation (FPAR) products
Other data issues Data products may not have the consistent spatial and temporal resolutions, upscaling and downscaling often cause problems; The generated remotely sensed products may not be consistent with what the model defines Remotely sensed skin temperature is an average measure of the mixed pixel, while many models define leaf/soil sunlit/shadow temperatures Some models calculate NDVI above the canopy, but many remotely sensed NDVI products are calculated from the top of the atmosphere The volume and format (e.g., HDF, map projections) of the products may be very challenging to many users
Final remarks NDVI is not everything, and remotes ensing can produce much more than just NDVI Interactions between remote sensing scientists and carbon modelers are critical Joint activities to bridge the disciplines, for example, joint workshop, training classes, joint textbooks, joint research projects
Thank you!