VEGETATION PREPARATORY PROGRAMME STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final report 15 July 1998 Convention N° 95/CNES/0427 Principal Investigator Gérard DEDIEU (CESBIO) Tel : (+33) 5.61.55.85.26 Fax : (+33) 5.61.55.85.00 Gerard.Dedieu@cesbio.cnes.fr Co-Investigators Jean-Claude GERARD (Université de Liège-LPAP) Dean GRAETZ (CSIRO-Earth Observation Centre) Participants A. CHEBOUNI (ORSTOM) A. FISCHER, W. CRAMER (PIK) L. FRANCOIS, C. PICCHI (LPAP) B. BERTHELOT, P. CAYROL, P.MAISONGRANDE, S. MOULIN (CESBIO) L. KERGOAT (LET) A. RUIMY, B. SAUGIER (LEV) STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Summary During the pre-launch phase of the project, we addressed the general topic of the use of VEGETATION measurements to improve the modeling of biosphere functioning. At the regional scales, the aims are the assessment of biomass production and possibly water balance. The aim of global scale studies is related to global change and global carbon cycle issues. We explored, and present in this report, three different directions for coupling biosphere models and remotely sensed measurements, namely : Test of vegetation model results : model outputs are compared to satellite observations. We present examples where the analysis of discrepancies provided a efficient guide to improve process modeling or model parameters. Driving of vegetation model : the TURC diagnostic model uses NDVI as input to determine absorption of solar radiation from which net primary productivity is estimated. Satellite NDVI put strong constraints on the model. By now, this approach is probably the more robust to estimate vegetation production. We show with some examples that the capabilities of this approach to detect year to year fluctuations of NPP are quite promissing. However, long term analysis of NPP is limited by the quality and consistency of the measurements provided by current satellite sensors. Assimilation techniques : this approach is based on optimization techniques that compare the distance between model outputs and satellite observations. It allows a fine tuning of model parameters, even if these parameters cannot be observed directly from space. Results we obtained with shortwave measurements are quite promising and open the path to regional assessment of vegetation functioning. The benefit of thermal infrared measurements is more difficult to assess and requires further work. In addition, we prepared satellite datasets to serve the project, including the reprocessing of five years of NOAA Global Vegetation Index dataset. The resulting product (LASUR) has been distributed to international teams for evaluation purpose and for use. The post-launch phase will mainly consist in the evaluation of the benefit gained from the use of VEGETATION data for vegetation modeling. We will continue to explore the three directions explained above, with a special emphasis on the development of assimilation techniques in the framework of the SALSA and EUROSIBERIAN CARBONFLUX experiments. July 1998 2 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Table of contents 1. OBJECTIVES OF THE INVESTIGATION ....................................................................................... 4 1.1 1.2 1.3 2. BACKGROUND ............................................................................................................................. 4 APPROACH.................................................................................................................................. 5 EXPECTED RESULTS .................................................................................................................... 6 DATA SIMULATION ........................................................................................................................ 7 2.1 2.2 3. EXISTING DATA SETS ................................................................................................................... 7 REPROCESSING OF NOAA/GVI : THE LASUR PRODUCT ............................................................... 8 BUILDING OF DATA SETS ........................................................................................................... 12 3.1 3.2 HAPEX-SAHEL ........................................................................................................................... 12 AUSTRALIA ................................................................................................................................ 12 4. USE OF SATELLITE DATA TO DRIVE TERRESTRIAL BIOSPHERE MODELS ........................ 15 5. USE OF SATELLITE DATA TO TEST PROCESS MODELS ....................................................... 21 5.1 5.2 6. CESBIO MODEL........................................................................................................................ 21 LPAP MODEL (CARAIB) ........................................................................................................... 26 ASSIMILATION OF REMOTELY SENSED MEASUREMENTS ................................................... 29 6.1 INTRODUCTION .......................................................................................................................... 29 6.2 MATERIAL AND METHOD ............................................................................................................. 30 6.3 RESULTS .................................................................................................................................. 30 6.3.1 Assimilation of shortwave data ........................................................................................ 30 6.3.2 Assimilation of longwave data .......................................................................................... 33 7. IMPACT OF DOUBLE CO2 ON CLIMATE AND VEGETATION ................................................... 36 7.1 7.2 8. RESULTS : LAI SIMULATION........................................................................................................ 37 RESULTS : EVAPOTRANSPIRATION SIMULATION ........................................................................... 40 WORKPLAN FOR THE POST-LAUNCH PHASE ......................................................................... 42 8.1 8.2 8.3 8.4 8.5 9. SATELLITE DATA TO IMPROVE VEGETATION PROCESS MODELS ..................................................... 42 ASSESSMENT OF CARBON BUDGET AT THE REGIONAL SCALE ....................................................... 44 ADAPTATION OF VEGETATION TO CLIMATE .................................................................................. 44 GLOBAL CARBON CYCLE AND CHANGE MONITORING .................................................................... 44 PRODUCT REQUEST ................................................................................................................... 45 CONCLUSION AND PERSPECTIVES .......................................................................................... 46 10. 10.1 10.2 REFERENCES ........................................................................................................................... 48 QUOTED PUBLICATIONS ............................................................................................................. 48 PUBLISHED WORKS THAT USE LASUR DATASET ......................................................................... 51 July 1998 3 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 1. OBJECTIVES OF THE INVESTIGATION Our general objective is to develop process models of vegetation functioning that can be used at regional and global scale for predicting carbon and water exchanges between land surface and the atmosphere. In this text, "vegetation functioning" refers to the temporal development of vegetation characterized by carbon uptake, phenology, carbon pools, biomass and its allocation within the various plant components (stems, roots, leaves, etc.). In a first step, we will focus on the assessment of fluxes at the seasonal time scale. Process models (TBMs: Terrestrial Biogeochemical Models) attempt to describe in detail most of biological processes but they require input parameters that are generally available for only a few ecosystems. The spatial patterns of these processes is not always well described within all ecosystems. The specific objective of this project is to develop and validate satellite-based methods for calibrating such spatially heterogeneous vegetation process models. We will first describe the work carried out to build the data sets that are needed to run TBMs. Then, we will present the use of satellite data to drive (section 4) and validate (section 5) the models. Assimilation techniques have been tested with both shortwaves and longwaves measurements and preliminary results are described in section 6. Finally, section presents our workplan for the postlaunch phase . 1.1 Background A number of statistical, parametric or process (mechanistic) vegetation models have been developed in view to assess crop yield, Net Primary Productivity (NPP) or energy and mass exchanges with the atmosphere. Each of the existing vegetation models has its advantages and drawbacks, but they all need to be further validated. Several compartments, or sub-models, like water balance and phenology, may be improved, as may be the time scale, space resolution and coupling with physical processes like energy and mass exchanges. A number of issues and assumptions have to be further investigated: i) most of the existing vegetation models are calibrated using a limited set of ground based measurements and then spatially extrapolated to similar ecosystems (e.g. Raich et al., 1991, Ruimy et al., 1994). Remotely sensed data could help us to validate models and to retrieve some of the parameters over large areas. ii) our interest is in prognostic models that could be used to predict vegetation functioning under various scenarios of climate change, or coupled to atmospheric circulation models. This means that we must assess the use of remotely sensed data for model validation or parameter estimation, not only for retrieving state variables such as Leaf Area Index. iii) Most of the models dealing with regional and global scales assume that every model grid-cell, generally corresponding to several hundreds of square kilometers, is an homogeneous medium. This July 1998 4 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report concept has to be checked and possibly improved, for example by investigating sub-grid processes and upscaling methods. These issues must be addressed in the framework of long term multidisciplinary projects, such as the ones developed under the auspice of IGBP or supported by the Environment and Climate Programme of the European Commission. We are contributing to several of these programmes (see the proposal for details). In the frame of the VEGETATION Preparatory Program, our objective is more a limited feasibility study i) to investigate the potential of VEGETATION and HRVIR sensors in solving some of the above mentionned issues and develop the required methods, and ii) to prepare the integration of VEGETATION data within terrestrial biosphere and carbon cycle models. 1.2 Approach Three different types of TBMs aimed at estimating Net Primary Productivity are being developed by research groups at CESBIO, LPAP and PIK. The first type is a diagnostic and parametric model, based on the Monteith's model (1977), which relies on satellite data inputs (Ruimy et al. 1994, Maisongrande et al. 1995, Ruimy et al., 1996-a). The second type of models are prognostic and process models which may use satellite data for initialization or validation but can be run alone (Kergoat et al. 1995-a & b, Warnant et al., 1994). These models operate with short time resolution (typically 1 day), space resolution of 0.5°x0.5 or 1°x1°, and are integrated over periods ranging from a few days to several years. Within the frame of this project, our objectives are to investigate the potential of VEGETATION measurements for contribution to these modelling activities. In order to improve soil moisture estimates, we also plan to supplement optical measurements by thermal infrared data that we will be used to constraint a surface flux model. We will address the following general topics : development, assessment, improvement of assimilation techniques of SPOT-4 HRVIR and VEGETATION measurements (actual or simulated) in vegetation models, including SWIR band ; most of vegetation processes such as phenology, growth, photosynthesis strongly depend on water availability. We plan to couple vegetation process models with a SVAT model in order: to improve soil moisture estimates to supplement satellite measurements in the shortwaves domain (HRVIR, VEGETATION) by thermal infrared data to add more constraints on the models through multispectral assimilation investigation of the use of combined high and low space resolution to account for sub-grid surface (vegetation, soil) heterogeneity assessment and improvement of "ecological rules", combined with assimilation techniques to further constraint the system or provide parameters which cannot be derived from satellite measurements. By "ecological rules", we mean a-priori hypothesis on the strategies that vegetation has developed during evolution to survive a given environment. July 1998 5 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 1.3 Expected results The investigation should provide a detailed assessment of the feasibility of constraining vegetation process model by simulated and then actual VEGETATION measurements in the shortwaves. Coupling vegetation and SVAT models would allow to investigate the use of SWIR band and the interest of thermal infrared measurements. This evaluation could be of interest for a possible evolution of the VEGETATION mission specification or of other sensors. It should provide a simulation tool for the specification of mission parameters such as the time and frequency of acquisition. The results of this investigation should also contribute to the improvement of global vegetation models used for global carbon cycle studies, by providing better initialization or validation. July 1998 6 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 2. DATA SIMULATION We are mainly using NOAA/AVHRR and SPOT-HRV data. The NOAA/AVHRR archive we use consists of four different data sets. 2.1 Existing data sets AVHRR measurements other Europe are routinely processed by the Institute of Remote Sensing Application (IRSA) of the Joint Research Center of Ispra, in the framework of the MARS project. IRSA kindly supplied us with daily, 1km, data acquired over France from 1991 to 1994. These data are geometrically and radiometrically corrected. Atmospheric corrections are based on a parameterization of the 5S code. Inputs are water vapor and ozone monthly climatologies. No correction is done for aerosols. We are also using AVHRR measurements acquired on a 5x5 degrees area centered near Niamey (Niger) where the Hapex Sahel took place in 1992. These data have been processed by the Hapex Sahel Information System (HSIS), and consist of daily, 1km, AVHRR data for 1991 and 1992 (from may to october). Both Top of the Atmosphere and atmospherically corrected data, using the SMAC code, are available. Detailed information are available on http://www.orstom.fr/hapex. The following AVHRR data sets have been acquired by Dean Graetz at CSIRO over Australia (continental coverage) : raw GAC (afternoon pass only) from 1981 - 1992 - thereafter HRPT until present; yet to be processed to a precision product. NASA processed GAC NDVI (ex Jim Tucker), 1981-1992 NASA Pathfinder GAC (8 km), daily datasets for 1982-1994 NASA Pathfinder GAC (8 km), 10 day composites for 1982-1994 NASA Pathfinder GAC (8 km), monthly composites for 1982-1994 In addition, GMS Pathfinder data have been acquired for the period July 1994 - December 1995. Unfortunately, the use of cuurent NOAA/DAAC pathfinder data for quantitative analysis is limited. A limited number of SPOT scenes have been acquired over the Hapex-Sahel site and geometrically and radiometrically corrected by LPAP in collaboration with VITO (F. Veroustraete). We plan to acquire new SPOT data in 1998 over the SALSA area. July 1998 7 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 2.2 Reprocessing of NOAA/GVI : the LASUR product Satellite measurements offer the unique opportunity to monitor vegetation at the global scale, but complex processing is required for a reliable analysis of interannual variability and use within assimilation procedures. In figure 1, zonal variation of Net Primary Productivity (NPP) is plotted for two years, 1989 and 1991. NPP is derived from a model (see section 4.) driven by NDVI computed from NOAA/AVHRR measurements (Maisongrande et al., 1995). The lower NPP of the southern hemisphere in 1991 is due to the eruption of the Pinatubo volcano which injected large amount of aerosols in the stratosphere in june 1991. Aerosol scattering and absorption lead to a decrease of NDVI, and therefore a decrease of NPP estimates. In other words, the variation of NPP between 1989 and 1991 as seen in figure 1 is mainly due to atmospheric effects on satellite measurements, and not to a change in actual productivity. Figure 2 presents time evolution of NPP as a function of latitude from 1986 to 1991. NPPs have been averaged weekly over zonal belt of 1°. Seasonal and geographical variations of NPP are clearly depicted, but a more quantitative assessment is difficult. For example, the change of satellite in november 1988 is visible, even if sensor calibration drift was taken into acount. This is due to the orbital drift of NOAA-9 which lead to later and later time of overpass, and consequently to an evolution of surface directional effects. When NOAA-11 was launched, time of overpass was again the nominal one, i.e. around 13 :30, and surface directionnal effects differed from the ones observed by NOAA-9. Another feature we already mentionned is the decrease of NPP in 1991 in the southern hemisphere. Figure 1 : zonal variation of Net Primary Productivity (NPP) for 1989 and 1991. NPP is derived from a model driven by NDVI computed from NOAA/AVHRR measurements (Maisongrande et al., 1995). July 1998 8 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 2 : time evolution of Net Primary Productivity (NPP) as a function of latitude, from 1986 to 1991. NPPs have been averaged weekly over zonal belt of 1°. (Maisongrande et al., 1995) Figures 1 and 2 illustrate the need for accurate datasets. Therfore, we have reprocessed NOAA/AVHRR dataset in order i) to test calibration, filtering and atmospheric corrections parameters and methods at the global scale, and ii) to obtain improved global data sets that can be used by global carbon cycle modelers and other applications. The Global Vegetation Index product distributed by NOAA (Tarpley, 1984) we have acquired consists of weekly global data at a 1/7x1/7 degrees resolution in Plate Carree projection. Seven channels are available, namely raw digital counts of channel 1 and 2 (visible and near-infrared), calibrated thermal infrared channels (Ch4 and 5), NDVI computed from raw counts, solar zenith angle and scan angle. This data set suffers from some weaknesses analysed by Goward (1993) In a first step, we reprocessed (Berthelot et al., 1997) two years of weekly data (1989 and 1990) that correspond to the first two years in orbit of NOAA-11. Two CD-ROMs have been produced, that include top of atmosphere and « surface » (atmospherically corrected) reflectances, temperatures and NDVI, viewing and sun angles, plus a channel (flag channel) that indicates the data that should be discarded and the reasons to discard (clouds, inconsistent angle, low solar elevation, ...). The product is called LASUR (LAnd SUrface Reflectance) and is available at both 1/7x1/7 and 1x1 degrees resolution. LASUR was designed to fit the needs of both remote sensing experts and scientific « end-users » who do not wish to apply corrections by themselves. A detailed description of the processing can be found in the documentation files distributed with the CD-ROMs. Shortly, filtering procedure is intended to identified clouds, clouds shadows, large aerosol optical depths, and abnormal data. It is based on improvements of the so-called Bise (Viovy et al., July 1998 9 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 1992) and INTUITIV methods (Loudjani et al., 1994). Atmospheric corrections are performed using the SMAC code (Rahman and Dedieu, 1994). Water vapor and ozone contents are derived from climatologies (Oort, 1983, and TOVS retrieval provided by CNRM). Aerosol optical depth is assumed to increase from the poles to the equator. The atmospheric corrections we applied is therefore rather close to the procedure used for actual VEGETATION measurements. The main steps of the reprocessing and example of results are given in figures 3-a and 3-b. The 1989-1990 LASUR dataset have been distributed to about 20 teams for evaluation purposes. The result of this evaluation was positive. LASUR results were also compared (figure 4) to FASIR product (Sellers et al, 1994). Following this evaluation, further years have been processed and the LASUR product is now available for the period 1986-1990. Information about LASUR, documentation file, and 0.5x0.5 degree resolution dataset are available on the web at the following address : http://www-sv.cict.fr/cesbio/lasur/. Figure 3-a : main steps of the LASUR reprocessing of GVI, illustrated by NDVI maps obtained for July 1990 (weekly composite). Figure 3-b : example of surface filtered NDVI for january, april, july and october 1990 produced by the LASUR reprocessing of GVI ((weekly composites). July 1998 10 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 4 : LASUR versus FASIR maximum annual NDVI. 13110 pixels. Correlation coefficient = 0.89. NDVI(LASUR) = 0.99 NDVI(FASIR) + 0.113. Lafont, 1998. See also Ouaidrari et al, 1997. July 1998 11 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 3. BUILDING OF DATA SETS The development of assimilation and upscaling techniques needs TBMs to be run at a finer resolution (ideally 1km or even less) than usually done for global carbon cycle study (50 to 100km), and then to integrate estimated flux at the scale of global model grid-cells. Consequently a significant amount of work has been devoted to the collection and processing (remapping, interpolation of weather fields, ...) of the data sets used as input to TBMs. 3.1 Hapex-Sahel LPAP has prepared several data sets for the Hapex-Sahel square degree area. Both HSIS data base and external information have been used, with the support of CESBIO for AVHRR data. Input variables of the water balance model IBM (Improved Bucket Model, Hubert et al., submitted) have been estimated at a 0.5x0.5 degrees resolution. IBM is coupled to the CARAIB model developed by LPAP (Warnant et al., 1994). Input parameters and variables include precipitation, air temperature, net radiation, rooting depth. Air humidity has been derived from radiosoundings performed every 6 hours. Interpolation at 0.5° relies on the use of ECMWF AGCM analysis and EPSAT-Niger data. NOAA/AVHRR data available in the HSIS data base are daily data that need further processing for eliminating clouds and other perturbing effects. The BISE procedure (Viovy et al., 1992) has been using to generate « clean » data every ten days. Best results are obtained with a sliding window of 30 days, especially during the rainy season. Exemple of results are given in figures 5 and 6. This AVHRR data set has been compared to NOAA Pathfinder data : similar time profiles are observed, but NDVI levels are different. 3.2 Australia The following digital datasets for landcover characterisation have been prepared by Dean Graetz at CSIRO at 5 x 5 km spatial resolution for the entire Australian continent : SOIL soil type (descriptive) density (kgs/m3) estimated max soil depth (cms) maximum profile water holding capacity (mms) minimum saturated hydraulic conductivity texture classes nutrient status rank July 1998 12 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report VEGETATION Type (descriptive; 31 classes) Overstorey LAI Overstorey projected foliage cover (pfc) Overstorey biomass (t/ha) Understorey LAI Understorey biomass (t/ha) Understorey projected foliage cover Biomass carbon density (t/ha) Estimated soil carbon density (t/ha) Estimated NPP (tC/ha/year) Estimated biomass carbon turnover time (years) Surface roughness (z0) Albedo (* month) Surface resistance (* month) Figure 5 : Time profile of atmospherically corrected NDVI derived from AVHRR daily measurements, and of NDVI values resulting from the BISE procedure. Data for a test site located at 13.50°N, 2.70°, in 1992, 1 km resolution. Figure 6 : Time profile of filtered surface NDVI for three different windows: 1 pixel (1km), 3x3 pixels and 5*5 pixels. Data for a test site centered on 13.50°N, 2.70°, in 1992. WEATHER 30 year monthly means for the following variables interpolated to 0.05, 0.01, and 0.005 ° : max temp min temp rainfall solar radiation with rainfall solar radiation July 1998 13 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report evaporation dew point 0900 and 1500 dry bulb 0900 and 1500 wet bulb 0900 and 1500 raindays wind run wind speed 0900 and 1500 PIK (A. Fischer and W. Cramer) in collaboration with CSIRO (D. Graetz) has prepared 10 days average of continental temperature (minimum and maximum) and precipitation fields from the interpolation of daily measurements performed by 645 australian stations. Interpolation procedure uses the digital elevation model (0.025° resolution) provided by the Australian National University. By now, temperatures have been processed for the period 1981-1990, and precipitation for 1981-1989, with a 8 km resolution. The processing of further years is in progress. Australian teams have identified on two large (>100 km2) and inert (rather than truly invariant) sites that can be used to check radiometric stability and consistency of satellite data sets. The first is Lake Argyle (128° 45'E; 16° 15'S): a dark (reflectance ~ 10%) target located in the tropical north under high and seasonal atmospheric loadings of water vapour and aerosols from biomass burning. The second is Strzleckie (139° 50'E; 29° 00'S); a bright (reflectance ~ 40%) white sand desert dune field under low and variable aerosol and water vapour loadings. CSIRO has instrumented both sites with Cimel sun photometers in conjunction with continuous or episodic surface reflectance measurements in the VIS and NIR - including the VMI blue band. July 1998 14 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 4. USE OF SATELLITE DATA TO DRIVE TERRESTRIAL BIOSPHERE MODELS The diagnostic global NPP model previously developed by Ruimy et al. (1994) has been improved by computing NPP as the difference between carbon uptake by the vegetation (or gross primary productivity GPP) and carbon released by autotrophic respiration Ra. Only GPP is dependent on solar radiation, while Ra has a maintenance component that depends on biomass and temperature, and a growth component that depends on C availability for growth. Thus, NPP is computed according to : NPP = e' f c S – Ra (1) which is the basis of the TURC (Terrestrial Uptake and Release of Carbon) model (see Ruimy et al., 1996, for a detailed description of the model). In this equation, e' is theconversion efficiency for GPP, S is the incoming solar radiation (300-4000 nm), c is the fraction of this radiation that is available for photosynthesis (cS is called PAR or photosynthetically active radiation), f is the fraction of incoming PAR that is absorbed by vegetation, derived from NDVI data. To get estimates of e' (conversion efficiency for GPP), Ruimy et al. (1995) compiled over one hundred relationships between CO2 flux measured above plant canopies (from micrometeorological or chamber methods) and incident solar radiation S. To compute Ra, a large number of respiration measurements (expressed per unit biomass) have been used with a separation into various plant organs : leaves, fine roots, wood. Thus average maintenance respiration rates have been computed at 20° and found to be 10.7 mgC gC-1 day-1 for leaves, 6.6 for fine roots and 0.5 for sapwood (same units). Sapwood is the living part of wood, and has to be estimated. In our case, we started from a map of vegetation biomass derived by Olson, derived leaf biomass from NDVI data, took fine root biomass equal to leaf biomass and calculated wood biomass as the remainder. Sapwood was then derived from total wood biomass using regression lines established on a few forest stands. The temperature dependence of maintenance respiration was taken as a linear one, rather than as an exponential for there is ample evidence that the Q10 values - often used in respiration - decreases markedly with increasing temperature. Finally, available carbon for growth was computed as the difference between GPP and total maintenance respiration, and growth respiration was taken as a constant fraction (0.28) of this available carbon. Putting all this information into eq. (1) leads to a GPP of 125 Gt C/yr, an autotrophic respiration of 74 GtC/yr (72% maintenance, 28% growth resp.) and thus a NPP of 51 GtC/yr (Fig. 7). It is important to stress that in the TURC model there is no calibration, i.e. all parameters have been derived from experimental data and no subsequent adjustment has been done. July 1998 15 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 2500 90 70 2000 50 30 1500 10 -10 1000 -30 -50 -70 500 GPP 0 -90 -180 -140 -100 -60 -20 20 60 100 140 180 90 1500 70 1250 50 30 1000 10 750 -10 -30 500 -50 -70 250 Ra -90 0 -180 -140 -100 -60 -20 20 60 100 140 90 180 1250 70 1000 50 30 750 10 -10 500 -30 -50 -70 250 NPP -90 -180 -140 -100 -60 -20 20 60 100 140 0 180 g(C) m-2 yr-1 Figure 7 : Maps of continental Gross Primary Productivity (GPP), leaf and sapwood maintenance respiration (autotrophic respiration, Ra), and Net Primary Productivity (NPP), as computed by TURC for 1986 (Ruimy et al., 1996) July 1998 16 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Improvements to the TURC model may come from better estimates of solar radiation, and of f (using better algorithms to derive it from satellite measurements). Sensitivity studies concerning the input data on solar radiation and NDVI as well as the slope of the fNDVI relationship have been performed (Ruimy et al., 1996). The global and zonal values for predicted incoming solar radiation are pretty close for the data sets based on GCM simulations, cloud cover climatologies and remote sensing (Fig. 8-a). The greatest deviation (+14% for the global average) from the bulk of the other data sets was found for solar irradiation predicted by the ISCCP cloud cover project. The quantitave effects of differences in NDVI caused by different data processing are rather important. The CESBIO product (an earlier version of LASUR) and the FASIR-NDVI have been compared (Fig. 8b). Annual global GPP was found to be 46 % higher for simulations based on FASIR-NDVI than for those based on the CESBIO product. These results are in accordance with a similar sensitivity study performed for an other diagnostic model of biospheric carbon exchange fluxes, CASA (Malmström et al., 1995). In addition to the type of NDVI processing the slope of the f-NDVI relationship contributes substantially to incertitudes of simulated GPP and NPP. The simulated carbon exchange fluxes show a relative sensitivity of about 1 with respect to this parameter. For an increase of 10 percent in the slope of the fNDVI relationship GPP increased by 10.7 %, autotrophe respiration by 10.0% and NPP by 11.2 %. Figure 8 : a) Incident PAR integrated over 10 degree latitude zones and one year. PIK: derived from cloud observations (Leemans and Cramer, 1991), ISCCP: derived from satellite observations (Bishop and Rossow, 1991), ISLSCP: derived from satellite observations (Meeson et al, 1995), CNRM: derived from GCM outputs (Planton et al., 1991) b) Fraction of PAR absorbed, f, integrated over 10 degree latitude zones and one year, as given by FASIR-NDVI product for 1987 (Sellers et al 1995) and derived from CESBIO-NDVI for 1986 (Berthelot et al. 1994). July 1998 17 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report The TURC model and the LASUR dataset have been used to estimate weekly global NPP from 1986 to 1990. Newly available global solar radiation dataset produced by the SRB-GEWEX project from meteorological satellite data were used instead of AGCM analysis. These SRB products agree rather well with AGCM estimates, excepted in convective regions, such as the tropics, where SRB dataset is expected to be more reliable. In order to drive autotrophic respiration, we used air temperature at 2m generated by ECMWF reanalysis activity. Compared to previous runs of the model, input datasets of NDVI, temperature and solar radiation have been improved and correspond to each year considered. Figure 9 is an example of the results we obtained. This figure maps over Africa the difference of NPP between 1990 and 1989, with surimposed isolines of precipitation anomalies derived from the 2.5x2.5° resolution dataset established by Huffman et al (1995). Differences of annual NPP in 1989 and 1990 are well correlated with precipation fluctuations over Sahel, Namibia/Angola and Botswana. The TURC NPP model does not use precipitation, and these correlations result from the sensitivity of carefully processed NDVI to the impact of drought situations. However, NPP over Tanzania is larger in 1990 than in 1989 despite a 30% decrease of precipitations. Further analysis is needed to explain this phenomenon which could result from different temporal distribution of rainfall during the year. The recent El-Niño 1997-1998 event drew attention on its impact on weather and biosphere. It is interesting to note that even if 1989 and 1990 are not El-Niño years, fluctuations of NPP are very significant over large areas. Figure 9 : Differences of yearly Net Primary Productivity between 1990 and 1989, with surimposed isolines of precipitations anomalies (e.g. +30 means that precipitations were 30% higher in 1990 compared to 1989). Resolution is 1x1°. Precipitations were interpolated from the 2.5x2.5° resolution dataset established by Huffman et al (1995). NPPs were computed using the TURC model (Ruimy et col., 1996) driven by the LASUR NOAA/AVHRR product and SRB-Gewex solar radiation. Precipitation data : Huffman et al (1995). July 1998 18 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report The assessment of global weekly NPP results for the period 1986-1990 is in progress (e.g. fig.10 and 11 for results over Africa and Australia). In a first step, NPP estimates and fluctuations are analysed together with weather data. We also started to use FAO agricultural statistics available at country level. Preliminary results are quite promising at country or regional level. Global long term trends of NPP are more difficult to assess. The main limitation is due to the orbital drift of NOAA satellite that induces complex bidirectional effects on the measurements that can hardly be corrected using NOAA/AVHRR data alone. Modeling and normalization of bidirectional effects is being studied by several teams, including colleagues in CESBIO involved in the POLDER programme. Normalization of bidirectional effects will be included in the LASUR processing chain as soon as possible. This issue is crucial to address long term trends of biosphere functioning. Finally, we also plan to use SPOT4-VEGETATION data and to work on specific algorithm to establish consistent long term AVHRR and VEGETATION data archives. Figure 10 : variations of NPP over tropical Africa between 1986 and 1987 (top) and 1989 and 1989 (bottom), respectively. The TURC NPP model was driven by LASUR dataset. July 1998 19 Figure 11 : NPP and its year to year variations over Australia. The TURC NPP model was driven by LASUR dataset. STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 5. USE OF SATELLITE DATA TO TEST PROCESS MODELS 5.1 CESBIO model We have develop a mechanistic and pronostic model (Kergoat, 1998) that is built on the knowledge of basic ecophysiological processes. This model does not require satellite data, even if it can be constrained by information retrieved from such data (e.g. phenology). It is based on the assumption that vegetation adapts to the local climate, leading to a development of the canopy which ensures a large light absorption but prevents severe soil water depletion. Particular attention has been paid to the processes which strengthen or weaken the basic LAI/evapotranspiration relationship, such as : soil moisture feedback on transpiration and evaporation, deciduousness of the canopy and interception losses. The model predicts LAI (fig. 12) and vegetation functioning (e.g. photosynthesis). It is driven by climate variables such as mean monthly temperature and precipitation and land characteristics such as soil water holding capacity and biome distribution. Phenology is currently derived from satellite data (Moulin et al., 1997). Our efforts have focused on the test of the model with available data, and, from the learnings of these tests, on the improvement of the model. We must stress that the model was never calibrated against data, and only rely on parameters and knowledge found in the litterature. Figure 12 : Global estimate of LAI, based on the availability of water (Kergoat, 1998). July 1998 21 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report A first test was to compare modeled annual water balance to stream flow data, since it is one of the rare large scale datasets available for TBM evaluation. Over a large river basin, the long term water balances reduces to : Pr - Etr = DR where DR is the annual long term average stream flow, Pr and Etr basin-averaged precipitation and actual evapotranspiration. Thirty years average rivers discharge were taken from Probst abd Tardy (1987). These data mainly follows UNESCO reports (UNESCO, 1977). The annual evapotranspiration and the basin contours (kindly provided by CGS) are shown in figure 13. For the 28 rivers we considered, the runoff simulated by the LAI model is plotted against the measured runoff (Fig. 14). The overall result is that the large variability of both runoff and evapotranspiration is well reproduced by the model. Two caveats however have to be considered : a) both runoff and precipitation observations can display significant errors and b) on a basin scale, climate alone exerts a strong constraint on the water balance and the surface processes effects are therefore bounded. Nevertheless, the general agreement found on Fig. 15 suggests that using an explicit canopy scheme and general physiology leads to reasonnable results. Figure 13 : Annual evapotranspiration estimated by the vegetation model and contours of river basins used for comparisons presend in figs. YY and YY (Kergoat, 1998). July 1998 22 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 14 : Modeled versus observed basin- Figure 15 : Modeled versus observed basinaveraged runoff, for 28 large rivers averaged evapotranspiration, for 28 large rivers A second test was to compare some of the pronostic variables of the model to satellite data. Model estimate of LAI are used to predict APAR (photosynthetically active radiation which is absorbed by the vegetation). We used satellite data to derive APAR values globally, with a weekly time step. Model and satellite APAR values have been compared. We present here the results for winter and summer months (Fig. 16). The general agreement between the two estimates, and especially over aridity gradients, supports our initial assumption that water constraint explains a great part of LAI variability at the global scale. However, the comparison also reveals that this constraint is not the major or unique limiting factor of canopy development, especially for northern latitudes biomes where light and its impact on carbon budget also plays an important role. The comparison also highlights some model weaknesses for dry tropical forests. The explanation, supported by ground experiments reported in the litterature, lies in an unsufficient rooting depth in the model. A third major discrepancy can be seen over China. This is probably due to the use of potential vegetation (deciduous forest for China) map by the model, while satellite observes the actual land use (mostly crops and grasslands). Following this comparison of model outputs to satellite estimates, we have implemented several improvements in the model (Lafont, 1997). First, we have increased from 1.5 m to 4.5m the rooting depth of forest biomes with suffer significant water stress, namely tropical evergreen forest, dry deciduous forests, savanna woodlands, and mediterranean vegetation. This modification is supported for example by the work of Nepstad (Nepstad et al , 1994) who report a rooting depth larger than 8 meters for evergreen tropical forest located at the border of the amazonian rainforest. July 1998 23 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Second, in order to improve model results for northern latitudes biomes, we have implemented in the model a carbon budget constraint. This mechanism limits LAI by computing the carbon cost of new leaves. Since this cost is assumed proportional to LAI while photosynthesis tends to saturate when LAI increases, an optimum exists that depends on local conditions. Equilibrium LAI is reached when the annual carbon cost of building new leaves is larger than the gain of carbon produced by the photosynthesis of additional leaves. Figures 17 and 18 present LAI maps obtained with the previous and the new version of the model, respectively. The carbon budget criterion reduces LAI of northern areas to more realistic values. For example, LAI decreases from 4 to 1 m 2.m-2 over Siberia. Change of the rooting depth of several biomes lead to an increase of LAI over several regions, such as south-east of Amazonia and southern Africa. APAR MJ/DAY Figure 16 : Comparison of satellite estimate of monthly APAR (left side) to model estimates (right side). Top : winter, Bottom :summer. July 1998 24 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 17 : Simulation of LAI by the previous version of the process model (Kergoat 1998). Figure 18 : Simulation of LAI by the modified version of the process model (Kergoat 1998). Change of parameters and process improvements were based on the analysis of the comparison of model results with satellite data (fig. 16). Lafont, 1997. Again, we used satellite data to assess LAI predicted by the new version of the model. The comparison is based on NDVI given by the FASIR (Sellers et al, 1994, results are not shown here) and LASUR (Berthelot et al ,1997) datasets derived from NOAA/AVHRR measurements. Modeled LAI is used to compute NDVI according to the following simple relationship : NDVI = NDVIinf + (NDVIs – NDVIinf).exp(-kLAI) Where NDVIinf is NDVI for an infinite canopy, determined for each dataset over Amazonia, NDVIs is bare soil NDVI, determined for each dataset over Sahara and australian desert, k is the light attenuation coefficient (k=0.5). Figure 19 : maximum yearly NDVI derived from model LAI. Figure 20 : maximum yearly NDVI derived from LASUR dataset July 1998 25 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figures 19 and 20 present annual maximum of NDVI derived from model LAI and LASUR reflectances, respectively. The agreement is improved, compared to the results obtained with the previous version of the model (figure 16). However, one must keep in mind that the way these two NDVI maps are obtained are rather different. The model is run with long term climate average and assumes potential vegetation, while satellite NDVI corresponds to a specific year (1989) and to actual land cover. Figures 21 and 22 present the scatterplots of the comparison with LASUR dataset. When using the previous version of the vegetation model, the correlation coefficient between predicted and LASUR NDVI is 0.62 (0.63 with FASIR). It increases to 0.76 with the new version of the model (0.79 with FASIR). This significant increase of the correlation coefficient shows the benefit of global satellite data to improve vegetation models. Depending on the model, further analysis could be done, for example to assess the way models predict vegetation phenology or respond to climate fluctuations. Figure 21 : Pixel by pixel comparison of LASUR NDVI and model NDVI derived from discrete LAI, using the previous version of the process model. Figure 22 : Pixel by pixel comparison of LASUR NDVI and model NDVI derived from discrete LAI, using the improved version of the process model. r=0.62, N=13000. r=0.76, N=13000. 5.2 LPAP Model (CARAIB) CARAIB (CARbon Assimilation In the Biosphere, Warnant et al., 1994, Nemry et al., 1996) is a global mechanistic model of the carbon cycle in the terrestrial biosphere which was entirely developed at LPAP. It couples various modules associated with leaf level physiology, canopy light attenuation and July 1998 26 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report CO2 assimilation, wood respiration and soil microbial oxidation of organic matter. It is also partially coupled with the IBM water model. The most important outputs of CARAIB are the net primary productivity (NPP) and the net ecosystem productivity (NEP) calculated with a 1°x1° resolution in latitude-longitude over the whole continental area. However, to achieve a formulation as mechanistic as possible, other fluxes or vegetation characteristics such as leaf and wood respiration, soil heterotrophic respiration (SHR) and leaf area index (LAI) are also calculated. The leaf carbon assimilation sub-model calculates photosynthetic rates using the models proposed by Farquhar et al. (1980) for C3 plants and Collatz et al. (1992) for C4 species. Due to the non-linearity of this system and the large variation of climatic conditions during the day, the leaf carbon assimilation is calculated every 2 hours. At the canopy level, the leaf area index of each vegetation class is estimated from criteria on the monthly average temperature and soil moisture. The distribution of light within the foliage is calculated assuming a spherical distribution of leaves and a quasi-exponential absorption of the light inside the canopy. Wood biomass in forest vegetation is estimated on an annual average basis from the stand age and the NPP fraction allocated to wood growth, using a simple parameterization proposed by Esser (1991). Daily mean wood respiration is taken as proportional to wood biomass, the proportionality factor varying as a Q10 function of the temperature integrated over the diurnal cycle. Soil heterotrophic respiration is estimated using a formulation depending on temperature (Q10 relationship) and soil moisture (Nemry et al., 1996). It is proportional to soil carbon content. At the present stage of development, SHR is assumed to be in equilibrium with NPP on an annual basis. One specificity of CARAIB compared to other global biosphere models is that each 1°x1° degree grid cell is subdivided into several fractions of different vegetation types, instead of considering only the dominant type in each grid element as done in existing global vegetation maps usually adopted by biospheric modellers. The fractional covers are estimated on every grid cell by combining the ecotypes distribution of Wilson and Henderson-Sellers (1985) with land use data at the country level published by the Food and Agricultural Organization (FAO, 1993). In a first step, CARAIB has been used to simulate vegetation phenology and leaf area index over the Hapex-Sahel area, without taking into account the vegetation carbon budget. Inputs data have been described in section 3.1. Since the availability of water is a major limiting factor in Sahel, most of the efforts have been devoted to check and improve the water budget model IBM. The sensitivity of LAI to a parameter that describes water availability relatively to the field capacity has been tested. When the model is run with parameter value used for global scale studies, the length of vegetation cycle of Hapex-Sahel is underestimated when compared to the phenology derived from AVHRR NDVI time series. The threshold that drives water stress has been modified in order to account for the adaptation of sahelian vegetation to dry conditions. In that case, phenology is well reproduced, but the maximum LAI predicted by the model is to high (LAI=4.5) compared to ground measurements available in the HSIS data base (and more qualitatively to NDVI). We are now investigating the reasons of this discrepancy. A possible explanation could be the effect of an inadequate description of vegetation cover (assumed to be 1) and distribution (proportion of grass, bare soils, bush and trees). In addition, the relative contribution of soil and canopy to evapotranspiration has probably to be improved, by a better coupling of carbon and water budgets. In order to achieve a more quantitative use of satellite data to test the model, the SAIL model is being implemented to predict reflectances and NDVI from simulated LAI. July 1998 27 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report NOAA/AVHRR data have also been used to asses model results obtained over the australian NATT 1 transect located in the Northern Territory. The study is based on climate (PIK), satellite (CSIRO) and land use and soil data (CSIRO) available for the period 1980 to 1989, with a 10 days sampling or average. Modeled fPAR have been compared to satellite estimates, using average of 10 days period over ten years. This comparison indicated a too large dynamics of vegetation canopy, which was attributed to a too high value of the so-called water threshold. In the CARAIB model, this threshold defines the level of sustainable water stress. Water threshold was increased by a factor of two in order to get a more realistic simulation of vegetation growth. This change lead to a decrease of productivity by a factor of two. 1 NATT: Northern Australia Tropical Transect July 1998 28 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 6. ASSIMILATION OF REMOTELY SENSED MEASUREMENTS (Note : the unpublished results presented in this section have been obtained by P. Cayrol in the framework of her PhD thesis) 6.1 Introduction Measurements acquired by Earth observing satellites provide global and consistent observations with high temporal and spatial sampling. Meteorological satellites as well as sensors dedicated to ocean topography monitoring, such as TOPEX/POSEIDON, are increasingly used to supply unique information that feed atmospheric and ocean models through assimilation procedures. We believe that satellite data will play in the future a similar role for biosphere studies and land monitoring. However, several reasons such as a wide range of issues and scientific objectives have delayed the development of a consistent combination of models and satellite observations for ecological research and land ressource management. The lack of satellite sensor dedicated to biosphere observation and the difficulty to make direct remote measurements of the variables of interest are two other major reasons for this delay. New sensors aimed at the observation of land surfaces and vegetation have been recently launched (e.g. SPOT-4 VEGETATION and POLDER on ADEOS) or will be available in a near future (e.g. MODIS and MISR onboard EOS, MERIS and ATSR onboard ENVISAT, SEVIRI on METEOSAT second generation). There is therefore a need to develop new techniques that maximize the benefit of these new sensors. The approach developed by CESBIO team is based on the assimilation of satellite data within process models. Assimilation is an inverse method that consists in controlling the time evolution of a vegetation model with radiometric observations, by modifying parameters, state variables or input data. This latter choice depends on the goal and on the variables which suffer the largest uncertainties. Two main reasons justify this approach : satellite measurements over land provide only spectral radiances that are not directly linked to the variables, such as soil moisture or Leaf Area Index (LAI), that drive or characterize vegetation functioning. The assimilation of satellite data within vegetation functioning models is probably the most consistent way to solve this issue. The use of satellite data alone, without coupling with process model, does not allow to build prognostic models and lacks the needed robustness, simply because satellite measurements are not always available, for example because of cloud cover. The question is to evaluate if the information contained in radiometric measurements can be used to determine non-radiometric parameters, such as photosynthates partitioning coefficients. As the objective is the spatial extension, the model must be simple and general enough. July 1998 29 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 6.2 Material and method Our work relies on ground-level, airborne and satellite datasets acquired during the Hapex-Sahel experiment held near Niamey (Niger) in 1992. We used the data available in the Hapex Sahel Information System (HSIS) database. In a first step, the focus was on the so-called perennial fallow vegetation type, that is grassland. The global vegetation process model developed by Kergoat (1998) has been first adapted to simulate the growth and senescence of grasslands in arid conditions. Then we coupled this vegetation process model with the Soil-Vegetation-Amosphere Tranfer Scheme (SVATs) developed by Chehbouni (1992, see also Lo Seen et al., 1997) for use in arid and semi-arid regions. The vegetation model is driven by weather data and predicts vegetation state variables (e.g. LAI, biomass). The SVAT model uses these variables to calculate surface resistance and latent and sensible heat fluxes. The resulting water fluxes allow to update soil moisture and therefore the water available for plant transpiration in the next step of vegetation model computations. Both models operate at a hourly time step in order to account for nonlinear processes, such as diurnal variation of stomatal resistance. Conventional approaches attempt to retrieve some vegetation parameters or state variables (for example leaf area index) by inverting the instantaneous satellite signal without using the knowledge incorporated in a vegetation model and the information of long (several months) time series. In our approach, the vegetation and SVAT models are coupled with radiative transfer models that allow to predict shortwave and longwave radiances measured by ground level radiometers or satellite sensors. We use optimization techniques (so-called data assimilation) to retrieve the set of parameters which minimizes the difference between predicted and observed reflectances over a given time period (e.g. a vegetation seasonal cycle). The main advantage of this technique is to allow the retrieval of model parameters which cannot be observed from space. The aim is to map model parameters that allow the best agreement between predicted and actual satellite observations. This best agreement is expected to lead to better model outputs in terms of, for example, LAI, photosynthesis and productivity. 6.3 Results 6.3.1 Assimilation of shortwave data The approach was first applied to radiances measured at ground level. The results were satisfying. They are not presented here, the focus being on the assimilation of satellite data. In a first step, we used only visible and near-infrared measurements acquired by NOAA-AVHRR over three sites of the Hapex-Sahel experiment. NDVIs were computed from both satellite and model shortwave reflectances for the whole rainy season (may-october 1992). In order to account for land cover heterogeneity, a land cover map derived from SPOT data was used to perform the so-called spectral unmixing of the satellite measurements (figs. 23 and 24). This method, described for example by Puyou et al. (1994) and Faivre and Fischer (1994), allows to retrieve the temporal evolution of reflectances of the main land use types. July 1998 30 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 23 : atmospherically corrected NDVI derived from daily NOAA/AVHRR data. Dotted line corresponds to the whole set of 1 km resolution NDVI. Stars correspond to NDVI for fallow land, obtained after cloud screening and spectral unmixing. Figure 24 : atmospherically corrected NDVI derived from daily NOAA/AVHRR data. Dotted line corresponds to the whole set of 1 km resolution NDVI. Stars correspond to NDVI for millet, obtained after cloud screening and spectral unmixing. Over each site, the assimilation was performed with « unmixed » NDVI derived from 1x1 km 2 data. Our initial plan was to use data assimilation to adjust the carbon allocation coefficient for grassland. This parameter drives the partitioning of photosynthesis products between roots and above-ground organs such as leaves and stems. This parameter is difficult to measure and has a strong impact on LAI development. However, first trials revealed that is was difficult to separate the impact on LAI (and therefore NDVI) of six parameters that have similar impact on growth (Table 1). Therefore, we adjust these parameters simultaneously through a weighting coefficient, . This approach may not be fully satisfactory since it assumes that the various parameters are correlated. More independant information and observations would be needed to adjust these parameters separately. Parameters First guess values Adjusted Values Ground Satellite Shoots respiration yield 0.6 0.56 0.57 Allocation to roots dimensionless 0.56 0.63 0.60 23 27 26 Mortality coefficient day-1 0.01 0.009 0.009 :quantum yield mol(CO2) mol-1(PAR) 0.044 0.049 0.047 0.00369 0.0045 0.0042 SLA:Specific Leaf Area m2 kg-1 (C) Initial storage kg m-2 Table 1 : List and values of parameters adjusted through the assimilation of shortwave measurements (NDVI). These parameters are calibrated simultaneously through a weighting coefficient, . July 1998 31 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report The weighting coefficient, , was adjusted by minimizing the distance between observed (i.e satellite) and predicted (i.e. modeled) NDVI. The minimization was applied to all the clear sky days of the rainy season. Root mean square error and correlation distance criteria have been tested. Table 1 presents first guess and adjusted values of parameters. These values are realistics, and results obtained with ground level and satellite data are very similar. Figure 25 shows model simulation of LAI obtained with first guess and adjusted parameters, respectively. The differences between first guess and adjusted parameters are rather weak, but lead to significant change of modeled LAI (fig.25). This illustrates the sensitivity of LAI modeling to growth parameters. Consequently satellite shortwave measurements, which strongly depend on LAI, provide an efficient way to retrieve growth parameters. Figure 25 : Assimilation of 1x1km satellite NDVI to adjust growth parameters of fallow grassland over the central west site of Hapex-Sahel experiment, Niger, 1992. Model simulation of LAI obtained with first guess (dotted line) and adjusted (solid line) parameters, respectively, and ground measurements of LAI (*). Sensitivity studies we performed showed that surface fluxes are highly sensitive to LAI, at least in arid regions. This is illustrated by figures 26 and 27 which present time profiles of latent heat flux before and after adjustement of growth parameters. Some discrepancies still exist at the end of the rainy season, and could be attributed to a poor description of senescence processes. Also, the accuracy of the ground measurements used to estimate latent heat with the Bowen ratio method decreases when soil moisture is low. However, these results show that improvement of LAI simulation lead to a better estimate of latent heat flux, with r.m.s. errors decreasing from 75 W.m -2 to 60 W.m-2. July 1998 32 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 26 : Model simulation of surface latent heat flux at 14:00 UTC obtained with LAI predicted with first guess (dotted line) and adjusted (solid line) growth parameters, respectively. Figure 27 : Model simulation of daily surface latent heat flux obtained with LAI predicted with first guess (dotted line) and adjusted (solid line) growth parameters, respectively. 6.3.2 Assimilation of longwave data 6.3.2.1 Assimilation of ground level measurements Assimilation of thermal infrared (TIR) data is a rather new topic, and we first explored it with ground measurements of brightness temperature. In a first step, we minimized the distance between the brightness temperatures predicted by the coupled vegetation-SVAT models and the measured temperatures. We did not attempt to perform simultaneous assimilation of short- and longwave measurements. In arid regions, vegetation cover and LAI are strong drivers of surface temperature. Consequently, we used the assimilation of thermal infrared data to adjust vegetation growth parameters. Results (not shown here) were only slightly less satisfactory than the ones obtained with ground level shortwave data. However, the main interest of longwave data should be to complement shortwave measurements. Therefore we used them to adjust the coefficients of the soil resistance parameterization, which plays a key role on surface energy budget but is poorly known : rss = a(wsat-wg)-b where rss is the surface resistance, wsat the saturated volumetric water content (m 3m-3), and wg the volumetric content water at the surface. The initial values of a and b coefficients were 7114 and 1515, respectively. Their adjustement with TIR assimilation lead to new values, respectively 11550 and 2543. Figures 28 and 29 present modeled and observed surface brightness temperatures. In order to be as close as possible to the characteristics of satellite observations, we only considered one measurement per day, at 14 :00 UTC. Assimilation of TIR data has a small but significant impact on temperatures July 1998 33 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report predicted after adjustement of surface resistance. This small impact is probably due to the fact that we used growth parameters determined with shortwave assimilation, i.e the model was already tuned against observations. From this experiment, we learned that the coupled vegetation-SVAT models allow to predict both LAI and brightness temperature time course, a result which was not certain. We also identified a possible difficulty for further development of the approach which results from the high spatial variability of surface temperatures. Temperatures were measured over four different but close targets. Depending on the target, temperatures can differ by 5°C to 10 °C (figure 30). In the experiment we described above this difficulty was successfully solved by using the average of these four temperatures. Figure 28: Comparison to ground measurements of surface brightness temperatures predicted by the coupled models. Period : may and october 1992, 14:00 UTC. Figure 29 : Ground measurements (green solid line) and model simulation of surface brightness temperatures obtained with first guess (black dotted line) and adjusted (red, solid line) surface resistance, respectively. Figure 30 : time profile of brightness temperatures measured over four different but close targets July 1998 34 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 6.3.2.2 Assimilation of satellite measurements We used 1km resolution NOAA/AVHRR thermal infrared data available in the Hapex-Sahel database. These data have been corrected for atmospheric emissivity effects according to the split-window approach of Kerr et al. (1992). Comparisons of modeled and NOAA/AVHRR surface brightness temperature are shown in figures 31 and 32 for two test sites. The model used to predict temperature was run with the growth parameters adjusted with shortwave data (section 6.3.1), and no further adjustement was done. Model and satellite temperatures exhibit rather similar seasonal variations, but differ in their absolute value by several degrees. No definite explanation of this discrepancy has been found yet. We believe that land surface heterogeneity and inaccurate modeling of radiative transfer in the thermal infrared domain are the best candidates to explain this discrepancy. We tried to perform the assimilation of TIR satellite data within the model in order to adjust surface resistance parameters. When the distance used is based on r.m.s.e criterion, the results are not realistic. The correlation distance criteria lead to more realistic adjustement but does not improve significantly model outputs. Figure 31 : time evolution of model and NOAA/AVHRR surface brightness temperatures. Surface air temperature is given for reference. Hapex-Sahel Western site, 1992. Figure 32 : time evolution of model and NOAA/AVHRR surface brightness temperatures. Surface air temperature is given for reference. Hapex-Sahel Eastern site, 1992. As a preliminary conclusion, we have shown that the assimilation of shortwave remotely sensed measurements allows to improve the outputs, such as LAI and evapotranspiration, of a coupled vegetation-SVAT model. In addition, this coupled model predicts realistic surface brightness temperatures at the local scale. However, but the assimilation of TIR satellite data deserves further work, especially to account for surface heterogeneity. July 1998 35 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 7. IMPACT OF DOUBLE CO2 ON CLIMATE AND VEGETATION Forewords : This section addresses the simulation of vegetation functioning in the future. Stricly speacking, this issue is not a part of our VEGETATION investigation. However, we believe it opens this report to scientific questions that partly motivated the work described in the previous sections. Since carbon dioxide is crucial to vegetation functioning, increased CO2 concentration is expected to deeply modify water and carbon exchanges at the Earth surface. A number of CO 2 enrichment experiments have shown a significant decrease of leave stomatal conductance, often of the order of 40% (Field et al. 1995, Drake et al. 1997). Simulteanously, leaf photosynthesis and biomass production increase, a phenomenon called the fertilizing effect of CO 2. It is difficult to extrapolate short term plant experiments to the response of whole ecosystems on the long term. However, it is important to test various assumptions in order to better understand the possible impact of a modified vegetation functioning on water and carbon cycles and energy fluxes. Most of the few published works dealing with this topic (e.g. Sellers et al.,1996) assume that vegetation under a double CO2 climate will have the same structure (e.g. LAI) than at present. However, one may expect that changes in plant physiology will combine with climate change and lead to change in vegetation structure (Betts et al., 1997). We have studied the response of vegetation to a doubling of atmospheric CO 2 concentration. The process model we developed (Kergoat, 1998) was driven by 2xCO 2 climate simulation produced by the Meteo France atmospheric model (ARPEGE, Douville et al., 1998). Leaf area index and water and carbon exchanges at the global scale were predicted. The vegetation process model used to predict LAI and vegetation functioning is the model we improved using satellite data. For every grid-cell, water and carbon balances are computed for a range of LAI. The iteration stops when one of the following conditions limits growth. C1 : Water stress for mean climatic conditions becomes critical (e.g. Woodward 1987). An optimum can be usually found since evapotranspiration increases with LAI . C2 : The annual carbon cost of building new leaves is larger than the gain of carbon produced by the photosynthesis of additional leaves. Since the carbon cost of new leaves is assumed proportional to LAI while photosynthesis tends to saturate when LAI increases, an optimum exists that depends on local conditions. C3 : Lower leaf layers receive an unsufficient amount of solar radiation to get a positive carbon balance. This could be a mechanism explaining limitation of LAI in tropical rainforests. The C2 criterion was added following the comparison of model results to satellite data (section 5.1) The result is the maximum LAI permitted by local ressources. July 1998 36 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report The climate model and double CO2 simulations are described in Douville et al. (1997). Climate model outputs drive vegetation model in off-line mode. We performed four simulations of LAI, namely : « CLIM » is the reference simulation for current climate conditions. Inputs are monthly climatologies, averaged over 30 years (IGBP-GAIM intercomparison dataset). « RAD » is the LAI simulation driven by the double CO2 climate, described in terms of climate anomalies. Only the impact of the physical climate on LAI is considered, that is the impact of variables such as temperature, precipitation, net radiation, and water vapor pressure deficit. « STOM » is the LAI simulation that assumes that leaf stomatal conductance decreases by 34% under double CO2 concentration (Field et al., 1995). Climate forcing is the same as for the « RAD » experiment. « FERT » considers an increase of 25 % of photosynthesis (Drake et al. 1997), in addition to the previous decrease of leaf stomatal conductance. The increase is applied to photosynthesis rate both under saturating light (maximum assimilation) and at low irradiance (quantum yield). Climate forcing is the same as for the « RAD » experiment. For deciduous biomes, the vegetative period is allowed to change, according to temporal shifts or changes in air temperature or precipitations. 7.1 Results : LAI simulation Figure 33 maps the regions where the various limiting factors operate. The carbon budget criterion (C2) limits LAI for high latitude biomes such as tundra and boreal forest. Leaf shadowing (C3) plays the major role to limit LAI of tropical rainforests and of some temperate forests. As expected, the water stress (C1) is dominant for all arid climates, including hot or cold continental climates. Figure 33 : map of dominant limiting factors. Gray : water stress (C1), black : carbon budget (C2), white : leaf shadowing (C3). July 1998 37 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 34 present differences between LAI of the CLIM reference simulation and LAI obtained with RAD, STOM and FERT simulations, repectively. When only the impact of physical climate (RAD simulation) is taken into account (fig.34, top), LAI increases at high latitudes. This increase of LAI under 2xCO 2 climate results of air temperature increase which favors photosynthesis and therefore improves the annual carbon budget and the amount of carbon available for leaf production. LAI slightly decreases over the southern part of Eurasia. This decrease results of the enhancement of potential evapotranspiration or of precipitation anomalies, leading to enhanced water stress. In tropical regions, LAI changes do not exhibit a unique trend. In the STOM experiment (fig. 34, middle), the main effect of reduced stomatal conductance is to compensate in most regions the impact of a warmer climate on water budget. As a consequence, most of LAI decreases obtained in the RAD experiment disappear in the STOM simulation. Combining CO2 fertilizing effect with reduced stomatal conductance (FERT experiment, fig. 34, bottom) leads to a general increase of LAI. This applies to biomes limited by the carbon budget criterion, mainly located in boreal and temperate latitudes, but also to tropical rainforest where the leaf shadowing effect is prevailing. In addition a positive feedback of LAI on photosynthesis exists. Over Alaska for example, the 25% increase of leaf photosynthesis efficiency, combined with reduced stomatal conductance, leads to an 50% increase of LAI and a final 80% increase of canopy photosynthesis. July 1998 38 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Figure 34 : differences between LAI of the CLIM reference simulation and LAI obtained with RAD (top), STOM (middle) and FERT (bottom) simulations, repectively. July 1998 39 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 7.2 Results : Evapotranspiration simulation When only the physical variables are taken into account, a warmer climate (RAD simulation) leads to increased evapotranspiration, ETR (fig. 35). The stomatal closure we assumed in the STOM simulation compensates this increase, and the latitudinal profile of ETR is rather similar to the ETR profile of current climate conditions (CLIM). In the FERT simulation, the increase of LAI due to the fertilization effect of CO 2 corresponds to an increase of evaporative surfaces that counteracts stomatal closure. As a result, ETR of boreal and temperate biomes slightly increases. However, increased LAI does not compensate stomatal closure effect in tropical rainforests. Figure 35 : Impact of climate and physiological processes on land evapotranspiration relative to current climate (CLIM simulation). Impact of physical climate change (RAD minus CLIM), stomatal closure (STOM minus CLIM), and of fertilization plus stomatal closure (FERT minus CLIM). Even if the magnitude of the effects are slightly different, results of the FERT simulation corroborates the ones obtained by Betts et al. (1997) who also considered changes of vegetation structure under 2xCO2 climate. On the contrary, our results differ from the ones presented by Sellers et al. (1996) who did not take into account LAI response to a modified climate. The simulation of LAI under a 2xCO2 climate will be improved to account for possible nitrogen limitations. However no clear effect of 2xCO2 climate on nitrogen cycle has been established yet. The results we obtained are probably in the upper range of vegetation response to climate change when physiological effects are accounted for. On the long term, biome dynamics and human impacts should July 1998 40 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report be taken into account. However, this experiment clearly shows that vegetation functioning has to be better understood to predict the impact of increased CO 2 concentration, since even the sign of water and carbon flux changes depends on vegetation processes that are considered. July 1998 41 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 8. WORKPLAN FOR THE POST-LAUNCH PHASE During the post-launch phase, we plan to use VEGETATION data to address the four topics described below. Some of them will not be completed within the rather short duration of the post-launch phase. Our short-term objective will be to obtain significant results demonstrating the capabilities and benefits of the VEGETATION system. 8.1 Satellite data to improve vegetation process models Our aim is to improve the results of vegetation process models by tuning model parameters against satellite data. Assimilation techniques developed over the Hapex-Sahel area will be tested and further developed over the SALSA experiment area, located at the border between Mexico and the USA (fig.38). Ground, airborne and satellite measurements started in 1997 and will continue at least until the end of 1999. This site will also serve the validation of EOS algorithms. The experimental plan defined by ORSTOM and CESBIO for the mexican part of the site was specifically dedicated to the study of upscaling issues and the development of multispectral assimilation within coupled vegetation and SVAT models. Among the various measurements performed over the SALSA area, the ones of the Large Aperture Scintillometer (LAS) are of great interest for our research. LAS provides estimate of areally averaged sensible heat flux. Figure 36 and 37 compare LAS heat flux to that obtained using an eddy correlation (EC) method over 300 and 900 m pathlengths, respectively (Chehbouni et al., 1998). Over the 300m pathlength LAS and the EC are integrating the same surface, which is no more the case with the 900 m pathlength. Therefore, LAS measurements will be usefull to test results of satellite data assimilation at 1x1km resolution within coupled models. Figure 36 : comparison between sensible heat flux measured by eddy correlation method (Solent) and that measured by the scintillometer (LAS) at a pathlength of 300 m (Chehbouni et al., 1998). r2= 0.90. Figure 37 : comparison between sensible heat flux measured by eddy correlation method (Solent) and that measured by the scintillometer (LAS) at a pathlength of 900 m (Chehbouni et al., 1998). r2= 0.75. July 1998 42 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report We plan to use both VEGETATION and NOAA/AVHRR data, in order i) to compare the respective benefit of the two sensors ii) to evaluate the interest of thermal infrared channels. The expected output of the research is a robust method allowing estimation of vegetation productivity, water balance and surface fluxes. We also plan to test assimilation techniques over the Hapex-Sahel area in order to check the method in different conditions. The Hapex-Sahel is more heterogeneous than the SALSA site, climate and vegetation gradient are different, and atmospheric turbidity is higher. This study will be performed in close collaboration with the participants of the investigation leaded by A. Bégué (CIRAD) and entitled « Application of VEGETATION data to ressource management in arid and semi-arid rangelands ». We also wish to acquire VEGETATION data over France and Northern Spain. These data will serve several ongoing activities and projects linked to the current investigation, such as : Crop growth monitoring and water ressources assessment for crop irrigation in the Ebre watershed (Saragosse area, Spain) monitoring of the phenology of french deciduous forests and modeling of forest functioning. This is a continuation of the work partly presented in the previous report. We also contemplate to submit proposals to the fifth framework programme of the EC. These proposals will address water ressources and vegetation productivity. The VEGETATION datasets we request, especially over France and Spain, will be used for preparatory work and will also serve as historical reference. Figure 38 : approximate location of the northern hemisphere study sites of the post-launch phase July 1998 43 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 8.2 Assessment of carbon budget at the regional scale In the framework of the EUROSIBERIAN CARBONFLUX project, theTURC diagnostic model we have developed will be used to estimate net primary productivity over two sites located in Russia (fig. 38). EUROSIBERIAN CARBONFLUX is coordinated by M. Heimann and supported by the Environnement and Climate Programme of the European Commission. “EUROSIBERIAN CARBONFLUX” is a feasibility study for the development of an observing system for the quantification of regional (1-2000 km) and continental scale CO2 and other long-lived biogeochemical trace gas fluxes over several years. The target region of EUROSIBERIAN-CARBONFLUX is eastern Europe and north-western Siberia. Amongst the various tasks of the project, remote sensing measurements and NPP model driven by VEGETATION data will be used to provide boundary conditions of a hierarchy of nested three-dimensional atmospheric models of varying horizontal and vertical resolution. These models will be used to relate atmospheric observations to surface fluxes on different spatial and temporal scales using forward and inverse modeling techniques. Using these tools, an optimal, cost-effective sampling strategy is to be determined which allows the quantification of regional scale surface fluxes on time scales of days to several years. 8.3 Adaptation of vegetation to climate The process model developed by Kergoat (1998) is based on several assumptions on the way plants adapt to climate. There is a need to test this model in a more detailed manner and to incorporate more processes. For example, we assume that LAI develops in order to maximize the use of ressources such as light and water. However, some studies have shown that a sub-optimal adaptation might be more efficient on the long-term to get over extreme weather fluctuations. We plan to perform this study over the whole Australian continent. Satellite data archive and a unique dataset on vegetation, soil and weather have been collected by australian colleagues. In addition, Australia has a large variety of biomes which periodically suffer from drought induced by El-Niño events. The study corresponds to a step in difficulty compared to our previous work on the model. We expect that VEGETATION data will provide more accurate data than existing sensors. VEGETATION dataset over Australia will also drive the TURC diagnostic model. The environmental conditions and availability of surface datasets we already mentionned will allow a detailled analysis of the results. 8.4 Global carbon cycle and change monitoring We wish to acquire global VEGETATION dataset in order to perform global Net Primary Productivity estimates, as a continuation of the work already performed with NOAA/AVHRR measurements in the framework of the ESCOBA-Biosphere project on global carbon cycle. NPP will be estimated with the TURC diagnostic model. July 1998 44 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report This global dataset will also be used to monitor possible change of vegetation functioning on the long term, i.e. back to the mid 80’s.This requires to make NOAA and VEGETATION archives consistent in terms of geometry, radiometry and spectral signatures. 8.5 Product request Apart from the topics described above, we also plan to work on radiometric correction algorithms, for example atmospheric corrections using the B0 band. Therefore, the products needed for topics 8.1 and 8.2 are daily P product. Over Australia (topic 8.3), we wish to receive S.1 products, plus daily P products over limited areas. For global scale study, our preliminary choice is S10.8 product. July 1998 45 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 9. CONCLUSION AND PERSPECTIVES During the pre-launch phase of the project, we addressed the general topic of the use of VEGETATION measurements to improve the modeling of biosphere functioning. At the regional scales, the aims are the assessment of biomass production and possibly water balance. The aim of global scale studies is related to global change and global carbon cycle issues. We explored, and presented in this report, three different directions for coupling biosphere models and remotely sensed measurements, namely : Test of vegetation model results : model outputs are compared to satellite observations. We presented examples where the analysis of discrepancies provided a efficient guide to improve process modeling or model parameters. Driving of vegetation model : the TURC diagnostic model uses NDVI as input to determine absorption of solar radiation from which net primary productivity is estimated. Satellite NDVI put strong constraints on the model. By now, this approach is probably the more robust to estimate vegetation production. As we shown with some examples, the capabilities of this approach to detect year to year fluctuations of NPP are quite promissing. However, long term analysis of NPP is limited by the quality and consistency of the measurements provided by current satellite sensors. Assimilation techniques : this approach is based on optimization techniques that compare the distance between model outputs and satellite observations. It allows a fine tuning of model parameters, even if these parameters cannot be observed directly from space. Results we obtained with shortwave measurements are quite promising and open the path to regional assessment of vegetation functioning. The benefit of thermal infrared measurements is more difficult to assess and requires further work. As explained in section 8, the post-launch phase will mainly consist in the evaluation of the benefit gained from the use of VEGETATION data for vegetation modeling. We will continue to explore the three directions explained above, with a special emphasis on the development of assimilation techniques in the framework of the SALSA and EUROSIBERIAN CARBONFLUX experiments. We anticipate that VEGETATION data will benefit to the development of terrestrial biosphere model in two different ways : the first benefit we expect relates to the service that will be provided by the VEGETATION system. It is currently difficult to get consistent datasets at 1km resolution over the studied areas (Sahel, France, Australia, ...). One may note that we have had to gather AVHRR datasets coming from various sources, depending of the site and time period. Geometrical and radiometric preprocessings, including compositing and filtering, of the data are not homogeneous, that leads to additional work or difficulties to calibrate models. July 1998 46 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report The second benefit we expect concerns the quality of VEGETATION data which is needed for assimilation procedures and to study long term phenomenon such as the carbon cycle and ecosystem dynamics. Currently, the development of most studies is difficult because of several problems related to AVHRR, mainly orbital drift and poor geometry, expecially for global products. Large AVHRR pixel size at large scan angles is a severe limitation to study small or heterogeneous areas. These issues are difficult to assess quantitatively. However, we can say that, for example, the analysis of global NPP for several years has been limited by the amount of work that would be needed to reprocess AVHRR data at the needed level of quality, assuming this quality can be reached. The third benefit is the availability, on the same platform, of high spatial resolution and high temporal repetitivity sensors. As explained in section 6, both type of data are needed for the assimilation of satellite measurements within vegetation process models. July 1998 47 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 10. REFERENCES 10.1 Quoted publications Berthelot, B., S. Adam, G. Dedieu, P. Maisongrande, L. Kergoat, and F. Cabot, A global dataset of surface reflectances and vegetation indices derived from AVHRR/GVI time series for 1989-1990: the Land SUrface Reflectances (LASUR) data, in 7th International Symposium Physical Measurements and Signature in Remote Sensing, 7-11 avril 1997, edited by G. Guyot, and T. Phulpin, pp. 685-690, A.A. Balkema, Rotterdam, Netherlands, Courchevel, France, 1997. Betts RA Cox PM Lee SL and Woodward FI, 1997, Contrasting physiological and structural vegetation feedbacks in climate change simulations, Nature, 387, 796-799 Chehbouni A., R.L. Scott , D.C. Goodrich, C. Watts, G. Boulet and Y.H. Kerr (1998) Aggregation of convective fluxes over patchy surfaces : application to Salsa data set. Submitted. to Journal of hydrology Chehbouni A., Lhomme J-P (1997) : Remotely Sensed Surface Temperature and Surface Fluxes in arid and Semi-arid Lands : 13th Conference on Hydrology, American Meteorological Society, Long Beach California, January 1997. pp179-182. Chehbouni A., Qi J., Lo Seen D., Dedieu G., S. Moran, Daubas M , Monteny B.M, (1996). Estimation of Real Evapotranspiration using Remotely Sensed Data, Proceeding of FAO workshop, « Remote sensing and water ressources management’’ B 2, pp 11-19, Montpellier, France. Chehbouni, A., 1992, Présentation d'un modèle de transfert couplé de masse et de chaleur dans le système solvégétation-atmosphère pour les zones arides et semi-arides. PhD thesis, Université Paul Sabatier, Toulouse, France, 1992. Chehbouni, A., D. Lo Seen, E.G. Njoku, and B. Monteny, 1996, Examination of the difference between radiative and aerodynamic surface temperature over sparsely vegetated surfaces, Remote Sens. Environ., 58, 177186. Chehbouni, A., D. Lo Seen, E.G. Njoku, J.-P. Lhomme, B. Monteny, and Y.H. Kerr, 1997, Estimation of sensible heat flux over sparsely vegetated surfaces, J. Hydrology (188-189), 855-868. Chehbouni, A., O. Hartogensis, Y.H. Kerr, L. Hipps, J.-P. Brunel, C.Watts, J. Rodriguez, G. Boulet, G. Dedieu, and H.D. Bruin, 1998, Sensible heat flux measurements using a large aperture scintillometer over heterogeneous surface, in 78th AMS Annual Meeting, American Meteorological Society, Special Symposium on Hydrology, Phoenix, Arizona, 11-16 January1998, edited by A.M. Society, pp. 20-23, Phoenix. Douville H., Planton S., Royer J-F., Stephenson D.B., Tyteca S., Kergoat L., Lafont S. and Betts R., 1998, Vegetation feedbacks and their role in doubled-CO2 time-slice experiments, Journal of Geophysical Research, submitted (02/98). Drake BG Gonzales-Meler MA and Long SP, 1997, More efficient plants : A conséquence of rising CO2 ? Annu. Rev. Plant. Mol. Biol., 48, 609-639 Faivre, R., and A. Fischer, Predicting crops reflectance using satellite data observing mixed pixels, in Biometrics International Congress, 1994.08.22-26, Hamlton, Canada, 1994. July 1998 48 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Field CB Jackson RB and Mooney HA, 1995, Stomatal response to increased CO2 : implication from the leaf to the global scale. Plant Cell Environ. 18, 1214-1225 Goward, S.N., Dye, D.G., Turner, S., and J. Yang, 1993, Objective assessment of the NOAA global vegetation index data product, Int. J. Remote Sensing, 14, 3365-3394. Huffman, G.J., R.F. Adler, B. Rudolf, U. Schneider, and P.R. Keehn, 1995: Global precipitation estimates based on a technique for combining satellite-based estimates, rain gauge analysis, and NWP model precipitation information. J. Climate, 8, 1284-1295. Kergoat L., 1998, A model of hydrological equilibrium of Leaf Area Index at the global scale, Journal of Hydrology, accepted. Kergoat, L., A. Fischer, S. Moulin, and G. Dedieu, Satellite measurements as a constraint on estimates of vegetation carbon budget., Tellus, 47B, 251-263, 1995-a. Kergoat, L., X. Le Roux, H. Gauthier and G. Dedieu, 1995-b : Assimilation of time series of satellite measurements in a vegetation model: application to a humid savannah site. Proceedings of the Photosynthesis and Remote Sensing, Satellite meeting of the 10th International Congress of Photosynthesis, 20-30 august 1995, Montpellier, France, 457-464. Kerr, Y.H., J.-P. Lagouarde, and J. Imbernon, Accurate Land Surface Temperature Retrieval from AVHRR Data with Use of an Improved Split Window Algorithm, Remote Sens. Environ., 41, 197-209, 1992. Körner, Ch., 1994, Leaf diffusive conductances in the major vegetation types of the globe, in Ecophysiology of photosynthesis, E.D. Schuze and M.M. Caldwell eds, 463-490, Ecological studies 100, Springer Verlag, Berlin Lafont S, 1997, modélisation de la végétation en condition de double CO2, mémoire de DEA OAB Université Paul Sabatier, Toulouse Lo Seen Chong, D., E. Mougin, S. Rambal, A. Gaston, and P. Hiernaux, A Regional Sahelian grassland model to be coupled with multispectral satellite data. II. Towards the control of its simulations by remotely sensed indices., Remote Sensing of Environment, 52, 194-206, 1995. Lo Seen, D., A. Chehbouni, E. Njoku, S. Saatchi, E. Mougin, and B. Monteny, An approach to couple vegetation functioning and soil-vegetation-atmosphere-transfer models for semiarid grasslands using the HAPEXSahel experiment, Agricultural and Forest Meteorology, 83 (1-2), 49-74, 1997. Loudjani, P., F. Cabot, V. Gond, and N. Viovy, Improving NDVI Time-Series Using Imposed Threshold on IRT, IR and Visible Values (INTUITIV): A method for reducing cloud contamination and noise in NDVI timeseries over tropical and sub-tropical regions., in 6th International Symposium "Physical Measurements and Signatures in Remote Sensing", 1994.01.17-21, pp. 93-102, Val d'Isère, France, 1994. Maisongrande, P., Ruimy, A., Dedieu, G. and B. Saugier, 1995 : Monitoring seasonal and interannual variations of Gross Primary Productivity, Net Primary Productivity and Net Ecosystem Productivity using a diagnostic model and remotely sensed data. Tellus, 47B, 178-190. Malmström, C.M., J.T. Randerson, M.V. Thompson, H.A. Mooney, and C.B. Field, The next dimension: extending the time axis of global NPP estimates. In: Proceedings of the International Colloquium Photosynthesis and Remote Sensing, edited by G. Guyot, pp. 407-411, 1995. Monteith, J.L. 1977 : Climate and the efficiency of crop production in Britain. Philosophical Transactions of the Royal Society of London B., 281, 277-294. Mougin, E., D. Lo Seen Chong, S. Rambal, A. Gaston, and P. Hiernaux, A Regional Sahelian grassland model to be coupled with multispectral satellite data. I. Model Description and validation., Remote Sensing of Environment, 52, 181-193, 1995. July 1998 49 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report Moulin, S., A. Fischer, and G. Dedieu, Assimilation of short wavelength remote sensing observations within a crop model: methodological development and interannual field scale study, in ISPRS Photosynthesis and Remote Sensing, pp. 333-338, Montpellier, France, 1995. Moulin, S., L. Kergoat, N. Viovy, and G. Dedieu, 1997 : Global scale assessment of vegetation phenology using NOAA/AVHRR satellite measurements. Journal of Climate, 10(6), 1154-1170 Ouaidrari, H., E. Vermote, N. El Saleous, and D. Roy, 1997, AVHRR Pathfinder II data set: evaluation and improvements, in 7th International Symposium Physical Measurements and Signature in Remote Sensing. Courchevel, France, 7-11 avril 1997, edited by G. Guyot, 1997., in 7th International Symposium Physical Measurements and Signature in Remote Sensing, 7-11 avril 1997, edited by G. Guyot, and T. Phulpin, pp. 131-137, A.A. Balkema, Rotterdam, Netherlands, Courchevel, France. Probst, J.L., and Y. Tardy, Long range streamflow and world continental runoff fluctuations since the beginning of this century, Journal of Hydrology, 94, 289-311, 1987. Puyou-Lascassies, P., G. Flouzat, M. Gay, and C. Vignolles, Validation of the Use of Multiple Linear Regression as a Tool for Unmixing Coarse Spatial Resolution Images, Remote Sensing of Environment, 49, 155-166, 1994. Rahman, H., and G. Dedieu, SMAC : A Simplified Method for the Atmospheric Correction of Satellite Measurements in the Solar Spectrum, Int. J. Remote Sens., 15 (1), 123-143, 1994. Raich J.W., Rastetter E.B., Melillo J.M., Kicklighter D.W., Steudler P.A., Peterson B.J., Grace A.L., Moore III B. and Vorosmarty C.J., Potential net primary productivity in South America. Application of a global model. 1991, Ecological Applications, 1, 4, 399-429 Ruimy A., Dedieu G., Saugier B. (1996). TURC - Terrestrial Uptake and Release of Carbon by vegetation, a diagnostic model of continental gross primary productivity and net primary productivity. Global Biogeochemical Cycles, 10, 269-285 . Ruimy A., Jarvis P., Baldocchi D.D. and Saugier B., 1995. CO2 fluxes over plant canopies and solar radiation: a review. Advances in Ecological Research, 26, 1-68. Ruimy, A., B. Saugier, G. Dedieu, TURC - a diagnostic model of terrestrial gross and net primary productivity based on remote sensing data. In: Proceedings of the International Colloquium Photosynthesis and Remote Sensing, edited by G. Guyot, pp. 261-267, 1995. Ruimy, A., L. Kergoat, C.B. Field, and B. Saugier (1996-b). The use of CO2 flux measurements in models of the global terrestrial carbon budget, Global Change Biology, 2 (3), 287-296. Ruimy, A., Saugier, B., and G. Dedieu, 1994 : Methodology for the estimation of terrestrial net primary production from remotely sensed data. J. Geophys. Res., 99, D3, 5263-5283. Sellers PJ Bounoua L Collatz GJ Randall DA Dazlich DA et al. 1996, Comparison of radiative and physiological-effects of doubled atmospheric CO2 on climate. Science 271(5254), 1402-1406 Sellers, P.J., S.O. Los, C.J. Tucker, C.O. Justice, D.A. Dazlich, G.J. Collatz, and D.A. Randall, A revised land surface parameterization (SiB2) for atmospheric GCMs. Part II: the generation of global fields of terrestrial biophysical parameters from satellite data, Journal of Climate, 9, 706-737, 1996. Tarpley, J.D., S.R. Schneider, and R.L. Money, Global Vegetation Indices from the NOAA-7 Meteorological Satellite, Journal of Climate and Applied Meteorology, 23 (3), 491-494, 1984. Viovy, N., O. Arino, and A.S. Belward, The Best Index Slope Extraction (BISE) : A method for reducing noise in NDVI time-series, Int. J. Remote Sens., 13 (8), 1585-1590, 1992. Woodward F.I., 1987, Climate and plant distribution, Cambridge University Press, Cambridge July 1998 50 STEM-VGT: Satellite Measurements and Terrestrial Ecosystem Modelling using VEGETATION instrument Prelaunch Phase Final Report 10.2 Published works that use LASUR dataset Gobron, N., B. Pinty, and M. M. Verstraete (1997) `A New Approach for the Characterization of Land Surfaces', American Geophysical Union Fall Meeting}, {\bf AGU}, San Francisco, California, USA, December 812. Gobron, N., B. Pinty, Verstraete, M. M., and Y. Govaerts (1997) `Presentation and Application of an Advanced Model for the Scattering of Light by Vegetation in the Solar Domain', in 7th International Symposium Physical Measurements and Signature in Remote Sensing. Courchevel, France, 7-11 avril 1997, edited by G. Guyot, 1997., in 7th International Symposium Physical Measurements and Signature in Remote Sensing, 7-11 avril 1997, edited by G. Guyot, and T. Phulpin, pp. 267-273, A.A. Balkema, Rotterdam, Netherlands, Courchevel, France. Gobron, N. : Caractérisation des surfaces terrestres par télédétection spatiale à partir de méthodes physiques avancées. Thèse de l’Université Blaise Pascal, UFR de Recherche Scientifique et Technique, Numéro d’Ordre : D.U. 906, N°140, S.P.I. 97.101, 29 mai 1997. Knorr, W., N. Gobron, Ph. Martin, B. Pinty, M. M. Verstraete, and G. Dedieu (1995) `Constraining a Climate Driven Vegetation Model with Satellite Data', in Proceedings of the International Colloquium and Remote Sensing, Edited by G. Guyot, Montpellier, France, 28-30 August, 269-279. Ouaidrari, H., E. Vermote, N. El Saleous, and D. Roy, 1997, AVHRR Pathfinder II data set: evaluation and improvements, in 7th International Symposium Physical Measurements and Signature in Remote Sensing. Courchevel, France, 7-11 avril 1997, edited by G. Guyot, 1997., in 7th International Symposium Physical Measurements and Signature in Remote Sensing, 7-11 avril 1997, edited by G. Guyot, and T. Phulpin, pp. 131-137, A.A. Balkema, Rotterdam, Netherlands, Courchevel, France. July 1998 51