Modeling CO 2 sinks and sources of European land vegetation using remote sensing data K. Wißkirchen, K. Günther German Aerospace Center (DLR), German Remote Sensing Data Center (DFD), Climate and Atmospheric Products. Münchner Straße 20, D-82234 Wessling, klaus.wisskirchen@dlr.de 1 INTRODUCTION Quantifying sources and sinks of climatological relevant trace gases as well as understanding their exchange between atmosphere and land surface have become essential research topics in atmospheric sciences during the last years. Modeling of the net CO 2 uptake by vegetation via photosynthesis (Net Primary Productivity, NPP, and Net Ecosystem Productivity, NEP) has become an important tool to study the mechanisms of CO 2 exchange and to quantify the magnitude of terrestrial sinks and sources. For Europe not many studies exist regarding NPP and NEP of the whole continent. Veroustraete et al. [1] used the C-Fix model, a Monteith type parametric model driven by temperature, radiation and fraction of Absorbed Photosynthetically Active Radiation (fapar). These type of models are based on the idea to estimate plant productivity by using the light use efficiency and the amount of absorbed photosynthetically active radiation. In order to make a comparison with the results of Veroustraete et al. simulations with a more complex model are presented for improving the fundamental knowledge of geographical distribution of European carbon sinks and sources. At the German Aerospace Center (DLR) the mechanistic vegetation model BETHY/DLR (Biosphere Energy Transfer Hydrology Model [2] [3]) is used to perform simulations of NPP and NEP over Europe. The model is driven by remote sensing data of leaf area index (LAI) and land cover classification and meteorological input from the ECMWF (European Center for Medium Range Weather Forecast). Two versions of the model are used for simulations: One with reduced resolution (27.5 km) where the original resolution is aggregated for simulations on continental scale and one for simulations on regional scale in the original resolution of 1.1km. 2 INPUT DATA As input data for BETHY/DLR a land cover classification, leaf area index, soil types [4] and meteorological input data are needed. For information about land cover the PELCOM classification [5] is used. It was the first pan- European land cover classification and is valid for the year 1997. PELCOM is based on satellite data from NOAA- AVHRR. 2.1 Meteorological forcing Meteorological forcing is given by data from the ECMWF (European Center for Medium Range Weather Forecast) in a spatial resolution of 0.5. Daily values of photosynthetically active radiation (PAR) are calculated using ECMWF cloud cover data and the radiation parametrization of Burridge and Gadd [6]. The advantage of this approach in contrast to the direct use of ECMWF-radiation data is the use of analysis data of cloud coverage which leads to more exact results than the direct use of radiation forecast data [3]. Daily maximum, minimum and average values of temperature are also calculated from ECMWF analysis data for the parametrization of the daily cycle of air temperature. Precipitation values are taken from forecast simulations. 2.2 Processing of leaf area index data An important input parameter for the calculation of net carbon uptake by photosynthesis of land vegetation is the time series of leaf area index (LAI). LAI is processed using the operational NDVI (Normalized Differences Vegetation Index) product of DLR-DFD [7] which is calculated from data of NOAA-AVHRR. Spatial resolution of this data is 1.1 km. In order to reduce cloud contamination 10day composites are generated. A land cover based correction of solar zenith angel is applied to the NDVI (Fig. 2). As land cover classification the NOAA-AVHRR based PELCOM classification is used [5] again. Finally a time series analysis (harmonic analysis) is then applied to the NDVI data set in order to eliminate data gaps and outliers (Fig. 2). Based on the smoothed NDVI time series, fapar (Fraction of Absorbed Photosynthetically Active Radiation) and LAI are calculated (Fig. 3) using the algorithm of Sellers et al. [8]. An example of the resulting LAI maps is given in Fig. 4.
4 SIMULATIONS ON CONTINENTAL SCALE For simulations on continental scale a data aggregation technique (DAG) is used to reduce resolution in order to keep the resolutions of input data compatible and to keep calculation time low. Original land cover and LAI data are aggregated to 25 by 25 pixel, resulting in a resolution of 27.5 km. For each land cover class the fraction of coverage area on the coarse grid is calculated. This information is used to calculate an area weighted average LAI for each class. Information on the coverage area is also used to calculate the overall sum of carbon uptake on each grid cell for each land cover class. Meteorological data from ECMWF are regridded using weighted area interpolation. As an example results of annual net carbon uptake (NPP) on continental scale for 1998 are shown in Fig. 6 for coniferous forest and grasland. The most important structures of this land cover types are well identifiable, f.e. the coniferous forest region around Bordeux in France or the black forest and the Bavarian Forest in Germany. The grasland areas in Ireland and in the UK are also covered as well as the areas in the Benelux countries and in northern Germany. Fig. 7 shows the overall NPP and NEP summarized over all land cover classes. NEP (Net Ecosystem Productivity) takes the additional carbon source of soil respiration into account. Negative values correspond to sources of CO 2 while positive values represent sinks. NEP for Europe is calculated to 0.596 PgC/a for all land cover classes while forest NEP is estimated to 0.41 PgC/a. Veroustraete et al. [1] estimated forest NEP of Europe to 0.735 PgC/a in 1997 which is in a comparable magnitude. 5 REGIONAL SIMULATIONS 5.1 Simulation example Data processing for regional simulations is performed by selecting subsets from the original datasets. For the application of the coarser resolved meteorological and soil data (0.5 ) a simple tiling approach is used. Regional CO 2 fluxes had been simulated for a 100 by 100 km subset around the permanent Eddy-Covariance station of Hyytiala which is a part of the CARBOEUROPE-network [9]. Simulations were carried out for the year 1998. Obviously the PELCOM land cover classification causes problems in regions with high structured water bodies, indicated by the overlay of a high resolved water mask [10]. The coarser resolved PELCOM classifies water bodies as land, leading to erraneous results on the borders of the lake regions. The overall effect of this misclassifications has to be estimated in further studies. The value of NEP calculated for the pixel corresponding to the location of Hyytiala station is 310 gc/m²/a which is in good agreement to the measured value of 296 gc/m²/a. 5.2 Accuracy of data aggregation compared to high resolution results The simulation on continental scale using the DAG method is compared to regional high resolution simulations in order to get information about the accuracy of the DAG technique. The DAG-gridboxes were overlayed over the high resolution grid. For each land cover class NPP and NEP of the high resolution pixels had been summarized within the corresponding DAG-gridbox. As an example for the region around Hyytiala-station the differences (in percent) between DAG and HR simulations for coniferous forest in the nine DAG gridboxes are given in Fig. 10 and Fig. 11. It s obvious, that for NPP the agreement is good with values around +/- 5% while NEP in DAG simulations is underestimated in the order of up to 25%. 6 CONCLUSIONS In this paper first results of NPP processing with a mechanistic model and satellite data as input had been presented. LAI data are processed from NOAA-AVHRR based NDVI data using the algorithm of Sellers et al. [8]. For simulations on continental scale a data aggregation method is used while regional simulations are performed for limited areas in the original resolution. Accuracy of the data aggregation is good for NPP while NEP is underestimated in a magnitude of 25%. These deviations can not be explained yet. Nevertheless forest NEP
calculated for Europe is comparable to previous results calculated using the C-Fix model or from measurements of the CARBOEUROPE network. In high resolution mode the use of the PELCOM classification is difficult in regions with high structured water bodies, leading to erraneous results due to misclassification of water bodies as coniferous forest. Values of NEP calculated in full resolution are in good agreement with measurements from the Hyytiala station. 7 REFERENCES 1. Veroustraete, F., Sabbe, H., Eerens, H., 2002: Estimation of carbon mass fluxes over Europe using the C-Fix model and Euroflux data. Remote Sensing of Environment, Vol. 83, 376-399. 2. Knorr, W. 1997: Satellite Remote Sensing and Modelling of the Global CO 2 Exchange of Land Vegetation: A Synthesis Study. Max-Planck-Institut für Meteorologie, Examensarbeit Nr. 49 (in German), Hamburg, Germany. ISSN 0938-5177. 3. Wißkirchen, K., 2005: Modellierung der regionalen CO 2 -Aufnahme durch Vegetation. Dissertation, Meteorologisches Institut der Rhein. Friedr.-Wilh.-Universität Bonn, 129 S. 4. Dunne, K.A., Willmott, C.J., 1996: Global dirstribution of plant-extractable water capacity of soil. International Journal of Climate, Vol. 16, 841-859. 5. Mücher, C.A., Steinnocher, K., Kressler, F., & Heunks, C., 2000: Land cover characterization and change detection for environmental monitoring of pan-europe. International Journal of Remote Sensing, Vol. 21, Nr. 6-7, 1159-1181. 6. Burridge & Gadd, 1974 in: Stull, R.B., 1988: An introduction to boundary layer meteorology. Kluwer Academic Publishers, Dordrecht, Boston, London, ISBN 90-277-2768-6. 7. Dech, S., Tungalagsaikhan, P., Preusser, C., & Meisner, R., 1998: Operational value-adding to AVHRR data over Europe: methods, results, and prospects. Aerospace Science and Technology, 5, 335-346. 8. Sellers, P.J. et al., 1996: A revised land surface parametrization (SiB2) for atmospheric GCM s. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. Journal of Climate, Vol. 9, 706-737. 9. Valentini, R. (Ed.), 2000: The Euroflux dataset 2000. In: Carbon, water and energy exchanges of European forests. Springer Verlag, Heidelberg, pp 300. 10. Wessel, P., and W. H. F. Smith, 1996: A global self-consistent, hierarchical, high-resolution shoreline database, Journal of Geophysical Research., 101, B4, 8741-8743.
Fig. 1: Flowchart of data processing for LAI time series. Fig. 2. Time series of NDVI before (black) and after (red) correction of solar zenith angle. The continous red line is the resulting time series of NDVI after harmonic analysis. LAI time series at Hyytiala-Pixel LAI 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 1 6 11 16 21 26 31 36 10day-Scenes Fig. 3. Resulting time series of LAI for the pixel corresponding to the location of the CARBOEUROPE station Hyytiala.
Fig. 4. Results of LAI calculation for the year 1998, 10day composite from day 18 to 190. Spatial resolution of the dataset is 1.1 km. Fig. 5. Flowchart of data processing for the simulations with BETHY/DLR-DAG on continental scale. Fig. 6.NPP (TgC/a) for coniferous forest (left) and grasland ecosystems (right)
Fig. 7. Overall sum of NPP (left) and NEP right (TgC/a) Fig. 8. Flowchart of data processing for high resolution simulations. Fig. 9. High resolution results for the region around the CARBOEUROPE-station Hyytiala/Finland (white circle). The high resolved water mask is indicated by the green lines. Coniferous Fig. 10. 2 3 1 4 5 6 7 8 9 Difference HR-DAG (%) 30 25 20 15 10 5 0-5 -10 1 2 3 4 5 6 7 8 9 DAG-Gridbox dnpp dnep Fig 11. Comparison of DAG and high resolution results. Positive deviation means an underestimation of carbon uptake by the DAG simulations