8. EXAMPLES OF THE USE OF SATELLITE DATA IN
NUMERICAL WEATHER PREDICTION MODELS
63
channels 1 and 2 were processed from April 1992 to September 1994 at a 2km resolution. After treatment for cloud detection and atmospheric
correction, monthly maps of the Normalized Difference Vegetation Index
(NDVI) values were assembled. For the purpose of this treatment, the
Holben (1986) composite technique was used; it involves the selection of the
maximum value of NDVI during the time period considered. Some 29
images have been utilized. Then, an automatic classification is performed on
monthly time series of NDVI to classify 11 classes (Figure 1): forests,
grasslands, orchards and bare ground, as well as 8 types of crops
characterized by distinct radiometric signatures, according to their seasonal
variations. In particular, this classification distinguishes very clearly the
areas with summer and winter crops. It is important to note that such
discrimination is not provided by high-resolution classification like the
European Corine database. The second phase of the treatment consists in
assigning values to the relevant vegetation parameters (roughness length,
leaf area index, minimum stomata 1 resistance) for each class of vegetation,
using look-up tables. As far as possible, the parameter values are derived
either from the local calibration of ISBA, or from the literature. In the last
step, the parameters are averaged within each model grid box using
aggregation rules described in Noilhan and Lacarrère (1995). Recently, the
method was improved by Habets et al. (1997) who assumed that the NDVI
profiles could be used to infer the seasonal evolution of the leaf area index
(LAI), independent of the vegetation class. Thus, Habets et al. (1997)
prescribed only the extreme values of LAI for each vegetation type, and the
monthly NDVI estimates were used to compute the seasonal variation of the
leaf area index between the minimum and maximum values. Such a method
has been tested to provide boundary conditions for a hydrological distributed
model in the Hapex-Mobilhy area (the Adour river basin). In an operational
context, high-resolution maps of vegetation are used over Europe. The
impact of either of these vegetation maps on ISBA (compared with the old
surface scheme with no vegetation and only one type of soil) has been
evaluated by Giard and Bazile (1997b). The new assimilation suite has
proved successful in improving the forecast scores against observed air
temperature and relative humidity at 2 m for synoptic stations. In the
subsequent assimilation process, the soil reservoirs and the soil temperatures
are corrected every 6 hours using the forecast errors of screen-level
temperature and humidity. The results of 20 consecutive days of assimilation
(6 hour cycles) with ISBA and the old surface scheme are shown figure 2 for
Spain and France. The statistics represent the mean bias and RMS for
temperature (K) and relative humidity (%) at 2 m. For both areas, the
forecast is improved, particularly for the daily cycle of humidity, which have
NUMERICAL WEATHER PREDICTION MODELS
63
channels 1 and 2 were processed from April 1992 to September 1994 at a 2km resolution. After treatment for cloud detection and atmospheric
correction, monthly maps of the Normalized Difference Vegetation Index
(NDVI) values were assembled. For the purpose of this treatment, the
Holben (1986) composite technique was used; it involves the selection of the
maximum value of NDVI during the time period considered. Some 29
images have been utilized. Then, an automatic classification is performed on
monthly time series of NDVI to classify 11 classes (Figure 1): forests,
grasslands, orchards and bare ground, as well as 8 types of crops
characterized by distinct radiometric signatures, according to their seasonal
variations. In particular, this classification distinguishes very clearly the
areas with summer and winter crops. It is important to note that such
discrimination is not provided by high-resolution classification like the
European Corine database. The second phase of the treatment consists in
assigning values to the relevant vegetation parameters (roughness length,
leaf area index, minimum stomata 1 resistance) for each class of vegetation,
using look-up tables. As far as possible, the parameter values are derived
either from the local calibration of ISBA, or from the literature. In the last
step, the parameters are averaged within each model grid box using
aggregation rules described in Noilhan and Lacarrère (1995). Recently, the
method was improved by Habets et al. (1997) who assumed that the NDVI
profiles could be used to infer the seasonal evolution of the leaf area index
(LAI), independent of the vegetation class. Thus, Habets et al. (1997)
prescribed only the extreme values of LAI for each vegetation type, and the
monthly NDVI estimates were used to compute the seasonal variation of the
leaf area index between the minimum and maximum values. Such a method
has been tested to provide boundary conditions for a hydrological distributed
model in the Hapex-Mobilhy area (the Adour river basin). In an operational
context, high-resolution maps of vegetation are used over Europe. The
impact of either of these vegetation maps on ISBA (compared with the old
surface scheme with no vegetation and only one type of soil) has been
evaluated by Giard and Bazile (1997b). The new assimilation suite has
proved successful in improving the forecast scores against observed air
temperature and relative humidity at 2 m for synoptic stations. In the
subsequent assimilation process, the soil reservoirs and the soil temperatures
are corrected every 6 hours using the forecast errors of screen-level
temperature and humidity. The results of 20 consecutive days of assimilation
(6 hour cycles) with ISBA and the old surface scheme are shown figure 2 for
Spain and France. The statistics represent the mean bias and RMS for
temperature (K) and relative humidity (%) at 2 m. For both areas, the
forecast is improved, particularly for the daily cycle of humidity, which have