909 
results is due to 
n created by the 
distribute pre- 
ocedure suggests 
al and chemical 
alysis linked 
ement practices 
ly influenced by 
mers and their 
vironment. 
and agricultural 
account such 
(van Keulen et 
e 1) detect and 
ion, reflected 
ce of the earth, 
hese satellites 
diverse spectral 
he atmosphere 
and dust) and of 
ypes and the 
- the spatial resolution of the RS satellites com 
pared with the pattern of the land use(e.g. layout 
of the arable fields); and 
- the occurrence of (nearly) cloudfree periods du 
ring the growing season. These pre-conditions limit 
the use of these interpretation techniques to the 
more developed areas of the world and/or to areas ha 
ving a semi-arid to semi-humid climate. 
Related to crop identification is crop monitoring 
and yield forecasting. These techniques include the 
acquisition and interpretation of synoptic meteoro 
logical data in relation to the development of the 
'green' surface. Assuming a quantitative relation 
between yield and the occurrence of meteorological 
anomalies (compared with an average year), particu 
larly water stress of the crop or vegetation due to 
lack of rainfall, statistical models may be applied 
to estimate the actual yields on the basis of rain 
fall predictions (see e.g. Ambroziak, 1985). If such 
models do not also include other variables as the 
meteorological variables, such as the natural ferti 
lity of the soil, the possible application of ferti 
lizer (see section 4), the occurrence of weeds, 
pests and diseases, the results may become to be re 
garded as doubtful. It is therefore necessary that 
such models incorporate a thorough knowledge of the 
existing farming systems in the area under study. 
ta sets are as- 
spatial resolu- 
pixel). Current- 
data obtained 
t detailed agro- 
Bgetation or crop 
ms of ground 
(as described 
strength of sat- 
Lbilities, en- 
is, i.e. the pos 
ts covering the 
1985). 
sciences, crop 
regarded an ope- 
i, that generates 
il limits. The 
llti-spectral and 
IS, are highly 
data obtained in 
is the current 
:s, including 
te growing sea- 
SPOT 
HRV-CCD 
832 km 
142 
variable 
20x20 m 
10x10 m (p) 
0.51-0.73 pm (p) 
0.50-0.59 (1) 
0.61-0.68 (2) 
0.79-0.89 (3) 
8 bit 
(256 levels) 
4 The combination of remote sensing and simulation 
models 
To link numerical simulation techniques with RS, it 
is necessary to determine those attributes that 
could be observed and quantified (or approximated) 
with RS and could act simultaneously as forcing var 
iables in the models, or at least could be used as 
calibration variables. Although the operational sat 
ellites have not been designed specifically for the 
quantitative assessment of most of the relevant var 
iables, such as the rainfall and soil-moisture, the 
following variables could be derived from RS infor 
mation (based on the current state of the art): 
i. actual precipitation 
Rainfall mapping requires sequences of satellite im 
ages, ground measurements of precipitation and oper 
ational definitions of cloud types (Griffiths, et 
al., 1978). The classification of cloud types is 
based on the reflected radiance (albedo) in the vis 
ible spectral channel(s) and the emitted radiance 
(cloud top temperature) in the infra-red spectral 
channel(s). Applying proper threshold values, depen 
ding on for example the actual temperature profile 
of the atmosphere, rain clouds could be identified 
(Rosema, et al., 1985). To obtain a more accurate 
discrimination between the various cloud types, an 
iterative cluster analysis procedure could be intro 
duced (Seldon, Hunt, 1985). Problems may always oc 
cur, due to e.g. the layered structures of certain 
cloud systems (as frontal systems). 
The relation between the presence of (rain) cloud 
types and observed spatial extent and amount of 
precipitation remains obscure. Intensive ground ob 
servations are required to establish statistical re 
lations between the precipitation patterns and the 
RS images (Milford, Dugdale, 1983). This emphasizes 
the region specific character of such rain mapping 
procedures. 
ii. actual irradiance 
The amount of solar radiation reaching the ground 
depends on the cloud cover and the degree of the ab- 
sorbtion and scattering in the atmosphere. Major ab 
sorbers are water vapour and aerosols. The radiation 
is scattered by atmospheric molecules and aerosols, 
producing diffuse light; the presence of clouds in 
creases largely the scattering. 
Irradiance estimation procedures (daily, hourly 
values) using RS data are based on quantitative re 
lations between actually observed irradiance and 
standardized cloud-free irradiance, the cloudiness, 
possibly augmented by the cloud type (RS data) and 
an atmospheric depletion characteristic, such as the 
atmospheric surface pressure (Tarpley, 1979). Again, 
such procedures are region-specific; slight modifi 
cations to be included are site-specific data on 
slope and exposition (Cappellini, et al., 1982). 
iii. air temperature 
Analogous procedures could be applied to estimate 
from satellite images the actual daily range of air 
temperatures and air temperature profiles. Region- 
specific relations have to be derived from observed 
shelter temperatures (i.e. the minimum and maximum 
air temperatures in a shelter at a given height a- 
bove the surface) and the surface temperature, as 
derived from satellite measurements in the infrared 
spectrum, after correction for atmospheric attenu 
ation (Davis, Tarley, 1983). 
iv. actual evapotranspiration (LE) 
The amount of water transpired by the plants and - 
in case of an incomplete cover - evaporated from the 
soil surface is a function of the moisture content 
of the top soil layer, the energy balance at the 
surface and the aerodynamic drying power of the at 
mosphere. The calculation procedure could be simpli 
fied (e.g. Soer, 1980, Menenti, 1984) for defined 
regions and crops by the introduction of calibrated 
relations between several variables, especially 
those specifying the aerodynamic component in the e- 
vapotranspiration equation and RS measurable vari 
ables, such as the surface temperature (Ts, see 
iii.) and the compound reflection coefficient a of 
the surface and the crop/vegetation 
a = a(veg) x Sc + a(soil) x (1-Sc), 
where Sc is the coverage of the soil by the crop, 
a(veg) and a(soil) the reflection coefficient of the 
vegetation and the soil. 
The crop cover could be deduced from the leaf area 
index (LAI), which in turn could be approximated 
with the RS vegetation index (VI) (Tucker, 1977). 
Such a procedure would require an a-priori avail 
ability of a crop identification procedure to deter 
mine cropping patterns. 
5 Geographic Information Systems 
Data Base Management Systems (DBMS) are developed to 
facilitate handling of large quantities of data in a 
computer system: it consists of facilities to store 
and change data, retrieve information by means of 
formal query rules and to present information in any 
type of format. If at least one entity in a DBMS is 
defined and identified by its spatial attributes 
(e.g. sets of geographical coordinates) and if fa 
cilities are available to map such a spatial entity 
as points, lines or bounded regions, the DBMS is re 
garded a GIS, i.e. Geographic Information System 
(Abel, 1983). 
In accordance with the main objective of a DBMS, 
the organization of the data within a GIS must allow 
the user to extract and analyse data sets in order 
to derive new information in respect to the selected 
geographical location or areas. In a standard GIS 
such procedures are relatively simply structured 
(Tomlin, 1983); for the purpose of in-depth agron 
omic analyses a possible combination with simulation 
models could be fruitfull. 
To exemplify such an approach, reference is made 
to a recent study of the Centre for World Food Stu 
dies, in which the effects of fertilizer application 
on food production were analyzed in a number of Af 
rican countries. One of these countries was Burkina 
Faso. First a selection was made of the crops and 
cultivars relevant to the study, i.e. several varie-