293 
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
Multitemporal analysis of LANDSAT Multispectral Scanner (MSS) 
and Thematic Mapper (TM) data to map crops in the Po valley 
(Italy) and in Mendoza (Argentina) 
M.Menenti & S.Azzali 
Institute for Land and Water Management Research (ICW), Wageningen, Netherlands 
D.A.Collado & S.Leguizamon 
Instituto de Investigaciones Aplicadas en Ciencias Espaciales (HACE), Mendoza, Argentina 
ABSTRACT: Practical applications of LAUDSAT Multispectral Scanner (MSS) and Thematic Mapper (TM) images in 
digital format to crop monitoring are presented. Image availability and timeliness are dealt with in relation 
with phenological variability and intercropping. Crop identification and mapping is done by establishing a 
indices and overpass dates. A.n application of this method is presented. Out of the different vegetation in 
dices, the transformed vegetation index (TVI) is chosen to assess the statistical significance of observed 
differences. It is concluded that the accuracy of the TM sensor onboard LANDSAT 5 is sufficient to guarantee 
pixels or more give statistically significant differences in TVI-values. Independent of the number of pixels, 
relative differences of about 4% are not significant. 
multitemporal discrimination scheme which makes use of crop labels defined in terms of different vegetation 
the accuracy of observed and significant differences in vegetation index values. MSS or TM image samples of 5 
Identification and mapping of agricultural crops with 
LANDSAT data is a classic piece of remote sensing 
research and application. In the first decade of 
monitoring of earth resources by satellites the ap 
pealing theoretical elegance of numerical classifica 
tion methods stimulated much research into crop map 
ping with single-date LANDSAT imagery. 
In the last years a clear trend seems to emerge 
(Crist 1984, Jackson et al. 1983, Miller et al. 1984, 
Hinzman et al. 1984), which emphasizes multitemporal 
analysis of satellite imagery to map crops. In our 
opinion this is due to the recognition of the limits 
of automatic classification methods and to the need 
of reducing the amount of processing required for 
each scene. The latter is a particularly important 
issue when one tries to use operationally satellite 
imagery in digital form. The potential of satellites 
to provide frequent and regular information on agri 
cultural crops within large areas can be spoiled by 
excessive costs and the processing time required for 
analysis. 
1 INTRODUCTION 
actually available images, i.e. those present in the 
Earthnet archive, and of suitable images, i.e. those 
available within the period most suited to crop 
identification (Azzali 1986) are relatively low. In 
1985, for example, only 17.5% of overpasses gave 
suitable images for our purpose. It should also be 
noted that 1985 was the best year in the time span 
1980-1985 (Table 1). Table 1 shows that applications 
requiring 3 suitable images per year were feasible 
in each year except 1983. Anderson (1986) underscored 
the commercial potential of crop monitoring by means 
of LANDSAT applications requiring 4 and 5 suitable 
MSS-images per year. Table 1 shows that such applica 
tions would not be operationally reliable, since 5 
MSS suitable images were available for only two 
years out of six for both irrigation districts. 
2.2 Phenological variability 
The practical aspects of crop mapping by means of 
satellites, therefore, are not only details to be 
dealt with after having developed a technique, but 
should be considered beforehand to establish which 
methodology best suits the operational requirements. 
In general terms one has to consider the practical 
aspects of image availability and processing and of 
agricultural reality, e.g. phenological variability 
within each crop and intercropping. 
The feasibility of crop mapping by means of multi 
temporal LANDSAT imagery relies entirely on accurate 
knowledge of crop phenology. It is especially impor 
tant to estimate precisely the period of occurrence 
of each phenological stage. Agricultural practices, 
such as choice of variety, seeding date and applica 
tion of fertilizers, increase the spread in the 
period of occurrence of phenological stages in in 
dividual fields where the same crop is being grown 
(Crist 1984). 
In this paper the approach and its application are 
briefly presented. We will focus on the underlying 
scientific issue of the statistical significance of 
the observed differences. Out of the seven TM-bands 
only two, i.e. TM 3 and TM 4, are applied because of 
the comparison with MSS 7 and MSS 5 and of the need 
to leave TM 5, TM 6 and TM 7 available for detection 
of crop-specific effects, e.g. water stress. 
Phenological observations in some 80 plots were 
collected in the two Italian irrigation districts 
during the growing season 1985. For each crop and 
phenological stage, a graph has been constructed of 
a function a (t) giving the area, on any given date, 
where that particular stage occurs. Then the function: 
t 
t 
2.1 Actually available and suitable LANDSAT images 
2 SOME PRACTICAL CONSTRAINTS IN OPERATIONAL CROP MAP 
PING BY SATELLITES 
We can, therefore, define a period of time 
with t2 being the last date of observed occurrence 
of each stage. The period [t^, 12] is, therefore, 
directly obtained from the field observations. The 
is calculated to obtain the total area where the par 
ticular stage has occurred before day = t. The day = 
t^ is the first date of occurrence according to the 
phenological field observations. 
is the first date of occurrence according to the 
The application of LANDSAT data to crop monitoring 
does in principle benefit from high temporal resolu 
tion. As Table 1 shows, however, the percentages of