the specification of the AVHRR sensor brings other
economies in that it only has three spectral bands. On
the other hand it has very much higher temporal
resolution in that daily coverage (16-18 times the
Landsat frequency) is acquired at radiometric resolutions
four times those familiar on the Landsat sensors (1024
levels instead of the 256 levels of Landsat MSS). It is
clear in these comparisons that it is in the spatial domain
that the largest economies are being made.
One very important area of research in the past twenty
years has been into the spectral properties of objects and
their detectability by remote sensing instruments. The
research to date into the spectral properties of the
complex soil, vegetation and crop surfaces which are the
concern of those who survey and manage renewable
natural resources shows that spectral information alone
will not be sufficient to discriminate many types of cover
of great significance to managers. The basis of this very
broad statement is the findings of the series of meetings
convened to exchange information on the spectral
signature of objects (INRA/CNES 1982 and 1984, ESA
1986). The results of the ten years of detailed research
into data derived from ground radiometer, as well as from
airborne and satellite instruments, show that only major
categories of cover at what have come to be known as
Anderson's Level 1 land cover classification for remotely
sensed data can be reliably detected. Otherwise only
tracts which have been managed to a high degree of
homogeneity and which are large in extent, that is many
times the pixel size of the sensor system, can be
distinguished from eachother and even some of these
homogeneous tracts may be difficult to discriminate.
This is particularly the case when such crops as wheat,
barley and grass are the crops to be detected. It has
been conclusively demonstrated that reliable
discrimination of cover will require multi-date information
and probably improved radiometric resolution. Happily
the management of large tracts (that is tracts bigger than
one hectare) to a high degree of homogeneity of cover is
normal in a large proportion of the rural areas of the
world and the major thrust of the argument of this paper
is that advantage should be taken of this normal
circumstance when designing a data capture system for
the detection of agricultural land cover. The discussion
will be restricted to the utilisation of visible and near
infrared data of the Landsat MSS/TM types as these
data have proved to be most useful in rural surveys to
date. At the same time the rich information environment
which the MSS and TM types of data represent enable
the comparison of various sampling strategies in a
search for an economical approach to the potentially
prohibitively expensive and data engorged problem of
acquiring and handling comprehensive information on
the land surface.
2 RELEVANT APPLICATIONS AND INSTITUTIONAL
ENVIRONMENTS
The majority of the land surface which it is likely to be
worth surveying for information on soil, water, vegetation
including forest resources and crops falls into two
categories. First those tracts which are managed by man
intensively and secondly those tracts which are managed
less intensively which grade into the regions enduring
low rainfall and which are as a consequence of no
■'terest from a resource management viewpoint unless
they can be commanded by irrigation water. The first
type of land, the intensively managed tracts, to which we
shall refer to as HISTAB Land (high intensity with stable
parcels) is characterised by:
• high or at least adequate rainfall for one season or
perennially.
• a limited range of crops grown in one or more
growing seasons.
• a relatively static arrangement of parcels
(agricultural fields) in which crops are managed.
The parcels will generally be more than one
hectare, but there are many areas, especially in the
tropical world where parcellation is much more
fragmented than the one hectare level.
• a high level of inputs and outputs suggesting the
usefulness of monitoring activities and an economic
justification for expenditure on agricultural censuses,
including remote sensing.
The second type of land, the less intensively managed
land, to which we shall refer as LOSTAB Land (low
intensity wih unstable parcels), is characterised by:
• low and unreliable rainfall with the intensity of use
determined by the level of rainfall. There are usually
very clear gradients in the intensity of use which
change slowly over long distances.
t a limited range of crops and livestock usually
restricted to one season.
• irregular and ill-defined parcels for crop and
livestock management except in those regions where
fencing has been installed.
• low levels of inputs and outputs and no economic
justification for the deployment of expensive
monitoring including high resolution remote sensing
systems.
The exact global extents of the above two types of land
are impossible to estimate but it is suggested that these
two types of environment include ninety percent or more
of the agriculturally (including rangeland) managed
environment. The other ten per cent or so of land falls
into an intermediate type which for reasons of
fragementation and low intensity of use falls outside the
areas susceptible to monitoring by remote sensing. The
discussion here does not refer to this type of cover which
albeit comprises a minority of global cover.
The second type of land, the lostab land, is associated
with low levels of output and any type of monitoring,
including remote sensing, can only be justified if it is
inexpensive. The gradual nature of the changes in land
cover characteristic of these areas indicate that only low
resolution systems of the AVHRR type will be appropriate
and economically viable in these areas Drobably in
association with low level aenal survey tecnniques.
(Watson 1981)
It is in surveys of the the first type of cover that remote
sensing is most likely to be afforded and where at the
same time remote sensing is most likely to have the
greatest impact. This type of cover produces at least
eighty per cent by value of the world's crop and livestock
production although by area it probably comprises only
twenty per cent of total agricultural and rangeland cover.
One matter that has become clear as a result of the
research of the past twenty years is that because of the