901 
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
How few data do we need: Some radical thoughts 
on renewable natural resources surveys 
J.A.Allan 
School of Oriental and African Studies, University of London, UK 
ABSTRACT: The experience of applying remote sensing 
techniques to renewable natural resources surveys and 
evaluation has been to show that the necessary is too 
expensive and the affordable inadequate. The promise 
of remote sensing has been that it could provide 
overviews sufficiently frequently to monitor and predict 
agricultural change and production. In the event the 
remotely sensed data sets assembled for predictive 
purposes, such as that which provided a proportion of 
the data used in the LACIE experiment proved to be 
unwieldly even though the samples handled were only a 
tiny fraction of the data sensed. 
It is argued here that the time has come for remote 
sensing specialists to identify economic activities which 
attract returns to institutions which will benefit by advance 
warning of agricultural performance at regional level, 
and then to determine the minimum level of sample 
coverage and frequency necessary to provide predictive 
information at a price which the economic activities will 
bear. The experience of users of the general purpose 
remotely sensed satellite data available to date for 
renewable natural resources surveys (viz Landsat data) 
has led many users, and especially potential users, to 
acquire extravagant expectations; they have accepted 
the marvel of the comprehensive high resolution 
overview enabled by the satellite perspective without 
recognising the consequences of needing in addition 
high radiometric and temporal resolution data to detect 
information of concern. It will be concluded that any 
economically viable operational system will require 
economies in spatial cover through sampling, and 
suggests principles according to which such economies 
might be decided. The implications for sensor design 
and data processing are also discussed briefly. 
Keywords: sampling, remotely sensed data, land cover, 
crops 
1 INTRODUCTION 
The fifteen years of handling satellite remote sensing 
data for renewable resources studies have been more 
remarkable for the research papers generated than for 
the economic benefit of the activity. Such initial 
disappointment in a new area of technology is not 
unusual, however, and there are recent precedents for 
example in a similar sphere, that of satellite space 
communication. The 1950s and the 1960s were 
decades when the optimism held out by the proponents 
of satellite communication was not confirmed by effective 
technological innovation. In applying remotely sensed 
data to renewable natural resources studies we are 
currently in a similar phase where the up-take of remote 
sensing data capture and data processing systems by 
professionals in relevant fields has been very partial. 
Impediments to the adoption of remote sensing 
procedures have been partly: 
• technological, 
• to do with the inertia of potential 
adopting institutions 
t the unreasonable expectations of the users of 
general purpose satellite systems 
t the lack of sampling and data reducing procedures 
for the vast data sets generated in studies of the 
dynamic environments of concern to renewable 
natural resource surveyors. 
This last aspect will be the main concern here; the other 
issues will be referred to in the discussion but will not be 
treated systematically. 
It has been shown elsewhere (Allan 1984) that with the 
launch of the Landsat satellites in the 1970s it became 
possible for the first time to address the demanding 
temporal features which have to be taken into account if 
land-cover, agriculture and cropping are to be surveyed 
comprehensively . The same the study emphasised that 
there was a clear data-volume- threshold at which 
current computing technology, and even future 
technology, would be inadequate to handle the huge 
volumes of digital information generated by sensing 
systems resolving the multi-dimensional remotely sensed 
data in the spatial, spectral, radiometric and temporal 
domains. The study also showed that most of the useful 
applications in renewable natural resources studies, 
namely crop area and production monitoring, were on 
the wrong side of the data volume threshold because 
relatively high spatial, temporal and radiometric 
resolutions were necessary to enable the reliable 
detection of such features as crop extent and condition. 
The only rational approach to the type of study which will 
inevitably generate huge data sets is some method of 
sampling or data reduction. The most obvious method is 
to relax the spatial resolution reducing the data volume in 
this domain by acquiring the data at a much larger pixel 
size and the 1980s have seen some important 
developments in the use of AVHRR data (one to four 
kilometres spatial resolution depending on whether 
captured at the nadir or the swath edge) as described by 
Justice et al. (1985) At the one kilometre spatial 
resolution the use of such data represents a reduction of 
226 times in the volume of data in the spatial domain 
compared with Landsat MSS; at the four kilometre level 
its use represents a 3616 times reduction. Meanwhile