1) The identity of the cover type 
2) The size of the individual element (patch) in which 
resource levels are assessed (Forman and Godron, 
1987; Forman, 1995). For example, a large 
deciduous forest patch may contain tigers, while a 
smaller forest patch may not be able to support 
such species. 
3) The shape of the patch. To illustrate, an elongated 
stream will have different sedimentation levels 
from a more spherical lake with the same total 
amount of water. 
4) The identities of surrounding cover types. For 
example, a garden in a city surrounded by high 
traffic roads will have higher levels of atmospheric 
pollutants compared to a similar garden that does 
not have such roads near by. 
The extent to which factors other than the identity 
of the cover type influence the distribution of resource 
levels needs to be understood, as these are crucial inputs 
required while modeling the levels of distribution from 
cover maps. Again, in this context, we have carried out 
preliminary investigations on the extent to which various 
factors influence the distribution of flowering plant 
species in the sample landscape discussed previously. 
We found that the major factor influencing the set of 
plant species found at particular points was the 
ecosystem type. Flowever, about 30% of the variation in 
plant species composition could not be explained by the 
identity of the ecosystem patch and was due to other 
factors like size, shape, and neighboring element’s. 
Similar investigations are required for distribution of 
other natural resources, and for water cover types as 
well as land cover types. As water is redistributed from 
place to place far more than soil, variables like water 
quality levels will have dynamics quite different from 
that of say soil quality. The relative importance of 
various factors like patch size, shape and neighbouring 
elements will hence differ considerably depending on 
whether one is dealing with land or water systems. 
Organizing an efficient system of mapping and 
collection of field data over a country as large and 
varied as India requires the consideration of several 
issues. At the first level, of using remotely sensed data 
tor mapping land and water cover, one needs to prepare 
a set of cover types which can be identified accurately 
by the sensors, relate well to the levels of various natural 
resources being monitored, and reduce inter-observer 
variation in identification. Collecting training data for 
supervised classification also involves several statistical 
issues, notably regarding the distribution of training sites 
(Atkinson, 1991; Dobertin and Biging, 1996). Ideally, 
training sites should be spatially well distributed over 
the area to be mapped, so that the range of variation in 
the cover type can be included. However, logistically, 
collecting information from areas clustered close to each 
other is far easier than collecting the same amount of 
data from sites scattered far apart. Similar issues arise 
while assessing map accuracy (Janssen and van der Wei. 
1994). 
In addition, the seasons at which remotely sensed 
data is ideal for monitoring a particular resource need to 
be determined separately for each resource. For 
example, to monitor carbon utilization in a landscape it 
may be ideal to use the monsoon months as deciduous 
trees would also have their full complement of leaves. 
However, to differentiate deciduous from evergreen 
forest cover it would be better to use pre-rainfall data, 
where deciduous trees are bare and can be differentiated 
from evergreen more easily (Roy, 1993). Again crop 
monitoring may require certain data collected at specific 
dates. 
Data collection in the field has also to be 
organized, taking into consideration various factors. The 
number, size and distribution of samples in the field will 
crucially affect the quality of information gathered. This 
is an entire statistical area in itself, of sampling theory, 
which we shall not elaborate on here for lack of space. 
However, these issues need to be given serious thought 
as data collection poses logistic as well as scientific 
constraints, and both of these may have varying 
requirements. 
Finally, analyzing the correlation between these 
two kinds of data collected by different means and at 
different scales is not a trivial task. This data has to be 
used to model the distribution of natural resources in the 
future too, based on cover maps, and in order to do this 
the extent to which various factors like type identity, 
shape or neighboring elements affect these distributions 
needs to be understood. 
ORGANIZING PRE-EXISTING INFORMATION 
The task before us, to assess the distribution of 
natural resources over space and time, is fairly large, as 
well as extremely important. Means have to be devised 
for carrying out this task as efficiently as possible. A 
large number of maps, at different scales and using 
different sets of cover types, already exist with some 
information about natural resource distribution. If such 
information, already existing but of different kinds, can 
be collated and analyzed then it would make our task 
much easier. As an exercise, therefore, we looked at 
recent land cover maps published by various individuals