Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
709 
Simple classifiers of satellite data for hydrologic modelling 
R.S.Drayton 
University College, Cardiff, UK 
T.R.E.Chidley & W.C.Collins 
University of Aston, Birmingham, UK 
ABSTRACT: Hydrological models are normally based on parameters extracted from conventional maps such as 
drainage density and land cover. In this paper we examine the use of parameters derived from thematic classif 
ications of satellite imagery. The advantage of this technique is that the satellite image becomes an up-to- 
date, easily managed source of data for monitoring of water resources. The methods of classification range 
■'from simple density slicing to supervised classification. A model was constructed in which run-off character 
istics were regressed on thematic classifications. The problem of achieving sufficient variance in the hydro- 
logical parameters within one satellite image are discussed, and recommendations for distributed modelling 
are made. 
1 INTRODUCTION 
One of the principal aims of applied hydrology is to 
make estimates of streamflow. Frequently, the hydro 
logist is asked to mak- such estimates in the absence 
of observed, historic streamflows, and is required 
to use his skills to construct mathematical models 
which can predict future flows. These fall into two 
broad categories: statistical models which provide 
estimates of river flows having a specified probabi 
lity, or process models which provide estimates of 
streamflow resulting from certain specified rainfall 
sequences. In this paper we investigate the ways in 
which satellite data can be employed in such models. 
The advantage of this approach is that the rectified 
and classified image becomes an up-to-date, easily 
manageable source of data for monitoring water 
resources. 
The obvious but rather uninspired way to use satel 
lite data is to construct the conventional maps from 
which existing hydrological models draw their inform 
ation. The benefits of timelines and global coverage 
have been claimed many times, while the limitations 
of satellite sensors in evaluating hydrological 
variables have been explored by Rango et al, 1983 
(thematic features) and by Drayton and Chidley, 1985 
(geomorphic features). The conclusion appears to be 
that satellite sensors may be used to evaluate con 
ventional parameters to a precision suitable for 
regional models. 
A far more promising approach is to construct new 
models which are appropriate to the view of the world 
offered by the satellites. The fundamental difference 
in this view is that it is holistic; it brings to 
gether the various aspects of the earth's surface 
rather than separating them, as happens with maps. 
Thus, the reflectance of a patch of ground depends 
not only on its vegetal cover, its soil type and 
its moisture content, but also on the interactions 
which occur between them, and with climate. All of 
this information is also relevant to its hydrological 
response, and is conveniently integrated or compres 
sed into this single characteristic, namely its 
spectral signature. Thus the judicious use of para 
meters based on spectral characteristics is likely 
to yield hydrological parameters which are far more 
powerful than any parameters derived from maps. 
Thus, our aim was to produce interpretations of 
our study catchments, based on their spectral 
properties, to build hydrological models around 
them, and to compare their efficiency with existing 
models using conventional classifications. This is 
different from the approach by Gurnell et al (1985) 
in which it was heathland vegetation which was 
considered to be the mechanism which brought the 
hydrological information together, and the satellite 
systems were used only to monitor the heathland. 
The benefits of using satellite data are not con 
fined to thematic considerations alone. There is an 
enormous benefit to be gained from the structure and 
organisation of the data, which is distributed 
uniformly over the scene, and is organised in a 
regular grid cell structure. Distributed hydrologic 
models do exist but their usefulness has been 
diminished by the logistical problems of estimating 
values at many node points. Satellite data is clearly 
well provided with the means to overcome such 
problems. 
Our current work is concerned with ways in which 
the holistic nature of satellite data can be pres 
erved in hydrological models. As a first step we 
have examined its use in "lumped" models, so that 
the thematic benefits can be clearly distinguished 
from those of distribution. Work is in hand to 
examine the benefits of data structure, in relation 
to finite difference groundwater models, and will be 
reported later. 
2 BACKGROUND TO MODEL 
The study reported here was based on a group of river 
basins in the South Wales area, for which hydrologic 
data were available. (For details see Chidley and 
Drayton, 1985). Although lying within the same region 
the basins show a good variety of vegetal, geologic 
and climatic characteristics. 
Our aim was to evaluate the usefulness of the 
thematic classifications which are unique to satel 
lite data. So, as a first step we considered a 
lumped model in which we could eliminate the benefits 
derived from distributed data. We considered a gen 
eralised lumped rainfall-runoff model, which would 
provide estimates of annual runoff (catchment yield), 
based on the inputs of annual rainfall and a set of 
characteristics evaluated from the satellite data. 
We assumed that any one year's runoff could be 
considered as being composed of the sum of runoff 
from n different runoff types, and that each type 
has a response which depends on the magnitude of 
rainfall. Then total runoff, 
RO = E f(RF) A 
n n 
n 
In this exploratory model we assume that the 
distribution of run off types is constant from year 
to year. Then the model reduces to the simple 
statement