907
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
The potential of numerical agronomic simulation models
in remote sensing
J.A.A.Berkhout
Centre for World Food Studies, Wageningen-Amsterdam, Netherlands
ABSTRACT: Long-term experience with numerical agronomic simulation techniques has resulted in a methodology that
is specifically application-oriented. Incorporated in a geographic information system, which consists of static
data on land, the simulation model has proven to be a valuable tool in a quantitative land evaluation (i.e. the
identification and quantification of possible land use developments). Introduction of Remote Sensing Images will
introduce the possibility to monitor actual crop growth situations (land use, crop development). The ability to
monitor actual crop growth using RS images enhances the detection of 'problem areas' and thus the system can be
used for early warning purposes. The paper will describe a GIS system including crop growth models. The possible
uses of RS images in such a system and the possibilities of such a system as a partial substitute of ground
truth in RS analysis will be developed and demonstrated in the paper.
1 Introduction
Satellite Remote Sensing (RS) provides a scientific
tool to analyze the spatial variability in the momen
tary state of the surface of a defined area on the
earth, and applying the multi-temporal possibilities
of Remote Sensing, the differential change of the
state in space and time. The usefulness of the infor
mation obtained by RS depends on the one hand on the
type of RS techniques applied, including their spa
tial resolution and on the other hand on the avail
ability of real information about the observed re
gion, i.e. the ground truth, regarding the actual
land use and vegetation, the landscape and soils,
the hydrological and meteorological conditions, etc.
This paper discusses the possibilities for reduc
ing the required ground truth data for RS image in
terpretation purposes by the introduction of numer
ical simulation models on plant- and crop growth
within the framework of a Geographic Information Sys
tem (GIS). The disciplines involved are treated and
it is argued that all applications will benefit from
a close cooperation.
2 A systems approach to plant- and crop growth
To study the complex, continuous reality of the
world, a meaningful (as related to the goal of the
study) section of the reality has to be identified
and separated from its environment. To obtain rele
vant and useful results, such a section, i.e. a so
called system - in the systems approach terminology
- must be sufficiently complex to exhibit a high de
gree of internal coherence. But on the other hand,
it must be simple enough for comprehension and in
vestigation (Chorley, Kennedy, 1971). For a region
al, agricultural land use analysis the conceptual
model of the system must at least cover the two main
objects involved, i.e.;
i. land, which comprises according to the F.A.O.
description (Brinkman, Smyth, 1973) all the earth-
related features such as landscape, soil, hydrology,
weather, vegetation and man made structures; and
ii. the farming system, or all the human activi
ties that are directly and indirectly, related to
agriculture in the region.
The selected features or object attributes, inclu
ding their observed or assumed relations, determine
the type of model applied. A schematic distinction
can be made between:
i. stochastic models, that contain statistical re
lations between some relevant and perceptible at
tributes of the system and lead indirectly to the
required results. The functioning of the system in
terms of the flow of energy, mass and information
within the system is considered a black box.
Specimens of this approach are the commonly used
(multiple) regression models, such as that by
Wiegand and Richardson (1984) and Ambroziak (1985).
The former describing the relation between incident
photosynthetic active radiation, leaf area index and
yield of defined crops. Ambroziak estimates yields
using actual monthly total precipitation and the
condition of the crop as deduced from its reflec
tance, in combination with historical records on the
performance of a crop in the selected regions.
A disadvantage of this type of models is their in
herent specific character with respect to the site
and crop type; they do not separate causes and can
not be applied in evaluation modules, assuming pos
sible changes within the current agricultural sys
tem.
ii. deterministic models, where at a predefined lev
el of generalization the variables and their relat
ions are formulated and quantified, based on insight
in and knowledge of the underlying basic processes.
Such models are applicable under a wide range of
conditions after a sound validation and calibration
procedure (van Keulen, 1976).
The Centre for World Food Studies has developed
such a 'cause-and-effect' model for agricultural
production, following a hierarchical approach (van
Keulen, de Wit, 1982). A top-down approach is ap
plied to generate production estimates (expressed in
kg/ha of dry matter of various components of the
crops as roots, stems, leaves and storage organ) for
specific crops, cultivated at specific locations,
with specific growth periods (van Keulen, Wolf,
1986; Rappoldt, 1986). The latter are characterized
by their specific soil and weather conditions. For
any crop, characterized by its genetic and physio
logical properties, the model (fig. 1) starts with
the calculation of the potential production as a
function of the incident photo-synthetic active ra
diation (PAR, roughly 50 percent of the global ir
radiance) and the temperature only.
This potential can subsequently be reduced by the
negative influence on crop production of the lack or
excess of water (using a water balance model with
time steps of one day), the lack of plant nutrients
and the occurrence of weeds, pests and diseases.
Fig. 2 shows some model results: for example produc
tion of Pearl Millet calculated for Dori, Burkina
Faso, using long-term mean monthly meteorological
