International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Service Page or Jsp, so that it can be used over the Internet (or
any other large network) without other additional requirements
for the client computer but general website browser software as
International Explore or Netscape. This is the browser layer.
The second layer is the key component that supports the whole
function of GZ-AgriGIS, including problem interpretation,
operation control, knowledge reasoning, model realization and
their integration in problem solving. This layer forms the web
pages that will forward to browser layer dynamically or
statically. Those web pages either accept data from the browser
or forward decision result to the browser. The accepted data
then will serve as parameters of analysis models, conditions for
knowledge or basic information for GIS. The model
grammar/implication interpretation module will be triggered
and interprets the input parameter value to choose, forms a
suitable model or models the complete numerical calculation.
Similarly the knowledge formalization/resolution constructor
will interpret some input data to form conditions for knowledge
and thus relative knowledge objects can be constructed to
realize knowledge reasoning. As for GIS functions, the input
data also includes spatial information as farm field ID, soil type
relating to the field, etc., so GIS will locate and produce a map-
based analysis result to users with the help of model analysis
and knowledge reasoning. In order to get map-based analysis
result, Esri’s ArcIMS 4.0 is used as webGIS engine '. ArcIMS
supports java connector that passes request of java-programmed
program to the spatial server and thus generates the result
expressed in map. All these processes are universally controlled .
by the controller component. The controller component is also
responsible for interaction with system data through data access
interface Jdbc. The bottom layer is system data. Here the
system data represents models, knowledge and spatial database
as well. Spatial data are accessed by SDE API offered by Esri's
ArcSDE 8.1 while non-spatial data are accessed by Jdbc.
Models are organized in the way of DE or LE as described in
Section 3.2. Knowledge is represented in Objected-oriented
method and can be constructed dynamically at system running
time. All parameters and grammar/implication interpretation
used in models as well as knowledge information are stored and
retrieved by database. Knowledge objects and models will be
constructed by. model grammar/implication interpretation
module and knowledge formalization/resolution constructor
module respectively.
4.3 Decision-making: Procedures
The agriculture area is divided into several districts, with each
equipped with a plant and soil sample analysis extension
service responsible for the corresponding district. This mode
facilitates users (usually farmers) since they don't need to cover
a long way to send soil or plant samples to the service for
chemical or physical analysis, which avoid delaying in
implementing decision recommendations. Any farm-level field
can be recommended with appropriate ways of field
management (as fertilizer application, water irrigation, or
herbicide application, etc.) made by GZ-AgriGIS. As the
decision is made based on specialist knowledge and models
gathered from experts or literatures, it has more credibility than
that made by an individual farmer experience.
There is two ways that prompt the system to make the decision
process. One is for user to offer ID number of a farm field or
the user can select a farm field or a region that contains series
of farm fields based on digital map. The system will then
prompt a list of variables some of which have been stored in
! See http://www.esri.com
190
database while others may require new input through graphic
user interface. In this case, we demonstrate the decision made
on fertilizer application (nitrogen application). The user selects
a region by drawing a selecting box on the digital map, and this
operation will highlight the selected fields and return the
selected field IDs. Some basic data as field location, field soil
type or some constant data can directly retrieved from the
system database while others that are more transient such as soil
water content, active nitrogen content have to be input by user,
The system then will use those data, supported by analysis
models and domain knowledge, to give -an understandable
decision result. Take nitrogen application as an example. Some
factors that affect nitrogen application of the spatial decision
unit should be input from user or retrieved from system
database, then appropriate fertilizer type as well as the amount
of the fertilizer that should be applied will be calculated by
fertilizer application models and crop requirement knowledge.
As some fields may lack data or some crop need knowledge
may lack, lacking information or lacking knowledge will be
returned to user (Fig. 5).
5. CONCLUSIONS
GIS originally is developed to store, retrieve and display spatial
data and domain models are combined with GIS to simulate
some complex phenomena later. The use of domain models in
GIS greatly expanded its application domain and improved its
application level. Applications as environmental pollution
simulation, shortest route selection and material distribution
plan, flood submersion prediction, etc are benefited a lot from
GIS and domain models. Some special spatial tasks which are
semi- or ill structured, however, are beyond either GIS itself or
domain models. This put GIS use in difficulty. Fortunately
Artificial Intelligence (AI) shows that expert knowledge can be
used to solve semi- or ill-structured problems, thus the
integration of GIS and expert knowledge is our research
consideration. The advantage of GIS and expert knowledge
integration is its power to support people in decision-making
with reliable and comprehensible map-based format. The
critical factors in this integration include expert knowledge
representation, model organization, the integration of GIS
models and knowledge, and the proper use of model and
knowledge.
fertilizer: Urea
20-30 kg/hr
15-20 kg/hr
10-15 kg/hr
5-10 kg/hr
0-5 kg/hr
Lack information"
Lack knowledge”
waler
Note: Ithe field lacks support information for decision
2)the field lacks domain knowledge for decision
Fig. 5 Nitrogen application decision made by GZ-AgriGIS
The fact that the topologic features and uneven surface of
agricultural land in most region makes farm fields small in arca,
International
HET
irregular in
overpopulatio
usually has to
need for shai
overpopulate«
large land ev
those regions
calculates fer
grids while
Knowledge le
extension of
farm fields (g
for enhancen
management
To offer an a
users, a web-
of GIS and
Expert knowl
human expert
management
The novelty
which contaii
With the sys!
decision-mak
of manageme
It fits with ur
esp. in moul
irregular farm
References
A GIS applic
Josef Benedil
Sciences 142
Andre Zerger
for real-time :
and Urban Sy
Chinese Aca
Devel-opmen
(Nong Ye Ji :
Science and 1
Geoffrey I. V
evaluation. oi
acquisition. Iv
Guo, W.T., 1
Technology i
Science and 1
Harald Vacil
decision supr
Vienna for si
and Managen
Mchiael P. J.
housing mob
3:49-67
Peter B. Kee
routing. Deci:
Shane Run
Development
Computers ar
