201 G. Russell et al. European Journal of Agronomy 11 1999 187–206
Fig. 9. Results of a query about the type of agriculture carried out on Eutric Regosols in Pinhal Litoral, Portugal.
later in the large region than in one of its constitu- to the specification right up to the end of the
project. Although the project has now finished, ents. This may be a genuine difference of opinion
due to differences in the primary sources. However, subsequent experience in using the system will
prove valuable in assessing its functionality in the term ‘earliest’ does not refer to the absolute
earliest date ever recorded but to the ten percentile practice and in drawing up a specification for
further developments. value averaged over a 5-year period. This is an
expert interpretation of the data given in the The tests showed that the system could perform
the tasks specified in Section 3.1 and highlighted original source. It is thus theoretically possible,
although unlikely in this case, for both statements some of the benefits of using this approach rather
than normal database operations. For example: to be true. The first and third solutions also differ.
These come from independent sources, and the 1. The system can infer solutions to a query where
there is no direct match. user must decide which to accept.
2. Conditional information, qualifying the validity of the information found, is given with each
solution.
4. Discussion
3. Conflicting solutions can be presented together with the associated information that should
The novelty of the system meant that the final form of the knowledge base shell evolved over
allow the user to decide which is preferred. 4. Proof trees allow the train of reasoning leading
time. Frequent meetings and discussions with the potential users allowed modifications to be made
to the solution to be checked.
202 G. Russell et al. European Journal of Agronomy 11 1999 187–206
Fig. 10. Proof tree for the query of Fig. 9.
The work raised several interesting epistemological ment are widely recognised by crop physiologists
and there is general agreement about the broad issues, of which three are discussed below.
phenological stages, the actual phenological stages at which farming operations are carried out or
4.1. Reasoning with phenological information when the crop is vulnerable to particular hazards
are often not specified using standard terminology. It was initially thought that the crop phenologi-
cal phases could be structured in a shallow hierar- The names for analogous phases may also differ
from crop to crop. This is important, as facts chy with sub-divisions of major phases. However,
it soon became apparent that the terms used in added to the knowledge base must use terms that
are recognised by it if they are to be used in the the literature would not fit neatly into such a
structure. Although key events in crop develop- reasoning process.
203 G. Russell et al. European Journal of Agronomy 11 1999 187–206
Fig. 11. Results of a query about the date of snowing of winter wheat in Oost Nederland.
204 G. Russell et al. European Journal of Agronomy 11 1999 187–206
The first way we overcame this problem was to inheritance rules and rules for identifying nearby
use expert knowledge to convert the terms in the regions. The two part approach was needed
source publications to a single set of phenological because using simple inheritance rules on their
stages for each crop group. The user could then own can be ineffective. For example, in an area
browse through the phenological hierarchy and such as the Moselle valley, the area most like the
choose the stage of interest. However, this region of interest may be in a different country.
approach conflicted with the principle of using Also, primary administrative regions often corre-
non-specialists to add information as close to the spond with river basins and so span a wide range
original form as possible. The hierarchical struc- of topography and thus farming systems. The
ture was abandoned, and the inference engine was NUTS I region of Nord Ovest in Italy, for exam-
modified to show the phenological phases for ple, includes the NUTS III regions of Valle d’Aosta
which information was available. In this way, and Mantova. The former consists of alpine valleys
matches would not be missed, and the phenological with vineyards and livestock rearing, whereas the
phases displayed for selection would only be those latter contains the intensive farming systems of the
appropriate to the crop. However, inferring solu- Po valley. The yield-limiting factors are rather
tions from related crops was inefficient when infor- different in these two regions, and a model of
mation was available for different phases of maize production would thus require different
development or when the same phase went by parameters.
different names, for example ear emergence and During the development of the Crop Knowledge
heading of cereals. Base, a utility was developed for identifying similar
We now feel that the best approach would be regions on the basis of objective geographical
to use a hybrid approach in which the original attributes. These attributes, which included lati-
names are used but additional rules are developed tude, longitude, lowest altitude, river basin, and
to relate them to a generic scheme of phenological selected climatic data, were chosen because of their
description, such
as the
extended BBCH
effect on phenology or farming system type. After Biologische Bundesanstalt Bundessortenamt and
a check that these attributes could be considered Chemical industry scheme described by Meier
independent, a cumulative score was calculated in 1997, which enables the temporal relationship of
which a one or zero was given for each attribute one stage to another to be deduced. However,
depending on whether or not the value lay within even reasoning with a monotonically increasing
a pre-defined range. Regions with high scores were numerical system as internally consistent as BBCH
very much like the regions of interest. The utility is complicated by the scale not being mathemati-
successfully grouped NUTS III regions in two test cally continuous, and by phases proceeding in
areas, a transect from the north Sea to Bavaria parallel or even not in their strict sequence.
and northern Italy. However, it was recognised that the degree of similarity between two regions
4.2. Reasoning by similarity might vary with crop and indeed with crop attri-
bute and that economic factors operating at the Some important crop attributes vary geographi-
country scale could lead to significant changes in cally. These are mainly those related to phenology
farming practice at frontiers, even though there and those that influence the ratio between actual
was no difference in the physical environment. and potential production. The latter category
There was insufficient time either to investigate includes factors related to farming system, such as
this aspect further or indeed to create the large crop rotation and cultivar. An important feature
database of attributes of the agricultural parts of of the Crop Knowledge Base is that the inference
the regions that would be required to implement engine is able to offer answers to queries about
the utility. The approach allows more flexibility attributes like these where there is no direct solu-
than a simple agroclimatic classification and would tion for the region of interest. As explained in
Section 3.5.3, this is carried out using a mix of ideally be implemented by linking the Crop
205 G. Russell et al. European Journal of Agronomy 11 1999 187–206
Knowledge Base with a Geographic Information improved by using problem-specific information
to decide on the most promising way to proceed System.
As explained earlier, the current system contains at each stage of the search. This has the effect of
targeting the search process towards a goal, avoid- a simplified version of this type of reasoning in
which nearness of one region to another is the ing fruitless paths. Algorithms that use these
heuristic methods already exist Bratko, 1990. main criterion.
Temporal reasoning should be expanded to enable the system to reason with the concept of time such
4.3. Alarm warnings as ‘during’, ‘before’ and ‘after’. The system, for
example, should know that agricultural practice of Whereas some query attributes Fig. 6 are easy
to represent, others, such as those related to factors ploughing takes place before sowing.
At the moment, it is relatively easy to add extra responsible for yield reduction below the potential,
posed considerable problems. One of the difficul- information to an existing template in the knowl-
edge base. However, adding new rules requires ties with representing crop warnings effectively was
that there is very little consistent information about specialist knowledge of Prolog. Further develop-
ment should include utilities, i.e. a knowledge base the threshold levels of environmental factors
beyond which yield losses become significant. This editor, to make these operations easier and to
provide checks on the integrity of the system. It problem is compounded by variation of thresholds
with phenological stage, and the dependence of would be worthwhile developing normalisation
rules, such as those used in database development, the consequences on previous conditions and the
possibility of recovery. Consequently, crop warn- to facilitate making problem-free additions, modi-
fications or deletions to the data Rothwell, 1993. ing information is described in many different ways
in the literature and incorporation into the The data dictionary should also be expanded to
include permitted values for attributes to allow Knowledge Model was not easy. The major limita-
tion is not actually in the programming but in the automatic flagging of dubious entries.
knowledge possessed by experts, which tends to be empirical and location-specific. Authors used a
wide range of ways of describing these effects and, since it had been decided to include information
5. Conclusion