, Change and rerun , Quit/store
, New sort criteria
, rules used , Help , Done :
You may reenter the system by pressing C and then following the same procedure as given in the previous section for changing an entry. In that case, the line you wish to change is
1 You prefer to see more information about this expert system
If you do not wish to continue by re-entering the system, you may type and have the option of either stopping or rerunning the system from the beginning.
F. DISPLAY OF RULES
There are at least three ways to see the rules tested for validity by the program:
1. Respond affirmatively to the question Do you wish to have the rules displayed as they are used (Y/N). Unlike the example used in C above, rules that have been found to be true will be displayed as the program proceeds.
2. When indicated by the program information line at the bottom of the paper, you may type WHY to see the rules used. Pressing when a rule is displayed will allow you to see previous rules in the chain.
3. When recommendations are given at the end of the
program, the program information line at the bottom of the page will instruct you to type the line number in order to see the rules used in making that recommendation.
RULE NUMBER; 2 2
(1) You prefer to continue
and (2) Soil order is some Alfisols or Inceptisols or Oxisols
and (3) Soil information: soil great group is known or soil
great group of not known
and (4) Al or Mn toxicity is probable or not likely
and (5) [LIME FACTOR] > 0.5
and (6) [LIME REQ] < 0.3
No lime is recommended. Probability=95/100
Input IF line number for info on how it was derived for reference for the topic, for help or to continue:
Typing the line number in the IF part will allow you see previous rules on the chain. The probability value assigned indicates the likelihood of the solution. For any rule, it is possible to see what reference materials were used by pressing . will give you more detailed information on the program.
G. SITIUNG TRANSMIGRATION AREA, INDONESIA
If in section C Step 12, we had selected 1 soil location is in the Sitiung transmigration area, the procedure would have been somewhat different. Because this program was developed using data from this area, and much more data is available, you are asked the location of the sample rather than soil order or great group. For example,
Soil sample was obtained from
1 Sitiung I (all Blocks)
2 Sitiung II (Blocks A, B or C)
3 Sitiung II (Block D or E)
4 Sitiung IV or V
5 none of the above
The data asked for here is different than for soils not in the Sitiung transmigration area. You will be asked:
Extractable K is
1 less than 0.12-0.15 cmol K/kg
2 greater than 0.12-0.15 cmol K/kg
3 not known
Olsen extractable P is
1 less than 10-12 mg P/kg
2 greater than 10-12 mg P/kg
3 not known
This information is used in determining P and K fertilizer requirement for areas within Sitiung.
EVALUATION OF ACID3B
Attached is a table with 10 test sets of soil conditions. Please enter these into the expert system and note the recommendations made. Your answers to the following questions together with any comments, data, and publications you think appropriate would be very much appreciated. (Responses may be sent to Russell S. Yost, Dept. of Agronomy and Soil Science, 1910 EastWest Road, Honolulu, HI 96822)
1. Do you agree with the predicted lime requirements given for the text soil and crop data? How do these compare with lime requirements you would make using the same data?
2. Do you feel that the interpretations, cautions and informative notes included in the output were sufficient?
3. Do you think the system asks the right questions in the right order? If not, please explain the sequence you would have preferred.
4. Do you have a different recommendation philosophy which you feel would be more appropriate?
5. Under what conditions do you think this system would be most useful? What modifications or additional considerations are important in your situation?
6. Did you have any particular difficulty using this system?
1 2 3 4 5
Soil order Ultisol Oxisol Inceptisol Ultisol Ultisol
Great group Paleudult Haplorthox Sulfaquept Hapludult Hapludult
Crop Willis, soy. rice maize peanut cassava
Depth Incorp 15 10 20 12 10
Bulk Density 1.0 1.1 1.3 0.9 1.1
ECEC 4.0 3.2 4.2 5.5 2.2
EXTRAL 3.2 2.8 13.5 2.5 1.0
Green Manure 0 0 5 10 0
Fine. Factor 0.7 0.8 0.8 0.5 0.6
CaCO3 equiv. 1.1 0.7 0.7 0.8 0.9
Soil pH 5.0-5.5 4.0-5.0 4.0-5.0 5.0-5.5 4.0-5.0
P orK -- -- -- -- -Recommend.:
6 7 8 9 10
Soil order Ultisol Ultisol Ultisol Ultisol Mollisol
Great group Paleudult Paleudult Paleudult Paleudult -Crop soybean soybean soybean cassava soybean
Depth Incorp 15 15 15 15 15
Bulk density 1.1 1.1 1.1 1.1 1.1
ECEC 1.0 4.5 3.0 4.5 4.5
EXTRAL 0.3 3.8 4.5 3.8 3.8
Green manure 0 0 0 5 0
Fine. factor 0.8 0.8 0.9 0.7 0.5
CaCO3 equiv. 1.0 1.0 0.9 0.6 1.1
Soil pH 4.0-5.0 4.0-5.0 5.5-6.0 4.0-5.0 6.0
P and K -- -- -- -- -Recommend.:
A BRIEF SUTMMIARY OF THE KNOWLEDGE BASE
Expert systems are designed to simulate a human expert's approach, knowledge and experience in making a diagnosis or recommendation. ACID3B is a prototype system for making lime recommendations in the humid tropics. The knowledge base is being developed from existing information and research experience. The primary objective, as part of the Tropsoils project, has been to address soil acidity problems in highly weathered soils of Sumatra, Indonesia. We have focused on extractable acidity (mostly exchangeable Al) as the primary cause of yield reduction due to soil acidity. A summary reference for the knowledge base is a review paper by Kamprath (1984). The main concepts in the database are:
1. Growth limiting effects are primarily due to exchangeable Al + H (exchangeable acidity) although if all cations are present in very small quantities some lime is probably needed to provide Ca. It is assumed that toxicity to exchangeable acidity is closely related with Al + H saturation.
2. crops vary considerably in their tolerance to
exchangeable acidity, extremes are represented by mung bean (very intolerant, tolerating no more than 0 % Al saturation) and cassava (very tolerant -- about 75% Al + H saturation.)
3. Organic material seems to reduce lime requirements. The approximation is currently 10 ton/ha of fresh organic material reduces lime requirement by 1 ton/ha.
4. Lime requirements need to be based on soil analyses in order to accurately reflect the soil conditions.
5. Although data is sparse, an attempt is made to determine the approximate effects of lime quality on the lime requirement. Included are the neutralization value relative to calcium carbonate and an estimate of physical reactivity as related to the particle size. The estimate of neutralization value is a relatively well defined laboratory procedure in which an excess of acid is added to the lime and allowed to fully react. The excess acid is back-titrated to estimate the unreacted acid for calculation.
Particle size fractions have been used to estimate the
physical reactivity of the limestone. Here many factors have been discussed and are relevant to field estimation of time required after application before crops can be planted. One of the simpler measurements of lime quality as affected by particle size is given by the estimate of the amount of lime of various particle sizes needed to give approximately equivalent yields.
The calculation of lime requirement is based on the need to neutralize sufficient Al to reduce Aluminum saturation to the 'critical Aluminum saturation' which has been established for the
various crops (Cochrane et al., 1980). Our modified form of the equation is :
Lime requirement (t/ha) = 1.4(exch. acidity (CAS*ECEC/100)
exch. acidity = 1 N KCl extractable Al + H
CAS = critical Aluminum saturation of the crop
ECEC = 'effective cation exchange capacity'
1.4 represents the relation of the cmol of CaCO3 required to
neutralize 1 cmole of Al + H in field studies adjusted for both bulk density and depth of incorporation. In this case 1.9 cmol of Ca was required for each cmol of Al + H, the bulk density was assumed to be 1.0, and the depth of incorporation was assumed to be 15 cm.
Preliminary data suggest approximately 0.53 cmol KClextractable acidity is neutralized for each cmol of Ca added as CaCO3 (Wade et al., 1985). This corresponds to a relation of 1.9 cmol of CaCO3 being required for each cmol of extractable acidity, a value much in keeping with results reported elsewhere (Kamprath, 1985). This reference points out the need to consider the effectiveness of lime in neutralizing the extractable acidity. Such data need to be obtained in field studies if possible because of the need to ensure that one is testing the liming material and soil reactivity in conditions that are representative of the situation or group of farmers to which the eventual recommendation is intended to apply.
The system is designed to apply to the Humid Tropics with
soils of the Ultisol, Oxisol, and Inceptisol orders. In addition, the system has additional information pertinent to the Sitiung region, Indonesia. The general recommendations are based on other relationships such as a general reactivity of 2 cmol of CaCO3 for each cmol of extractable acidity. Levels of critical aluminum saturation are, so far, the same for the general recommendation as for the specific location in Sitiung. Other data and results from the Tropsoils work in Sitiung are incorporated such as minimal requirements of P and K for soybean, rice, cowpea and peanut.
INTRODUCTION TO A RULE-BASED SYSTEM
ACID3B has been developed on a rule-based expert system
development shell, EXSYS. The shell provides editing facilities to design output formats, to run test datasets, and to ensure modifications have not disrupted the core logic flow of the system. The inference engine is backward chaining and provides probability accumulation in dependent, independent, and averaging modes. Only relatively simply WHY capability is provided which displays the rules which are being evaluated in the information input mode or provides the chain of fired rules in support of a recommendation. Programming effort is minimal with this software,
however with the usual loss in flexibility for certain types of expert system construction.
In EXSYS the search procedure follows several relatively
simple rules. These search rules will be dicussed in the sequence in which they operate.
1. Rule selection. The first search or "pattern matching" that is done on the knowledge base begins with the choiceses. Choices in EXSYS are all the potential conclusions from which the system can choose in presenting final results. None, one or several choices are possible with any consultation (or "run" of the system).
The first choice is selected to determine if it can be proved true or false. The search routine determines which rules have this choice in their THEN part. If there are more than one, the first rule (in numerical order) is the one chosen for analysis.
You prefer to continue
The soil great group is Paleudults
No lime is recommended probability (100/100) <- This is the first choice of the list of choices. This choice will cause RULE 11 to be selected first for analysis.
2. Condition selection. Once the rule is selected EXSYS then proceeds to analyze components of the rule -- the "conditions". In EXSYS a condition is composed of two parts: the "Qualifier" and the "Value". After selecting which rule to analyze EXSYS then determines which "condition" to analyze. This selection is quite logical -- the first condition in the IF part of the rule is selected for evaluation. The purpose of the evaluation is to determine if the condition is true or false. The system determines if the condition is true by first searching the file of facts or input already concluded to be true, if the condition is not in the list of facts, then the system searches rules which have the condition in their THEN parts. If the system finds a rule with the condition in the THEN part, it then determines whether the conditions in the IF part of that rule are true or can be concluded to be true from other rules by chaining. If our first condition can be determined to be true then the system will proceed otherwise it will, as a last resort, display the first condition in the rule that had our first condition in its THEN part. These conditions will be displayed in the form of the qualifier and values, asking the user to indicate which ones are successive conditions. For example:
IF: (qualifier) (value) (search order)
You prefer to continue <- first searched
The soil great group is Paleudults <- second searched
No lime is recommended probability (100/100) <- This is the first choice of the list of choices. This choice caused this rule to be selected first for analysis.
If you have use this system before and you don't need to see details of this system. <-This condition will be searched
for and, if not found, will be
displayed for user selection
You prefer to continue <-This condition matches the condition in Rule 11 and is
why RULE 22 was selected. (This relationship between rules is
also known as chaining.)
The search begins with the selection of a rule to analyze. The first "Choice" in the list of Choices is selected. A rule which has this choice in its THEN part is then selected. If there is more than one rule with the choice in its THEN part, the rule that has the smallest rule number will be selected. The IF part of the selected rule will then be looked at. The first condition will be selected and determined whether it can be proved true or false from information already determined to be true. If the condition cannot be determined to be true or false based on information already in the system, the system will put the condition on screen and ask the user directly which combination of qualifiers and values are true.
The order of the conditions in the first selected rule will determine the order of the questions asked of the user. For example, one would prefer that the most likely conditions be asked first. This can be done by placing the most commonly selected choice as the first in the choice list (Choice 1). This order should match the "directed graph" or decision tree that we recommend you first construct to document the logical organization of the expert system. In other words, the first condition in the rule should ask the most general information. This will ensure that the system will reduce the number of rules to be searched in subsequent steps. If a very specific condition were placed first, it might ask for information that is irrelevant to most of the system.