Premium
WEEDER: an Advisory System for the Identification of Grasses in Turf
Author(s) -
Fermanian T. W.,
Michalski R. S.
Publication year - 1989
Publication title -
agronomy journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.752
H-Index - 131
eISSN - 1435-0645
pISSN - 0002-1962
DOI - 10.2134/agronj1989.00021962008100020034x
Subject(s) - digitaria sanguinalis , agrostis , perennial plant , biology , agronomy , identification (biology) , weed , botany , poaceae
To effectively control weeds found in a turf it is first necessary to correctly identify them. A computer program, WEEDER, was built using the artificial intelligence system AGASSISTANT to provide a means for effectively identifying grass weed and turf species through the recognition of selected variables. WEEDER has a rule‐based, non‐hierarchical knowledge base concerning 37 grass species commonly found in turfs throughout the USA. Each species is represented by 11 or fewer variables. In order to measure the value of WEEDER for identifying unknown grasses in comparison to a commonly used method, the dichotomous identification key, 41 volunteers were assigned to one of two groups; (i) those with any previous experience in plant diagnosis or any formal training in plant science; and (ii) those with no experience or training. Each individual identified four unknown grasses; creeping bentgrass ( Agrostis palustris Huds.); perennial ryegrass ( Lolium perrene L.); zoysiagrass ( Zoysia japonica L.); and large crabgrass ( Digitaria sanguinalis [L.] Scop.) using WEEDER or a printed identification key. The maximum mean of either group to identify a grass species was 55% of the specimens, which were examined by participants with plant science training using WEEDER. Participants with some plant science training had a higher mean identification of each species (23% identified) than participants with no training (18%) when using the identification key. Little difference in their ability to identify the unknown species was found between the two groups when they were using WEEDER. There was a significant increase in the mean ability of all participants to identify an unknown grass using WEEDER (50%) rather than the identification key (20%) after rules for the four species were modified. A demonstrated advantage of WEEDER over the printed key was its ability to be easily modified to increase its usefulness. The mean percentage of correctly identified grasses by all participants increased from 11 to 50% after rules pertaining to the unknown grasses were modified to reflect variable values most consistently selected. No significant dependency on a participant group was found for correctly identifying a grass species when using WEEDER after the rules were modified. Further testing of WEEDER is required to determine if the modified rules are consistent with additional grass sample and user populations.