A rule-based vs. a set-covering implementation of the knowledge system LIMPACT and its significance for maintenance and discovery of ecological knowledge

The knowledge system LIMPACT estimates the pesticide contamination of small lowland streams within agricultural catchment areas. The system considers the abundance of 39 macroinvertebrate taxa during four time frames (T1: March/April, T2: May/June, T3: July/August and T4: September/October) within a year. The four diagnoses Not Detected (ND), Low (L), Moderate (M) and High (H) pesticide contamination represent a calculated annual toxic sum without any specification of the chemical agents. In this paper, we present a new model-based implementation of the existing knowledge system LIMPACT using set-covering relations including diagnosis exclusions. This type of knowledge base outperforms the former rule-based implementation in size and complexity, knowledge acquisition costs and explanatory characteristics. We were able to extract a common and average appearance of taxa in the specific group of streams. A wide range of common taxa with a tendency to more taxa in less severely contaminated streams was observed. Only a few taxa indicate exclusively a specific contamination class. For the exclusion conditions there was a clear trend for more taxa to exclude streams in the High pesticide contamination category than in the other classes.

[1]  G. Friedrich.  Eine Revision des Saprobiensystems , 1990 .

[2]  Joachim Baumeister,et al.  Diagnostic Reasoning with Multilevel Set-Covering Models , 2002 .

[3]  J. Wright,et al.  River classification using invertebrates: RIVPACS applications , 1998 .

[4]  M. Liess,et al.  DETERMINATION OF INSECTICIDE CONTAMINATION IN AGRICULTURAL HEADWATER STREAMS , 1999 .

[5]  R. Norris,et al.  Design and implementation of rapid assessment approaches for water resource monitoring using benthic macroinvertebrates , 1995 .

[6]  C. M. Cooper,et al.  Biological Effects of Agriculturally Derived Surface Water Pollutants on Aquatic Systems—A Review , 1993 .

[7]  Joachim Baumeister,et al.  Erratum to “ An expert system to estimate the pesticide contamination of small streams using benthic macroinvertebrates as bioindicators Part 2 : The knowledge base of LIMPACT ” [ Ecological Indicators 2 ( 2002 ) 239 – 249 ] , , 2003 .

[8]  J. Baker,et al.  Group report: Physiological and ecological effects of acidification on aquatic biota (Chapter 19). Book chapter , 1994 .

[9]  B. Wachs.  A QUALITATIVE CLASSIFICATION FOR THE EVALUATION OF THE HEAVY METAL CONTAMINATION IN RIVER ECOSYSTEMS , 1998 .

[10]  C. Steinberg,et al.  Acidification of freshwater ecosystems: Implications for the future , 1994 .

[11]  Susanne Ziegler,et al.  Wissensbasierte Diagnosesysteme im Service-Support , 2001 .

[12]  Frank Puppe.  Knowledge reuse among diagnostic problem-solving methods in the Shell-Kit D3 , 1998, Int. J. Hum. Comput. Stud..

[13]  Ralf Schulz,et al.  A field study of the effects of agriculturally derived insecticide input on stream macroinvertebrate dynamics , 1999 .

[14]  E.T.H.M. Peeters,et al.  New methods to assess the ecological status of surface waters in The Netherlands. Part 1: Running waters. , 1994 .

[15]  D. Dudgeon,et al.  The impact of agricultural runoff on stream benthos in Hong Kong, China. , 2002, Water research.

[16]  Reid Simmons,et al.  Second Generation Expert Systems , 1993, Springer Berlin Heidelberg.

[17]  R. Schulz,et al.  Linking insecticide contamination and population response in an agricultural stream , 1999 .

[18]  Frank Puppe,et al.  Incremental Development of Diagnostic Set-Covering Models with Therapy Effects , 2003, Int. J. Uncertain. Fuzziness Knowl. Based Syst..