Applications of Machine Learning to Ecological Modelling

1
Applications of Machine Learning to Ecological
Modelling

• Saso Dzeroski
• Jozef Stefan Institute
• Ljubljana, Slovenia

2
Ecological modelling and machine learning

• The goals of modelling include
• understanding the domain studied
• predicting future values of system variables of
  interest
• decision support for environmental management
• Machine learning can be used to
• automate modelling
• discover knowledge that meets some or all of the
  above goals

3
Analysis of water quality data

• Biological classification
• British rivers
• Slovenian rivers
• Predicting chemical parameters of water quality
  from bioindicator data
• British rivers
• Slovenian rivers
• Determining ecological requirements
  of some organisms in Slovenian rivers

4
Modelling

• Modelling algal growth
• Lagoon of Venice
• Lake of Bled
• Modelling phytoplankton growth
• Modelling a red deer population

5
Environmental applications of machine learning

• Analysis of the influence of environmental
  factors on respiratory diseases
• Analysis of the influence of soil habitat
  features on the abundance of Collembola
• Predicting biodegradability of chemical
  compounds
• Runoff prediction from rainfall and past runoff

6
A regression tree for predicting algal growth in
the Venice lagoon
7
Rules for classifying British Midland rivers into
quality classes based on the community of
benthic macroinvertebrates

• IF Hydrobiidae lt 3
• AND Planorbidae lt 0
• AND Gammaridae lt 5
• AND Leuctridae gt 0
• THEN Class B1a 42 0 0 0 0
• IF Asellidae gt 2
• AND 0 lt Gammaridae lt 4
• AND Scirtidae lt 0
• THEN Class B2 0 0 41 0 0

IF Planariidae lt 0 AND Tubificidae gt 0
AND Lumbricidae lt 0 AND Glossiphoniidae lt 2
AND Asellidae gt 0 AND Gammaridae lt 0 AND
Veliidae lt 0 AND Hydropsychidae lt 0 AND
Simulidae lt 0 AND Muscidae lt 0 THEN Class B3
0 0 3 28 10
8
Rate of change equation for phytoplankton growth
in Lake Glumsoe, Denmark

• Variables in the model are the concentrations of
• phytoplankton phyt
• zooplankton zoo
• soluble nitrogen nitro
• soluble phosphorus phosp
• water temperature temp

9
Analysis of environmental data with machine
learning methods 22-25 April 2002, Ljubljana

• http//www-ai.ijs.si/SasoDzeroski/aep/
• Introduction to machine learning
• and its environmental applications
• Data mining and knowledge discovery

10
Contents of course

• Induction of decision and regression trees
• Induction of classification rules
• Bayesian classification
• Nearest neighbor classification
• Evaluating, selecting and combining classifiers
• Equation discovery
• Practical hands-on exercises on environmental
  datasets
• Applications of machine learning to environmental
  problems

11
Recent applications
(joint work with participants from previous
seminars)

• Topics considered at workshops
• Modelling a red deer population (data cleaning,
  body-weight model for calves of the year,
  two year olds and hinds)
• Influence of environmental and social factors on
  acute respiratory diseases in children
• Influence of various parameters on alkalinity of
  an artificial lake near an ashes dump
• Modelling the transport of concrete through pipes

12
Recent applications
(joint work with participants from previous
seminars)

• Habitat-suitability modelling (using GIS data and
  animal locations - sightings/radio-tracking)
• red deer (Debeljak et al. 1999)
• brown bears (A. Kobler and M. Adamic 1999) used
  to identify locations for wildlife bridges across
  highways
• Influence on concentrations of dissolved reactive
  phosphorus in surface runoff from arable land
  (Weissroth and Deroski 1999)
• Diagnosis of a waste-water treatment plant
  (Deroski and Comas 1999)