A data mining approach to improve multiple regression models of soil nitrate concentration predictions in Quercus rotundifolia montados (Portugal)

摘要

The changes in the soil nitrate concentrationwere studied during 2 years in a ‘‘montado’’ecosystem, in the South of Portugal. Total rainfall,air and soil temperature and soil water content underand outside Quercus rotundifolia canopy were alsoevaluated. A cluster analysis was carried out usingclimatic and microclimatic parameters in order tomaximize the intraclass similarity and minimize theinterclass similarity. It was used the k-Means ClusteringMethod. Several cluster models were developedusing k values ranging between 2 and 5. Thereafter, ineach cluster, the data were divided according to theirorigin (soil under canopy and open areas, and fromsurface and deep layers). Multiple regression modelswere tested for each cluster, to assess the relationshipbetween soil nitrate concentration and a set of climaticand microclimatic parameters and the results werecompared with models assessed without clustering.The models achieved with data grouped in result ofclustering analysis showed better performance thanthe models achieved without clustering, mostly fordata from open areas soils. When temperature is lowand/or water presents excess or scarcity levels, thedata from soils in undercanopy areas, give rise tomodels with worst performance than models fromopen soil areas data. The results obtained for undercanopyarea suggest that nitrification process in soilunder Quercus rotundifolia trees influence is morecomplex than for open areas, and subject to otherrelevant factors beyond water and temperature.