DOI: 10.1016/J.ECOINF.2013.02.003
关键词: Statistics 、 Random forest 、 Resolution (electron density) 、 Relative species abundance 、 Mathematics 、 Image resolution 、 Species distribution 、 Sample (statistics) 、 Modifiable areal unit problem 、 Grain size
摘要: Abstract Spatial resolution and zoning affect models predictions of species distribution models. I compared grain sizes 90 m grid cells to ecological units soil polygons (approximately 209 ha composed discontinuous 16 ha), then introduced error into samples examined influence topographic variables. used random forests, which is a machine learning classifier, open access data. Predictions based on were slightly more accurate than coarser-sized polygons, particularly false positive rates (mean values 0.11 0.16, respectively). The trade-off for accuracy was the number mapping required increase resolution. Probability presence decreased with Similarly size comparisons, affected probability prediction. Unlike relationship between count each (i.e., relative abundance) area predicted as present lost addition error. Introduction absences modeling sample presences through plot location increased introduction use background pseudoabsences presence. Finer amplified effect absences; reduced by factor 5.4 1.4 polygons. choice fine or coarser shaped resulted in different models, due varying variables Use (tens hundreds hectares) may be worthwhile exchange greater spatial extent ecologically zoned appeared avoid modifiable areal unit problem.