作者: Ryan A. McManamay , Mark S. Bevelhimer , Shih-Chieh Kao
DOI: 10.1002/ECO.1410
关键词: Computer science 、 Data mining 、 Outlier 、 Sample size determination 、 Random forest 、 Ecohydrology 、 Curse of dimensionality 、 Cluster analysis 、 Class (biology) 、 Hydrology 、 Streamflow
摘要: Hydrologic classifications unveil the structure of relationships among groups streams with differing stream flow and provide a foundation for drawing inferences about principles that govern those relationships. classes template to describe ecological patterns, generalize hydrologic responses disturbance, stratify research management needs applicable ecohydrology. We developed two updated continental US using streamflow datasets varying reference standards. Using only reference-quality gages, we classified 1715 gages into 12 across US. By including more (n=2618) in separate classification, increased dimensionality (i.e. classes) distinctiveness within regions at expense decreasing natural standards quality). Greater numbers higher regional affiliation our compared previous classification (Poff, 1996) suggested level variation resolution was not completely represented smaller sample sizes. Part utility systems rests their ability classify new objects analyses. constructed random forests predict class membership based on indices or landscape variables. In addition, an approach assessingmore » potential outliers due alteration assignment. Departures from disturbance take account multiple simultaneously; thus, can be used determine if disturbed are functioning realm hydrology.« less