Comparison of parametric and non-parametric estimations of the annual date of positive water temperature onset

摘要: Summary The onset date of positive water temperature in the annual thermal cycle North-American streams is modeled using parametric (regression) and non-parametric (artificial neural networks) approaches. Physiographic, land cover weather-related variables are used to predict for 191 station-years at 48 locations Canada Northern US. Preliminary correlation analysis performed order test relationships between physiographic/land cover/weather onset. Moreover, several different subsets tested as inputs each model type. Artificial networks can a given station-year, its longitude, lake coverage drainage basin, two January–February daily indices, with split-sample validation root mean square error (RMSE) ∼8.8 days. Ordinary least (OLS) regression models allow RMSE ∼9.5 days, station’s latitude, one index. OLS adjusted on canonical variates combining 13 weather achieve prediction performance ∼9.1 days. precipitation does not impact much all models.