Regression model-based predictions of diel, diurnal and nocturnal dissolved oxygen dynamics after wavelet denoising of noisy time series

Continuous time-series measurements of diel dissolved oxygen (DO) through online sensors are vital to better understanding and management of metabolism of lake ecosystems, but are prone to noise. Discrete wavelet transforms (DWT) with the orthogonal Symmlet and the semiorthogonal Chui–Wang B-spline were compared in denoising diel, daytime and nighttime dynamics of DO, water temperature, pH, and chlorophyll-a. Predictive efficacies of multiple non-linear regression (MNLR) models of DO dynamics were evaluated with or without DWT denoising of either the response variable alone or all the response and explanatory variables. The combined use of the B-spline-based denoising of all the variables and the temporally partitioned data improved both the predictive power and the errors of the MNLR models better than the use of Symmlet DWT denoising of DO only or all the variables with or without the temporal partitioning.