Spatiotemporal modeling of saturated dissolved oxygen through regressions after wavelet denoising of remotely and proximally sensed data

A few limnological studies have attempted to combine wavelet denoising, time- and space-series data, and regression models for the purposes of reliable predictions and reconstructions of time-space variations. This study was focused on median and linear regression models of saturated dissolved oxygen (DO _sat ) after denoising by using discrete wavelet transform (DWT) with Chui-Wang B-spline and Coiflet wavelets and was based on remotely and proximally sensed noisy time series during 144 days. The aim was to explore effects on predictive accuracies of (1) applying multiple median or linear regression models after DWT denoising with the orthogonal Coiflet or the semiorthogonal Chui-Wang B-spline to proximally and remotely sensed noisy data, (2) adding spatial heterogeneity, and (3) including more explanatory variables. Multiple linear regression (MLR) models after DWT denoising with Chui-Wang B-spline performed better in elucidating spatiotemporal DO _sat dynamics than median regressions and MLR models denoised with Coiflet. The best agreement between measured and predicted values based on an independent validation dataset was obtained by a median regression model and by a MLR model after DWT denoising with Chui-Wang B-spline for spatially homogeneous or heterogeneous DO _sat estimates, respectively. Spatiotemporally increased predictive capabilities of the wavelet-augmented regression models can yield more realistic estimates, thus further bridging the gap between public policies and environmental models in the process of decision-making.