ABSTRACT
A time series is composed of a collection of samples taken at discrete time intervals from an intrinsically continuous process. Environmental time series may be acquired either manually or via automatic equipment. The analysis of environmental time series could be more easily carried out by restricting the scales to powers of 2, thus performing a dyadic signal decomposition, without loss of accuracy. Noise affects data accuracy and may masks important information, essential for modelling, calibration, and forecasting. Further, noisy signals cannot be numerically differentiated, and this may become a serious limitation whenever the information we are seeking is contained in the data time derivative. Data smoothing prior to numerical differentiation is essential, and deals two smoothing techniques, based either on spline approximation or on wavelet filtering. Smoothing is a numerical technique to decrease the data roughness and represents a simple and effective way of reducing the unwanted disturbance in the data, especially in anticipation of their numerical differentiation.
