UV-visible spectrometers are instruments that register the absorbance of emitted light by particles suspended in water for several wavelengths and deliver continuous measurements that can be interpreted as concentrations of parameters commonly used to evaluate physico-chemical status of water bodies. Classical parameters that indicate presence of pollutants are total suspended solids (TSS) and chemical demand of oxygen (CDO). Flexible and efficient methods to relate the instruments’s multivariate registers and classical measurements are needed in order to extract useful information for management and monitoring. Analysis methods such as Partial Least Squares (PLS) are used in order to calibrate an instrument for a water matrix taking into account cross-sensitivity. Several authors have shown that it is necessary to undertake specific instrument calibrations for the studied hydro-system and explore linear and non-linear statistical methods for the UV-visible data analysis and its relationship with chemical and physical parameters. In this work we apply classical linear multivariate data analysis and nonlinear kernel methods in order to mine UV-vis high dimensional data, which turn out to be useful for detecting relationships between UV-vis data and classical parameters and outliers, as well as revealing non-linear data structures.