Photocatalysis has long been touted as one of the most promising technologies for environmental remediation. The ability of photocatalysts to degrade a host of different pollutants, especially recalcitrant molecules, is certainly appealing. Titanium dioxide (TiO₂) has been used extensively for this purpose. Anodic oxidation allows for the synthesis of a highly ordered nanotubular structure with a high degree of tunability. In this study, a series of TiO₂ arrays were synthesised using different electrolytes and different potentials. Mixed anatase-rutile photocatalysts with excellent wettability were achieved with all the experimental iterations. Under UVA light, all the materials showed significant photoactivity towards different organic pollutants. The nanotubes synthesised in the ethylene glycol-based electrolyte exhibited the best performance, with near complete degradation of all the pollutants. The antibacterial activity of this same material was similarly high, with extremely low bacterial survival rates. Increasing the voltage resulted in wider and longer nanotubes, characteristics which increase the level of photocatalytic activity. The ease of synthesis coupled with the excellent activity makes this a viable material that can be used in flat-plate reactors and that is suitable for photocatalytic water treatment.

The potential application of multivariate three-way data analysis techniques, namely parallel factor analysis (PARAFAC) and discriminant multi-way partial least squares regression (DN-PLSR), on three-dimensional excitation emission matrix (3D-EEM) fluorescent data were used to identify the uniqueness and authenticity of Maltese extra virgin olive oil (EVOO). A non-negativity constrained PARAFAC model revealed that a four-component model provided the most appropriate solution. Examination of the extracted components in mode 2 and 3 showed that these belonged to different fluorophores present in extra virgin olive oil. Application of linear discriminate analysis (LDA) and binary logistic regression analysis on the concentration of the four extracted fluorophores, showed that it is possible to discriminate Maltese EVOOs from non-Maltese EVOOs. The application of DN-PLSR provided superior means for discrimination of Maltese EVOOs. Further inspection of the extracted latent variables and their variable importance plots (VIPs) provided strong proof of the existence of four types of fluorophores present in EVOOs and their potential application for the discrimination of Maltese EVOOs.

The price of honey, as a highly consumed natural product, depends on its botanical source and its production environment, causing honey to be vulnerable to adulteration through mislabeling and inappropriate, fraudulent production. In this study, a fast and simple approach is proposed to tackle this issue through non-target one dimensional zg30 and noesypr1d ¹H NMR fingerprint analysis, in combination with multivariate data analysis. Results suggest that composition differences in sugars, amino acids, and carboxylic acid were sufficient to discriminate between the tested honey of Maltese origin and that of non-local origin. Indeed, all chemometric models based on noesypr1d analysis of the whole fraction honey showed better prediction in geographical discrimination. The possibility of discrimination was further investigated through analysis of the honey’s phenolic extract composition. The partial least squares models were deemed unsuccessful to discriminate, however, some of the linear discriminant analysis models achieved a prediction accuracy of 100%. Lastly, the best performing models of both the whole fraction and the phenolic extracts were tested on five samples of unknown geographic for market surveillance, which attained a high agreement within the models. Thus, suggesting the use of non-target ¹H NMR coupled with the multivariate-data analysis and machine learning as a potential alternative to the current time-consuming analytical methods.

Elemental analysis using energy-dispersive X-ray fluorescence on extra virgin olive oils and seed oils revealed the presence of two major concentration related clusters, one containing elements of pedological origin, whilst the other consisted of heavy metals. Seed oils were found to contain a higher concentration of titanium when compared to extra virgin olive oils, whilst extra virgin olive oils derived from the Maltese Islands had a significantly higher concentration of barium and phosphorus on using the Kruskal–Wallis one-way ANOVA (p-value < 0.05 for both elements). Application of stepwise linear canonical discriminate analysis proved to be highly superior to PCA, as it was able to distinguish between seed oils from extra virgin olive oils and distinguish between foreign and locally produced extra virgin olive oils.

The application of ¹H and ¹³C nuclear magnetic resonance (NMR) in conjunction with chemometric methods was applied for the discrimination and authentication of Maltese extra virgin olive oils (EVOOs). A total of 65 extra virgin olive oil samples, consisting of 30 Maltese and 35 foreign samples, were collected and analysed over four harvest seasons between 2013 and 2016. A preliminary examination of ¹H NMR spectra using unsupervised principle component analysis (PCA) models revealed no significant clustering reflecting the geographical origin. In comparison, PCA carried out on ¹³C NMR spectra revealed clustering approximating the geographical origin. The application of supervised methods, namely partial least squares discriminate analysis (PLS-DA) and artificial neural network (ANN), on ¹H and ¹³C NMR spectra proved to be effective in discriminating Maltese and non-Maltese EVOO samples. The application of variable selection methods significantly increased the effectiveness of the different classification models. The application of ¹³C NMR was found to be more effective in the discrimination of Maltese EVOOs when compared to ¹H NMR. Furthermore, results showed that different ¹H NMR pulse methods can greatly affect the discrimination of EVOOs. In the case of ¹H NMR, the Nuclear Overhauser Effect (NOESY) pulse sequence was more informative when compared to the zg30 pulse sequence, since the latter required extensive spectral manipulation for the models to reach a satisfactory level of discrimination.