Understanding and predicting Taste and Odour events is as difficult as critical for drinking water treatment plants. Following a number of events in recent years, a comprehensive statistical analysis of data from Lake Tingalpa (Queensland, Australia) was conducted. Historical manual sampling data, as well as data remotely collected by a vertical profiler, were collected; regression analysis and self-organising maps were the used to determine correlations between Taste and Odour compounds and

more »

... ential input variables. Results showed that the predominant Taste and Odour compound was geosmin. Although one of the main predictors was the occurrence of cyanobacteria blooms, it was noticed that the cyanobacteria species was also critical. Additionally, water temperature, reservoir volume and oxidised nitrogen availability, were key inputs determining the occurrence and magnitude of the geosmin peak events. Based on the results of the statistical analysis, a predictive regression model was developed to provide indications on the potential occurrence, and magnitude, of peaks in geosmin concentration. Additionally, it was found that the blue green algae probe of the lake's vertical profiler has the potential to be used as one of the inputs for an automated geosmin early warning system. Environments 2016, 3, 22 2 of 14 compounds [2, 12] . However, the production of metabolites is related to the species and strain of the cyanobacteria blooming [13] , and there is still large uncertainty related to which species can produce these compounds, since newer studies often prove older studies wrong (as explained in [14] ). During a bloom, different species and strains are in competition and interacting with each other, through a number of nonlinear behaviours determined by factors, such as nutrient availability, presence of grazing zooplankton, or physical factors [5] . For example, in Hinze dam (South-East Queensland, Australia), Uwins et al. [11] reported steadily increasing geosmin concentration following an early spring bloom in Anabaena sp. Following the bloom, and decay of precipitating cells, geosmin was released. Actinobacteria on the other hand, despite potential for contributing to T&O compounds release, were inhibited in this production by high water temperature, high dissolved oxygen and low phosphorus levels. As a result of this complexity and high uncertainty, certain variables which are surrogate estimators of algal counts, such as chlorophyll-a, have been correlated with geosmin both positively, i.e., high chlr-a levels linking to high geosmin concentrations (e.g., [15] although based on only few data points; [4, 16] ), and negatively, with higher geosmin levels measured where lower chlr-a was detected (e.g., [1]). In general, the presence of geosmin and MIB, which in many cases is linked to algal blooms, has been correlated to a large number of possible predictors. Aside the already mentioned chlr-a, also the sum of green algae [4], regardless of the species, sometimes proved to be a good predictor. In that particular study however, single species and strains were not measured, thus it is unknown if geosmin was caused by the same specific type of algae, or if different ones result in similar geosmin concentrations. Other possible predictors include nutrients such as nitrogen and phosphorus [4, 10, [17] [18] [19] , as well as metallic micronutrients, such as copper or manganese [4, 17] . Other critical factors proved to be water temperature [17, 19, 20] , light intensity [17, 21] , turbidity and water clarity [9,10], dissolved oxygen [7], rainfall [11] and oxidation-reduction potential [4] . The importance of light availability is related to the energy that light provides to enable photosynthetic fixation of dissolved inorganic carbon, which can be subsequently routed into the cellular synthesis of geosmin [1] . Additionally, although algal blooms, and thus T&O events, typically occurred in warm stratified seasons, some studies [22] also proved how high geosmin levels can be detected during lake circulation periods; interestingly, other studies [23] also found how low, instead of high, temperatures can stimulate the production of geosmin and their accumulation in cells due to lower chlr-a demand, although high temperature or optimum light intensity would be necessary for more intracellular geosmin release. This study aims to exploit historical sets of relevant data, and use cutting-edge data analytics to better understand, and model, the occurrence of T&O events in a relatively shallow, subtropical reservoir in Australia. As already mentioned, there is large uncertainty around T&O events, with the understanding and prediction of such events being site and season specific. Therefore, a full statistical analysis of the available historical data was performed to gain a specific understanding of the behaviour of the reservoir of interest. Given the relatively large amount of data, and recent advancements in the hydroinformatics field, it was possible to identify potential predictor variables of T&O events. Based on the correlations found, a simple statistical model was also developed which enables a prediction of the magnitude of possible future high geosmin concentrations. Despite limited by the number of historical events available for analysis and the complexity of the system, the model can assist water treatment operators for an improved understanding and preparedness towards geosmin peak events; the results of this analysis also provide an example of potential geosmin production behaviour in similar reservoirs. Environments 2016, 3, 22 3 of 14 Materials and Methods Study Location Lake Tingalpa (Figure 1) , bounded by Leslie Harrison dam, is located about 20 km south-east of Brisbane, in Queensland (Australia). The main purpose of the reservoir is to supply water to the Redland City, which has around 150,000 inhabitants. On average, it supplies 20% of the water demand in this area. The main inflow to the reservoir is Tingalpa Creek. The 535-m-long dam wall was completed in 1984, and the reservoir could supply 24,868 ML at full capacity; however, its capacity decreased to 13,206 ML on Friday, 1 August 2014, due to safety concerns with the integrity of the wall. The catchment area covers 87 km 2 . An intake tower allows the withdrawal of raw water, which is redirected to the Capalaba water treatment plant (WTP).