Cyanobacterial harmful algal blooms are a critical risk for public health since the release of cyanotoxins can reach hazardous levels in drinking water. In this study, a cost-effective monitoring programme has been developed and applied by a Spanish water operator to detect cyanobacteria presence in a drinking water plant reservoir; a total of 65 phytoplankton genera and seven cyanobacterial taxa were identified. Based on these monitoring data, different analysis techniques such as classification trees have been used to identify the potential situations of risk and to forecast cyanobacteria evolution in surface water. As a result, a new operational risk management plan has been developed and applied by the water operator since 2019. The main improvements concern the classification of the situations of risk in six groups (from low to very high, according to chlorophyl-a and cyanobacteria concentrations), the planning of the actions required under each situation (monitoring, management practice and treatment), and the use of the risk forecast to perform some of these actions from 2–3 days in advance. The methodology can be replicated to other case studies to facilitate the management of cyanobacteria risk and comply with the requirements of the revised European Drinking Water Directive adopted in 2020.
A robust forecasting of cyanobacteria risk in reservoir was developed.
An upgrade of the operational cyanotoxin risk management procedure was performed.
The risk control methodology can be replicated to other drinking water treatment plants.
The production of drinking water from surface water needs appropriate risk control measures to avoid impacts of toxic cyanobacteria to human health. While the presence of cyanobacteria occurs naturally in lake and rivers, concentrations hazardous to human health are usually due to human activity in the catchment (Chorus & Welker 2021). In such conditions, the toxins that they can produce could become carcinogenic and even lethal for humans (He et al. 2016). Considering that cyanobacteria and cyanotoxins occur worldwide in surface water used as sources of drinking-water (Chorus & Welker 2021) it is necessary to implement an holistic risk-based approach that allows to identify the potential risks of toxins release and develop strategies to mitigate them at different stages (avoid favourable conditions in the catchment or in the storage reservoir, adapt water treatment, etc.).
In the catchment, favourable conditions promoting cyanobacteria growth include sunlight, high nutrient levels (phosphorous (P) and nitrogen (N), low turbulence and warm weather (World Health Organization 2017). The risk of elevated cyanobacterial biomass is generally higher under eutrophic conditions (visible algal blooms), but this is too restrictive for a good understanding of toxin risks since certain cyanobacteria such as Planktothrix rubescens tend to decrease under eutrophic conditions (Ibelings et al. 2021). Cyanobacteria can both develop suspended in the water (planktonic cyanobacteria) and grow on surfaces such as rocks, sediments, or submersed vegetation (benthic cyanobacteria); they compete with other phytoplankton and with each other for their necessary resources (light energy and nutrients). In addition, several subsets of cyanobacteria have a few specific traits that favour their dominance over algal phytoplankton as well as bloom formation such as the capability to fix atmospheric nitrogen or the regulation of buoyancy, for example in buoyant taxa like Microcystis (Ibelings et al. 2021). The development of cyanobacteria is not restrained to rivers and lakes, since they can also develop in some drinking water treatment plant (DWTP) processes such as washing waters or sludge (Almuhtaram et al. 2018).
The critical issue for the protection of public health is whether concentrations are likely to exceed hazardous levels at points of human exposure (Chorus & Welker 2021). In the European Union, national drinking-water legislation is abided by the Drinking Water Directive, which in its recent revised version (EU Council 2020) follows the provisional WHO guideline value of 1 μg/L for total microcystin-LR (free plus cell-bound) to assess whether the concentration in treated drinking water would be qualified as ‘hazardous’ (Li et al. 2017). The guideline value is provisional, as it covers only microcystin-LR, the family of cyanotoxins more widely distributed and better known (World Health Organization 2017). In Spain, where 67% of the drinking water come from surface water (Spanish Water and Wastewater Association 2016), the Spanish national legislation (Royal Decree 140/2003), determines since 2003 threshold of 1 μg/L in treated water for microcystin. This monitoring of microcystin (Microcystin-LR) starts when there is a risk of algal bloom in the water source; this requirement is now also considered in the revised drinking water directive (EU Council 2020).
The monitoring and forecasting of algal blooms, cyanobacterial biomass and toxins concentration, and especially in a climate change context, has been the subject of different studies in the past years. Several studies suggest a plausible increase in algal biomass because of temperature increase (Visser et al. 2016) and a reduction in the time of species generation resulting in cyanobacteria becoming more competitive (Wells et al. 2020). An example assessment of the potential for high biomass of cyanobacteria based on environmental conditions is presented in the WHO technical brief on cyanobacteria management for drinking water (World Health Organization 2015b); the potential is estimated from low to high according to the values of total phosphorus, water residence time, pH, transparency and temperature. However, to derive practical action to control risk at DWTP this type of assessment should be adapted to the specific conditions that promote the growth of cyanobacteria in one area and to the available data.
The release of toxins associated to a bloom of cyanobacteria would occur in some specific situations. Release of toxins into water can occur during cell senescence and death or in other circumstances such as the competition between species in specific environmental conditions (allelopathy in sudden nutrient limitation) (Merel et al. 2013). Cyanobacteria death can be linked to the disturbance caused by heavy rains, a lack of nutrients or space during a dry season, or a sudden change in temperature conditions (Merel et al. 2013). However, the drivers of toxins release are complex and not sufficiently known by DWTP managers to derive practical actions.
Regarding the monitoring methods, the pigment chlorophyll-a (Chl-a) is a widely used and accepted measure of total phytoplankton biomass and can be useful to detect algal blooms caused mainly by cyanobacteria (Chorus & Welker 2021). If cyanobacteria are suspected, species identification and cell count by microscopic observation are recommended (World Health Organization 2015b); it is also common that the phycobilin group of pigments is used as cyanobacteria species biomarkers. However, the monitoring methods are not sufficiently applied and standardized in most DWTPs and catchment to provide precise information to better manage risk.
An initiative started in 2003 at La Contraparada DWTP (Aguas de Murcia, Murcia, Spain) to determine the dynamics of phytoplankton communities, to detect microcystin, and to define control measures. The results demonstrated the occurrence of cyanobacteria in the upstream water bodies, which produced a small but stable amount of microcystin in raw water (Hurtado et al. 2008). Operational risk control measures were determined at that time, consisting principally in the monitoring of surface water (including the DWTP regulation reservoir and an upstream reservoir in the catchment), the control of the water level of the DWTP reservoir to avoid the development of benthic communities (fundamentally cyanobacteria), and the study of adapted treatment processes to eliminate potential toxins. The main limitations were the limited forewarning of the strategy and the lack of information on the effectiveness of the measures to be able to tune them.
From 2016 to 2019 the H2020 IMPREX Project (Improving PRedictions of Hydrological Extremes) (van den Hurk et al. 2016) was developed to ‘take better water management decision now to be prepared for the future’. In the project, the technical center CETAQUA coordinated the Spanish case studies and the data analysis for La Contraparada DWTP, and the company AQUATEC provided technical assistance in algal monitoring to EMUASA that managed La Contraparada DWTP. The work carried out at La Contraparada DWTP facilities aimed at addressing the current gap in knowledge and by such addressing the current limitation of the risk control in place: (1) Improve the monitoring program and complement the monitoring data gathered in the last years to achieve a better monitoring cost-efficiency and to obtain a large dataset; (2) Use adapted data analysis method to understand better the factors driving cyanobacteria development in the case study, and especially the impact of climate change; (3) Improve the risk management protocol for the DWTP and propose recommendations for other DWTPs.
La Contraparada DWTP is in the Segura River Basin (Murcia, SE Spain) and receives water from the Segura River and the water transfer from the Tajo River through a complex system of infrastructures. Its principal component is the Ojós Reservoir, from which proceed the main canal of the left riverbank, the intake pipe from this canal and the regulation reservoir adjacent to La Contraparada DWTP (Figure 1).
Ojós Reservoir, with a capacity of 2.8 hm3 and a maximum water level of 15 m, supplies water for agricultural use as well as for human consumption in different towns in the provinces of Murcia, Alicante and Almería. Nutrient load to the dam is quite limited (average nitrate of 2.7 mg/L in 2018) but the climatic condition is quite favorable to algal development with high temperature and insolation (average water temperature of 17.4 °C in 2018), resulting in recurrent algal bloom.
La Contraparada Reservoir, which has a total capacity of 175,000 m3 and a maximum height of 8 m, presents an average water retention time ranging from 5 to 10 days. La Contraparada Reservoir was built in 1996 to avoid fluctuations in water flow from Tajo-Segura River Transfer. Algal development in the La Contraparada Reservoir is limited (value Chl-a generally inferior to 20 μg/L during 2016–2018).
La Contraparada DWTP has a nominal treatment capacity of 2000 m3/h, which supplies the city of Murcia and its municipal districts. It has a daily average flow of around 1000 m3/h. The DWTP generally operates during a continuous period of nine months (January–September) and is stopped when the water concession is reached or during maintenance periods (usually October–December). A stop in the DWTP means for the reservoir an increase in the retention time and a certain stratification of water in the deep zone. The reservoir water is mixed with new water before each start-up of the treatment plant. The treatment process consists of pre-ozonation (1.17 mgO3/L, contact time 1.32 min), optional pre-chlorination (depending on the raw water quality), coagulation/flocculation with aluminum sulfate (59–95 ppm) and polyDADMAC (1.2–2.5 ppm), sedimentation, post-ozonation in two steps (doses 1.46 and 0.49 mgO3/L, and contact times 2.34 and 2.1 min, respectively), GAC filtration by means of 12 automatic filters placed in parallel and a final chlorination (chlorine gas added in watery solution). Molecular ozone, generated in the plant from liquid oxygen, is used as a principal oxidizing agent guaranteeing the disinfection and the absence of toxins in the water treated with 0.4 mg O3/L of residual ozone. The sludge coming from the sedimentation enters the sludge treatment line of the DWTP.
The intake from the left riverbank canal, the La Contraparada Reservoir and the DWTP are all operated by EMUASA, the Murcia Municipal Water Supply System Operator (Empresa Municipal de Abastecimiento y Saneamiento de Murcia, S.A-Aguas de Murcia).
Sampling and sample preparation
During the IMPREX project (2016–2018), analyses and tests were performed to increase knowledge on algal development and to update the monitoring campaign procedure to be more cost-effective. Diverse sampling points were selected at the Ojós Reservoir, the inflow point of La Contraparada Reservoir, the surface of the La Contraparada Reservoir, the raw water (inflow of the treatment process) and drinking water (final water). Two types of sampling methods were carried out in La Contraparada Reservoir, first a surface monitoring of the algal community every 10 days approximately, to analyse algal pigments, concentrations and biovolume, and secondly a vertical profile sampling of temperature and dissolved oxygen in the water column every three months on average. In addition, a total of 11 samples were taken from the upstream Ojós Reservoir.
Three litres of water were taken with a Van Dorn bottle at every sampling point and were then transported to the laboratory and kept in the dark under refrigerated conditions. The water samples for the detection of cyanotoxins were frozen.
From each of the 3-L samples the following subsamples were obtained: 500 mL for Chl-a determination, 500 mL for phycobilin determination, 500 mL for genus determination and concentrations and biovolume calculation, 500 mL for physicochemical analyses, and 1000 mL for cyanotoxins analyses.
Ammonium nitrite and soluble reactive phosphorus (SRP) were quantified by means of spectrophotometry (Thermo ScientificTM Evolution 201 UV/vis), whereas nitrate was determined using ionic chromatography (Metrohm 881 Compact IC pro). The amount of organic matter was quantified via oxidation by potassium permanganate.
The pH was analyzed by parametric probe (CRISON GLP22), such as electrical conductivity (CRISON GLP 31). Dissolved oxygen and oxygen saturation were analyzed by oximeter (CRISON OXI330), and turbidity was quantified by turbidimeter (HACH 2100N).
To analyze the phytobenthos, 4 cm2 samples were taken from the biofilm that carpets the walls of La Contraparada Reservoir, and in Ojós Reservoir the scratching of the mat was done on the banks free of vegetation.
The calculation of the algal concentration was carried out in a Sedgewick-Rafter counting chamber in an optical microscope (Olympus) once the sample had been passed through a 0.7 μm pore fiberglass filter (Filter Lab MFV5), which was centrifuged at 1000 rpm for 3 minutes and finally fixed in 4% formaldehyde. The algal biovolume was calculated following the indications of TAXAGUA (the taxonomic thesaurus commonly used in Spain).
Chlorophyll-a extraction was performed following acetone-mediated extraction and determination of the optical density (absorbance) of the extract by means of a spectrophotometer. For phycobilins (phycocyanin, allophycocyanin and phycoerythrin), the phosphate buffer-mediated pigment extraction procedure with spectrophotometric determination was used.
The quantification of the dissolved microcystin in the water samples and in the benthic mats was performed by immunological test, specifically ELISA ADDA (ENZO LIFE SCIENCES MICROCYSTIN-ADDA ELISA), with a detection level of 0.10 ppb in water and 0.10 μg/g dry weight in benthos.
Due to the low algal development in the La Contraparada Reservoir and the influence of different management factors (intake control, great variation in retention time), the use of a data-driven model was preferred instead of the use of dynamic models (the latter are more adapted to simulate variation of algal with higher range and driven by nutrients input). Several factors, represented by variables, can provide some explanation on algal development, cyanobacteria presence and resulting toxins. The purpose of the data analysis is to find these key predictor variables from the ones that are easily accessible to the water operators. By doing so, regular predictions can be provided to support decision-making. In a first step, empirical knowledge is used to select the variables that are potentially the best predictors of algal development and cyanobacteria risk in the area (supplementary material –Table S1) and an overview of the important processes is drawn to understand the relationship between variables (Figure 2).
|Model (A) .||Predicted|
|Model (B) .||Predicted|
|<1000 cells/mL .||>1000 cells/mL .||<1000 cells/mL .||>1000 cells/mL .|
|Observed||<1000 cells/mL||207||36||Observed||<1000 cells/mL||15||5|
|>1000 cells/mL||43||54||>1000 cells/mL||1||4|
|Model (A) .||Predicted|
|Model (B) .||Predicted|
|<1000 cells/mL .||>1000 cells/mL .||<1000 cells/mL .||>1000 cells/mL .|
|Observed||<1000 cells/mL||207||36||Observed||<1000 cells/mL||15||5|
|>1000 cells/mL||43||54||>1000 cells/mL||1||4|
A database is created covering the period 2009–2018. For the data analysis the open-source data visualization and machine learning toolkit Orange is used (Demšar et al. 2013). The explanatory variables are the ones shown in Figure 2, with some additional derived variables (e.g. 5 days average), and the response variables are the algal concentration (estimated by Chl-a), cyanobacteria hazard (estimated by concentration, dominance, biovolume) and cyanotoxin concentration. The analysis consists first in studying the correlation between variables and secondly in training and testing some decision tree learning models to provide a daily alert on the risk of cyanobacteria. Decision tree learning is one of the predictive modelling approaches used in machine learning; observations are classified in branches to infer about the target value (for example, if a certain substance has been observed repeatedly as very high in a certain situation, if a similar situation occurs again then it is likely that this substance will be very high again). Random forests is another predictive method based on a multitude of decision trees (following the example, the identification of the situations linked to the concentration of the substance are more complex and consider different combination of variables).
Monitoring of algal development
From 2016 to 2018, generally low concentration of algae (Chl-a) were observed in La Contraparada (Figure S2 in supplementary material) with a mean of 7.1 μg/L compared to the mean of 13.1 μg/L in the period 2009–2015. The maximum algal cells concentration reaches 18,000 cells/mL in the period 2016–2018, significantly lower than the maximum observed in 2009–2015. The general decrease in algal concentration potentially demonstrates an improvement in recent years due to the preventive management in the operation of the reservoir.
The higher algal concentration generally occurred in the periods when the DWTP was stopped, involving an increase in water retention time (average Chl-a of 10.5 μg/L compared to 6.4 μg/L when the DWTP is working). When the plant started again the concentrations of algal get back to low value in a few days because of dilution.
Monitoring of cyanobacteria presence
During the 2016–2018 period, 65 phytoplankton genera were identified. The seven detected cyanobacterial taxa did not exceed 1% of the biovolume in 76% of the samples. This represents an improvement compared to the period 2009–2015 with almost constant cyanobacteria presence on the surface of La Contraparada Reservoir.
The most abundant cyanobacteria genus during the study period was Merismopedia sp (average of 1426 cells/mL considering all samples, abundance is also shown in Figures 3 and 6). While this genus is potentially toxic (Izaguirre et al. 2007) it did not produce toxins during the period of study and has a limited total biovolume, accordingly the risk was limited and no changes were done in the reagent doses in the purification treatment process.
From the end of September 2018 to January 2019, La Contraparada DWTP was stopped since the water concession was reached. At the beginning of this period, the cyanobacteria Planktothrix agardhii was detected in October 2018 at the plant reservoir and Ojós Reservoir. Planktothrix is a rare genus in the south of Spain, more typical of algal blooms of northern Spain and the rest of Europe and had not been found massively in the Segura Basin so far. Monitoring was carried out until its disappearance in a natural way (Figure 3). Its persistence time in the water was very short, and it did not affect the quality of the water stored. Others rare genus in La Contraparada such as Woronichinia naegeliana and Microcystis aeruginosa were also identified.
Monitoring of cyanotoxins presence
In the period 2016–2018, microcystin (MC) were only detected in May 2016 in the inflow point of La Contraparada Reservoir (MC=2.96 ppb) but were never detected in the raw water at the entrance of the plant (results in Table S1 in supplementary material). In the case of the Ojós Reservoir, the presence of MC was detected in March, September, and November 2016, but with a value inferior to 1 ppb, the threshold of the RD140/2003. Anatoxin-a was only detected in benthic mats, for both Ojós Reservoir and La Contraparada Reservoir and has not generated any issues for drinking water production (Table S1 in supplementary material).
During the presence of Planktothrix in October 2018 no cyanotoxins were detected in La Contraparada Reservoir or in the Ojós Reservoir. Also, no cyanotoxins were detected in any of the analyses performed on treated water during the study period.
Forecast of algal growth
Considering the full period 2009–2018, the strength of the correlation is weak between Chl-a concentration and the different environmental variables theoretically linked to algal growth (Table S3 in supplementary material). Among the statistically significant correlations (determined by the False Discovery Rate, FDR <5%) the Pearson's r coefficient is between 0 and −0.16 for water temperature, air temperature, sun irradiation and wind velocity. This might indicate that the temperature is not a limiting factor of algal growth in this region. The higher coefficient is obtained for oxidability (r of 0.29) which is a parameter measured during the water treatment process and linked to the content in organic matter.
Due to the difference in the range of observations in Chl-a concentration before and after the beginning of 2016, the correlation study is also performed separately for period November 2009–March 2016 (average Chl-a 13.1 μg/L, with first and third quantile of 5.1 and 15.3 μg/L) and the period April 2016–January 2018 (average Chl-a 6.2 μg/L, with first and third quantile of 3.2 and 7.3 μg/L). In the first period the strength of the correlation is moderate with oxidability (r of 0.56) and turbidity (r of −0.40); in the second period the strength of the correlation is weak for all the variables. The low correlation with turbidity might indicate that most suspended solids are made up of inorganic materials. Also, the low range of Chl-a concentration observed in the period 2016–2018 limit the suitability of further analysis for this period.
Different classification trees have been developed to identify better the links between the different factors explaining high concentration of Chl-a (Chl-a>10 μg/L) in the period 2009–2016. An example is provided in Figure 4, among the 360 measurements of Chl-a performed when the DWTP was working, 133 had high concentrations (Chl-a>10 μg/L), so a total of 36.9% (this is shown in the root node on the top of the tree). These situations with high Chl-a occurred very frequently when turbidity <4 NTU and oxidability >2.18 (in other words high organic matter and very low suspended matter, such as period without rainfall erosion), and almost never when turbidity >4 NTU and oxidability <2 (high turbidity principally associated to rainfall erosion). This is an important result since the inexpensive and daily measurement of oxidability and turbidity, performed to control the water treatment process, informs about potential high value of Chl-a in the reservoir. Accordingly, a first daily estimate of Chl-a can be derived to complement the weekly measurement of Chl-a done in the reservoir. The thresholds identified by the classification tree in the period 2009–2016 show coherent results if they are applied to 2016–2018 (when turbidity >4 NTU and oxidability <2, value of Chl-a >10 μg/L only represent 9% of total compared to 20% when turbidity >4 NTU and oxidability >2). Nevertheless, the very limited number of situations with high Chl-a (eight values over a total of 73) and the absence of situation with turbidity <4 limit the validation of the thresholds in the period 2016–2018.
In addition, the use of classification trees helped in identifying algal development in adverse conditions, this will allow the plant operator to be more cautious about certain signals in such period. For example, it has been observed that in cold periods (Tª air <12 °C) and with low turbidity (<4 UNF) there are fast-growing algal genera that cause Chl-a >10 μg/L, such as Diatoms (Asterionella and Cyclotella) and Chlorophyceae (Tetraselmis and Coelastrum).
Forecast of cyanobacteria occurrence
The classification tree allows to identify in which situation or combination of situations the cyanobacteria concentrations were the highest. Since the parameters that govern the classification are easily available, they can be used to predict cyanobacteria concentrations (e.g. in situations A or B the concentration is likely to be high). Cyanobacteria occurrence in La Contraparada Reservoir in the period April 2016–November 2018 also seems to follow some general rules as shown in the classification tree presented in Figure 5. It was observed that elevated concentration (superior to 2000 cells/mL) is very infrequent below an air temperature of 23 °C (5% of sampling) and unobserved if moreover the DWTP is working (low retention time of the water). Above 23 °C high concentrations are quite common (38%) but they have not been observed during peak insolation days (>66 hours of cumulated sunshine hours in the last 5 days).
In conditions of abundant sun and heat cyanobacteria spread quicker than other algae and become dominant in the reservoir (i.e. more cyanobacteria cells of this genus than all other algae genera). This is illustrated in Figure 6 showing that when in the previous 5 days there is more than 45 hours of sunshine and an average air temperature superior to 25 °C, a high dominance of cyanobacteria genus Merismopeda and Aphanocapsa (alone or the two genus together) is observed. In other words, there is a high risk of cyanobacteria bloom in these conditions. This situation can be predicted since temperature and sunshine hours are two variables that are considered in meteorological forecasts.
These results have been used to build a predictive model to estimate the risk of high concentration of cyanobacteria (situation with total cyanobacteria >1000 cells/mL), taking as explanatory variables the ones that are easily available by the DWTP manager such as meteorological information (sun hours, air temperature, precipitation) and routine water quality analysis done at the plant entrance (turbidity, oxidability and water temperature). Classification tree and random forest techniques have been used to build these predictive models. A first set of models (A) is trained with 50% of the dataset and evaluated with the remaining data, this operation is repeated 10 times with a random sampling, and a second set of models (B) is trained by using 2016 and 2017 data and tested with 2018 data. For (A) the classification accuracy for tree and random forest was 75 and 78% respectively, and the area under the ROC of 0.81 and 0.86 respectively, while for (B) accuracy of 76 and 76% and ROC of 0.87 and 0.83 respectively; for each model the three most important predictors were 5 days temperature minimum, DWTP water temperature, and 5 days air temperature average. The confusion matrixes for the random forest models and the training sets are presented in Table 1; most of the situations with high cyanobacteria concentration are well identified but additional data would be needed to produce more significant results. Indeed, only few situations with elevated cyanobacteria concentration are used to train and test the models (only 19 events) and all of them can be grouped in approximately three periods of time that last a few weeks each (August and September 2016, July and August 2017 and August 2018).
Improvement of risk management strategy
The additional monitoring performed during this study have made it possible to update the risk management strategy for the control of algal growth in La Contraparada Reservoir. According to the level of concentration of Chl-a and cyanobacteria, five alert levels are defined with corresponding actions at the reservoir and DWTP.
The actions are based on the guidelines of the Word Heath Organization (World Health Organization 2017). Considering the level of risk, they consist of realizing additional in-situ monitoring, perturbing the reservoir water level, and upgrading the treatment of the DWTP (Figure 7).
The estimation of the concentration of Chl-a and cyanobacteria triggering all the actions are based on the routine analysis performed in the reservoir (weekly analysis of Chl-a) and on the estimation from the different classification tree developed. For example, if at the plant water intake the turbidity becomes superior to 4 NTU and the oxidability >2 mg O2/L, a sampling in the reservoir is immediately taken to confirm the situation of alert. The same occurs when the intake water temperature exceed 25 °C and the 5 days minimum temperature is superior to 22 °C.
From the monitoring and data analysis performed in the reservoir of La Contraparada DWTP and upstream reservoir of Azud de Ojós we can highlight some relevant implications.
First, the additional monitoring campaign highlights the variety of phytoplankton genera (total 65) and cyanobacterial taxa (seven) in this Spanish basin and allows better understanding of their behavior in the period of study. When the DWTP is stopped, the algal biomass increases in the reservoir confirming the role of the water residence time, in coherence with previous studies (World Health Organization 2015a). Also, during the plant stop, the algal growth may have been influenced by the transformation of calcium phosphate from the sediments into soluble phosphate and released into the environment caused by the anoxia of the deep layers and the appearance of a very negative redox potential. In turn, the nitrogen forms are displaced to ammonium and not to nitrates. This makes nutrients available for a rapid growth of phytoplankton.
The management of the reservoir and the water level changes performed was probably effective to limit algal growth and toxin production, as observed in other studies (Merel et al. 2013); as a result the most abundant cyanobacteria (Merismopedia) has not been observed to produce toxin in the period of interest. The rapid development of rare taxa has been observed (Planktothrix) confirming that the cyanobacterial community can quickly change to adapt to new environmental conditions; this is particularly relevant considering the potential effects of climate change. In general, very low values of cyanotoxins have been detected and no toxins have been detected in the treated water suggesting an efficient degradation at these levels.
Secondly, the data analysis allowed to identify useful relationships between algal growth, and especially cyanobacteria growth, and several variables easily accessible by the plant operator. Only weak correlation between Chl-a concentration and the different environmental variables theoretically linked to algal growth (World Health Organization 2015a) have been observed, supporting the need for local study. The use of a classification tree technique allows a better identification of the situations of risk, recognized as a combination of different variables that the plant operator can monitor or access easily (such as DWTP water analysis, weather forecast). The situations of cyanobacteria dominance were mostly observed in cases of high temperature and sunshine hours, supporting the conclusions of previous studies (Visser et al. 2016), and indicating potential additional risks due to climate change (the number of hot days will considerably increase under all the scenarios of climate change for 2050 in the Mediterranean regions).
Finally, the identification of the situations of risk of cyanobacteria in the reservoir allow an adjustment of the risk management strategy and a better implementation of the revised drinking water directive (EU Council 2020). The predictive models developed using classification tree and random forest are promising to anticipate the risk of cyanobacteria but still need some additional work to be fully operational.
Several limitations prevent a precise forecasting and a better adjustment of risk management decisions. A first limitation is that the impacts of the disturbance intentionally caused on the reservoir water level and water inflow could not have been quantified with the available data since those disturbances occurred together with other environmental changes (nutrient concentration, irradiation, etc.) and that relatively few data are available to make comparisons. A second limitation is that the role of the upstream reservoir (transfer of algae and toxins) has not been fully considered, since the amount of data (Chl-a, cyanobacteria) is very limited (typically one sample per month for Ojós reservoir). A third limitation is that the automatization of the alerts has not been completed, so the DWTP managers still need to access manually different sources of data to assess the situation and realize the actions planned in the risk management strategy.
The additional monitoring performed and the different data analysis techniques applied such as classification trees allowed a better identification of the situation at risk of cyanobacteria in La Contraparada Reservoir. As a result, a new operational risk management plan has been developed and is now applied by the water operator. This has a positive impact on the management of the DWTP and the safety of drinking water supply in the Murcia region, and comply with the recommendations of the revised drinking water Directive (EU Council 2020).
The limitations that prevent a precise forecasting and a better adjustment of risk management decision are currently being considered in a new European project called WQeMS that started in January 2021. The project is about the use of satellite images to get historical data on algal development and cyanobacteria bloom in La Contraparada Reservoir and in the upstream reservoirs to develop an automatic forecast system based on in-situ data and satellite imagery. These improvements will be detailed in another article.
The authors are grateful for the funding contribution of the H2020 financial instrument of the European Community and would like to thank all the project stakeholders for their collaboration. The authors are also grateful to the project's partners involved in some activities described in this study, and particularly the IIAMA-UPV and KNMI. Special thanks go to the anonymous reviewers.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICTS OF INTEREST
The authors declare there is no conflict.