Use of artificial neural network in the prediction of algal blooms
Introduction
The vast proliferation of cyanobacteria in lakes and reservoirs has become an emergent aquatic environmental issue owing to fish mortalities caused by oxygen-depletion, toxins such as microcystin and anatoxin, the occurrence of an unpalatable taste and odor, as well as the loss of recreational resources (Pinckney et al., 1997; Inamori et al., 1998; Sugiura et al., 1998). The development of cyanobacteria and green algae blooms are a common phenomenon in eutrophicated waters (Smith, 1990). Research on preventive technologies and monitoring methods of algal bloom have accordingly received worldwide attention.
To be able to control algal blooms, it is necessary to be able to determine the key factors governing the algal dynamics and to establish an algal response model which can effectively simulate the timing and magnitude of algal blooms. Recently, artificial neural network (ANN) technology has been applied in the prediction of algal blooms (Recknagel et al., 1997; Yabunaka et al., 1997), and the succession of several dominant cyanobacteria (Recknagel, 1997). However, the effects of different factor combinations have not been assessed in detail. Observations have shown that some selected factors in the input layer lead to “noise” affecting the output results. Therefore, an evaluation of the factors affecting the algal growth is necessary before creating an artificial neural network. Further, it is important for successful lake management to find the optimal combinations of these factors based on the availability of data and measures to regulate pollution inputs into the system.
In this study, the most commonly used computational algorithm, back-propagation, was used in ANN model to determine the nonlinear relationships between the water quality factors and the dominant algal genera. Relative importance of factors affecting algal growth was evaluated by sensitivity analysis. Finally, the optimized combinations were determined by observing the change of algal densities using the factorial orthogonal design.
Section snippets
Study site and water quality data
Lake Kasumigaura is the second largest lake in Japan with an area of 22,000 ha and water mean depth of 4.0 m. Mean hydrological retention time is 200 days. From 1982 to 1996, near the center of lake, the annual average water temperature and rainfall was 16.0°C and 1202 mm, respectively. Water quality data from 1982 to 1996 in Kakezaki sampling station used in this study were provided by the Enterprise Bureau of Ibaraki Prefecture (Annual Reports of Water Quality from 1982 to 1996).
Identification of factors affecting the algal growth
Factors affecting the growth of algae in lake ecosystems are multidimensional, including physical factors (water temperature, light radiation, and the conditioning of water environment), stimulatory or inhibitory chemicals (nutrient loading, aquatic humic substances) as well as biotic interactions (species-species competition, predation). Among these factors, temperature, light irradiance, concentration and composition of nutrients, the conditioning of water, and the rate of cell division were
Artificial neural network technology
ANN is an appropriate technique to develop models because of its abilities to assign significance to the input parameters and map the inputs to outputs when the relationships between parameters are unknown. Especially, the backpropagation learning technique has been applied successfully in nonlinear complex systems, as it can arbitrarily approximate functions by using the gradient descent algorithm or faster algorithm (Hagan et al., 1996). Recently, ANN has been widely used in water resource
Prediction of the timing and magnitude of algal blooms
MSE decreased rapidly and remained around 0. 0195 for Microcystis, 0.0216 for Oscillatoria, 0.0265 for Phormidium and 0.0262 for Synedra after 1000 epochs. The post-processed results for the four representative algae using the developed network are shown in Table 1. The trained values were significantly correlated to the measured data (p<0.001) for Microcystis, Oscillatoria, Phormidium and Synedra, viewing from the correlation coefficient and trained data points.
Further, the trained networks
Conclusions
An ANN model to simulate the algal proliferation for the four dominant genera, Microcystis, Oscillatoria, Phormidium and Synedra was created. The factors affecting the algal growth were selected to be input parameters based on previous studies.
The proliferation of algal blooms of Microcystis, Phormidium and Synedra in Lake Kasumigaura was successfully predicted and validated using neural network technology. Comparatively, the prediction of Oscillatoria blooms, a new dominant phytoplankton since
Acknowledgements
The study was supported by the Scientific and Technology Promotion Society of Ibaraki Prefecture, Japan. We are also very grateful for water quality monitoring data provided by the Enterprise Bureau of Ibaraki Prefecture.
References (25)
- et al.
Artificial neural network approach for modeling and prediction of algal blooms
Ecol. Modell.
(1997) - et al.
Significance of attached cyanobacteria relevant to the occurrence of musty odor in Lake Kasumigaura
Wat. Res.
(1998) - et al.
Novel application of a back-propagation artificial neural network model formulated to predict algal bloom
Wat. Sci. Tech.
(1997) Marine Plankton Ecology
(1976)- et al.
Using neural networks to assess the influence of changing seasonal climates in modifying discharge, dissolved organic carbon, and nitrogen export in eastern Canadian rivers
Water Resour. Res.
(1998) - Enterprise Bureau of Ibaraki Prefecture, Japan. (1982–1996) Annual Reports of Water Quality of Enterprise Bureau, Nos....
- et al.
Structure and dynamics of zooplankton communities maintained in closed systems, with special reference to the algal food supply
J. Plankton Res.
(1982) - et al.
Nutrient-limited growth of Microcystis aeruginosa and Phormidium tenue and competition under various NP supply ratios and temperatures
Limnol. Oceanogr.
(1997) - et al.
Neural network design
(1996) Phytoplankton EcologyStructure, Function and Fluctuation
(1986)
Particulate light attenuation in a large subtropical lake
Can. J. Fish. Aquat. Sci.
Effects of aquatic humic substances on the growth the cyanobacterium Microcystis aeruginosa
Japan. J. Water Environ. Sci.
Cited by (121)
Assessing physical and biological lake oxygen indicators using simulated environmental variables and machine learning algorithms
2024, Environmental Modelling and SoftwarePrediction of Aureococcus anophageffens using machine learning and deep learning
2024, Marine Pollution BulletinOptimizing biogas production from palm oil mill effluent utilizing integrated machine learning and response surface methodology framework
2023, Journal of Cleaner ProductionCharacterizing ocean surface contamination: Composition, film thickness, and rheology
2023, Marine Pollution BulletinCitation Excerpt :When present in large numbers, they can lead to the formation of thin patchy layers on the ocean surface (Durham and Stocker, 2012). A common type of cyanobacteria present in such layers is Phormidium sp. (Carmichael and Boyer, 2016; Wei et al., 2001), which we use in our experiments on rheological investigation (see §3). When conditions for their growth are favourable, there could be an explosion in the population of some species, which leads to the formation of dense blooms that resemble scum floating on the water surface.
Remote sensing to detect harmful algal blooms in inland waterbodies
2022, Science of the Total EnvironmentCitation Excerpt :These approaches leverage the strength of satellite imagery to reveal relationships with water quality variables, including HABs. While some studies have demonstrated reasonable effectiveness in using Neural Networks in HAB detection (Bui et al., 2017; French and Recknagel, 1970; Wei et al., 2001), other machine learning techniques can provide improvements in prediction and detection skill (X. Li et al., 2017b; Pahlevan et al., 2020). Nonetheless, machine learning methods require extensive validation, and are influenced by uncertainties within satellite correction models and/or waterbody type (Neil et al., 2019).