Welcome to NOAH
Forecasting Algal Blooms in Drinking Water Supplies
New Jersey

Problem Definition

Algal blooms (AB) in potable water supplies are becoming an increasingly prevalent and serious water quality problem around the world. AB events can cause taste and odor problems, damage the environment, and some algal classes like cyanobacteria (blue-green algae) may release toxins that can cause human illness or even death. There is a need to develop models that can accurately forecast algal bloom events on the basis of predictive water quality and meteorological information. Given the multitude, interplay, and complexity of the various controlling environmental factors, modeling and forecasting AB is a daunting challenge. This research focused on the feasibility of using artificial neuralnetwork (ANN) technology as an accurate, real-time modeling and forecasting tool. Previously-collected data from a NJ water utility served as the test case. AB forecasting periods included one-week and two-weeks prior to the event. Despite a less than ideal number of historical AB events, the high predictive accuracy achieved in this study indicates that with sufficient data, both in terms of the number of historical AB events and availability of important predictor data, ANNs can serve as reliable, accurate, real-time AB forecasting


Despite the relatively small number of data events available for training, the ANN models (several hundred models developed and assessed) generally performed well during validation, achieving relatively high correlation coefficients and accurately predicting sudden and significant changes in algal populations. The models developed with both one-week and two-week ahead prediction periods accurately predicted formation and dissipation of AB events, as well as the relative increase and decrease in cell counts.

Figures 1, 2, and 3 provide a visual assessment of model performance for three representative cases. The validation figures also depict the initial algal count corresponding to the beginning of the prediction event, so that relative changes in algal counts (Final), and evaluation of ANN prediction performance (ANN) is more transparent.

One of the critical issues of this study was a rudimentary assessment of the existing data and analysis as to what sampling strategies might improve forecasting capability. It was found that select water quality variables, including so-called “limiting nutrients”, were identified as relatively non-essential for accurately forecasting algal counts in this system. These findings agree with the experience of water utility personnel, who believe that these constituents generally exist within a range of concentrations that neither diminish nor propagate algal populations. Despite a very limited number of available data sets, the ANN models performed well in most cases during validation, accurately predicting large changes in algal cell populations. The degree of accuracy was surprising, given the complexity and non-linear behavior of algal populations, inherent data “noise”, and the relatively small number of historical events available for model training. While not definitive, the results strongly indicate that the ANN models learned underlying relationship between select water quality and meteorological parameters, and algal cell concentrations at this WTP. This is supported by: 1) relatively high model accuracy and overall consistency between training and validation results; 2) consistency in performance for different types of models (single value outputs and classification) and input structures (original and revised); 3) consistency between modeling results and physical intuition/system understanding; and 4) comparatively poor performance of linear models.

This study was funded by the New Jersey Department of Environmental Protection, Division of Science, Research and Technology. The Research Project Summary is available in PDF format for downloading, and the final report (NOAH, 2006) may be accessed at http://www.state.nj.us/dep/dsr/wq/finalreport-june2006.pdf.