Seasonal forecasting of lake water quality and algal bloom risk using a continuous Gaussian Bayesian network

Abstract. Freshwater management is challenging, and advance warning that poor water quality was likely, a season ahead, could allow for preventative measures to be put in place. To this end, we developed Bayesian network (BN) seasonal lake prediction. BNs have become popular recent years, but the vast majority are discrete. Here, Gaussian (GBN), simple class of continuous BN. The aim forecast, spring, mean total phosphorus (TP) chlorophyll (chl a) concentration, colour, maximum cyanobacteria biovolume upcoming growing (May–October) Vansjø, shallow nutrient-rich southeastern Norway. develop model, first identified controls on interannual variability seasonally aggregated quality. These variables were then included GBN, conditional probability densities fit using observations (≤39 years). GBN predictions had R2 values 0.37 0.75 (colour) classification errors 32 % 17 (cyanobacteria). For all including weather did not improve predictive performance (assessed through cross-validation). Overall, found approach well suited forecasting. It straightforward produce probabilistic predictions, exceeding management-relevant thresholds. sensibly parameterised only observed data, despite small dataset. Developing comparable discrete BN much more subjective time-consuming. Although low high temporal autocorrelation study meant performed slightly better than naïve forecast (where forecasted value simply previous season), believe forecasting presented here particularly useful areas with higher sensitivity catchment nutrient delivery climate at shorter (daily or monthly) timescales. Despite parametric constraints GBNs, their simplicity, together relative accessibility software handling, means they good choice development variables.