Development of a decision tree model for the prediction of the limitation potential of phytoplankton in Lake Okeechobee, Florida, USA
East, Therese L.; Sharfstein, Bruce
published: Jan 23, 2006
ArtNo. ESP141016571008, Price: 29.00 €
Conducting long-term algal bioassays in large, complex systems such as Lake Okeechobee is an expensive and time-intensive undertaking, especially in comparison with physical and chemical monitoring. This paper describes a water quality-based decision tree model for predicting whether the phytoplankton in Lake Okeechobee is limited by light or nutrients. The model was developed and validated using the results of algal bioassays coupled with routinely monitored water quality data. Algal bioassays indicated that the factor most commonly limiting phytoplankton production in Lake Okeechobee for the period of October 1997 to November 2000 was light (59 %) followed by nitrogen (41 %). Limitation status of the phytoplankton was positively correlated with irradiance (in terms of Secchi depth/total depth) and phytoplankton biomass (in terms of chlorophyll-a) and negatively related to dissolved inorganic nitrogen and soluble reactive phosphorus concentrations. A cross-tabulation procedure was used to examine how the frequency of occurrence of light limitation and nutrient limitation varied as a function of these variables. The cross-tabulation procedure was also used to derive the empirical threshold values used to construct the model. This result supports both the accuracy of the derived critical threshold values and the validity of using chemical measurements in predicting whether light is limiting or nutrients are limiting in Lake Okeechobee. The model successfully predicted light limitation versus nutrient limitation in three independent validation data sets 70 % to 85 % of the time. When nutrient limiting conditions prevailed, the model did not successfully predict which nutrient (nitrogen, phosphorus, or a combination of nitrogen and phosphorus) was limiting. Our results suggest that the predictive abilities of the model could be enhanced by using time-specific data rather than averaged monthly data.