Premium
An empirical modeling approach to predict and understand phytoplankton dynamics in a reservoir affected by interbasin water transfers
Author(s) -
Fornarelli Roberta,
Galelli Stefano,
Castelletti Andrea,
Antenucci Jason P.,
Marti Clelia L.
Publication year - 2013
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1002/wrcr.20268
Subject(s) - environmental science , phytoplankton , hydrology (agriculture) , water resource management , geology , ecology , geotechnical engineering , nutrient , biology
In this paper, we use empirical modeling to predict and understand phytoplankton dynamics in a reservoir affected by water transfers. Prediction of phytoplankton biovolume is central to the management of water resources, particularly given the significant impacts on quality of the water‐quantity oriented management of transfers between reservoirs. A novel tree‐based iterative input variable selection algorithm is applied for the first time in an ecological context, and identifies a maximum of eight driving factors out of 77 candidates to explain the biovolume of chlorophytes, cyanobacteria and diatoms. The stepwise forward‐selection to iteratively identify the most important inputs leads to a physically interpretable model able to infer the physical processes controlling phytoplankton biovolume. Reservoir inflows and outflows are found to exert a strong control over diatom and chlorophyte dynamics while water temperature, nitrate and phosphorus determine the biovolume of cyanobacteria. Following the selection of the most relevant inputs, the 1 week ahead predictions of four different data‐driven model classes, i.e., neural networks, extra trees (ETs), model trees and linear regressions, are compared based on performance indices and statistical tests. ETs are found to outperform the other models by providing accurate predictions of cyanobacteria, chlorophyte and diatom biovolume by explaining 66.6%, 66.9%, and 80.5% of the variance, respectively. The methodology is applicable to different environmental studies and combines the strength of empirical modeling, i.e., compact models and accurate predictions, with a good understanding of the physical processes involved.