Indiana University-Purdue University Indianapolis (IUPUI) / In this research, Environmental Fluid Dynamic Code (EFDC) and Adaptive- Networkbased
Fuzzy Inference System Models (ANFIS) were developed and implemented to
determine the spatial-temporal distribution of cyanobacterial metabolites: 2-MIB and
geosmin, in Eagle Creek Reservoir, IN. The research is based on the current need for
understanding algae dynamics and developing prediction methods for algal taste and odor
release events.
In this research the methodology for prediction of 2-MIB and geosmin production was
explored. The approach incorporated a combination of numerical and heuristic modeling
to show its capabilities in prediction of cyanobacteria metabolites. The reservoir’s
variable data measured at monitoring stations and consisting of chemical/physical and
biological parameters with the addition of calculated mixing conditions within the
reservoir were used to train and validate the models. The Adaptive – Network based
Fuzzy Inference System performed satisfactorily in predicting the metabolites, in spite of
multiple model constraints. The predictions followed the generally observed trends of
algal metabolites during the three seasons over three years (2008-2010). The randomly
selected data pairs for geosmin for validation achieved coefficient of determination of
0.78, while 2-MIB validation was not accepted due to large differences between two
observations and their model prediction. Although, these ANFIS results were accepted,
the further application of the ANFIS model coupled with the numerical models to predict
spatio-temporal distribution of metabolites showed serious limitations, due to numerical
model calibration errors. The EFDC-ANFIS model over-predicted Pseudanabaena spp.
biovolumes for selected stations. The predicted value was 18,386,540 mm3/m3, while
observed values were 942,478 mm3/m3. The model simulating Planktothrix agardhii gave
negative biovolumes, which were assumed to represent zero values observed at the
station. The taste and odor metabolite, geosmin, was under-predicted as the predicted
v
concentration was 3.43 ng/L in comparison to observed value of 11.35 ng/l. The 2-MIB
model did not validate during EFDC to ANFIS model evaluation.
The proposed approach and developed methodology could be used for future applications
if the limitations are appropriately addressed.
| Identifer | oai:union.ndltd.org:IUPUI/oai:scholarworks.iupui.edu:1805/3043 |
| Date | 29 October 2012 |
| Creators | Bruder, Slawa Romana |
| Contributors | Babbar-Sebens, Meghna, Jacinthe, Pierre-Andre, Tedesco, Lenore P. |
| Source Sets | Indiana University-Purdue University Indianapolis |
| Language | English |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0021 seconds