Neural network based modeling of environmental variables: A
summary of successful applications
The Boreal Plain ecozone of the Canadian Boreal Forest is experiencing both natural,
mainly wildfires, and anthropogenic, primarily forest harvesting, watershed disturbances.
Such disturbances can significantly alter the timing and intensity of water, nutrients, and
sediment outputs from these watersheds. A measurable increase in nutrient loading to
water bodies may increase dissolved oxygen depletion, cyanobacteria biomass formation,
and cyanobacterial toxin production potentially disrupting fish habitat and possibly
deteriorating the performance of downstream water treatment plants. Therefore, nutrient
and sediment modelling, is critical to the protection of aquatic ecosystems and the
preservation of source water quality. Most of the currently available models, for
watershed modeling, are undermined in practice because of the extensive landscape data
required for model calibration. However, satellite remote sensing (RS) has recently made
available cost-effective data over the entire landscape rather than providing a sampling of
it, as would be the case with ground-based measurements. This study proposed an
artificial neural network (ANN) modelling algorithm that relies on low cost readily
available meteorological data as well as RS information for simulating streamflow (Q),
total suspended solids (TSS) concentration, and total phosphorus (TP) concentration. The
developed models were applied to four forested watersheds in the Canadian Boreal Plain.
Our results demonstrated that through careful manipulation of time series analysis and
rigorous optimization of ANN configuration, it is possible to simulate Q, TSS, and TP
reasonably well. R2 values exceeding 0.8 were obtained for all modelled data cases. The
proposed models can provide real time predictions of the modelled parameters, can
answer questions related to the impact of climate change scenarios on water quantity and
quality, and can be implemented in water resources management through Monte Carlo
simulations.