Natural and anthropogenic watershed disturbances may expose soils to erosion, resulting in the potential for increased export of nutrients and sediment to surface waters. The resulting excessive nutrient loads can potentially lead to a variety of problems ranging from anoxic waters to toxic algal blooms and a decrease in habitat diversity, and thus leading to habitat destruction. Algal blooms' impacts adversely affect not only the health of people, animals, and aquatic organisms, but also the "health" of local and regional economies. Hence, nutrient modelling and, in particular, is critical to provide the necessary information for responsive watershed management practices. Most of the currently available models for watershed modelling are limited in practice because of the extensive requirement for landscape data (e.g., soils, vegetation, precipitation) needed for model calibration. Therefore, a class of models that can simulate the response of ungauged watersheds with reasonable accuracy is critical to provide the necessary information for responsive watershed management practices.
With the advent of remote sensing (RS) techniques, the availability of high quality time- and space- variant data at an affordable cost has been made real. Developing a class of watershed models, that can utilize this RS information and that is less reliant on ground-based watershed specific measurements, is expected to overcome the drawbacks of more readily available models that are ground-based-data-collection and time-intensive. It will provide a substitute approach using inexpensive RS data with few ground truthing requirements that can move this class of models from a research base to possible industrial applications.
This study is an effort to incorporate low-cost time-variant RS information in watershed-scale nutrients modelling. Artificial neural network (ANN) total phosphorus and total nitrogen models, which rely on a time series of Moderate resolution Imaging Spectroradiometer (MODIS)-derived vegetation indices (VIs), were developed. The output of a previously-developed ANN streamflow model, which utilizes widely available meteorological data as model inputs, was used along with the MODIS-derived VIs for nutrients modelling. Cross-correlation analysis was employed to identify possible time-lagged inputs based on the strength of the correlation between the output variable and each of the time-lagged inputs. Spectral analysis was utilized to account for nutrients/flow hystereses in the modelled processes. The proposed approach was tested on four watersheds (5 to 130 km2) in the Canadian Boreal forest and was found to provide an efficient modelling alternative for water-phase nutrients predictions.
In order to assess the possibility of successful model transferability from a gauged watershed to a hydrologically similar ungauged watershed, new remotely sensed hydrologic similarity measures, that rely on MODIS-derived enhanced vegetation index (EVI), normalized difference vegetation index (NDVI), greenness fraction vegetation index (GFVI), and normalized difference water index (NDWI) were examined. The NDWI-based hydrologic similarity index was found to provide a successful indicator of basin similarity. The Pearson's correlation coefficient was evaluated to exceed 0.7 when NDWI-based hydrologic similarity index was regressed to models' "goodness-of-fit" statistics, reflecting the usefulness of the approach.