Time-Series Analysis for Watershed Scale Predictions of Water Quantity and Quality Export from Agricultural Watersheds
MetadataShow full item record
Complex hydrological models are widely used to predict overall watershed responses by incorporating knowledge acquired on field or plot scales. However, processes or complexity critical at smaller scales may not necessarily be important at larger scales. Consequently, it is unnecessary, and could potentially be problematic, to predict hydrological responses at the watershed scale using driving variables acquired at the field scale. With long-term detailed water quality data and a comprehensive set of forcing, state, and flux variables at the watershed scale, dominant variables were able to be successfully identified for watershed scale streamflow, suspended sediment (SS), particulate phosphorus (PP), and soluble reactive phosphorus (SRP). The identification of dominant variables and their relative importance was conducted through the establishment of time series seasonal autoregressive integrated moving average (SARIMA) models. I found that catchment scale hydrological responses including streamflow, SS, PP, and SRP had different dominant variables. The results showed that models based on the dominant variables were capable of replicating watershed scale hydrological responses. As such, simple models were sufficient for watershed hydrological response simulations and it appeared that identification of dominant variables was the first step to achieve simple models. The application of predefined model complexity and model structure developed for one watershed may not guarantee successful predictions in another watershed. To address this problem, I tested how model complexity, as expressed through differences in the number and configuration of flux and state equations, affects hydrological processes, and to evaluate the validity of current water quality models’ assumption that driving variables must include those implied by the plot or field scale empirical studies. Four models with different model complexity were used to generate runoff and test the needs of model complexity. By removing the assumption, dominant variables of water quality models were identified based entirely on their statistical significance as determined from the SARIMA analysis. The results suggested that the more complex models did not generate better predictions. Simple models were sufficient to generate total runoff at different time scales for water quality modeling purpose. Without runoff flux variable, water quality models with identified forcing and state variables still presented reasonable predictions of hydrological responses. It is difficult to transfer all model details in terms of model structure and parameters from where a model was developed to another watershed especially ungauged ones. When essential and important features of the watershed hydrological dynamics could be reliably represented using only a few dominant variables, it may be sufficient to transfer dominant variables among watersheds. Model transferability was compared using models based on flux variables and based on dominant variables. The results showed more credible model transferability for streamflow, SS, PP, and SRP across watersheds when models were based on dominant variables. It suggested that simple models that simulate one flux may be easier to move among watersheds without a lot of calibration.