Identifying Seasonal Trends In Dissolved Phosphorus In a Mixed Land Use Watershed Using a Mixed Linear Model Approach

In a study of the Kalamazoo River Watershed, Michigan, USA, we attempted to identify patterns in suites of chemicals and their relationship to land use to understand the cycling (sources, pathways, fate) of chemicals in the watershed. Although such work is being done for phospohorus (P), patterns are not easily identified in part because of potential seasonal influences (e.g., climate, fertilizer application, algal productivity). Therefore, relationships in chemical and land use patterns might change over the course of a growing season. We hypothesized that dissolved P (DP) concentrations are correlated to land use and stream chemistry, but the specific pattern changes as a function of time. If correct, then changes in these relationships would be difficult to recognize by commonly used multi-variate techniques. To test this hypothesis we used mixed linear models (MLM) to identify seasonal effects and the relationships between DP and stream chemistry and subwatershed land use.

In the Kalamazoo River study, weekly stream chemistry measurements and land use data were segregated into landscape and algal influenced datasets. Each was analyzed using a MLM incorporating variables for year and week within year. The landscape MLM identified a DP seasonal trend related to a similar trend in baseflow. The MLM yielded statistical significance (p<0.05) for nitrate, potassium, magnesium, sodium, pH, alkalinity, specific conductance and urban, agricultural and lowland forest land uses. The algal MLM identified a year effect related to differences in algal productivity and statistical significance (p<0.05) for particulate P, nitrate, sulphate, calcium, pH and specific conductance. These results indicate that the MLM provides a method to model not only the means of data but their variances and covariances as well.

Fitting a MLM that includes the seasonal effects allows inferences to be drawn about the DP cycling from the significant fixed effects and the covariance parameters.