Danielle Winter
Purdue University
Co-Authors: S. McMillan, G. Noe, S. Donohue
Restoration of impaired and hydrologically disconnected floodplains is seen as having great promise for alleviating water quality concerns in river corridors given the capacity of floodplains to remove inorganic nitrogen from surface water and groundwater via denitrification. Despite high investment into floodplain restoration, little is known about how different restoration strategies and designs impact denitrification rates. Our research fills this gap by linking denitrification to the environmental factors that can incorporated into restoration design. We measured denitrification potential across hydrologic and geomorphic gradients and across levels of restoration intervention in riverine floodplains in the agricultural Midwest. Simultaneously, we measured environmental factors, such as soil and hydrologic characteristics, to elucidate environment factors that are predictive controls of denitrification. Preliminary findings suggest that vegetation type and hydrologic regime are strong controls of denitrification capacity and high levels of subsurface connectivity are necessary to restore denitrification capacity. Our findings will aid watershed managers and regulatory agencies in identifying floodplains that will likely exhibit high denitrification upon restoration and will assist practitioners to develop future floodplain designs that maximize water quality benefits.
Virtual Poster Symposium 
That was a very nice presentation with interesting results. Were there any significant correlations between the explanatory variables in your BRT analysis? For instance, I would expect vegetation type and soil organic matter might be correlated and could explain part of why there is a significant interaction effect. Also, did you include site (e.g. ag, restored wetland, etc.) as an explanatory variable?
LikeLike
Thank you!
I had not previously assessed the significance of correlations between the explanatory variables, only between denitrification rates and each explanatory variable. I investigated some relationships between explanatory variables this morning, and organic matter differs significantly across geomorphic features and vegetation communities as based on ANOVA like you suggested it may.
I did include restoration type as an explanatory variable by dividing the sites into wetland, agriculture, and prairie, but restoration type was not selected as variable for my final model. My final model predictors were selected by performing k-fold cross validation. For each fold, the least contributing predictor was removed, and, after this removal, the change in predictive deviance between when this set of reduced variables is used and when all predictors are used is calculated. This process is repeated within each fold. At the end of cross-validation, the best set of variables to remove is identified. Despite not being included in my final model, from previous analyses, I have already seen that denitrification rates are significantly different across restoration types. I am still in the preliminary stages of my data analysis. It is quite possible that one of the other variables that is included in the final model may serve as a proxy for restoration type, which I think is likely to be one of the most important predictor of denitrification rates. Continuing to explore the relationships between explanatory variables is definitely an important next step in continuing to refine my model!
LikeLike
Thank you for that detailed response. I was recently involved on a project using BRT, which is one of the reasons I asked my questions. BRT is able to handle some correlation between explanatory variables, but we screened ours to remove variables that had correlation coefficients above 0.8. That way we wouldn’t have variables “masking” each other in terms of importance. Just something to consider. Very cool results so far and I’m excited to see what you find with your additional analysis!
LikeLike
Very nice presentation and interesting results. Did you consider correlations between the explanatory variables in your BRT analysis? For example, I would expect vegetation type and soil organic matter might be correlated. Also, did you include site (e.g. ag floodplain or restored wetland) as an explanatory categorical variable in your analysis?
LikeLike
Sorry for the double posting.
LikeLike