In coastal regions with extensive, non-linear water bodies, large scale regional macrophyte surveys are rarely done due to logistical difficulties and high costs. This study proposes to examine whether remote sensing can be used for regional monitoring of submerged aquatic vegetation (SAV) using a field study in the Currituck Sound of North Carolina. The objectives are: (1) to determine if different levels of macrophyte cover, different growth forms or specific species could be detected using the Digital Globe Quickbird, high-resolution satellite sensor, and (2) to determine if predictions of macrophyte abundance and distribution in the sound can be improved by including sediment type or measures of water clarity (Secchi disk transparency, total nitrogen, total phosphorus, or water color) in our models. Using binomial and multinomial logistic regression models, we will develop statistically significant relationships between SAV measures and Quickbird spectral values using multinomial and binomial logit (logistical regression) models. Significant correlations between water quality characteristics and the Quickbird spectral values within pelagic zones of the sound will be used to adjust model predictions. Model validation will be developed using SAV and water quality data not included in the SAV abundance and distribution model. The research products from this study will provide a distribution map of SAV distribution and status within the Currituck Sound of North Carolina. NC DOT will use the model results of this study to assess impacts to SAV from proposed projects in an effort to determine appropriate avoidance, minimization, and compensatory mitigation alternatives. Additionally, models results will reduce permit processing time, and reduce the potential for unintentional violations. The model database that will be developed in this study will further serve as a baseline database, compiled in a GIS format, which is easily updated as future data becomes available. New model scenarios will have the advantage of continually improved and updated data.
Submerged Aquatic Vegetation (SAV) is an important component in any estuarine ecosystem. As such, it is regulated by federal and state agencies as a jurisdictional resource, where impacts to SAV are compensated through mitigation. Historically, traditional detection methodologies have been proven to be ineffective or inappropriate for SAV mitigation over very large areas. These tasks are further complicated in that the location and density of SAV can change from year to year depending on variances in weather and water quality. Satellite remote sensing holds great promise for providing a labor and cost-effective means of monitoring and quantifying SAV distribution. For this analysis, sensor specific models based on multinomial logit procedures proved to be the best approach for predicting SAV presence or absence. No models could be developed for low distribution occurrence categories due to a low ratio of events to non-events. Statistical automated selection methods were developed to produce the final models we selected for each sensor. The use of the automated best-subsets method allowed for exploration of a number of potential candidate models based on the number of variables input in the model. The automated stepwise selection method led to the final, most reasonable model as decided upon in the best-subset procedure. For a variable to enter into or remain in the model, a p-value of <0.01 was necessary. A model was considered fit if the Hosmer and Lemeshow test yielded an insignificant difference in groups (p>0.05). Sensor specific models were developed for both the Quickbird and Worldview-II sensors; however LANDSAT 5 specific models were inconclusive largely due to quality of the data.
