In Before-After monitoring studies, statistical models are used to characterize baseline (i.e., predisturbance) conditions, and to detect, quantify, and forecast change during operational monitoring (i.e., post-disturbance). To establish best practices for analyzing monitoring data, a model evaluation was developed and applied using Marine Renewable Energy (MRE); a case study of a disturbance with no best practice monitoring methods. The evaluation was performed on normal and non-normal acoustic metrics representative of MRE monitoring data. Evaluated models included: generalized regression models, time series models, and nonparametric models. 10-fold Cross Validation was used to evaluate baseline model fit. Models were then fit to 5 simulated Before-After change scenarios using Intervention Analysis. A power analysis was used to evaluate model ability to detect change. Residual error diagnostics were used to quantify model fit and forecast accuracy. State-space models are recommended for baseline characterization. Deterministic Parametric models are recommended to detect change. Time series and semi-parametric models are recommended to quantify change. Nonparametric models are recommended to forecast change. These recommendations form best practices for analyzing MRE monitoring data, which enables comparisons among MRE sites and reduces uncertainty in environmental effects. The evaluation approach is applicable to any monitoring program.
Evaluating Statistical Models for Baseline Characterization and Measuring Change in Environmental Monitoring Data
Title: Evaluating Statistical Models for Baseline Characterization and Measuring Change in Environmental Monitoring Data
January 01, 2016
Thesis Type: Master's Thesis
Academic Department: Aquatic and Fishery Sciences
Linder, H. (2016). Evaluating Statistical Models for Baseline Characterization and Measuring Change in Environmental Monitoring Data. Master's Thesis, University of Washington.