Assessing estuary condition using Chiu’s Latent Health Factor Index
Citation:
Power, Jim. Assessing estuary condition using Chiu’s Latent Health Factor Index. Ecological Society of America Annual Meeting, New Orleans, Louisiana, August 05 - 10, 2018.
Impact/Purpose:
Effective management of estuaries requires a comprehensive assessment of estuary condition. However, the data available for such an assessment is often comprised of limited disparate measures collected at point locations. A traditional approach is that specific metrics, such as oxygen and nutrient concentrations, can be regarded as measures that indicate estuary health. However, integrating this data to evaluate an estuary’s overall ecological condition is difficult. Traditional approaches often incorporate subjective assessments and may lack statistical rigor, especially when spatial or temporal comparisons of estuary condition are desired. Recently Chiu et al. (2011; 2013) presented an approach that reverses the traditional perspective. Chiu’s approach considers the observed values of an estuary’s metrics as response values, and that these are the outcome of an unobserved indicator of overall estuary. Chiu terms this unobserved (latent) variable as the Latent Health Factor Index, or LHFI, and they describe how a hierarchical Bayesian analysis can be used to obtain a distribution of a region’s LHFI. This LHFI distribution can then be used in comparisons to the LHFI’s from other regions, or to evaluate the change in a region’s LHFI over time. Because the LHFI is obtained as a distribution, the level of uncertainty associated with a LFHI score can be evaluated. Finally, the LHFI can in turn be evaluated as the response to external factors, such as watershed urbanization.
Description:
Effective management of estuaries requires a comprehensive assessment of estuary condition. However, data available for such an assessment is often comprised of limited and disparate measures collected at point locations. A traditional approach is that specific metrics, such as oxygen and nutrient concentrations, can be regarded as measures of estuary health. However, integrating this data to evaluate an estuary’s overall ecological condition is difficult. Chiu et al. (2011; 2013) present a different perspective, in which the observed metrics are response variables whose values reflect an unobserved latent explanatory variable (the Latent Health Factor Index, or LHFI) as an indicator of the estuary’s health. This latent variable can in turn be modeled as a response variable that depends on other factors, such as watershed characteristics. The research presented here utilized five water quality measures (bottom dissolved oxygen, surface chlorophyll a concentration, near-surface light transmissivity, and the total concentrations of surface nitrogen and phosphate) that were collected in 2011 from six regions of San Francisco Bay area as part of the Environmental Protection Agency’s National Coastal Condition Assessment. These metrics were used in a hierarchical Bayesian model to determine the LHFI values for the six San Francisco Bay regions. The hierarchical model presented here considered the LHFI as the response variable in an ANOVA context, where the explanatory factor consisted of regions of San Francisco Bay. The LHFI for each region was in turn combined with parameters indexing each metric’s identity (dissolved oxygen, nitrogen, etc.) to serve as predictors of the observed metric values. A Monte Carlo Markov Chain simulation was run for 5,000 samples of the LFHI and associated parameters. The resulting parameter distributions showed that the South Bay and Lower South Bay had the poorest LHFI, a result consistent with other assessments of the region. Advantages of the LHFI approach are that it provides a statistically rigorous numerical assessment of a region’s heath for which credible intervals can be evaluated to judge the uncertainty present in the region’s LHFI. Additionally, the contributing metrics can be evaluated as to which are important contributors to the health score. Finally, such models can be extended to include watershed characteristics, such as urbanization, so that their influence on estuary LHFI score can also be evaluated.