Regression Estimation of Trends in Temperature When Time and Date of Sampling Are Haphazard

• Activity Number: 619
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Brian R. Gray*+ and Yulia R. Gel and Vyacheslav Lyubchich
• Companies: U.S. Geological Survey and University of Waterloo and University of Waterloo
• Keywords: confounding ; multilevel model ; river temperature ; stream temperature ; trend estimation

Abstract

The estimation of trends in river and stream temperature from measurements from haphazard times of day or haphazard dates is discouraged because water temperatures typically vary with time of day and date. However, temperature data from haphazard times and dates may arise under broadly defensible designs. For example, logistic, cost or legal considerations may prevent the placement of automated samplers at a random or systematic sample of locations in the main stem of a river. Confounding of temperature trends with trends in sampling times or dates that may occur under haphazard sampling of times or dates may be addressed using multilevel regression models. Such models may need to address that temperature-time of day associations may vary with day, temperature-date-of-sampling associations may vary with year, sample sizes within days may often represent singletons, and trends in time and/or date of sampling may be confounded with longitudinal trends; few days or years may lead to the expection of biases in variance estimators of time(day), location(day) and date(year) associations. We address these concerns using simulated and river temperature data.