The Effect of Data Granularity on Temperature Gradient Modeling in Michigan’s Streams

Abstract

Stream temperature is a critical characteristic for aquatic ecosystems. Many physical, chemical and biological components are influenced by this environmental variable; therefore, it is crucial to understand the factors that take place in thermodynamic processes in these ecosystems. Regression models are useful tools that help us comprehend and explain the drivers of these thermal processes since they can be used for quantifying the magnitude and the type of the relationship between the independent variables (i.e., air temperature, discharge) and the response variable (i.e., stream temperature). However, selection of data granularity (or time aggregation) of data may often be a key decision for modelers. Although granularity of data is selected based on the ecological relevance of data to the question of interest in many cases, it may arbitrarily be selected by the researchers in many other cases. However, data granularity can substantially influence model coefficients, can affect the model predictions, and influence evaluation of model fitness and interpretation of model outputs. In this article, we adopted regression models and applied different data granularity scenarios to investigate the consequences of data granularity selection in modeling approaches. Our findings showed that using different data granularities resulted in considerable changes in regression coefficients in the models. Our results also revealed that overall model fitness increased with coarser-scale data granularity and model selection was influenced by the type of data granularity. This study might be helpful for modelers and environmental managers since it highlights the significance of selection of data granularity, and proposes a different point of view in model design, evaluation and application from the perspective of data selection.