All Times EDT
Keywords: watershed, high-resolution sensor data, gap filling, generalized additive model, distributed non-linear lag model
The riverine flux data obtained via high-resolution sensor provide detailed information that is novel to watershed science and also provides opportunities for fine-scale statistical and predictive modeling that is not readily discussed in the classical literature. One common problem with sensor data is the occurrence of data gaps, caused by?power failure, extreme weather, instrument malfunction, maintenance, or other causes. We first propose a generalized additive model (GAM) gap filling method. For high frequency multivariate environmental time series data, GAM effectively incorporates temporal characteristics such a seasonal patterns and trends with smooth terms, and also correlation among variables by linear terms. An important issue is the effect of antecedent conditions created by extreme weather on riverine loads of nitrogen and dissolved organic matter. High-resolution sensor data enables the joint modeling of load and many lags of one or several driver variables. Distributed non-linear lag models (DNLMs) is a framework to describe such a relationship which is potentially varying non-linearly between the response at time t and the predictors at a large number of time lags. We apply DNLMs to hourly watershed data after gap filling in order to explore the effect of runoff on nitrate and dissolved organic matter.