Optimizing the flow adjustment of constituent concentrations via LOESS for trend analysis

Area of Science:

• Environmental Science
• Hydrology
• Statistical Modeling

Background:

• Stream constituent concentrations are influenced by discharge, necessitating flow adjustment for accurate trend analysis.
• Locally Weighted Scatterplot Smoothing (LOESS) is a common method for flow-adjusting water quality data, using a smoothing parameter (f).
• Current practice often uses a single, non-optimized f value, potentially affecting analysis accuracy.

Purpose of the Study:

• To develop and validate a robust, automated method for determining the optimal smoothing parameter (fopt) for LOESS flow-adjustment.
• To minimize prediction error in LOESS fitting for stream water quality data.
• To provide a reliable approach for selecting the best f value for trend analysis.

Main Methods:

• Utilized an iterative K-fold cross-validation procedure to identify the optimal f value (fopt).
• Applied the method to 119 datasets of seven different constituents across 17 stream monitoring sites.
• Recommended a 10x10 cross-validation approach for selecting fopt.

Main Results:

• The study developed an automated method to determine the optimal f value (fopt) for LOESS flow-adjustment.
• The 10x10 cross-validation procedure effectively minimized prediction error in LOESS fits.
• While default f values did not alter trend interpretations in this specific dataset, the method offers enhanced accuracy.

Conclusions:

• An automated, cross-validation-based method for selecting the optimal LOESS smoothing parameter (fopt) was successfully developed.
• The proposed method provides a more rigorous approach to flow-adjusting water quality data for trend analysis.
• Implementation in R is provided, facilitating broader application in water quality studies.