Multi-model analysis of lake chlorophyll-a predictions enabled by open, automated data and modeling workflows

Chlorophyll-a is a key indicator of lake water quality and phytoplankton blooms, motivating the need for predictions of future chlorophyll-a concentrations for management. Advance notice of blooms could enable preemptive actions to mitigate water quality impacts. Despite decades of research, however, accurate prediction of future lake chlorophyll-a remains difficult. Moreover, the application of various predictive modeling approaches across different datasets, waterbodies, and prediction windows (e.g., one-week-ahead vs. seasonal predictions) makes model inter-comparison challenging and hampers identification of the most accurate modeling methods. Here, we compared 25 models predicting chlorophyll-a from 1 to 35 days ahead at a drinking water reservoir, using exactly the same data for each model. Models were fit using 3.5 years of data and validated by generating predictions for an additional year. We found that regardless of model type (e.g., machine learning, process-based), models with an autoregressive component performed best, indicating that real-time chlorophyll-a measurements, which are quickly assimilated into models, may be important for accurate lake chlorophyll-a predictions. Notably, this work was enabled by our team’s implementation of automated data quality assurance and access workflows, which permit easy ingestion of new data into models and ongoing analysis of model performance. Our analysis exemplifies how integration of data science and open science into aquatic research can lead to insight regarding the predictability of lake chlorophyll-a.

Primary Presenter: Mary Lofton, Virginia Tech (melofton@vt.edu)

Authors:

Mary Lofton, Virginia Tech  (melofton@vt.edu)

R. Quinn Thomas, Virginia Tech  (rqthomas@vt.edu)

Abhilash Neog, Virginia Tech  (abhilash22@vt.edu)

Arka Daw, Virginia Tech  (darka@vt.edu)

Anuj Karpatne, Virginia Tech  (karpatne@vt.edu)

Adrienne Breef-Pilz, Virginia Tech  (abreefpilz@vt.edu)

Austin Delany, Virginia Tech  (addelany@vt.edu)

Dexter Howard, Virginia Tech  (dwh1998@vt.edu)

Heather Wander, Virginia Tech  (hwander@vt.edu)

Freya Olsson, Virginia Tech  (freyao@vt.edu)

Paul Hanson, University of Wisconsin-Madison  (pchanson@wisc.edu)

Cayelan Carey, Virginia Tech  (cayelan@vt.edu)