What is the purpose of calibrating and validating environmental models, and what are two common performance metrics?

• A. Calibration tests predictive ability; Validation tunes parameters; metrics: p-values and t-tests.
• B. Calibration tunes model parameters to fit observed data; Validation tests predictive ability on independent data; metrics: R^2 (or NSE), RMSE, MAE.
• C. Calibration and validation are the same; metrics: AIC and BIC.
• D. Calibration adjusts data collection frequency; validation calibrates instruments; metrics: correlation only.

Answer: (B)

Calibrating and validating an environmental model serves two linked purposes: making the model's outputs match observed data as closely as possible, and then testing whether those outputs can reliably predict new, unseen data. Calibration adjusts the model’s parameters so the simulated results line up with measurements you’ve collected. Validation takes that calibrated model and checks its predictive ability using data that wasn’t used during calibration, which helps ensure the model isn’t just overfitting the calibration data but can generalize to real-world conditions.

For evaluating how well the model performs, several metrics are commonly used. R^2 (or NSE in hydrology contexts) shows how much of the variability in observations the model explains. RMSE and MAE quantify the size of prediction errors, with RMSE being more sensitive to larger errors and MAE representing the average error magnitude. It's typical to report more than one metric to capture different aspects of performance.

P-values and t-tests aren’t standard measures of predictive accuracy, and information criteria like AIC or BIC are more about comparing model structures than directly assessing how well predictions match observed data.