Calibration reveals limitations in modeling rainfall interception at the storm scale

• Published in: Journal of hydrology (Amsterdam) 2020-05, Vol.584, p.124624, Article 124624
• Main Authors: Linhoss, Anna C., Siegert, Courtney M.
• Format: Article
• Language: English
• Subjects: comparative study; deciduous forests; Fagus grandifolia; Forest canopy; hydrologic cycle; Interception; Liriodendron tulipifera; mechanics; Model evaluation; model validation; PEST, parameterization; prediction; rain; Rainfall; rivers; storms; uncertainty analysis; water budget; water interception
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2020.124624

•This paper compares five rainfall interception models using an automatic calibration routine.•The models accurately simulate cumulative canopy interception.•Simulations of canopy interception during single storm events are problematic.•During single storm events, simulated canopy interception is less variable than measured interception.•Poor model performance is from errors in parameter ranges, measurements, and/or model conceptualization. Rainfall interception by the forest canopy plays an important role in the water budget by removing water from the terrestrial hydrologic cycle as an atmospheric loss. Effective models of canopy interception are critical for simulating the water budget and river flows. Over the years, several models have been developed to simulate canopy interception. Few comparative studies have been conducted that assess how well these models simulate measured interception. This study compared five mechanistic canopy interception models including the Rutter, Rutter Sparse, Gash, Gash Sparse, and Liu models. The objective of the study was to use an objective automatic calibration to compare how the models behave relative to each other. The objective of the study was not to establish definitive optimal parameter sets. Each interception model was calibrated independently using the Parameter Estimation and Uncertainty Analysis program (PEST), an automatic parameter estimation routine. The five models were calibrated using field measurements from an eastern deciduous forest dominated by American beech and yellow-poplar under leafed and leafless conditions. The models performed well in predicting cumulative interception, with errors ranging between 0.0% and 14.9%. However, model performance declined when assessing individual rainfall events where the coefficient of determination (R2) between measured and modeled interception events ranged between 0.21 and 0.48. The models were unable to simulate very low or very high levels of interception. Measured interception ranged between 0.2 and 12.2 mm while modeled interception only ranged between 1.2 and 6.9 mm. The inability of the models to satisfactorily simulate measured interception is likely due to: 1) errors in parameter ranges, 2) measurement errors, and/or 3) a fundamental misunderstanding of interception mechanics. These results indicate an important gap in our ability to understand a substantial portion of the water budget.