Influence of Interannual Leaf Phenology Dynamics on Evapotranspiration Predictions

Abstract Evapotranspiration (ET) plays a crucial role in determining water and energy partitioning of precipitation between land and atmosphere. A recognized uncertainty in modeling of ET is a proper representation of leaf phenology dynamics. Many existing land surface models often employ a “static scheme” of leaf phenology while performing ET simulations, i.e., they assume identical intra-annual variation in leaf phenology, often quantified as the variation in leaf area index (LAI), from year to year. There is a need to assess the influence of this “static” scheme on ET prediction errors. Furthermore, it is important to determine where improvements in predictions of leaf phenology dynamic models will yield the largest improvements in ET predictions. Utilizing site-specific artificial neural network models for the estimation of ET, this study investigates the influence of static versus dynamic representation of leaf phenology dynamics on ET estimation errors across 30 flux network (FLUXNET) sites. Results indicate that neglecting the interannual LAI dynamics in the ET prediction model led to an additional loss of accuracy of up to 29% (0.18) with the average loss being about 1.7% (0.01) based on the percentage bias (coefficient of determination) metric. Our findings also show that interannual leaf phenology dynamics have a larger impact on ET estimates in warmer and drier regions. Phenological traits such as green-up onset anomaly and maximum annual LAI anomaly are identified as key contributors to model error in static phenology scenarios.