Comparison of the Two Most Common Phenology Algorithms Imbedded in Land Surface Models

A reasonable representation of plant phenology in land surface models (LSMs) is necessary to accurately simulate the momentum, heat, and mass interchanges between land and the atmosphere from ecosystem to global scales. Many process-based phenology algorithms have been developed and coupled to LSMs to describe seasonal vegetation changes. The growing degree day (GDD) and the growing season index (GSI) algorithms are the two most well-known algorithms used in LSMs for simulating phenophases. However, assessments of these two most commonly used phenology algorithms in LSMs are quite scarce. In this study, these two phenology algorithms were respectively coupled with the Community Land Model (CLM) and the Dynamic Land Model (DLM) to obtain four modeling scenarios. The simulation accuracy of phenophases and gross primary production (GPP) in the four scenarios was assessed against observations at the site scale, focusing on deciduous forests and grasses. The three main findings were as follows: (a) the difference in simulated phenological events between different LSMs coupled with the same phenological algorithm was small and less than 1 day, DLM performed better than CLM; (b) compared with the GSI algorithm and regardless of whether it was coupled with the DLM or CLM model, the GDD model performance was better for spring phenology and worse for autumn phenology; (c) GSI performance was better than GDD for GPP simulation over different vegetation function types across different bioclimatic zones: on average, the root mean square error and the index of agreement were about 8.0% higher and about 6.5% lower, respectively.