An Observation‐Driven Approach to Improve Vegetation Phenology in a Global Land Surface Model

• Main Authors: Jana Kolassa, Rolf H. Reichle, Randal D. Koster, Qing Liu, Sarith Mahanama, Fan‐Wei Zeng
• Format: Article
• Language: English
• Published: American Geophysical Union (AGU) 2020-09-01
• Series: Journal of Advances in Modeling Earth Systems
• Subjects: vegetation modeling; model calibration; plant functional types; photosynthesis; particle swarm optimization
• Online Access: https://doi.org/10.1029/2020MS002083

Abstract An empirical model calibration approach is presented that aims to approximate missing biosphere processes in a global land surface model without the need for substantial model structural changes. The strategy is implemented here using the NASA Catchment‐CN land surface model and Moderate Resolution Imaging Spectroradiometer (MODIS) observations of the fraction of absorbed photosynthetically active radiation (FPAR). Existing plant functional types (PFTs) of the Catchment‐CN model are divided into three subtypes, based on the bias between the model‐simulated and MODIS‐observed FPAR. Separate sets of vegetation parameters for each subtype are then calibrated at a small number of grid cells with homogeneous, single‐PFT land cover, using MODIS FPAR reference observations from 2003 to 2009. The effectiveness of the empirical approach at improving the realism of modeled vegetation dynamics is investigated with two global model simulations for the period 2010–2016, one using the newly calibrated parameter values and the other using the original values. Globally, the calibrated parameters reduce the root mean square error (RMSE) of the modeled FPAR with respect to MODIS by 0.029 (∼10%) on average. In some regions, substantially larger RMSE reductions are achieved. RMSE reductions are primarily driven by model bias reductions, with neutral effects on the temporal correlation skill. While the empirical approach is suitable for achieving consistent model improvements, it is shown to be sensitive to the characteristics of the model error, specifically a dominance of the bias component in the case of Catchment‐CN. Ultimately, more fundamental model structural changes may be required to achieve better improvements.