Parameter sensitivity analysis for a biochemically-based photosynthesis model

Abstract

A challenge for the development of Land Surface Models (LSMs) is improving transpiration of water exchange and photosynthesis of carbon exchange between terrestrial plants and the atmosphere, both of which are governed by stoma in leaves. In the photosynthesis module of these LSMs, variations of parameters arising from diversity in plant functional types (PFTs) and climate remain unclear. Identifying sensitive parameters among all photosynthetic parameters before parameter estimation can not only reduce operation cost, but also improve the usability of photosynthesis models worldwide. Here, we analyzed 13 parameters of a biochemically-based photosynthesis model (FvCB), implemented in many LSMs, using two sensitivity analysis (SA) methods (i.e., the Sobol’ method and the Morris method) for setting up the parameter ensemble. Three different model performance metrics, i.e., Root Mean Squared Error (RMSE), Nash Sutcliffe efficiency (NSE), and Standard Deviation (STDEV) were introduced for model assessment and sensitive parameters identification. The results showed that among all photosynthetic parameters only a small portion of parameters were sensitive, and the sensitive parameters were different across plant functional types: maximum rate of Rubisco activity (V_cmax25), maximum electron transport rate (J_max25), triose phosphate use rate (TPU) and dark respiration in light (R_d) were sensitive in broad leaf-evergreen trees (BET), broad leaf-deciduous trees (BDT) and needle leaf-evergreen trees (NET), while only V_cmax25 and TPU are sensitive in short vegetation (SV), dwarf trees and shrubs (DTS), and agriculture and grassland (AG). The two sensitivity analysis methods suggested a strong SA coherence; in contrast, different model performance metrics led to different SA results. This misfit suggests that more accurate values of sensitive parameters, specifically, species specific and seasonal variable parameters, are required to improve the performance of the FvCB model.