Accelerated photosynthesis routine in LPJmL4

Abstract

The increasing impacts of climate change require strategies for climate adaptation. Dynamic global vegetation models (DGVMs) are one type of multi-sectorial impact model with which the effects of multiple interacting processes in the terrestrial biosphere under climate change can be studied. The complexity of DGVMs is increasing as more and more processes, especially for plant physiology, are implemented. Therefore, there is a growing demand for increasing the computational performance of the underlying algorithms as well as ensuring their numerical accuracy. One way to approach this issue is to analyse the routines which have the potential for improved computational efficiency and/or increased accuracy when applying sophisticated mathematical methods. In this paper, the Farquhar–Collatz photosynthesis model under water stress as implemented in the Lund–Potsdam–Jena managed Land DGVM (4.0.002) was examined. We additionally tested the uncertainty of most important parameter of photosynthesis as an additional approach to improve model quality. We found that the numerical solution of a nonlinear equation, so far solved with the bisection method, could be significantly improved by using Newton's method instead. The latter requires the computation of the derivative of the underlying function which is presented. Model simulations show a significantly lower number of iterations to solve the equation numerically and an overall run time reduction of the model of about 16 % depending on the chosen accuracy. Increasing the parameters θ and αC3 by 10 %, respectively, while keeping all other parameters at their original value, increased global gross primary production (GPP) by 2.384 and 9.542 GtC yr−1, respectively. The Farquhar–Collatz photosynthesis model forms the core component in many DGVMs and land surface models. An update in the numerical solution of the nonlinear equation in connection with adjusting globally important parameters to best known values can therefore be applied to similar photosynthesis models. Furthermore, this exercise can serve as an example for improving computationally costly routines while improving their mathematical accuracy.