Combining multiple statistical methods to evaluate the performance of process-based vegetation models across three forest stands

Abstract

Process-based vegetation models are crucial tools to better understand biosphere-atmosphere exchanges and ecophysiological responses to climate change. In this contribution the performance of two global dynamic vegetation models, i.e. CARAIB and ISBA CC , and one stand-scale forest model, i.e. 4C, was compared to long-term observed net ecosystem carbon exchange (NEE) time series from eddy covariance monitoring stations at three old-grown European beech (Fagus sylvatica L.) forest stands. Residual analysis, wavelet analysis and singular spectrum analysis were used beside conventional scalar statistical measures to assess model performance with the aim of defining future targets for model improvement. We found that the most important errors for all three models occurred at the edges of the observed NEE distribution and the model errors were correlated with environmental variables on a daily scale. These observations point to possible projection issues under more extreme future climate conditions. Recurrent patterns in the residuals over the course of the year were linked to the approach to simulate phenology and physiological evolution during leaf development and senescence. Substantial model errors occurred on the multi-annual time scale, possibly caused by the lack of inclusion of management actions and disturbances. Other crucial processes defined were the forest structure and the vertical light partitioning through the canopy. Further, model errors were shown not to be transmitted from one time scale to another. We proved that models should be evaluated across multiple sites, preferably using multiple evaluation methods, to identify processes that request reconsideration.

Accuracy, realism and general applicability of European forest models

“…Earlier findings by Kramer et al (2002) and Morales et al (2005) suggested that forest models have an adequate accuracy regarding daily carbon and water fluxes. Yet, on the multi‐annual time scale, Horemans et al (2017) found larger uncertainties for NEE than on the daily time scale. Our findings using a much larger ensemble of models confirm these earlier findings.…”

“…In addition, only few of these evaluation studies include forest structure variables (e.g., LAI: Richardson et al, 2012; biomass: Klesse et al, 2018). Earlier model evaluations have either focused on selected processes (e.g., NPP: Morales et al, 2005; mortality: Bugmann et al, 2019), relied on short time series of observed data (Kramer et al, 2002), or investigated only few models and sites (Horemans et al, 2017). Yet, the increasing amount of harmonized data recently becoming available across Europe (e.g., Reyer et al, 2020a, 2020b) allows for a rigorous evaluation of the state‐of‐the‐art in forest modeling across different biogeographical regions, forest types and types of data.…”

Description and evaluation of the process-based forest model 4C at four European forest sites

“…The mismatches in phenology were also discussed by Collalti et al (2016). For Sorø, Horemans et al (2017) discussed in great detail the differences between simulated and observed NEE for 4C and concluded that 4C overestimates the importance of high frequency variability because 4C uses the daily temperature to redistribute the weekly calculated NEE and the applied dependency is possibly too sensitive. These daily calculated values are only used for comparison reasons.…”

Improving Phenology Representation of Deciduous Forests in the Community Land Model: Evaluation and Modification Using Long‐Term Observations in China

“…Scholars indicated that the heat requirement could break the dormancy (Murray et al., 1989). This method has been widely used to predict the leaf unfolding dates in most terrestrial biosphere models, such as BIOME‐BGC (White et al., 1997), CARAB (Horemans et al., 2017), CLASS 2.7 (Verseghy et al., 1993), Community Land Model (CLM) 4.5 (Oleson et al., 2013), DLEM (Tian et al., 2011), IBIS (Foley et al., 1996), LPJ‐DGVM (Sitch et al., 2003), and ISAM (El Masri et al., 2015). These models predict earlier leaf unfolding dates in colder regions but later leaf unfolding dates in warmer regions (Chen et al., 2016; Chuine et al., 2010).…”

“…Contrary to the enormous amount of research on leaf unfolding dates, we still have little understanding of the mechanism of leaf falling dates (Chuine et al., 2010; Y. H. Fu et al., 2022; M. Wang et al., 2022). Most models predicted the dates of leaf falling based on the fixed temperature threshold (Chuine et al., 2013; Krinner et al., 2005) as in CARAIB (Horemans et al., 2017), CLASS 2.7 (Verseghy et al., 1993), DLEM (Tian et al., 2011), IBIS (Foley et al., 1996), LPJ‐DGVM (Sitch et al., 2003), and ORCHIDEE (MacBean et al., 2015), the fixed daylength threshold (White et al., 1997) as in CLM4.5 (Oleson et al., 2013; Verseghy et al., 1993), and both of them (Wareing, 1956) as in BIOME‐BGC (White et al., 1997), CEVSA2 (F. Gu, 2007; White et al., 1997), CTEM (White et al., 1997), and ISAM (El Masri et al., 2015). The above methods exhibited large deviations up to 15 ± 17 days in simulated end of photosynthetic uptake (Richardson et al., 2012).…”