Performance of the ecosystem demography model (EDv2.2) in simulating gross primary production capacity and activity in a dryland study area

Highlights

• Performance of the Ecosystem Demography (EDv2.2) model in drylands is evaluated.

• The EDv2.2 model captures the GPP capacity at the low productive site and years.

• The simulated GPP follows the general trend of GPP estimated from satellites.

• The model exaggerates the greening and senescing trends.

• EDv2.2 model needs improvement in representing the heterogeneity of drylands.

Abstract

Dryland ecosystems play an important role in the global carbon cycle, including regulating the inter-annual global carbon sink. Dynamic global vegetation models (DGVMs) are essential tools that can help us better understand carbon cycling in different ecosystems. Currently, there is limited knowledge of the performance of these models in drylands partly due to characterizing the heterogeneity of the vegetation and hydrometeorological conditions. The aim of this study is to evaluate the performance of a DGVM for drylands to facilitate improved understanding of gross primary production (GPP) as one of the important components of the carbon cycle. We performed a sensitivity analysis and calibrated the Ecosystem Demography (EDv2.2) DGVM to simulate GPP in a dryland watershed (Reynolds Creek Experimental Watershed, Idaho) in the western US for the years 2000-2017. GPP capacity and activity were investigated by comparing model simulations with GPP estimated from eddy covariance data (available from 2015-2017) and remote sensing products (2000-2017). Our results show good performance of EDv2.2 at daily timesteps ($RMSE\approx 0.38\left[\text{kgC}/m^2/\text{year}\right])$between simulated and measured GPP in lower elevations of the watershed. Moreover, remote sensing analysis show that EDv2.2 captures the long-term trends in this ecosystem and performs relatively well in capturing phenometrics (start/end of the season). The performance of the model degrades in more productive sites with greater GPP (located at higher elevations in the watershed). To improve model performance, future studies will need to introduce additional plant functional types for drylands such as our study area, and modify plant processes (e.g., plant hydraulics and phenology) in the model.

Introduction

Dryland vegetation plays an important role in the global carbon budget, including regulating the global land carbon sink (Ahlström et al., 2015; Metcalfe, 2014; Poulter et al., 2014). Vegetation dynamics in drylands are largely a function of spatial and temporal variability in climate. For example, characteristic vegetation structure, composition, and function (e.g. photosynthesis) can differ markedly between lower and higher elevations in drylands, largely in response to corresponding differences in precipitation and temperature. Climate variability can also drive long-term changes in vegetation dynamics in dryland ecosystems. For example, climate change in the drylands of the Great Basin, US, may lead to a shift from winter snow to rain-dominated precipitation regimes, which in turn may favor fire-prone invasive species, such as cheatgrass (Bromus tectorum), that can convert native shrub-steppe communities to exotic annual grasslands (Concilio et al., 2013; Polley et al., 2013; Scott et al., 2015). These changes in structure and function ultimately affect ecosystem-scale vegetation productivity. Indeed, studies have shown that up to 60 percent of the global carbon sink anomaly can be explained by vegetation dynamics in dryland ecosystems (Ahlström et al., 2015; Poulter et al., 2014). This carbon sink variability is mostly associated with changes in gross primary production (GPP) (Yao et al., 2020). Thus, modelling the spatial and temporal dynamics of GPP in drylands is essential for global-scale studies on carbon balance and atmospheric CO₂.

GPP represents ecosystem-scale apparent photosynthesis and is a primary indicator of the vegetation state of an ecosystem. GPP can be assessed in terms of capacity (i.e., amount) and activity (i.e., dynamics) (Medvigy et al., 2013; Smith et al., 2018). Estimation of GPP at the ecosystem scale in drylands is helpful in understanding food and fiber availability, livestock grazing resources, and long-term processes related to the global carbon cycle (Ryu et al., 2019). However, there is a poor understanding of GPP in drylands due to variable hydrometeorological conditions (e.g., along an elevation gradient) and different plant functional types and photosynthetic pathways (e.g., C3 vs. C4) (Yan et al., 2019), among other factors. Direct measurement of GPP is a challenging task (Ryu et al., 2019; Yan et al., 2019) and commonly used products are based on data from eddy covariance (EC) towers, remote sensing, and dynamic global vegetation models (DGVMs) (for the latest review refer to Ryu et al., 2019). EC data provide the most reliable estimates of GPP capacity; however, in drylands the spatial distribution of EC towers is limited, and their time series may not be long enough to represent GPP dynamics. Remote sensing-based GPP derived from space-based sensors, such as MODIS (Running et al., 2004), provides long-term estimates that can be used for analysis of photosynthetic activity; however, GPP derived from remotely sensed data may be under- or overestimated depending on the ecosystem (Stocker et al., 2019; Verma et al., 2014), thus limiting our understanding of GPP capacity.

Process-based DGVMs are important tools to study GPP capacity and activity that can provide complementary observations to EC tower and remote sensing data. These models can provide simulations of photosynthesis at the leaf scale, as well as at the canopy and ecosystem scales. A wealth of studies have implemented DGVMs in different ecosystem types (see review in Fisher et al., 2018) to estimate GPP from local (EC towers) to regional scales. An evaluation of the DGVMs’ performance prior to implementation should take place and include model parameterization, sensitivity analysis, calibration, and validation (Fer et al., 2018; Keenan et al., 2013; Kuppel et al., 2012; Pandit et al., 2019a; Post et al., 2017; Renwick et al., 2019; Santaren et al., 2007; Wang et al., 2001). However, there is an information gap in the evaluation of DGVMs in drylands and, more specifically, in dryland regions where vegetation productivity rapidly changes, especially across elevation gradients. Many studies have investigated the correlation between simulated GPP and GPP estimated from EC towers or remote sensing (Antonarakis et al., 2014; Pandit et al., 2019a; Renwick et al., 2019; Trugman et al., 2016). Comparing simulated GPP of drylands with EC towers is largely applicable to assessing GPP capacity rather than activity due to limited years of EC data. To fully capture GPP activity more specific metrics are required. Phenometrics, such as start of season (SOS) and end of season (EOS), and long-term trend analyses are examples of criteria that can be used for studying GPP activity (Chen et al., 2016; Forkel et al., 2015; Zhao et al., 2019).

The focus of this paper is on the Ecosystem Demography model version 2 (EDv2.2; Medvigy et al., 2009; Moorcroft et al., 2001). This model has been implemented in a variety of ecosystems (Antonarakis et al., 2014; Davidson et al., 2011; Kim et al., 2012; Levine et al., 2016; Lokupitiya et al., 2016; Medvigy et al., 2013, 2012; Trugman et al., 2016; Xu et al., 2016; Zhang et al., 2015) and, unlike most DGVMs, EDv2.2 represents the vertical and horizontal heterogeneity of terrestrial ecosystems (Longo et al., 2019b, 2019a). A further strength is that EDv2.2 accommodates the local variability of vegetation composition and structure. However, a thorough evaluation of this model in drylands is lacking. The heterogeneity and in many areas, the sparsity of vegetation in dryland communities, makes modeling these ecosystems challenging.

The objective of this study is to explore the ability of EDv2.2 to predict GPP capacity and activity in a dryland study area. To address this objective we (i) perform a sensitivity analysis and a state-of-the-art calibration method; (ii) assess GPP capacity by comparing the model output with EC tower data with varying vegetation productivity; and (iii) assess GPP activity by comparing long-term trends and phenometrics (SOS and EOS) between model GPP simulations and remote sensing data.