Simulating the net ecosystem CO2 exchange and its components over winter wheat cultivation sites across a large climate gradient in Europe using the ORCHIDEE-STICS generic model

Nicolas Vuichard, Philippe Ciais, Nicolas Viovy, Longhui Li, Éric Ceschia, M. Wattenbach, Christian Bernhofer, Carmen Emmel, Thomas Grünwald, W.w.p. Jans, Benjamin Loubet, Xiuchen WuAgriculture 2016 · DOI

TL;DR

This study evaluates the ORCHIDEE-STICS model's ability to simulate CO2 exchange in winter wheat fields across Europe. The model shows good performance for gross primary production (GPP) but struggles with ecosystem respiration (TER) and net ecosystem exchange (NEE), particularly for annual budgets. Fertilization emerges as the most sensitive management practice impacting model accuracy.

What Problem Does This Paper Address?

Accurately modeling the carbon cycle is crucial for understanding and mitigating climate change. Croplands, which cover a significant portion of the Earth's surface, play a vital role in this cycle, but their carbon exchange is complex, influenced by both climate and intensive human management. Global terrestrial ecosystem models (GTEMs) have historically lacked detailed representations of agricultural systems, leading to uncertainties in global carbon budget estimates. This paper addresses this gap by evaluating a specific GTEM, ORCHIDEE-STICS, which incorporates crop-specific processes, to simulate the net ecosystem CO2 exchange (NEE) and its components (GPP and TER) across a diverse range of European winter wheat cultivation sites. The goal is to assess the model's performance and identify areas for improvement, particularly concerning the impact of management practices.

Methods

The researchers employed the ORCHIDEE-STICS model, a coupled system combining the ORCHIDEE biophysical model with the STICS agronomic model. STICS handles crop phenology, development stages, and management practices (like fertilization and sowing dates), providing inputs such as Leaf Area Index (LAI) to ORCHIDEE. ORCHIDEE then simulates carbon assimilation (GPP), respiration (autotrophic and heterotrophic), and ultimately, NEE.

To evaluate the model, they used a suite of statistical metrics, including:

  • Root Mean Square Error (RMSE): To quantify the average magnitude of errors.
  • Figure of Merit (FMT): For LAI, measuring the overlap between observed and modeled curves.
  • Correlation coefficient (R) and Normalized Standard Deviation (NSTD): Used in Taylor diagrams to assess the agreement in seasonal variations of GPP, TER, and NEE.
  • Bias and Error Percentage (EP): For annual budgets, to quantify systematic over- or underestimation.

A 1000-year spin-up simulation was performed to stabilize soil carbon pools. Sensitivity analyses were conducted by varying plant variety, sowing date, and fertilization intensity to understand their impact on model outputs.

Datasets

The study utilized data from seven European winter wheat sites collected during the CarboEurope-IP project (2003-2007). These sites spanned a large climate gradient. The datasets included:

  • Meteorological data: Hourly measurements of temperature, precipitation, radiation, humidity, and wind speed.
  • CO2 flux data: Eddy-covariance measurements of Net Ecosystem Exchange (NEE).
  • Leaf Area Index (LAI) and Above Ground Biomass (AGB): Measured at various sites.
  • Management information: Details on sowing dates, harvest dates, fertilization (type and amount), and ploughing events.

The NEE data underwent gap-filling using the Marginal Distribution Sampling (MDS) method, and NEE was partitioned into GPP and TER using the algorithm by Reichstein et al. (2005).

Key Insights and Ideas

  • GPP is generally well-simulated, but TER and NEE are challenging: The model excels at capturing the seasonal dynamics of GPP, but its performance significantly drops for TER and especially NEE, particularly when looking at annual totals. This highlights that respiration processes are more difficult to model accurately.
  • Phenology is a key driver of GPP errors: Discrepancies in GPP simulation are often linked to errors in simulating crop phenology, such as LAI development and emergence timing.
  • Fertilization is the most influential management factor: The sensitivity analysis clearly showed that fertilization intensity has the largest impact on model performance for all fluxes (GPP, TER, and NEE), both seasonally and annually. This underscores the importance of accurate fertilization data for crop modeling.
  • Soil carbon disequilibrium is a critical missing piece: The study suggests that the lack of accounting for soil carbon disequilibrium (due to past land-use and management) is a major limitation for accurately simulating annual NEE budgets.
  • Model struggles with post-harvest and winter fluxes: The model often fails to capture CO2 fluxes outside the main growing season, such as re-growth after harvest or winter CO2 exchange, which can be significant.

Key Results

  • Seasonal GPP: The model reproduced seasonal GPP variations well, with 5 out of 7 sites showing a correlation coefficient (R) > 0.9 and a Normalized Standard Deviation (NSTD) between 0.8 and 1.2.
  • Seasonal TER: Model performance for TER was lower, especially in terms of NSTD.
  • Annual Budgets:
    • Mean simulated deviations for GPP and TER were within 10% and 15% of observed annual budgets, respectively.
  • For NEE, the model's mean deviation was around 35% of the observed annual value, indicating a significant limitation in predicting annual NEE budgets.
  • Sensitivity Analysis:
    • Plant variety had a moderate impact, affecting GPP at only 4 sites.
  • Sowing date had a lesser impact, primarily modulating the phase rather than the amplitude of seasonal cycles.
  • Fertilization showed the strongest influence, with significant changes in GPP, TER, and NEE performance.

Limitations and Future Work

  • Soil Carbon Disequilibrium: The model's initialization did not account for soil carbon disequilibrium, which is crucial for accurate long-term carbon balance simulations. Future work should incorporate site history and tillage effects.
  • Post-Harvest and Winter Fluxes: The model's inability to simulate CO2 exchange outside the main growing season (e.g., re-growth, winter respiration) is a limitation.
  • Management Interactions: The sensitivity analysis treated management practices independently, neglecting potential interactions between them. Future studies could explore these combined effects.
  • Variety Parameterization: The study used a generic variety parameterization for STICS due to a lack of specific data for all sites, which might have influenced results.
  • TER and HR Representation: The study suggests that errors in representing heterotrophic respiration (HR) and the climate response of litter decomposition contribute to TER discrepancies.

The authors suggest that improving the representation of soil carbon dynamics and incorporating more detailed management information (especially fertilization and land-use history) are key steps for enhancing the accuracy of crop models like ORCHIDEE-STICS for regional carbon budget estimations.

Summary updated

BibTeX
@inproceedings{vuichard2016simulating,
  author = {Nicolas Vuichard and Philippe Ciais and Nicolas Viovy and Longhui Li and Éric Ceschia and M. Wattenbach and Christian Bernhofer and Carmen Emmel and Thomas Grünwald and W.W.P. Jans and Benjamin Loubet and Xiuchen Wu},
  title = {{Simulating the net ecosystem CO2 exchange and its components over winter wheat cultivation sites across a large climate gradient in Europe using the ORCHIDEE-STICS generic model}},
  booktitle = {Agriculture},
  year = {2016},
  doi = {10.1016/j.agee.2016.04.017},
}
Figure