Using satellite data to improve the leaf phenology of a global Terrestrial Biosphere Model

Natasha MacBean, Fabienne Maignan, Philippe Peylin, Cédric Bacour, François‐Marie Bréon, Philippe Ciais2015 · DOI

TL;DR

This paper tackles a critical challenge in climate modeling: accurately simulating plant phenology (leaf onset and senescence). By using satellite-derived vegetation data (NDVI) to optimize the ORCHIDEE Terrestrial Biosphere Model, the researchers significantly improved its representation of seasonal plant cycles, particularly for temperate and boreal forests. This advancement is crucial for better predictions of carbon cycles and climate feedbacks.

What Problem Does This Paper Address?

Terrestrial Biosphere Models (TBMs) are essential tools for understanding how ecosystems interact with the climate, particularly concerning carbon, water, and energy cycles. A key component of these models is the accurate representation of plant phenology – the timing of leaf onset, growth, and senescence. However, many existing TBMs struggle to accurately reproduce observed phenological patterns, leading to biases in simulations of ecosystem processes. This paper addresses this gap by developing and applying a method to optimize phenology-related parameters within the ORCHIDEE TBM using satellite observations, aiming to improve its global simulation capabilities.

Methods

The core of this study is the application of a multi-site (MS) data assimilation (DA) framework to optimize the phenology parameters of the ORCHIDEE TBM. This approach uses a variational DA procedure, iteratively minimizing a cost function that quantures the misfit between model outputs and satellite observations. A key aspect is the use of a multi-site optimization, where data from multiple locations are assimilated simultaneously to derive a single, generic parameter set for each Plant Functional Type (PFT). This contrasts with single-site (SS) optimization, which can lead to parameter sets that are too specific to individual locations. The study also employed a curve-fitting method (Thoning et al., 1989) to derive phenological metrics like Start of Season (SOS) and End of Season (EOS) from time-series data.

Datasets

  • MODIS NDVI v5: This dataset, derived from the MODerate resolution Imaging Spectrometer (MODIS) on NASA's Terra satellite, provided daily global surface reflectance data. It was processed to derive the Normalized Difference Vegetation Index (NDVI) for the period 2000-2008.
  • ERA-Interim meteorological fields: These 3-hourly meteorological data were used to drive the ORCHIDEE model in forced offline mode for simulations.
  • Olson land cover classification: Used to derive PFT fractions at the model's grid resolution.
  • Zobler soil texture classes: Used for soil characterization within the model.

Key Insights and Ideas

  • Satellite data is a powerful tool for TBM calibration: The study demonstrates that readily available satellite vegetation indices like NDVI can be effectively used to constrain and improve the phenology parameters of complex TBMs.
  • Multi-site optimization yields generalizable parameters: The MS approach successfully produced a single set of phenology parameters for each PFT that performed well across multiple sites and globally, reducing the issue of site-specific parameterization.
  • Phenology optimization significantly impacts model bias: The optimized parameters led to a substantial reduction in model-data misfit, particularly by advancing leaf senescence in temperate and boreal forests.
  • Model structure limitations persist for certain ecosystems: While improvements were seen for many PFTs, tropical deciduous trees and C4 grasses showed limited improvement, suggesting potential structural deficiencies in the phenology models for these ecosystems.
  • Optimized phenology affects global carbon cycle estimates: The improved phenology led to a decrease in simulated global mean annual Gross Primary Production (GPP), highlighting the sensitivity of carbon cycle estimates to phenological representation.

Key Results

  • Improved model-data fit: The optimization significantly reduced the Root Mean Square Error (RMSE) and increased the correlation (R) between simulated fAPAR and observed NDVI for temperate and boreal PFTs. For example, for boreal PFTs, the median RMSE reduction was 50-60%, and R increased from 0.41 to 0.9.
  • Global correlation increase: The global median correlation between modeled fAPAR and MODIS NDVI increased from 0.56 to 0.67 after optimization.
  • Reduced phenology timing bias: The global median bias for End of Season (EOS) was dramatically reduced from 33 days to 5 days. Start of Season (SOS) bias also decreased from 22 to 14 days.
  • Impact on GPP: The simulated global mean annual GPP decreased by approximately 10 Pg C yr⁻¹ over the 1990-2010 period due to a shortened Growing Season Length (GSL).
  • Limited improvement for specific PFTs: Tropical broadleaved raingreen (TrBR) and natural C4 grasses (NC4) showed less significant improvements, with NC4 grasses exhibiting no discernible improvement in model-data misfit.

Limitations and Future Work

  • Model structural deficiencies: The limited improvement for tropical deciduous trees and C4 grasses suggests that the current phenology model structures may be inadequate for these water-limited ecosystems. Further modifications are needed.
  • Spatial variability: The study acknowledges that coarse-resolution observations may not fully capture fine-scale spatial variability within PFTs, which can influence phenology. Future work could explore species-level optimization or incorporating explicit spatial variability terms.
  • Data assimilation limitations: The authors note that the DA system's assumptions of linearity and Gaussian errors might underestimate true posterior parameter uncertainty.
  • Future work suggestions: The authors propose that an elicitation exercise involving experts could help derive better prior probability distributions for phenology parameters. They also suggest testing models under future warming scenarios to assess their predictive capabilities. The framework presented could be extended to improve phenology in all global TBMs.

Summary updated

Abstract

Correct representation of seasonal leaf dynamics is crucial for Terrestrial Biosphere Models (TBMs), but many such models cannot accurately reproduce observations of leaf onset and senescence. Here we optimized the phenology-related parameters of the ORCHIDEE TBM using satellite-derived Normalized Difference Vegetation Index data (MODIS NDVI v5). We found the misfit between the observations and the model decreased after optimisation for all boreal and temperate deciduous Plant Functional Types, primarily due to an earlier onset of leaf senescence. The model bias was only partially reduced for tropical deciduous trees and no improvement was seen for natural C4 grasses. Spatial validation demonstrated the generality of the posterior parameters for use in global simulations, with an increase in global median correlation of 0.56 to 0.67. The simulated global mean annual GPP decreased by ~10 Pg C yr−1 over the 1990–2010 period due to the substantially shortened Growing Season Length (up to 30 days in the Northern Hemisphere), thus reducing the positive bias and improving the seasonal dynamics of ORCHIDEE compared to independent data-based estimates. Finally, the optimisations led to changes in the strength and location of the trends in the simulated GSL and mean annual fAPAR, suggesting care should be taken when using un-calibrated models in attribution studies. We suggest that the framework presented here can be applied for improving the phenology of all global TBMs.
BibTeX
@misc{macbean2015using,
  author = {Natasha MacBean and Fabienne Maignan and Philippe Peylin and Cédric Bacour and François‐Marie Bréon and Philippe Ciais},
  title = {{Using satellite data to improve the leaf phenology of a global Terrestrial Biosphere Model}},
  year = {2015},
  doi = {10.5194/bgd-12-13311-2015},
}
Figure