Using Data Assimilation of Leaf Area Index to Constrain Decadal Global Carbon Dynamics in the Community Land Model 5.0

Abstract:

The Community Land Model (CLM) is a powerful tool to investigate how the global carbon cycle is changing, and predict its feedback to future climate. To address inaccuracies in the model due to the inadequate representation of terrestrial biogeochemical processes, we employ an Ensemble Adjustment Kalman Filter in the Data Assimilation Research Testbed (DART) coupled with the CLM5.0 to ensure that modeled leaf area index (LAI) agrees with historical global LAI observations. LAI influences latent heat (LE) and has a strong control over gross primary productivity (GPP), so constraining LAI also potentially constrains these fluxes. We conducted a 60-ensemble experiment that assimilates biweekly GIMMS LAI3g observations into CLM5 globally from 2000 to 2010. A corresponding decade-long freerun experiment (in which DA is not implemented) is conducted for comparison.

Globally, LAI was reduced by 0.4 m²/m² (25.6%) between the freerun and the assimilation run. This resulted in a reduction of gross primary productivity from 859.26 gC/m² per year to 707.44 gC/m² per year, or 17.7%. The impact on global latent heat flux (LE) was less marked, but still overall LE was reduced by 6.6%. The root mean square error (RMSE) of LAI was reduced from 0.42 m²/m² to 0.05 m²/m², and RMSE of GPP decreased from 129.07 gC/m² per year to 43.77 gC/m².

Data assimilation corrects LAI in different regions of the globe to different extents. In particular, DA significantly reduces the overestimation of LAI at high latitudes, which are dominated by needleleaf evergreen boreal trees and broadleaf deciduous boreal shrub plant functional types (PFTs), suggesting these are poorly parameterized in the model. Generally, DA of LAI significantly improves the modeled LAI in almost every natural PFT but has less impact in regions dominated by crop PFTs.

We found our system works very well despite large model bias in LAI and attribute this to the advanced spatially-varying time-adaptive inflation employed by DART and highlight how this works.

To investigate the impact of initial conditions on forecasting LAI for different PFTs, we conducted a forecast run which starts from the analysis produced by the assimilation run in 2006 and runs to 2010. The result shows that DA improves the accuracy of forecast LAI in evergreen PFTs more than in PFTs with annual leaf turnover.