Evaluation of the ecosystem dynamics of Amazonian rainforests with 4 global vegetation models

Transcription

1 Evaluation of the ecosystem dynamics of Amazonian rainforests with 4 global vegetation models David Galbraith, Naomi Levine, Brad Christofferson, Hewlley Imbuzeiro, Tom Powell, Yadvinder Malhi, Patrick Meir, Antonio Carlos Lola da Costa, Marcos Costa, Scott Saleska, Paul Moorcroft d.r.galbraith@leeds.ac.uk

2 Background: Amazon Die-back White et al. 999, Global Environmental Change; Hybrid model forced with HadCM3 data Present-day 2 Cox et al. (2), Nature

3 The land surface is an important element of overall uncertainty in future changes in Amazonian C stocks Number of GCM/SRES combinations Frequency (Number of GCM/SRES combinations) GHG EMISSIONS CLIMATE RESPONSE (GCMs) ECOSYSTEM RESPONSE (DGVMs) (c) dctotal by DGVM ECOSYSTEM FEEDBACKS 5 4 HYLAND LPJ TRIFFID 3 24-member ensemble (2 GCMs, 4 SRES Scenarios, 3 DGVMs) 2 Key Result: DGVMs explain 2/3 of the variance in modelled changes in Amazonian C stocks dctotal (%) % ΔC total (23-2) 2 3

4 How can we reduce uncertainty in simulations of climate change impacts on Amazonian rainforests? Answer: Bringing global vegetation models closer to field data Historically, there has been little evaluation of the performance of global vegetation models in the tropics There is now a large amount of field data from the Amazon that can be used to evaluate/improve global vegetation models These data now allow model processes to be probed at a level of detail that was not possible before now

5 The Amazon-Andes Initiative Project Approach: Comprehensive evaluation and further development of the predictive capabilities of four state-ofthe-art terrestrial ecosystem models against a suite of field measurements from Amazonia collected over a range of spatial and temporal scales. 2 Phases: ) Site-level simulations, including drought experiments. 2) Regional simulations. Principal Investigator: Paul Moorcroft, Harvard Modelling Groups: Arizona (CLM-DGVM), Harvard (ED2), Oxford (JULES), Viçosa /Woods Hole (IBIS)

6 Site-level Evaluation Data for Amazonian Rainforests Flux tower data (Present-day carbon, water and energy exchange) (LBA-MIP) Forest plot inventory data (Biomass trends) Data from Intensive C-cycle sites (Individual components of the carbon cycle) Data from drought experiments (Rainforest resilience to drought)

7 Intensive C-cycle sites () Malhi et al. 29 Malhi et al. 29 GPP TOO HIGH, RESPIRATION TOO HIGH, CUE TOO LOW RIGHT NPP GPP TOO LOW, RESPIRATION TOO LOW, CUE TOO HIGH RIGHT NPP GPP RIGHT, RESPIRATION RIGHT, CUE RIGHT RIGHT NPP

8 Mean annual NPP (t C ha - yr - ) NPP for 4 lowland Amazonian sites CAX-6 K34 K67 TAM CLM ED2 IBIS JULES Bottom-Up Obs Models generally over-simulate NPP at the less productive forest sites, but many models are within the error of the observations at the more productive sites. ED2 NPP always higher than the other models. High NPP relative to obs is better than low NPP relative to obs!

9 Mean annual Carbon Use Efficiency Carbon Use Efficiency (NPP/GPP) for 4 lowland sites CAX-6 K34 K67 TAM CLM ED2 IBIS JULES Bottom-Up Obs Most models simulate CUE within the error bounds of the observations. CUE simulated by ED2 generally higher than observations (except in K67). Modelled CUE shows little variation across sites.

10 Results: NPP Components for slower-growing, less dynamic forests Fraction of Total NPP CAX-6 K CLM ED2 JULES IBIS OBS CLM ED2 JULES IBIS OBS NPPleaf NPPfroot NPPwood General tendency towards underestimation of canopy NPP and overestimation of wood NPP fraction. Woody NPP usually the largest simulated NPP component.

11 Carbon Allocation in tropical forests as simulated by 3 terrestrial ecosystem models Predicted Standing Biomass (Mg C ha - ) Woody NPP Terrestrial Ecosystem Models Fine root NPP Canopy NPP Malhi et al., Philosophical Transactions of the Royal Society (2)

12 Fraction of Total Respiration Autotrophic Respiration Components for slowgrowing, less dynamic forests CAX-6 K CLM ED2 JULES OBS CLM ED2 JULES OBS LeafResp RootResp WoodResp

13 Amazonian Drought Experiments Caxiuanã National Forest, Pará, Brazil 5 % Rainfall Exclusion (22 present) from -ha treatment plot Images from Antonio Lola da Costa

14 AGB fraction relative to start of experiment AGB fraction relative to start of experiment Drought impacts on aboveground biomass CAX TNF JULES CLM ED2 IBIS Observations Models fail to capture observed biomass dynamics at the TFE sites. 3 models (CLM, IBIS, JULES) are too insensitive, model (ED2) is over-sensitive.

15 Fractional change relative to start of experiment Fractional change relative to start of experiment Timing of process response to drought: NPP/LAI/Biomass Responses at Tapajós CLM-DGVM ED IBIS Date JULES LAI Biomass NPP Date

16 Fractional Cover LAI fraction relative to start of experiment Aboveground Biomass (kg C m -2 ) Fractional Cover LAI fraction relative to start of experiment AGB fraction relative to start of experiment LAI fraction relative to start of experiment AGB fraction relative to start of experiment AGB fraction relativ to start of experimen Mortality processes are key for capturing drought impacts on biomass CAX da Costa et al. 2; reduction of ~ 2% following years of drought TNF d d d2 d3 d d d d2 d3 d4 d5 d6 d7 d8 d9 7 6 CLM λ background, λ heat, λ light, λ NPP ED2 CLM-DGVM.5 JULES - TNF Background disturbance, wood, Broadleaf Tree Shrub 2 early successional late successional d d d2 d

17 Cluster-based Coarse Woody Biomass Residence Time (Years) Probability Woody Biomass Residence Time in Global Vegetation Models.25.2 (a) CTEM CARLUC, GTEC, IBIS (original) FBM LM3V ED SIB-CASA CASA.5 BIOME-BGC MOSES-TRIFFID ORCHIDEE. Hybrid VISIT.5 adgvm IBIS (current) JULES-TRIFFID CARAIB HRBM JSBACH CN-CLASS Hyland Cluster-based Coarse Woody Biomass Residence Time (Years) 2 (b) HRBM 5 CARAIB JULES-TRIFFID JSBACH MOSES-TRIFFID SIB-CASA ED Hybrid ORCHIDEE FBM 5 CTEM VISIT CASA CARLUC, LM3V BIOME-BGC GTEC, IBIS (current) IBIS (original) CN-CLASS, Hyland Observed Percentile Galbraith et al., in review

18 Woody Biomass Residence Time (Years) Edaphic Controls on Woody Biomass Residence Times of Tropical Forests Woody Biomass Residence Time (Years) 4 All plots 4 Lowland plots Soil Category Soil Category Galbraith et al., in review

19 Summary There is now more Amazonian field data than ever and it can be used to evaluate vegetation models Intensive C-cycle data allows comprehensive evaluation of respiration and NPP allocation that has hitherto not been possible there is considerable divergence across models in the simulation of these processes Current models struggle to reproduce results from drought experiments some processes are too sensitive while others are too insensitive Improved representation of mortality is an immediate priority for DGVM development