Dynamic Global Vegetation Models. Rosie Fisher Terrestrial Sciences Section, NCAR

Transcription

1 Dynamic Global Vegetation Models Rosie Fisher Terrestrial Sciences Section, NCAR

2 What is the D in DGVM? Recruitment Assimilation Growth Competition Movement of vegetation in space predicted by model Mortality & Disturbance Decomposition

3 How do ecological systems organize the diversity of plant life? Recruitment Growth & Competition Co-existence or Exclusion Mortality Text n.b This is for a light-limited system!

4 Recruitment

5 Reproductive Allocation Rauto Tissue Turnover Storage GPP NPP C Balance Growth

6 Reproductive Allocation Reproduction? Reproduction? Rauto Tissue Turnover Storage GPP NPP C Balance Growth Reproduction?

7 Recruitment: Bioclimatic Envelopes (conditions needed for establishment) The standard assumption is that all seeds are everywhere Sitch et al (The LPJ model & CLM-Dynamic Vegetation Model)

8 Recruitment: Migration in TREEMIG Lischke et al. 2009

9 Present Day Implementing migration-induced lags in vegetation establishment has large impacts on biomass of expanding ecosystems n.b. this is not standard in DGVMs 2100: no lag 2100: migration Epstein, Yu, Kaplan & Lischke 2007

10 Competition

11 Area-based Models (e.g. CLM, TRIFFID, LPJ, IBIS - models used in IPCC assessments) Cell divided into plant type tiles 1 average tree per plant type NL%% tree% BL%% tree% C3% C4% Deterministic Computationally efficient No competition for light Expansion via relative growth rates Bare%Gd% Shrub% Widely used in climate simulations

12 Climate models don t represent competition realistically (most living plant ecologists)

13 How do ecological systems organize the diversity of plant life? Recruitment Growth Competition Co-existence Exclusion Mortality Text

14 Modeling competition for light requires that we have different plant functional types in the same vertical profile

15 Individual-Based Models (e.g. SORTIE, LPJ-GUESS, SEIB, adgvm) Individual Based 3D light environment Simulates: recruitment competition disturbance Stochastic demographics Computationally intensive Ensemble approach required

16 Co-existence in LPJ vs LPJ-GUESS (Smith et al. 2001)

17 Ecosystem Demography Model (ED) Moorcroft, Hurtt and Pacala Landscape divided into successional age classes Text x x

18 Ecosystem Demography Model (ED) Moorcroft, Hurtt and Pacala Landscape divided into successional age classes Vegetation divided into height and plant type classes Text x x

19 Benefits of ED approach to competition Computationally plausible simulations of ecological dynamics Represents vertical competition for light: Representation of multiple niches & the possibility of plant co-existence Simulation of recovery from human and natural disturbance events. BUT - the extra ability to simulate ecological dynamics with a stochastic model is lost

20 Modeling Tree Mortality

21 The status quo in DGVM world Mortality algorithms are typically empirical, poorly tested, and based on proxies of plant health! CLM takes its mortality model from LPJ

22 What about more detailed forest models?

23 Can we do any better? Why makes plants die? Carbon Budget Failure? Hydraulic Failure? Phloem Transport Failure?

24 A conceptual revolution: Internalizing plant physiology models Soil Moisture Stomatal Conductance Soil Moisture Leaf Moisture Stomatal Conductance NPP Mortality NPP Carbon Store Mortality

25 Recent developments in mortality models... ED(X) Carbon Storage Dynamics CLM(ED) MuSICA TREES SperryM Xylem Water Dynamics FINNSIM Not an exhaustive list... Phloem transport & xylem repair (Water/carbon interdependency)

26 Modeling drought mortality Ensemble difference (dying - surviving) NSC Greater PLC in dying trees 0.2 Pinon piòon pine Juniper juniper NSC decline = -0.47*PLC increase , r 2 = 0.32, p< Lower NSC in dying trees Ensemble difference (dying-surviving) PLC McDowell, Fisher, Xu, Domec, Holtta, Mackay, Sperry et al. New Phytologist (2014)

27 Why bother with the extra complication? Mortality models can be parameterized with real observations (carbon pool sizes, plant hydraulic properties, LWP obs, etc.) Some features (lags, relations to other plant traits) cannot be predicted from average productivity metrics (NPP/LAI).

28 How do we model plant diversity?

29 Sitch et al Plant Diversity in DGVMs

30 Why dieback : aggregation of plant diversity? There are only ~10 kinds of plant. Dieback events occur at the physiological thresholds of single plant types. Is it realistic that, e.g. all boreal trees, have the same physiological thresholds?

31 There are not enough plant types in climate models (every living plant ecologist) Low (functional) diversity causes low resilience to change.

32 Improved resolution of plant functional types? What do we want to represent?

33 Plant Traits Functional properties of plants are called traits Models define plant properties according to a set of trait values wood density, leaf lifespan, photosynthetic capacity, root depth, allometry, reflectance, nitrogen content, etc. Representing diversity involves increased sampling of trait space. This is made easier by trade-off s between plant traits.

34 Plant variation through multi-dimensional trait space ALL THEORETICAL PLANTS better Trait 2 worse worse Trait 1 better

35 Plant variation through multi-dimensional trait space ALL THEORETICAL PLANTS better better PLANTS THAT EXIST Trait 2 Trait 2 worse worse Trait 1 better worse worse Trait 1 better

36 Plant variation through multi-dimensional trait space These plants do not exist because they are eliminated by natural selection better PLANTS THAT EXIST Trait 2 worse worse Trait 1 better

37 Plant variation through multi-dimensional trait space These plants do not exist because they are eliminated by natural selection better PLANTS THAT EXIST Trait 2 worse worse Trait 1 better These plants do not exist because they are outside physiological limitations

38 Plant variation through multi-dimensional trait space resource rich environments better Growth PLANTS THAT EXIST worse worse Survival better hazardous environments

39 An example empirical trade-off Wood density Minimum operating ѱsoil Example from Bolivia: Markesteijn, Poorter, et al. 2011

40 Our knowledge of trait space is increasing

41 Specific Leaf Area Text Leaf N content

42 How might we use all of this data?! Alternative approaches to plant trait modeling

43 How quickly do plant traits vary? Model 1: Plant traits are static, adaptation happens via change in plant types Model 2: Plant traits optimize to prevailing environmental conditions Model 3: Plant traits explicitly evolve through time

44 Trait filtering in CLM(ED) High WD Low WD

45 Wood Density Emerging from Competition Increasing wood density Increasing drought tolerance Changes in plant traits occur via changes in plant type

46 Optimality: an emergent property of evolution? All existing species are the winners of evolution Competition selects the fittest species Sub-optimal plants should be eliminated What should a fit plant do? Optimality Models are hypotheses for how competitive evolution might shape plant function Changes in traits occur via changes in the environment

47 Optimal models of plant function

48 The adgvm2 model Traits can t just optimize, they need to evolve through time Individual, population and community trait values adapt to conditions

49 Modelling plant diversity Trait filtering models allow traits to vary with changing frequency of plant types Optimal models allow traits to vary as the environment changes Evolving models allow plant traits to vary in space and time No models have a concept of phenotypic plasticity

50 Summary Vegetation Dynamics models vary according to how they aggregate plants And according to whether they include climate envelope concepts And depending on how they model recruitment and mortality Biome boundaries are actually extremely poorly understood, but without testing them we have limited confidence in future predictions.