Parameter identifiability, constraints, and equifinality in data assimilation with ecosystem models...

download Parameter identifiability, constraints, and equifinality in data assimilation with ecosystem models Dr. Yiqi Luo Botany and microbiology department University

of 34

  • date post

    14-Dec-2015
  • Category

    Documents

  • view

    214
  • download

    1

Embed Size (px)

Transcript of Parameter identifiability, constraints, and equifinality in data assimilation with ecosystem models...

  • Slide 1

Parameter identifiability, constraints, and equifinality in data assimilation with ecosystem models Dr. Yiqi Luo Botany and microbiology department University of Oklahoma, USA Land surface models and FluxNET data Edinburgh, 4-6 June 2008 (Luo et al. Ecol Appl. In press) Slide 2 Observed Data Prior knowledgePosterior distribution Parameter identifiability Inverse model Constrained Edge-hitting Equifinality Slide 3 Wang et al. (2001) ------ a maximum of 3 or 4 parameters can be determined. Wang et al. (2001) ------ a maximum of 3 or 4 parameters can be determined. Braswell et al. (2005) ------ 13 out of 23 parameters were well-constrained. Braswell et al. (2005) ------ 13 out of 23 parameters were well-constrained. Xu et al. (2006) ------ 4 or 3 out of 7 parameters can be constrained, respectively under ambient and elevated CO 2. Xu et al. (2006) ------ 4 or 3 out of 7 parameters can be constrained, respectively under ambient and elevated CO 2. Identiable parameters Slide 4 Three methods to examine parameter identifiability 1.Search method 2.Model structure 3.Data variability Slide 5 Harvard Forest EMS-Tower Eddy flux data Slide 6 CO 2 flux CO 2 flux H 2 O flux H 2 O flux Wind speed Wind speed Temperature Temperature PAR PAR Relative humidity Relative humidity Hourly or half-hourly Eddy flux technology Slide 7 Leaf-level Photosynthesis Sub-model Canopy-level Photosynthesis Sub-model System-level C balance Sub-model Model Slide 8 Table 1 Parameters information Slide 9 Develop prior distribution Develop prior distribution Apply Metropolis-Hasting algorithm Apply Metropolis-Hasting algorithm a) generate candidate p from sample space b) input to model and calculate cost function c) select according to decision criterion d) repeat Construct posterior distribution Construct posterior distribution Bayesian inversion Slide 10 Slide 11 Slide 12 Conditional Bayesian inversion Bayesian inversion Bayesian inversion Bayesian inversion Bayesian inversion Slide 13 Slide 14 Fig. 2 Decrease of cost function with each step of conditional inversion Slide 15 Slide 16 Conclusions Conditional inversion can substantially increase the number of constrained parameters. Conditional inversion can substantially increase the number of constrained parameters. Cost function and information loss decrease with each step of conditional inversion. Cost function and information loss decrease with each step of conditional inversion. Slide 17 Measurement errors and parameter identifiability Slide 18 Leaves X1Woody X2Fine Roots X3 Metabolic Litter X4Structural Litter X5 Microbes X6 Slow SOM X7 Passive SOM X8 GPP TECO biogeochemical model Slide 19 No. of parameter 8 12 8 3 Slide 20 Exit rates Slide 21 Transfer coefficients Slide 22 Initial values Slide 23 Pool sizes without data Slide 24 Pool sizes with data and different SD Slide 25 Conclusion Magnitudes of measurement errors do not affect parameter identifiability but influence relative constraints of parameters Slide 26 Base model GPP Leaves X1Stems X2Roots X3 Metabolic L. X4Struct. L. X5 Microbes X6 Slow SOM X7 Passive SOM X8 Slide 27 Simplified models Plant C Litter C GPP CO 2 Soil C Plant C Litter C GPP CO 2 O Soil C Miner. C 3P model 4P model Slide 28 Simplified models 6P model7P model GPP Leaves X1Stems X2Roots X3 Litter X4 Slow C X5 Miner. Soil C X6 GPP Leaves X1Stems X2Roots X3 Metabolic L. X4 Struct. L. X5 Microbes X6 Soil C X7 Slide 29 3P model-parameter constraints Plant CLitter C Soil C Slide 30 4P model-parameter constraints Plant CLitter C Slow Soil C Passive Soil C Slide 31 6P model-parameter constraints Foliage Litter CSlow Soil CPassive Soil C WoodyFine roots Slide 32 7PM model-parameter constraints Foliage Metabolic L. C Structure L. CMicrobes C Woody Fine roots Soil C Slide 33 8P model-parameter constraints Slide 34 Conclusion Differences in model structure are corresponding to different sets of parameters. The number of constrained parameters varies with data availability