Data-model integration: Examples from belowground ecosystem ecology Kiona Ogle University of Wyoming Departments of Botany & Statistics www.uwyo.edu/oglelab Today’s Task • What are some ecological questions to which sensor network data could be applied? • How would those data be used in models? • Overview modeling of ecological data and processes. Types of Questions • What are some ecological questions to which sensor network data could be applied? – Spatial & temporal processes • Improved ecological understanding • More accurate prediction & forecasting – Example problems • “Biogeochemical exchanges between the atmosphere & biosphere” • How do environmental perturbations affect carbon & water exchange? • Partitioning ecosystem processes & components • Linking processes & mechanisms operating at multiple temporal & spatial scales How to Address Such Questions? • Couple data and models – Sensor network data • Very rich – Real-time; large datasets; spatially extensive and/or temporally intensive • Heterogeneous – Different locations, processes, and conditions – Models & data analysis • Less appropriate: – “Classical” analyses that assume linearity and normality of data – Design-based inference about patterns • More appropriate: – Coupling of process-based models with diverse and rich datasets – Model-based inference about patterns and mechanisms Why Couple Data & Process Models? – Parameter estimation (or “model parameterization”) • Quantification of uncertainty • Improved predictions and forecasts • Decision support, management, conservation – Synthesize multiple types of data • Relate different system components to each other • Learn about important mechanisms – Hypothesis generation • Use data-informed models to generate testable hypotheses • Inform sampling and network design – Data analysis • Go beyond simple “classical” analyses • Explicit integration of multiple data types, diverse scales, and nonlinear and non-Gaussian processes How to Couple Data & Process Models? – Multiple approaches, for example: • Maximum likelihood-based models • Least squares, minimization of objective functions • Hierarchical Bayesian models – Hierarchical Bayesian approach • Recall, from Jennifer’s talk … Unknown quantities Observed data P( D , P , Process | Data) Posterior Process parameters Latent (or true) process Data parameters P( Data | Process, D ) P( Process | P ) P( D , P ) Likelihood Probabilistic process model Prior(s) Outline • The process model: – Types of ecological models – Building process models • Examples from belowground ecosystem ecology: – Motivating issues – Ex 1: Estimating components of soil organic matter decomposition – Ex 2: Deconvolution of soil respiration (i.e., CO2 efflux) – In both examples, highlight: • Data sources • Process models • Data-model integration • Implications of data-model integration for sensor network data & applications Hierarchical Bayesian Model Unknown quantities Observed data P( D , P , Process | Data) Posterior Process parameters Latent (or true) process Data parameters P( Data | Process, D ) P( Process | P ) P( D , P ) Likelihood Probabilistic process model Prior(s) Data model (likelihood) P( Data | Process, D ) : Data = Latent process + observation error Probabilistic process model P( Process | P ) : Latent process = Expected process + process error The “process model” The Process Model • Conceptual model: – Systems diagrams – Graphical models • Model formulation: – Explicit, mathematical eqn’s • Systems equations • State-space equations Inputs Unobserved quantities (parameters) Observed quantities (driving variables) Outputs “Compare” Conceptual model Mathematical model Simulation model The “process model” Analytical output Numerical/ simulation output Observed quantities (data) “Predict” Unobserved or latent quantities Types of Process Models Deterministic Stochastic Compartment models (differential or difference equn’s) Matrix models Reductionist models (include lots of details & components) Holistic models (use general principles) Static models Dynamic models Distributed models (system depends on space & time) Lumped models Linear models Nonlinear models Causal/mechanistic models Black box models Analytical models Numerical/simulation models Jorgensen (1986) Fundamentals of Ecological Modelling. 389 pp. Elsevier, Amsterdam. Upcoming Example: Soil Carbon Cycle Model Deterministic Stochastic Compartment models (differential or difference equn’s) Matrix models Reductionist models (include lots of details & components) Holistic models (use general principles) Static models Dynamic models Distributed models (system depends on space & time) Lumped models Linear models Nonlinear models Causal/mechanistic models Black box models Analytical models Numerical/simulation models Example Process Model Pools or state variables Simplified systems diagram of the soil carbon cycle in a temperate forest Source: Xu et al. (2006) Global Biogeochemical Cycles Vol. 20 GB2007. Flows of carbon Model Formulation d X( t ) A X( t ) B u( t ) dt A: matrix of flux rates or “carbon transfer coefficients” (parameters) u(t): flux of carbon into the system (e.g., photosynthetic flux) (driving variable or modeled quantity) B: vector of ‘allocation fractions’ (parameters) X: vector of state variables (unobservable latent quantities, outputs) Source: Xu et al. (2006) Global Biogeochemical Cycles Vol. 20 GB2007. Model Formulation d X( t ) A X( t ) B u( t ) dt X (t ) d X (t ) Observable (data) Expected dt A X (t ) process u (t ) B u (t ) Source: Xu et al. (2006) Global Biogeochemical Cycles Vol. 20 GB2007. How to Couple Data & Process Models? – Hierarchical Bayesian approach Unknown quantities Observed data P( D , P , Process | Data) Posterior Process parameters Latent (or true) process Data parameters P( Data | Process, D ) P( Process | P ) P( D , P ) Likelihood Probabilistic process model Prior(s) Data model (likelihood) P( Data | Process, D ) : Data = Latent process + observation error Probabilistic process model P( Process | P ) : Latent process = Expected process + process error Outline • The process model: – Types of ecological models – Building process models • Examples from belowground ecosystem ecology: – Motivating issues – Ex 1: Estimating components of soil organic matter decomposition – Ex 2: Deconvolution of soil respiration (i.e., CO2 efflux) – In both examples, highlight: • Data sources • Process models • Data-model integration • Implications of data-model integration for sensor network data & applications Ecosystem Processes Emphasis on aboveground What about belowground? Biogeochemical Cycles N H 20 N H 20 C C P H 20 Biogeochemical Cycles N H 20 Belowground system is critical N Tightly linked to aboveground system H 20 C C P H 20 Belowground “Issues” Aboveground • Lots of info • Easy to measure Belowground • Little info • Difficult to measure • Aboveground measurements (helpful but limited) Outstanding issues • • • • Partitioning above- & belowground Quantifying & partitioning belowground Implications for ecosystem function Examples: arid & semiarid systems Figure from Kieft et al. (1998) Ecology 79:671-683 Motivating Questions: Soil Carbon Cycle • • • • From where in the soil is CO2 coming from? What are the relative contributions of autotrophs vs. heterotrophs? What factors control decomposition rates & heterotrophic activity? How does pulse precipitation affect sources of respired CO2? • Implications of climate change for desert soil carbon cycling? Integrative Approach • Diverse data sources – – – – Experimental & observational Lab & field studies Multiple scales Varying “amounts” & “completeness” • Process-based models – Key mechanisms, processes, components – Balance detail & simplicity – Multiple scales & interactions • Statistical models: data-model integration – Hierarchical Bayesian framework – Mark chain Monte Carlo Examples Presented Today Deterministic Stochastic Compartment models (differential or difference equn’s) Matrix models Reductionist models (include lots of details & components) Holistic models (use general principles) Static models Dynamic models (implicit dependence on time) Distributed models (implicit dependence on space & time) Lumped models Linear models Nonlinear models Causal/mechanistic models Black box models Analytical models Numerical/simulation models Ex 1: Soil organic matter decomposition Objectives: 1. Identify soil & microbial processes affecting decomposition 2. Learn how vegetation (i.e., microsite) controls these processes Experimental Design Mesquite shrubland in southern Arizona Microsite types: 1. bare ground 2. grass 3. small mesquite 4. big mesquite Bare ground 3 cores (reps) Grass Small mesquite Big mesquite Experimental Design Add water ... ... Incubate at 25 oC Add sugar + water CO2 Measure CO2 efflux (soil respiration rate) at 24 & 48 hours CO2 8 depths (layers) CO2 Experimental Design Add water ... ... Measure: Microbial biomass Soil organic carbon Soil nitrogen Incubate at 25 oC Add sugar + water CO2 Measure CO2 efflux (soil respiration rate) at 24 & 48 hours CO2 8 depths (layers) CO2 Design & Data Overview • Full-factorial design: • Microsite • 4 levels: bare, grass, small mesq, big mesq • Stochastic data: • Soil respiration rate • N = 359 (25 missing) • Soil layer • Microbial biomass • Substrate addition type • Soil organic carbon • 8 levels: 0-2, 2-5, ..., 40-50 cm • 2 levels: water only, sugar + water • Incubation time • 2 levels: 24, 48 hrs • Soil core or rep • 3 cores per microsite • N = 18 (14 missing) • N = 89 (7 missing) Some Data Analysis Objectives microbes soil C CO2 flux Soil depth Estimate microbial respiration (decomposition) parameters (i.e., process parameters) ? biomass & activity ? data Process Model: Soil Respiration Estimate microbial respiration (decomposition) parameters (i.e., process parameters) Respiration (R) Saturating carbon (C) Low C Microbial biomass (B) Michaelis-Menton type model: Ab B Ac C Ab B Ac C Ac max 0, c 0 c 1 N R Ab microbial “base-line” metabolic rate Ac microbial carbon-use efficiency Assume Ac related to “substrate quality”: Ac max 0, c 0 c 1 N Data-Model Integration Ab B Ac C R Ab B Ac C Ac max 0, c 0 c 1 N • Full-factorial design: • Microsite • Soil layer • Substrate addition type • Incubation time • Soil core or rep • Things to consider: • Multiple data types • Nonlinear model • Missing data • Experimental design microbes Soil depth • Stochastic data: • Soil respiration rate • Microbial biomass • Soil organic carbon B ? some data C soil C ? some data R N CO2 N data fixed Data Model (Likelihood) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) 1. Let LR = log(R) 2. For microsite m, soil depth d, soil core r, substrateaddition type s, and time period t: Observed rate Mean (“truth”) (latent process) Observation precision (= 1/variance) Data Model (Likelihood) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) 1. Now, for the covariates... 2. For microsite m, soil depth d, and soil core r: C{m ,d ,r } ~ Normal C {m ,d } , C B{m ,d ,r } ~ Normal B{m ,d } , B Observed Mean (“truth”) (latent process) Observation precision (= 1/variance) 3. Note: the likelihoods are for both the observed and missing data Data Model (Likelihood) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Likelihood components LR{m ,d ,r ,s ,t } ~ Normal LR{m ,d ,r , s } , LR C{m ,d ,r } ~ Normal C {m ,d } , C B{m ,d ,r } ~ Normal B{m ,d } , B Data parameters D LR , C , B Latent processes Latent processes LR{m ,d ,r ,s } , C {m ,d } , B{m ,d } Probabilistic Process Model P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Latent processes Latent processes LR{m ,d ,r ,s } , C {m ,d } , B{m ,d } Deterministic model for soil microbes & carbon contents C {m ,d } c{m ,d } C{*m } B{m ,d } b{m ,d } B{*m } Stochastic model for latent respiration LR{m ,d ,r ,s } ~ Normal . LR{m ,d ,s } , LR Probabilistic Process Model P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Stochastic model for latent respiration LR{m ,d ,r ,s } ~ Normal . LR{m ,d ,s } , LR Specify expected process: Michaelis-Menten (process) model Ab B{m ,d } Ac{m ,d } C {m ,d } Ab B{ m ,d } Ac{ m ,d } C { m ,d } . LR{m ,d ,s } Ab B{m ,d } s water only if s sugar + water Saturating carbon (C) Respiration (R) Ac{m ,d } max 0, c 0 c 1 N{ m ,d } if Low C Microbial biomass (B) Probabilistic Process Model P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Process components LR{m ,d ,r ,s } ~ Normal .LR{m ,d ,s } , LR Ab B{m ,d } Ac{m ,d } C {m ,d } .LR{m ,d ,s } Ab B{m ,d } Ac{m ,d } C {m ,d } Ab M {m ,d } water sugar Ac{m ,d } max 0, c 0 c 1 N{m ,d } C {m ,d } c{m ,d } C{*m } B{m ,d } b{m ,d } B{*m} Process parameters P , c{m ,d } , b{m ,d } , C{*m } , B{*m } , Ab , c 0 , c 1 LR Parameter Model (Priors) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Data parameters D LR , C , B Process parameters P , c{m ,d } , b{m ,d } , C{*m } , B{*m } , Ab , c 0 , c 1 LR Conjugate, relatively non-informative priors for precision terms LR , C , B , ~ Gamma 0.01, 0.001 LR Parameter Model (Priors) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Data parameters D LR , C , B Process parameters P , c{m ,d } , b{m ,d } , C{*m } , B{*m } , Ab , c 0 , c 1 LR Non-informative Dirichlet priors for relative distributions of microbes and carbon c{m ,.} , b{m ,.} ~ Dirichlet 1,1,1,...,1 Multivariate version of the beta distribution (with all parameters set to 1: multidimensional uniform) Parameter Model (Priors) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Data parameters D LR , C , B Process parameters P , c{m ,d } , b{m ,d } , C{*m } , B{*m } , Ab , c 0 , c 1 LR Relatively non-informative (diffuse) normal priors for the rest: c 0 , c 1 , ln C{*m } , ln B{*m} , ln Ab ~ Normal 0, 0.0001 The Posterior P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) P ( Data |Process , D ) 4 8 3 2 2 LR 2 LR exp LR {m ,d ,r ,s ,t } LR{m ,d ,r ,s } 2 2 m 1 d 1 r 1 s 1 t 1 4 8 3 C 2 2 C B B exp C exp B {m ,d ,r } C {m ,d } 2 {m ,d ,r } B{m ,d } 2 2 2 m 1 d 1 r 1 4 8 3 2 2 LR LR exp LR{m ,d ,r ,s } .LR{m ,d ,s } 2 m 1 d 1 r 1 s 1 2 LR 0.99 0.001 LR e C 0.99 0.001 C e B 0.99 0.001 B e LR 0.99 0.001 LR e 0.0001 0.0001 0.0001 0.0001 2 2 exp exp c 0 0 c1 0 2 2 2 2 0.0001 0.0001 exp Ab 0 2 2 2 0.0001 2 2 0.0001 * 0.0001 0.0001 * exp B 0 exp C 0 {m} {m} 2 2 2 2 m 1 4 The Posterior P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) P ( Data |Process , D ) 4 8 3 2 2 LR 2 LR exp LR {m ,d ,r ,s ,t } LR{m ,d ,r ,s } 2 2 m 1 d 1 r 1 s 1 t 1 4 8 3 C 2 2 C B B exp C exp B {m ,d ,r } C {m ,d } 2 {m ,d ,r } B{m ,d } 2 2 2 d 1 r 1 m 1 analytical No solution for the joint posterior distribution 4 8 3 2 2 LR LR exp LR{m ,d ,r ,s } .LR{m ,d ,s } No analytical solution 2for most m 1 d 1 r 1 s 1 2 of the marginal distributions e e the e Markov chain Monte e Carlo methods, Approximate posterior: LR 0.99 0.001 LR C 0.99 0.001 C B 0.99 0.001 B LR 0.99 0.001 LR 0.0001 implemented WinBUGS 0.0001 0.0001 0.0001 2 in 2 exp exp c 0 0 c1 0 2 2 2 2 0.0001 0.0001 exp Ab 0 2 2 2 0.0001 2 2 0.0001 * 0.0001 0.0001 * exp B 0 exp C 0 {m} {m} 2 2 2 2 m 1 4 Model Implementation: WinBUGS Model Goodness-of-fit Example Results C* (total soil carbon, g C/m2) B* (microbial biomass, g dw/m2) box plot: parms.m[1,] 5.00E+3 box plot: parms.m[2,] 10.0 [1,2] [2,2] [2,3] 4.00E+3 [1,3] [1,4] 3.00E+3 [2,4] 5.0 [1,1] [2,1] 2.00E+3 1.00E+3 0.0 Bare Big Med. Grass mesq. Mesq. Bare Big Med. Grass mesq. Mesq. Example Results Bare ground Big mesquite Relative amount of microbial biomass box plot: parms.md[5,1,] box plot: parms.md[5,2,] 0.2 0.2 [5,2,1] 0.15 0.15 0.1 0.1 [5,1,1] [5,2,2] [5,1,2] [5,1,3] 0.05 [5,2,4] 0.05 [5,1,4] [5,1,7] [5,1,5] [5,2,3] [5,2,5] [5,1,8] [5,1,6] [5,2,8] [5,2,6] [5,2,7] 0.0 0.0 Surface Deep Surface Soil depth (or layer) Deep Sensitivity to Data Sources Ex 2: Deconvolution of Soil Respiration • • • • From where in the soil is CO2 coming from? What are the relative contributions of autotrophs vs. heterotrophs? What factors control decomposition rates & heterotrophic activity? How does pulse precipitation affect sources data of respired data data CO2? data data data data • Multiple data sources • lots • limited data The Field Sites San Pedro River Basin Sonoran Desert Santa Rita Experimental Range Stable Isotope Tracers CO2 Respired CO2 signature CO2 12C 13C Source isotope signatures 12C 12C Important data source: facilitates “partitioning” Data Source Examples stochastic data Pool Isotopes (δ13Ci) (roots, soil, litter; Keeling plots) Soil CO2 flux (automated chambers) Soil samples (carbon content, C:N, root mass) Literature data Soil Isotopes (δ13CTot) (automated chambers & Keeling plots) Root mass (manual chambers) (arid systems; total mass) Datasets: field/lab pubs Soil carbon (arid systems; total C) (in situ gas exchange) (root-free, carbon substrate, microbial mass, heterotrophic activity) (arid systems; total mass, carbon, microbes) Soil CO2 flux Root respiration Soil incubations Litter covariate data Soil temp & water (automated, multiple locations, many depths) Root distributions (arid systems, different functional types) Microbial mass (arid systems; total mass) Root respiration (arid systems, different functional types) Potential sensor network data Example Data Santa Rita pulse experiment San Pedro automated flux measurements 5 Respiration (mol / m2 / s) Pre-monsoon Dry Monsoon Wet Monsoon 2 Respiration (mol / m / s) 6 4 3 2 1 0 -1 0 2 6 14 Day San Pedro incubation experiment -1 Open Grass Medium Mesquite Big Mesquite 0.010 -1 -17 24 hours post-sugar addition -19 -21 -23 -25 0.008 0.006 0.004 0.002 -2 0 2 4 6 8 Day 10 12 14 16 0.010 0.008 0.006 0.004 0.002 0.000 0.000 -27 0.012 -1 -15 Respiration (mol g s ) 13 o d C of respired CO2 ( /oo) 0.012 -1 -13 Respiration (mol g s ) Santa Rita pulse experiment – d13C 0-2 2-5 5-10 0-15 5-20 0-30 0-40 0-50 1 1 2 3 4 Depth (cm) 0- Hierarchical Bayesian Model: Deconvolution Approach • • • Integrate multiple sources of information • Diverse data sources • Different temporal & spatial scales • Literature information • Lab & field studies Detailed flux models • Respiration rates by source type & soil depth • Dynamic models Mechanistic isotope mixing models • Multiple sources Data Source Examples stochastic data Pool Isotopes (δ13Ci) (roots, soil, litter; Keeling plots) Soil CO2 flux (automated chambers) Soil samples (carbon content, C:N, root mass) Soil Isotopes (δ13CTot) (automated chambers & Keeling plots) (arid systems; total mass, carbon, microbes) Root mass (manual chambers) (arid systems; total mass) P ( j | Data ) L (Data | j ) P ( j ) Microbial mass Soil carbon (arid systems; total C) (in situ gas exchange) (root-free, carbon substrate, microbial mass, heterotrophic activity) Litter Soil CO2 flux Root respiration Soil incubations Literature data covariate data Soil temp & water (automated, multiple locations, many depths) Root distributions (arid systems, different functional types) (arid systems; total mass) Root respiration (arid systems, different functional types) Bayesian Deconvolution The Hierarchical Bayesian Model P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Obs Obs Data d 13CTot (t ), RTot (t ),SWC (z , t ),T (z , t ), M i (z , t ) Some Likelihood Components Obs d 13CTot (t ) ~ No d 13CTot (t ), C2 Obs 2 RTo t (t ) ~ No RTot (t ), R Observations (data) Likelihood of data (isotopes & soil flux) Latent processes: from isotope mixing model & flux models Define process models… Functions of parameters The Deconvolution Problem Theory & Process Models d 13CTot (t ) N source i 1 B 13 d C ( z , t ) p ( z , t ) dz i i 0 Isotope mixing model (multiple sources & depths) ?? Contributions by source (i ) and depth (z )? Temporal variability? pi (z , t ) ri (z , t ) RTot (t ) Relative contributions (by source & depth) ?? Source-specific respiration? Spatial & temporal variability? RTot (t ) N source i 1 ?? B r ( z , t ) dz i 0 ri (z, t ) f i ,SWC(z, t ),T (z, t ), M i (z, t ) (Q10 Function, Energy of Activation) Total flux (at soil surface) Flux model (source- & depth- specific) From previous “incubation/decomposition” study (Ex 1) M i (z , t ) known /measured /estimated Mass profiles (substrate, microbes, roots) The Deconvolution Problem Objectives ri (z, t ) f i ,SWC(z, t ),T (z, t ), M i (z, t ) Covariate data Flux model (source- & depth- specific) What is i? (source-specific parameters) i ri (z , t ) Component fluxes ri (z , t ) RTot (t ) Total soil flux RTot (t ) pi (z , t ) Contributions How to estimate i? Bayesian Deconvolution The Parameter Model (Priors) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Example: Lloyd & Taylor (1994) model ri (z , t ) f i ,SWC (z , t ),T (z , t ), M i (z , t ) ri (z , t ) r i (z , t ) exp Eo 1 1 To T (z , t ) To Informative priors for Eo and To: Eo ~ No 308.56,2 304 To ~ No 227.13,10 308 312 316 215 220 225 230 235 240 Implementation • Markov chain Monte Carlo (MCMC) • Sample parameters (θi ) from posterior • Posteriors for: θi’s, ri(z,t)’s, pi(z,t)’s, etc. • Means, medians, uncertainty • WinBUGS Results: Dynamic Source Contributions San Pedro Site – Monsoon Season Proportional Contribution 0.025 Proportional Contribution of Respiration Sources Heterotrophs (0-5 cm) Grass Roots (5-50 cm) Mesquite Roots (5-50 cm) 0.020 0.015 0.010 0.005 VWC (v/v) 0.000 0.12 Soil Moisture 0.08 Zoom-in 0.04 o Soil T ( C) 30 Soil Temperature 25 20 15 190 200 210 220 230 240 Day of Year 250 260 270 Results: Root Respiration Responses Rain (mm) Zoom-in: July 27 – August 4 30 25 20 15 10 5 0 5.0 4.0 0.30 Mesquite (C3 shrub) 0.25 0.20 3.0 0.15 2.0 1.0 Soil water Sacaton (C4 grass) 0.05 0.0 209 Jul 27 0.10 0.00 210 211 212 213 Date 214 215 216 217 Aug 4 Soil water (v/v) Total root respiration (umol m-2 s-1) 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 5.0 0.30 Mesquite (C3 shrub) 4.0 0.20 3.0 0.15 2.0 Soil water 1.0 Depth (cm) 0.10 0.10 Sacaton (C4 grass) 0.05 0.0 0.00 209 0.00 0.25 Soil water (v/v) Total root respiration (umol m-2 s-1) Results: Contributions Vary by Depth 210 211 212 213 214 215 216 Date Relative contributions by depth 0.20 0.00 0.10 0.20 0.00 0-5 5-10 10-15 15-20 20-25 25-30 0-5 5-10 10-15 15-20 20-25 25-30 0-5 5-10 10-15 15-20 20-25 25-30 30-40 30-40 30-40 40-50 40-50 40-50 Day 210 Day 213 217 0.10 0.20 Day 216 Summary • Sources of soil CO2 efflux • Mesquite (shrub): major contributor, stable source • Sacton (grass): minor contributor, threshold response • Microbes (bare): minor contributor, coupled to pulses • Deconvolution & data-model integration • • • • Soil depth (including litter) By species or functional groups Quantify spatial & temporal variability Incorporate environmental drivers • Implications & applications • Identify mechanisms • Predictions & forward modeling Outline • The process model: – Types of ecological models – Building process models • Examples from belowground ecosystem ecology: – Motivating issues – Ex 1: Estimating components of soil organic matter decomposition – Ex 2: Deconvolution of soil respiration (i.e., CO2 efflux) – In both examples, highlight: • Data sources • Process models • Data-model integration • Implications of data-model integration for sensor network data & applications Implications for Sensor Networks – Parameter estimation (or “model parameterization”) • Process models related to “biogeochemical exchanges between the atmosphere & biosphere” • Quantification of uncertainty • Improved predictions and forecasts – Synthesize data • • • • Go beyond simple “classical” analyses Explicit integration of multiple data types & scales Relate different system components to each other Learn about important mechanisms – Hypothesis generation & sampling design • Use data-informed models to generate testable hypotheses • Inform sampling and network design – Where (spatial), when (temporal), what (components)? Questions? Photo by Travis Huxman Monsoon flood, San Pedro River Basin; Sonoran desert Results: Dynamic Source Contributions Proportional Contribution 0.025 Proportional Contribution of Respiration Sources Heterotrophs (0-5 cm) Grass Roots (5-50 cm) Mesquite Roots (5-50 cm) 0.020 0.015 0.010 0.005 VWC (v/v) 0.000 0.12 Soil Moisture 0.08 0.04 o Soil T ( C) 30 Soil Temperature 25 20 15 190 200 210 220 230 240 Day of Year 250 260 270 Example WinBUGS Output parms[1] chains 1:2 parms[1] chains 1:2 sample: 100 315.0 EO 0.4 0.3 0.2 0.1 0.0 310.0 305.0 302.5 300.0 1 2000 4000 305.0 307.5 310.0 6000 iteration parms[2] chains 1:2 parms[2] chains 1:2 sample: 100 230.0 TO 0.2 0.15 0.1 0.05 0.0 220.0 210.0 200.0 190.0 190.0 1 2000 4000 195.0 6000 iteration Posterior statistics parameter node Eo parms[1] To parms[2] .... contribution pSc[1,1,1] by pSc[1,1,2] [date, pSc[1,1,3] plot, pSc[1,2,1] source] pSc[1,2,2] pSc[1,2,3] .... mean 307.7 201.1 sd 1.205 2.335 0.00145 0.8854 0.1132 0.00158 0.9721 0.02636 0.002904 0.009694 0.009998 0.003157 0.008952 0.008649 2.5% 305.2 195.8 -0.03455 0.8433 0.06992 -0.03353 0.9297 -0.01523 median 97.5% 307.7 309.9 201.0 205.1 sample 1000 1000 6.34E-4 0.8853 0.1131 6.685E-4 0.9725 0.02622 1000 1000 1000 1000 1000 1000 0.03565 0.9288 0.1558 0.03607 1.013 0.0654 200.0 205.0 The Inverse Problem Plant water uptake Soil respiration B d Dstem (t ) d D (z ) q (z , t )dz 0 ?? B d 18Ostem (t ) d 18O (z ) q (z , t )dz Isotope mixing model d 13CTot (t ) N source i 1 d 13Ci B pi (z , t )dz 0 ?? 0 q (z , t ) U (z , t ) U Tot (t ) Fractional contributions pi (z , t ) B U Tot (t ) U (z , t )dz Total flux RTot (t ) 0 U (z , t ) RA(z ) ln a RA(z ) (z , t ) k (z , t ) root (t ) kroot (t ) RA(z ) Ga(1, 1 ) (1 ) Ga(2 , 2 ) Flux model N source i 1 ri (z , t ) M i (z , t ) RTot (t ) B r ( z , t ) M ( z , t ) dz i i 0 ri (z, t ) f i ,SWC(z, t ),T (z, t ) (Q10 Function, Energy of Activation) M i (z , t ) known /measured ? Substrate or root profiles The Inverse Problem d CTot (t ) 13 N source i 1 B 13 d Ci (z , t ) pi (z , t )dz 0 Isotope mixing model (multiple sources & depths) ?? Contributions by source (i ) and depth (z )? Temporal variability? pi (z , t ) ri (z , t ) M i (z , t ) RTot (t ) ?? Relative contributions (by source & depth) Source-specific respiration? Spatial & temporal variability? RTot (t ) N source i 1 B r ( z , t ) M ( z , t ) dz i i 0 ri (z, t ) f i ,SWC(z, t ),T (z, t ) (Q10 Function, Energy of Activation) M i (z , t ) known /measured ? Total flux (at soil surface) Flux model (source- & depth- specific) Mass profiles (substrate, microbes, roots) The Deconvolution Problem Data-Model Integration ri (z, t ) f i ,SWC(z, t ),T (z, t ), M i (z, t ) Covariate data Flux model (source- & depth- specific) What is i? (source-specific parameters) ri (z , t ) RTot (t ) Total soil flux RTot (t ) pi (z , t ) Contributions Obs d 13CTot (t ) ~ No d 13CTot (t ), C2 Obs 2 RTo t (t ) ~ No RTot (t ), R From isotope mixing model & flux models Likelihood of data (isotopes & soil flux) Depend on i Data Source Examples stochastic data Pool Isotopes (δ13Ci) (roots, soil, litter; Keeling plots) Soil CO2 flux (automated chambers) Soil samples (carbon content, C:N, root mass) Soil Isotopes (δ13CTot) (automated chambers & Keeling plots) (arid systems; total mass, carbon, microbes) Root mass (manual chambers) (arid systems; total mass) P ( j | Data ) P (Data | j ) P ( j ) Microbial mass Soil carbon (arid systems; total C) (in situ gas exchange) (root-free, carbon substrate, microbial mass, heterotrophic activity) Litter Soil CO2 flux Root respiration Soil incubations Literature data covariate data Soil temp & water (automated, multiple locations, many depths) Root distributions (arid systems, different functional types) (arid systems; total mass) Root respiration (arid systems, different functional types) The Deconvolution Problem Plant water uptake Soil respiration B d Dstem (t ) d D (z ) q (z , t )dz 0 ?? B d 18Ostem (t ) d 18O (z ) q (z , t )dz Isotope mixing model d 13CTot (t ) N source i 1 d 13Ci B pi (z , t )dz 0 ?? 0 q (z , t ) U (z , t ) U Tot (t ) Fractional contributions pi (z , t ) B U Tot (t ) U (z , t )dz Total flux RTot (t ) 0 U (z , t ) RA(z ) ln a RA(z ) (z , t ) k (z , t ) root (t ) kroot (t ) RA(z ) Ga(1, 1 ) (1 ) Ga(2 , 2 ) Flux model N source i 1 ri (z , t ) M i (z , t ) RTot (t ) B r ( z , t ) M ( z , t ) dz i i 0 ri (z, t ) f i ,SWC(z, t ),T (z, t ) (Q10 Function, Energy of Activation) M i (z , t ) known /measured ? Substrate or root profiles The Deconvolution Problem Plant water uptake Soil respiration What are ω, 1, 1, 2, 2? What is i? RA(z ) Ga(1, 1 ) (1 ) Ga( 2 , 2 ) ri (z , t ) f i ,SWC (z , t ),T (z , t ) U (z , t ) q (z , t ) RTot (t ) pi (z , t ) Obs d Dstem (t ) ~ No d Dstem (t ), D2 Obs d 13CTot (t ) ~ No d 13CTot (t ), C2 Obs 18 2 d 18Oste m (t ) ~ No d Ostem (t ), O Obs 2 RTo t (t ) ~ No RTot (t ), R Likelihood of data From isotope mixing model & flux model Types of data provides by sensor networks • high-frequency tunable diode laser (TDL) measurement of the stable isotope • eddy covariance for measuring concentrations and fluxes of gases (e.g., water vapor and CO2) • soil environmental data: temperature, water content, water potential, etc. • micro-met data: air temp, RH, vpd, light, wind speed, etc. • plant ecophys/ecosystem data: sapflux, ET, albedo & reflectance Key components ID TABLENM 229 TREE 407 TREE 408 TREE 1059 TREE 6768 TREE 7019 TREE 7111 TREE 8105 TREE 8539 TREE 12808 TREE 12810 TREE 13315 TREE 19399 TREE 19445 TREE 22050 TREE 22053 TREE 22060 TREE 22137 TREE 23519 TREE 26415 TREE 26783 TREE 28623 TREE 29299 TREE 29320 TREE 30129 TREE 30139 TREE 32119 TREE 34017 TREE 34329 TREE 34716 TREE 35041 TREE 36375 TREE 36410 TREE 36411 TREE "PLT_CN" PHYSCLCD "STATECD" "CYCLE""SUBCYCLE""UNITCD""COUNTYCD" "PLOT" "SUBP" "TREE" "CONDID" 23854928010661.00 21 51 3 2 1 1 13 2 1 1 23958646010661.00 21 51 3 4 1 1 19 1 5 1 23958646010661.00 21 51 3 4 1 1 19 1 6 1 23958861010661.00 21 51 3 4 1 1 54 2 6 2 23997965010661.00 22 51 3 5 2 7 9 2 4 1 23906005010661.00 22 51 3 3 2 7 15 2 9 1 23857072010661.00 22 51 3 2 2 7 16 2 7 1 23805545010661.00 22 51 3 1 2 7 39 2 1 1 23906968010661.00 22 51 3 3 3 9 9 1 3 1 23807296010661.00 23 51 3 1 4 15 29 3 8 1 23807296010661.00 23 51 3 1 4 15 29 3 10 1 23858135010661.00 22 51 3 2 4 15 54 1 1 1 23809332010661.00 22 51 3 1 2 19 23 3 4 1 23909859010661.00 22 51 3 3 2 19 26 1 8 1 23861227010661.00 23 51 3 2 5 21 8 2 5 1 23861227010661.00 23 51 3 2 5 21 8 2 2 1 23861227010661.00 23 51 3 2 5 21 8 1 4 1 23910676010661.00 23 51 3 3 5 21 12 2 6 1 23910590010661.00 23 51 3 3 5 21 39 2 5 1 23911957010661.00 22 51 3 3 1 25 1 2 8 2 23863007010661.00 21 51 3 2 1 25 7 2 3 1 23862849010661.00 22 51 3 2 1 25 41 2 17 1 24002221010661.00 24 51 3 5 1 25 54 3 6 1 24002221010661.00 24 51 3 5 1 25 54 4 14 1 23862787010661.00 22 51 3 2 1 25 69 3 6 1 23862787010661.00 22 51 3 2 1 25 69 3 11 1 23913201010661.00 23 51 3 3 5 27 42 3 2 1 23813210010661.00 22 51 3 1 2 29 23 4 9 1 23913514010661.00 22 51 3 3 2 29 29 1 4 1 23914198010661.00 22 51 3 3 2 29 35 1 20 1 24003030010661.00 22 51 3 5 2 29 41 3 2 1 23813999010661.00 22 51 3 1 2 29 68 2 5 1 23813999010661.00 22 51 3 1 2 29 68 3 7 1 23813999010661.00 22 51 3 1 2 29 68 3 8 1 Data 6 6 a c dP A Ri Si E i i dt i 1 i 1 ci d i dSi ai E d i Si dt ci d i Process models P ( | X ) P ( X | ) P( ) P ( X | ) P( )d P( | X ) State VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA VA Statistical tools data-model integration The Process Model • Conceptual models: – Systems diagrams – Graphical models Observations of real system Conceptual model • Model formulation: – Explicit, mathematical eqn’s • Systems equations • State-space equations Mathematical model Analytical output “Compare” Observational data Simulation model Numerical/ simulation output Examples Presented Today Deterministic Stochastic Compartment models (differential or difference equn’s) Matrix models Reductionist models (include lots of details & components) Holistic models (use general principles) Static models Dynamic models (implicit dependence on time) Distributed models (implicit dependence on space & time) Lumped models Linear models Nonlinear models Causal/mechanistic models Black box models Analytical models Numerical/simulation models Jorgensen (1986) Fundamentals of Ecological Modelling. 389 pp. Elsevier, Amsterdam. Data Model (Likelihood) P( D , P , Process | Data) P( Data | Process, D ) P( Process | P ) P( D , P ) Likelihood components LR{m ,d ,r ,s ,t } ~ Normal LR{m ,d ,r , s } , LR C{m ,d ,r } ~ Normal C {m ,d } , C B{m ,d ,r } ~ Normal B{m ,d } , B Assuming conditional independence, likelihood of all data is: 4 8 3 2 2 LR 2 P ( Data | Process , D ) exp LR LR{m ,d ,r ,s ,t } LR{m ,d ,r ,s } 2 m 1 d 1 r 1 s 1 t 1 2 4 8 3 C 2 2 C B B exp C{m ,d ,r } C {m ,d } exp B{m ,d ,r } B{ m ,d } 2 2 m 1 d 1 r 1 2 2