Presentation on theme: "EO data assimilation in land process models"— Presentation transcript:
1 EO data assimilation in land process models Mat Disney and Shaun QueganNo one trusts a model except the man who wrote it; everyone trusts an observation except the man who made it (Harlow Shapley)
2 Concept for Global Carbon Data Assimilation System NB carbon and water are inextricably linked, so this is a more generalised vegetation – soil – water- atmosphere schemeLand surface/Dynamic Global Vegetation Models are central to the entire TCO concept which envisages that such models are driven by gridded data inputs either directly or through landscape syntheses, with a two-way interaction with climate/atmosphereCiais et al IGOS-P Integrated Global Carbon Observing Strategy
3 Terrestrial Component Which model(s) should go here?Zoom into the diagram to give an indication of the place of DGVMs in land component of TCO with reference to input gridded data and in situ observations.Critically dependent on gridded data productsbut also on landscape synthesis from consolidated in situ observations+ Water components:SWEsoil moisture
4 Land process models Land models need to deal with transfers of - energy- matter- momentumbetween the land surface and the atmosphere.Three classes of land (coupled carbon-water) models:Models driven by radiation (light use efficiency models)Dynamic Vegetation Models: climate drivenSimple box modelsSome models emphasise hydrology (not discussed here)
5 Light Use Efficiency models IncomingPARCO2LUEAbsorptionfAPARPhoto-synthesisRespirationGPPNPPEfficiency coefficient: LUEThe LUE may depend on biome, soil moisture, temperature, nutrients, age,Modeled or measured by satellitesMeasured by satellites
6 Notes on LUE modelsModels built by ecologists tend to focus on leaves as the functional element (e.g. Leaf Area Index).Models built by remote sensors tend to focus on radiation.LUE models are driven by EO data, rather than geared to assimilating data.
7 Properties of DVMsDVMs originally designed to examine long-term trends under climate change so…Data-independent, except for varying climate data and static soil texture dataComprehensive description of biophysicsAll processes internalised, parameterisedComplex, non-linear, non-differentiable, (discontinuities, thresholds)Expensive to run
8 The Structure of a Dynamic Vegetation Model ParametersClimateSnSn+1DVMSoil textureProcessesTesting
9 How EO data can affect DVM calculations ClimateSoilsSnSn+1ProcessesObservableLand coverForest agePhenologySnow waterBurnt areaTesting:RadiancefAPARPossible feedbackfAPARParameters
10 Calibration– boreal budburst Offline setting of global parameters can be thought of as a form of DA, but is better described as model calibration.In the following e.g, we use new EO observations that are unaffected by snow-melt to parameterise the spring warming boreal phenology model.
11 The SDGVM budburst algorithm min(0, T – T0) > Threshold, budburst occurs.The sum is the red area. Optimise over the 2 parameters, Threshold and T0 (minimum effective temperature).WhenT0Start of budburst
12 The Date of budburst derived from minimum NDWI (VGT sensor, 2000) N The Date of budburst derived from minimum NDWI (VGT sensor, 2000) N. Delbart, CESBIODay of year
13 Testing SDGVM with EO data SDGVM can predict satellite ‘observations’ since it contains a canopy model and the concept of radiation interception
14 Model “skill”1999SDGVM fAPARAVHRR NDVISkillBadGood
15 Are derived parameters the problem? Is the problem the SDGVM or the derived parameter from the EO signal?The next slide shows the fAPAR derived from Seawifs (JRC) and from MODIS for a site in the UK. The large bias between the two is a general feature of these two datasets.
17 Assimilating products AssumptionsObservationsData Assimilation Scheme(KF, EnKF, 4DVAR, etc)AssumptionsObservationsMODELAssumptionsFor example: soil moisture from SMOS, surface temperature, LAI from MODIS
18 Low-level vs derived products similar products give substantially different values;assumptions used to derive products usually inconsistent with biospheric models;Product uncertainties are poorly knownCan we use low-level products (Reflectance? BOA radiance? TOA radiance?)
19 Assimilating reflectance Data Assimilation Scheme(KF, EnKF, 4DVAR, etc)ObservationsObservationsObservation OperatorAssumptionse.g. reflectance, backscatter, etc…MODELAssumptionsAssumptions in the observation operator are made to be consistent with those in the model
20 Observation operators This approach needs observation operators: translate ecosystem model state vector into observable properties e.g.reflectance data assimilated into DALEC;predicting radar coherence in ERS Tandem data from the SPA model;relating snowpack properties to SSM/I radiometer data;recognising burnt area and severity of burn.
21 Which is the right model? Complex DVM-type models never designed for DASo, pursuing another approach with a simplified box model designed from the start for DADALEC
22 The Structure of a Data Assimilation Model (DALEC) EO data (e.g. LAI, VI, reflectance)Observation modelEnsemble Kalman FilterBlue lines indicate integration of EO data with DALECPptRaRhETCfWS1ClWe have now created a new version of DALEC, adding a simple model of local hydrology, so that both carbon and water dynamics, and their coupling, are simulated. Following the DALEC philosophy, the model is kept simple, with precipitation inputs to the surface soil layer, drainage down through n soil layers with discharge from the lowest, and root water abstraction from a specified number of layers. Daily evapotranspiration is determined from an emulator created from the detailed SPA model. The emulator is driven by external factors such as air temperature, daily radiation etc but also by....[next slide]Blue parts of the diagram indicate the route via which EO data is integrated with the model. The blue circle enclosing the carbon stocks is intended to show that, in principal, the entire state vector may be used to drive the observation model. In practice only the foliar carbon is used currently.GPPCrWS2CwCsQWSnStocks and fluxes of carbon (left) and water (right)
24 Canopy foliage results No assimilationAssimilating MODIS(bands 1 and 2)
25 Canopy foliage results Assimilating MODISexc. snowAssimilating MODISinc. snowQuaife, Williams, Disney et al. RSE in press
26 EO land cover and carbon All EO land cover the same?DGVMs use land cover indirectlyHow do we translate land cover classes to PFTs?Quaife, Quegan, Disney et al., submitted
27 EO land cover and carbon Quaife, Quegan, Disney et al., submitted
28 How do we find best model-data framework? Use ‘God’ models to test assumptions of simpler modelsDVMs + DALEC-type modelsModel-data fusion inter-comparison e.g. REFLEX: Regional Flux Estimation ExperimentCompare strengths/weaknesses of various model-data fusion techniquesQuantify errors/biases introduced when extrapolating fluxes in both space and time using a model constrained by model-data fusion methods.
29 Key issues for DA in land models 1 Simple enough for effective DA but complex enough to capture biophysicsSuitable interface with observation operatorspreferably differentiable
30 Key issues for DA in land models 2 DataSame meaning of observed parameters as used in modelsProper characterisation of uncertainty i.e. PDFsUse OOs to make best use of all available data e.g. optical, LiDAR, RADAR, thermal ….We are still searching for the best model-data structure.
31 Key issues for DA in land models 3 DA through observation operators not only answer, for various practical reasons.Also pursue general concepts of how EO data can reduce the uncertainty in land modelsCalibration, testing etc.
36 LessonThe DVM as currently formulated only supports a simple observation operator. This allows meaningful estimates of time series of observables; absolute values of the observables are of dubious value.These time series permit the model to be interrogated with satellite data, and model failures to be identified.
37 Detecting incorrect land cover Crop class incorrectly setCrop class correctly set0.00.9Pearson’s product momentTemporal correlation
38 Forward operators may prove a powerful tool in land cover mapping LessonForward operators may prove a powerful tool in land cover mapping
39 Impact on Carbon Calculations 1 day advance: NPP increases by 10.1 gCm-2yr-115 days advance: 38% bias in annual NPPCalibrated model is unbiased,unlike methods based on NDVIObservationscalibrateCarbon CalculationDynamic VegetationModelPhenology modelPicard et al.,GCB, 2005
40 Comparison Model-EO: RMSE Model needs to be region specific,here include chilling requirement ?
42 A Dynamic Vegetation Model (SDGVM) ATMOSPHERICCO2BIOPHYSICSSoilPhotosynthesisGPPFireGROWTHMortalityNPPThinningLitterNBPBiomassDisturbanceLEACHED
43 Assimilating reflectance ObservationsData Assimilation Scheme(KF, EnKF, 4DVAR, etc)The real worldMODELAssumptionsBut how do we use a non-linear observation operator?
44 Comparing model and measured fAPAR August 99May 99SeawifsSDGVM
45 Model and predicted fAPAR Average overthe whole ofEurope for 1999and 2000Note: if SDGVM were driven by the Seawifs values, most model forests would die
46 ExperimentsState and parameter estimation. DE1 and EV1 sites, 3 years driving data, all available obsAs 1. but using synthetic data (DE2 and EV2)Within site forecasting. Another year of driving data for DE1 and EV1, but no observationsAs 3. but using synthetic data (DE2 and EV2)Between site extrapolation. DE3 and EV3 sites, 4 years driving data, MODIS LAI only
47 Integrated flux predictions (gC.m-2)Assimilated dataTotalStandardDeviationNEPAssimilation exc. snow373.0151.3Assimilation inc. snow404.8129.6Williams et al. (2005)406.027.8GPP2620.396.82525.642.72170.318.1
48 REFLEX data sets “Paired” sites to test extrapolation/estimation Brasschaat (DE2) and Vielsalm (EV2) (MF)Hainich (DE3) and Hesse (DE1) (DBF)Loobos (EV1) and Tharandt (EV3) (ENF)Meteorological drivers, fluxes, MODIS LAI and stocksAttempting to estimate “uncertainty” in fluxes and MODIS LAI