Faculty of Environmental Sciences, Department of Hydrosciences, Chair of Meteorology Effects of Measurement Uncertainties of Meteorological Data on Estimates of Site Water Balance Components PICO presentation Dr. U. Spank Tel.: +49(0)351-463 31341 Mail: Uwe.Spank@tu-dresden.de Vienna, 04/08/2013
Aim of this study: What is the effect of actual measurement uncertainties in meteorological variables on simulation results? investigations on site scale three sites with different land use reference data: evapotranspiration measured via eddy covariance (EC) technique two water balance models of different complexity physically based vs. statistical approach
Assessment of effects due to input uncertainties Monte-Carlo-Simulations with ensemble of artificially perturbed input data Assessment of reference uncertainty (annual evapotranspiration measured via EC-technique) deficit of energy balance closure uncertainty of net radiation (available energy AE) 0.85 ET R ET R 1.30 ET R Main Conclusions (1) Monte Carlo simulations suitable and reliable approach for assessment of typical effects due to input uncertainties in case of complex models (2) Reference uncertainty not negligible for evaluation of simulation results (3) Uncertainties in meteorological input data significant influence to simulation results typical effects: ± 30 mm to ± 45 mm (~ 5 %) for annual ET uncertainty of input data same magnitude as uncertainty of parameters uncertainty predominately caused by uncertainties of measured precipitation P but also by measured global radiation R G effects of input uncertain similar for complex and simple models Measured and simulated annual evapotranspiration ET
Faculty of Environmental Sciences, Department of Hydrosciences, Chair of Meteorology Effects of Measurement Uncertainties of Meteorological Data on Estimates of Site Water Balance Components Discussion Part Dr. U. Spank Tel.: +49(0)351-463 31341 Mail: Uwe.Spank@tu-dresden.de Vienna, 2/3/2016
5 Table of Content Contend... 5 back to this page (1) Introduction... 7 (2) Eddy Covariance Technique... 10 (3) Assessment of Reference Uncertainty... 14 (4) Test Sites... 16 (5) Water Balance Modelling... 19 (6) Analysis of Input Uncertainty... 23 (7) Results... 30 (8) Conclusions... 32 References... 34 go go go go go go go go go
6 1 Introduction How do measurement uncertainties affect hydrological simulations?
Reality Meteorological and Hydrological Observation Measurement of Physical Quantities Process Understanding and Modeling Phys. Parameter Input Data Reference Data necessary Simplifications Fit Parameter Measurement Uncertainties Uncertainties due to Regionalisation and Generalisation Calibration/ Validation Input Uncertainty Reference Uncertainty Models of Different Complexity Parameter Uncertainty Model Uncertainty Simulation Result and Uncertainty of Simulation Result
8 1 Introduction Aim of this study: What is the effect of actual measurement uncertainties in meteorological variables on simulation results? Effects of regionalisation and generalisation should be ruled out. Investigations have be on a smaller scale than catchment scale site scale Suitable reference data: lysimeter measurements seepage, evapotranspiration micrometeorological measurements (e.g., eddy covariance) evapotranspiration
9 1 Introduction What are eddy covariance (EC) measurements?
2 Eddy Covariance Technique EC technique = state of the art micrometeorological method to measure mass and energy fluxes (Burba and Anderson, 2010) Measurement principle recording of small turbulence elements = eddies high frequent measurement of vertical wind speed w and high frequent measurement of a scalar air property as temperature T or a gas concentration like water vapor atmosphere cold, dry Effective transport (=flux) wind up wind down covariance between vertical wind speed and scalar air property additional complex processing steps required flux processing surface warm, wet 10
Measurement Principle spatial measurement of mass and energy exchange over a source area = footprint [Figure: U. Eichelmann (2012)] size and position of footprint depending on: static parameters: measurement height, surface characteristic (vegetation height and roughness) meteorological conditions: wind speed, wind direction, atmospheric stability 11
12 2 Eddy Covariance Technique Limitations/ Restrictions of EC Measurements theoretical requirement: infinite homogeneous and plane surface vertical turbulent transport only measured NOT measured horizontal exchange non-turbulent transport ( advection ) systematic underestimation of actual fluxes other uncertainties spatial representativeness ( homogenous surface within the footprint) uncertainties due to non-stationary conditions ( half hourly data sets) uncertainty of estimated storage change ( tall vegetation) uncertainties of flux processing ( tilt and damping correction) device uncertainties,
13 How can we evaluate the reliability of measured evapotranspiration?
Two ways to achieve site evapotranspiration ET 14 (1) Directly measured water vapor flux F w ET EC Covariance of vertical wind speed and water vapor concentration (2) Derived from energy balance and measured sensible heat flux H ET EB AE = Rn G = H + LE ET EB AE H = L R n net radiation G soil heat flux AE available energy H sensible heat flux LE latent heat flux = L ET L latent heat of evaporation Measured fluxes (F w and H) underestimate actual fluxes! ET EC underestimation of actual ET = lower bound of plausibility ET EB overestimation of actual ET = upper bound of plausibility necessity of a central (reference) value Partionation of difference ET EB - ET EC by Bowen-Ratio ( H / LE ) reference or central value = ET R
15 3 Assessment of Reference Uncertainty Assessment of reference uncertainty on the example of annual evapotranspiration being measured with EC-technique at a spruce stand in the Tharandt Forest (Germany) (1) annual evapotranspiration derived from measured water vapour flux ET EC, energy balance ET EB as well as reference value ET R (2) additional uncertainty due to uncertainty of measured net radiation and measured soul heat flux ΔAE (3) estimated range of uncertainty (conservative approach) range between ET EB and ET EC worst case but also other uncertainties well founded assumption (conservative approach): 0.85 ET R ET R 1.30 ET R
4 Test Sites (1) Tharandter Wald (Tharandt Forest): biggest closed forest within the federal state Saxony (2) (1) Map of Germany; Wikipedia, 2013 [http://commons.wikimedia.org/wiki/ File:Deutschland_topo.jpg] (2) Tharandter Wald (Tharandt Forest); Google Maps, 2010 (3) Spruce site (4) Grassland site (5) Agricultural site 16
17 4 Test Sites Spruce site (Coniferous Forest) Agricultural site Grassland site land use different, but typical for the landscape of the lower Erzgebirge (Ore Mountains) soil and hydro-geological properties similar flat and smooth terrain lateral water movements negligible no influence due to groundwater small spatial distances, equal level of altitude comparable climatic conditions
18 Which hydrological model must be chosen to simulate the water balance on selected test sites?
19 5 Water Balance Modelling Restrictions/ Simplifications Spatial scale ca. 0.75 km² defined by the footprint of the EC-measurements Site properties sites are flat and homogenous no lateral water movements no influence due to groundwater only(!) vertical water movement 1D- Investigations Special focus: evapotranspiration ET (micro meteorological observation, EC- technique) ET is caused by root water uptake (transpiration), interception and soil evaporation effects due to geological properties are ruled out vertical watershed: below rooting zone
20 5 Water Balance Modelling Two models of different complexity (1) simple Black-Box-Model: HPTFs Hydro-Pedotransfer-Functions according to Wessolek et al. 2008 Hydrologischer Atlas von Deutschland (Hydrological Atlas of Germany) (2) complex water balance model: BROOK90 mainly driven by physically based approaches (Federer: http://home.roadrunner.com/~stfederer/brook/brook90.htm) both models require the same meteorological input: precipitation P, global radiation R G, minimum and maximum of air temperature T max / T min, air humidity RH, wind speed u BROOK90: daily values HPTFs: annual total of precipitation and total of precipitation in summer months annual total of grass-reference evapotranspiration ET 0 calculation according to Allen et al. (1998) derived from R g, T max, T min, RH and u long term totals derived from daily measurements actual input uncertainty identically for both models
21 5 Water Balance Modelling HPTFs: only two parameters Parameterisation type of land use (coniferous forest, deciduous forest, arable land, grassland) plant available water within rooting zone no groundwater influence = usable field capacity BROOK90: ~100 parameters 2 sets of parameters simple parameterisation (default parameter) effort for parameterisation identical to HPTFs advanced parameterisation/ calibration well adapted to site condition How does the input uncertainty influence the simulation performance?
22 How can we estimate the effects of input uncertainties on hydrological simulations?
6 Analysis of Input Uncertainty Fact: analytical analysis NOT possible too complex algorithms and model structures unknown functions Monte-Carlo-Simulations with artificially perturbed input data statistical analysis of spread of simulation results 23
24 6 Analysis of Input Uncertainty fully random or fully systematic effect independent from other variables and weather conditions constant for all values selective effect depend on other variables and therewith depending on weather conditions correlation between size of value and weather condition uncertainty/ measurement error often correlated with size of actual value simplification of actual error behavior to effects of offset and slope Model: Offset Model: Slope error characteristic of precipitation is more complex page 26 go Random Error Systematic Error
6 Analysis of Input Uncertainty native time series native time series systematic uncertainty native time series systematic uncertainty random uncertainty measurement quantity measurement quantity measurement quantity simplification of actual uncertainty characteristic slope term and offset term time time systematic discrepancy (slope and offset) randomly calculated for complete time series time random scattering (slope and offset) randomly and individually calculated for all data points parameterisation by usage of parallel measurements and producer information collective of 50000 individual data series water balance simulations ensemble of results statistical analysis (analysis of spread) measurement quantity time 25
26 6 Analysis of Input Uncertainty Meteorological data parameterization of slope A and offset term B for simulations of systematic and random measurement uncertainties in the Monte-Carlo-approach Precipitation data systematic underestimation of actual precipitation correction according Richter (1995) P = b ε P mes P cor = P mes + P systematic uncertainties: (i) thresholds: snow-sleet = 0 C, sleet-rain = 3.2 C uncertainty ± 1K (ii) stratiform vs. convective = winter/summer thresholds: April 1 st, October 31 st ± 30 days (iii) level of protection ± 50% of ΔP random uncertainties: offset ± 0.2 mm, slope 0.8 1.2
6 Analysis of Input Uncertainty ET 0 : affected by totality of all meteorological variables however, predominately determined by R G 70% of explained variability measurement uncertainties of R G significant influence to ET 0 however, effects partly compensated Guideline: uncertainty of measured of daily R G sum up to an uncertainty of 250 MJ (equivalent to 100 mm) on an annual scale uncertainty of annual ET 0 ~ 40 mm annual totals of global radiation R G and grass-reference evapotranspiration according Allen (1998) ET 0 at spruce site (periods April to March, 1997 2008) 27
6 Analysis of Input Uncertainty special weather conditions: 2002: extreme rainfall in August (precipitation of 264.3 mm between August 11 th and August 13 th ) 2003: extreme hot and dry summer annual precipitation Pa (April to March), precipitation in Summer month (April to September), annual grass-reverence evapotranspiration ET0 (April-March) Spruce site representative for all sites box plots: median, 5% and 95% quantile, minimum and maximum 28
29 Let s have a look to the results
7 Simulation Results Measured and simulated annual evapotranspiration ET box plots: median, 5% and 95% percentile, minimum and maximum of spread effect of input uncertainty similar for HPTFs and BROOK90 range between median and 5%/ 95% percentile ± 30 up to ± 45mm (~ 5%) uncertainty predominately caused by uncertainties of R G and P effects due to input uncertainty same magnitude as effects due to parameter uncertainties 30
31 7 Simulation Results rape winter wheat corn spring barley Measured and simulated annual seepage R box plots: median, 5% and 95% percentile, minimum and maximum of spread
8 Conclusions BROOK90 (advanced parameterisation) high reliability of simulated annual seepage R and annual evapotranspiration ET two outliers: discrepancy in 2004 at Grassland site missing information about vegetation (height and LAI) discrepancy in 2003 extreme dry and hot summer period partially unresolved: insufficient parameterisation ( changes of parameters or changes of parameter sensitivity under extreme conditions) insufficient model algorithms? very high effort for parameterisation in practical applications typically NOT feasible BROOK90 (simple parameterisation) simulation results less reliable than simulations with advanced parameterization BUT: median typically within range of tolerance (within range of reference uncertainty) input uncertainty decides if simulation passes or fails problems and large discrepancies: unusual weather conditions, e.g., droughts 2003 special agricultures, e.g., maize agricultural site (2007) general characteristics and tendencies well described but NOT usable for quantitative analysis HPTFS too simple and too generalised approaches for description of hydrological response on meteorological force spruce and agricultural site: significant overestimation of ET significant underestimation of R usable results only at grass site attention when using HPTFs ( Hydrologischer Atlas von Deutschland/ hydrological atlas of Germany) 32
33 8 Conclusions (1) Monte Carlo Simulations suitable and reliable approach for assessment of effects due to input uncertainties for complex models depending on method estimation of typical range of uncertainties, NOT for worst case approximations (2) Reference uncertainty not negligible for evaluation of simulation results uncertainty of eddy covariance measurements 0.85 ET R ET R 1.30 ET R (ET R gap between ET EB and ET EC portioned by Bowen ratio) (3) Uncertainties in meteorological input data significant influence to simulation results typical effects: ± 30 mm to ± 45 mm (~ 5 %) for annual ET uncertainty of input data same magnitude as uncertainty of parameters uncertainty predominately caused by uncertainties of measured precipitation P but also by measured global radiation R G effects of input uncertain similar for complex and simple models
34 References Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop Evaporation (guidelines for computing crop water requirements). FAO Irrigation and Drainage Paper No. 56. Burba, G.G., Anderson, D., 2010. A Brief Practical Guide to Eddy Covariance Flux Measurements: Principles and Workflow Examples for Scientific and Industrial Applications, LI-COR Inc. Federer, C.A., 2002. BROOK 90: A simulation model for evaporation, soil water, and streamflow. Available via dialog: http://home.roadrunner.com/~stfederer/brook/brook90.htm (accessed 16th Nov. 2009) Richter, D., 1995. Ergebnisse methodischer Untersuchungen zur Korrektur des systematischen Meßfehlers des Hellmann-Niederschlagsmessers, Berichte der Deutschen Wetterdienstes 194. Selbstverlag des Deutschen Wetterdienstes. Taylor, J.R, 1997. Introduction to Error Analysis: The Study of Uncertainties in Physical Measurements, Second Edition. University Science Books. Wessolek, G., Duijnisveld, W.H.M., Trinks, S., 2008. Hydro-pedotransfer functions (HPTFs) for predicting annual percolation rate on a regional scale. J. Hydrol. 356, 17-27. Spank, U., Schwärzel, K., Renner, M., Moderow, U., Bernhofer, C., 2013. Effects of Measurement Uncertainties of Meteorological Data on Estimates of Site Water Balance Components. J. Hydrol. (accepted)
35 There are trivial truths and the great truths. The opposite of a trivial truth is plainly false. The opposite of a great truth is also true. Niels Bohr