Docstoc

Welcome to the Gap Filling Compa

Document Sample
Welcome to the Gap Filling Compa Powered By Docstoc
					          Welcome
                to the
          QuickTime™ and a

Gap Filling Comparison
      TIFF (LZW) decompressor
   are needed to see this picture.

      Workshop
   September 18-20, 2006

            Antje Moffat
                                     1
           QuickT ime™ and a
      T IFF (LZW ) decom pressor
   are needed to see this picture.


                                     Welcome      T IFF (L
                                               are n
                                                       QuickT im
                                                     eed
                                                                e™ and a
                                                          ZW ) decom  presso
                                                        ed to see this p
                                                                             r
                                                                        icture .




• 12 of the 14 members from the gap filling
  comparison

• Over 30 participants - from all over Europe
  (Germany, Italy, Netherlands, Switzerland),
  US, Canada, Russia, and Australia



                                                                                   2
        QuickT im




                  Goals of the Workshop
                 e™ and a
   T IFF (LZW ) decom pressor
are needed to see this picture.
                                             T IFF (L
                                          are n
                                                  QuickT im
                                                eed
                                                           e™ and a
                                                     ZW ) decom  presso
                                                   ed to see this p
                                                                        r
                                                                   icture .




• Review of Gap Filling Techniques
• Completion of the Gap Filling Comparison
         – Discussion of the Results
         – Review of Paper
         – Evaluation of the Techniques
• Work Sessions and Plenary Debates to
  Exchange our Experiences and Expertise
• Generate Ideas for Further Improvement of
  the Gap Filling
• New Insights into the Eddy Flux Data


                                                                              3
        QuickT ime™ and a
   T IFF (LZW ) decom pressor
are needed to see this picture.


                                     Outline           T IFF (L
                                                    are n
                                                            QuickT im
                                                          eed
                                                                     e™ and a
                                                               ZW ) decom  presso
                                                             ed to see this p
                                                                                  r
                                                                             icture .




                                  Workshop Program


                                     GFC Analysis

                          Performance of Techniques

                                  Error on Annual Sum




                                                                                        4
Program: This Morning (Monday)
         QuickT ime™ and a
    T IFF (LZW ) decom pressor
 are needed to see this picture.
                                                        T IFF (L
                                                     are n
                                                             QuickT im
                                                           eed
                                                                      e™ and a
                                                                ZW ) decom  presso
                                                              ed to see this p
                                                                                   r
                                                                              icture .




9:00 Registration of the Participants
If you need any kind of help, please contact Ulli or Silvana!

9:30 Setting the Stage of the Workshop
• Antje Moffat: “Welcome”
• Martin Heimann: “Biogeochemistry Research at the MPI in Jena”
• Dario Papale: “The CarboeuropeIP Ecosystem Component Database: data
  processing and availability”
• Markus Reichstein: "Gap filling: Why and how?”

11:00 COFFEE BREAK (Foyer)

11:30 Review of Gap Filling Techniques: SPM and ANNs
• Vanessa Stauch: “Semi-parametric models”
• Dario Papale: “Gap filling of eddy fluxes with artificial neural networks”
• Rob Braswell: "Gap filling by iterative regression using a regularized neural
  network”

12:30 LUNCH (Campus Cafeteria)                                                           5
            QuickT ime™ and a
       T IFF (LZW ) decom pressor


                                  This Afternoon
    are needed to see this picture.
                                                         T IFF (L
                                                      are n
                                                              QuickT im
                                                            eed
                                                                       e™ and a
                                                                 ZW ) decom  presso
                                                               ed to see this p
                                                                                    r
                                                                               icture .




14:00 Review of Gap Filling Techniques (cont.): NLRs and UKF
• Ankur Desai: “Towards a robust, generalizable non-linear regression gap filling
  algorithm”
• Asko Noormets: “NLR_AM - AQRTa-Model” (10 min recording)
• Andrew Richardson: “Maximum likelihood non-linear regression model”
• David Hollinger: “Data assimilation for eddy flux filling: The unscented Kalman
  filter”

15:30 COFFEE BREAK

16:00 Review of Gap Filling Techniques (cont.): Models and comparison
• Zisheng Xing: “A gap-filling model for tower-based NEP measurements”
• Jens Kattge: “Model parameter inversion against Eddy Covariance Data using
  a Monte Carlo Technique”
• Bart Kruijt: “Comparing gap filling using neural networks and the CarboEurope
  tool, for Fluxes and Meteo data”

19:30 Dinner Suggestion: Restaurant Papiermuehle (Please sign up!)

                                                                                          6
              QuickT ime™ and a
         T IFF (LZW ) decom pressor

                                   Tuesday Morning
      are needed to see this picture.
                                                                   T IFF (L
                                                                are n
                                                                        QuickT im
                                                                      eed
                                                                                 e™ and a
                                                                           ZW ) decom  presso
                                                                         ed to see this p
                                                                                              r
                                                                                         icture .




9:00 Eddy and Component Flux Measurements
• Corinna Rebmann: “Eddy covariance measurements and their shortcomings for
   the determination of NEE”
• David Hollinger: “Uncertainty in eddy flux data and its relevance to gap filling”
• Eva van Gorsel: “Nocturnal Carbon Efflux: Can eddy covariance and chamber
   measurements be reconciled?”
• Pasi Kolari: “Gapfilling submodel selection based on measured component
   fluxes”

10:30 COFFEE BREAK

11:00 Gap Filling of Grassland and Agricultural Sites
• Christof Ammann: “Gap-Filling of CO2 Fluxes of Frequently Cut Grassland”
• Mauro Colavincenzo: “A gap filling methodology used at a agricultural site in
   Southern Italy”
• Irene Lehner: “Carbon balance of a maize canopy: comparison of different gap
   filling strategies”

12:00 LUNCH
                                        END of OPEN SESSIONS!                                       7
          QuickT ime™ and a
     T IFF (LZW ) decom

                            Tuesday Afternoon
                        pressor
  are needed to see this picture.
                                                         T IFF (L
                                                      are n
                                                              QuickT im
                                                            eed
                                                                       e™ and a
                                                                 ZW ) decom  presso
                                                               ed to see this p
                                                                                    r
                                                                               icture .




13:00-15:00 Parallel Work Sessions Part 1
• Group 1: “Analysis of the partitioned GPP/ER comparison results” (Ankur
  Desai)
• Group 2: “Gap filling of meteorological data and water and energy fluxes”
  (Dario Papale)

15:00 COFFEE BREAK

15:30-17:30 Parallel Work Sessions Part 2
• Group 3: "Gap filling of sites with non-steady time series, e.g. cut grassland,
  cropland" (Christof Ammann)
• Group 4: "Using gap-filling techniques for estimating random errors in eddy
  covariance data" (Andrew Richardson)

Please sign up for the work sessions!

19:30 WORKSHOP DINNER (Restaurant: Weinbauernhaus “Im Sack”)
                                                                                          8
              QuickT ime™ and a
         T IFF (LZW ) decom pressor
      are needed to see this picture.



                                        Wednesday             T IFF (L
                                                           are n
                                                                   QuickT im
                                                                 eed
                                                                            e™ and a
                                                                      ZW ) decom  presso
                                                                    ed to see this p
                                                                                         r
                                                                                    icture .




Roundtable on the Gap Filling Comparison

8:30 Review of Gap Filling Comparison Paper, Antje Moffat
  - Interpretation of the comparison results
  - Derivation of key findings
  - Evaluation of techniques

12:30 LUNCH
13:30 Minutes from the four Work Sessions
14:30 COFFEE BREAK

15:00 Plenary Debates
  - Site dependency of gap filling technique performances
  - Filling of long gaps using previous years
  - Conception of a public domain code library with filling routines
  - Extended gap filling comparison for urban and crop
  - Workshop resume and outlook

17:00 End of Workshop                                                                          9
                Handling of Presentations
           QuickT ime™ and a
      T IFF (LZW ) decom pressor
   are needed to see this picture.
                                                     T IFF (L
                                                  are n
                                                          QuickT im
                                                        eed
                                                                   e™ and a
                                                             ZW ) decom  presso
                                                           ed to see this p
                                                                                r
                                                                           icture .




• 20 min presentations: 15 min talk plus 5 min
  discussion
• Please transfer your presentation onto
  common laptop during coffee or lunch break
  (Ulli or Silvana)
• Publicized on GFC webpage after workshop



                                     Questions?
                                                                                      10
        QuickT ime™ and a                 QuickT ime™ and a
   T IFF (LZW ) decom pressor        T IFF (LZW ) decom  presso r
are needed to see this picture.   are n eeded to see this picture .




           Gap Filling Comparison
                   Analysis



                                                                      11
               QuickT ime™ and a
          T IFF (LZW ) decom pressor
       are needed to see this picture.

                                         Basic Principle T IFF (L
                                                      are n
                                                              QuickT im
                                                            eed
                                                                       e™ and a
                                                                 ZW ) decom  presso
                                                               ed to see this p
                                                                                    r
                                                                               icture .




   Keyfile: Artificial Gap Flags


   Golden File - fragmented:


   Superimposition


   Comparison of Observed NEP and Predicted NEP:


                                                                                          12
            QuickT ime™ and a
       T IFF (LZW ) decom pressor



                                 Statistical Metrics
    are needed to see this picture.
                                                                              T IFF (L
                                                                           are n
                                                                                   QuickT im
                                                                                 eed
                                                                                            e™ and a
                                                                                      ZW ) decom  presso
                                                                                    ed to see this p
                                                                                                         r
                                                                                                    icture .




• Bias Error                           BE 
                                               1
                                               N
                                                    p   i    oi 


                                                                                                  •p - predicted NEP
• Root Mean Square Error
                                                                                                •o - observed NEP



                                      RMSE 
                                                   1
                                                   N
                                                        p   i    oi 
                                                                       2




• Correlation Coefficient
         
                   ( pi  p )(oi  o)
                  
                                       2
              2
             R 
                  ( pi  p ) 2 (oi  o) 2
                                                                                                               13
            QuickT ime™ and a
       T IFF (LZW ) decom pressor
    are needed to see this picture.


                                      Analysis         T IFF (L
                                                    are n
                                                            QuickT im
                                                          eed
                                                                     e™ and a
                                                               ZW ) decom  presso
                                                             ed to see this p
                                                                                  r
                                                                             icture .




                       Predicted versus Observed
• Half-hourly basis

• Daily sum basis for full day artificial
  gaps                      •Daytime/Nighttime data


                                                     DSumd  xd            
                                                                             x         i

                                                                                Nd

                                                 •Weighted ALL data

                                                                       xd * hhd xn * hhn
                                                    DSumALL                     
                                                                            48       48
                                                                                     14
                  Challenge of the Analysis
             QuickT ime™ and a
        T IFF (LZW ) decom pressor
     are needed to see this picture.
                                                    T IFF (L
                                                 are n
                                                         QuickT im
                                                       eed
                                                                  e™ and a
                                                            ZW ) decom  presso
                                                          ed to see this p
                                                                               r
                                                                          icture .




    5 artificial gap length scenarios (single hh - 12 days)
*   10 permutations
*   3 subsets: day, night, all
*   12 golden sites
*   19 submissions
*   15 statistical metrics: RMS, R2, Bias, Daily Sum,
    normalized, benchmarked, …

           513,000 comparison results!
                                       (see selection on posters in foyer)
                                                                                     15
        QuickT ime™ and a                 QuickT ime™ and a
   T IFF (LZW ) decom pressor        T IFF (LZW ) decom  presso r
are needed to see this picture.   are n eeded to see this picture .




                                                                      16
        QuickT ime™ and a                 QuickT ime™ and a
   T IFF (LZW ) decom pressor        T IFF (LZW ) decom  presso r
are needed to see this picture.   are n eeded to see this picture .




                    Some Comparison
                       Findings




                                                                      17
        QuickT ime™ and a
   T IFF (LZW ) decom pressor
are needed to see this picture.
                                            RMSE and R2:                  T IFF (L
                                                                       are n
                                                                               QuickT im
                                                                             eed
                                                                                        e™ and a
                                                                                  ZW ) decom  presso
                                                                                ed to see this p
                                                                                                     r
                                                                                                icture .




                                           Half-hourly Basis
Correlation Coefficient R2


                                                              Daytime




                                   Nighttime




                                 Root Mean Square Error (gCm-2)

                             Performance of gap filling techniques from bottom
                             • MIM, MDV, UKF_LM, NLR & Others
                             • 3 ANNs leading                                                              18
                                              QuickT ime™ and a
                                         T IFF (LZW ) decom pressor
                                                                    RMSE and R2:
                                      are needed to see this picture.
                                                                                                    T IFF (L
                                                                                                 are n
                                                                                                         QuickT im
                                                                                                       eed
                                                                                                                  e™ and a
                                                                                                            ZW ) decom  presso
                                                                                                          ed to see this p
                                                                                                                               r
                                                                                                                          icture .




                                                               Half-hourly & Daily Sums
                                                                                                        Daytime
Correlation Coefficient R2




                                                                    Daytime
                                                                              Nighttime



                                  Nighttime


                                      Root Mean Square Error (gCm-2)              Root Mean Square Error (gCm-2)

                                          Half-hourly basis                                  Daily sum basis
                             Daytime                                          Daytime
                             •   R2: 0.6 - 0.8                                •   R2: 0.8 - 0.95
                             •   RMSE: 2.5 - 4.0 gCm-2                        •   RMSE: 1.0 - 1.8 gCm-2
                             Nighttime                                        Nighttime
                             •   R2: 0.2 - 0.4                                •   R2: 0.75 - 0.9
                             •   RMSE: 1.5 - 2.5 gCm-2                        •   RMSE: 0.5 - 1.0 gCm-2
                              Good filling performance for daytime            Very good filling performance for
                               but not for nighttime                            daytime and nighttime data

                             Techniques:                                      Techniques:
                             •   MIM, MDV, UKF_LM, NLR & Others, 3            •   MIM, Others, ANNs leading                          19
                                 ANNs leading
        QuickT im  DailySum Bias per Site Year:
                 e™ and a
   T IFF (LZW ) decom pressor
are needed to see this picture.
                                                                 T IFF (L
                                                              are n
                                                                      QuickT im
                                                                    eed
                                                                               e™ and a
                                                                         ZW ) decom  presso
                                                                       ed to see this p
                                                                                            r
                                                                                       icture .




                                  Medium Gaplength, ALL




                         Bias




                                   Medium gap length (1.5 days):
                                  Bias of <0.07 gCm-2 per filled day
                                                                                                  20
        QuickT im  DailySum Bias per Site Year:
                 e™ and a
   T IFF (LZW ) decom pressor
are needed to see this picture.
                                                                 T IFF (L
                                                              are n
                                                                      QuickT im
                                                                    eed
                                                                               e™ and a
                                                                         ZW ) decom  presso
                                                                       ed to see this p
                                                                                            r
                                                                                       icture .




                                   Long Gaplength, ALL




                    Bias




                                     Long gap length (12 days):
                                  Bias of <0.2 gCm-2 per filled day
                                                                                                  21
               Annual Sum Error Estimate
             QuickT ime™ and a
        T IFF (LZW ) decom pressor
     are needed to see this picture.
                                                T IFF (L
                                             are n
                                                     QuickT im
                                                   eed
                                                              e™ and a
                                                        ZW ) decom  presso
                                                      ed to see this p
                                                                           r
                                                                      icture .




Assumption:
• representative choice of golden sites
• good technique (red stars)
 Error estimate on the annual sum

Annual Sum Error
• Small to med gaps: <0.07 gCm-2 per filled day equivalent
• Periods of longer gaps: <0.2 gCm-2 per filled day equivalent
 Quality of long gap filling critical




                                                                                 22
        QuickT im




                       Calculation Example
                 e™ and a
   T IFF (LZW ) decom pressor
are needed to see this picture.
                                                            T IFF (L
                                                         are n
                                                                 QuickT im
                                                               eed
                                                                          e™ and a
                                                                    ZW ) decom  presso
                                                                  ed to see this p
                                                                                       r
                                                                                  icture .




       QuickTime™ and a                                         QuickTime™ and a
   TIFF (LZW) decompressor                                  TIFF (LZW) decompressor
are needed to see this picture.                          are needed to see this picture.
                     Example for average file with 35% gaps:
                     • 18% small to medium gaps
                     • 18% periods of longer gaps of 5-10 days
                         =
                     18% ^ 66 filled days

                     Error estimation
                     • 66 x 0.07 gCm-2: 5 gCm-2
                     • 66 x 0.2 gCm-2: 13 gCm-2
                      Total error induced by filling of the
                         gaps on the annual sum:
                                      ±18 gCm-2                                              23
                  Test using Real Gap Filling Results
                        QuickT ime™ and a
                   T IFF (LZW ) decom pressor
                are needed to see this picture.
                                                                                  T IFF (L
                                                                               are n
                                                                                       QuickT im
                                                                                     eed
                                                                                                e™ and a
                                                                                          ZW ) decom  presso
                                                                                        ed to see this p
                                                                                                             r
                                                                                                        icture .




                                Are 18gCm-2 an appropriate estimate of the
                                    error on the annual sum prediction?

  Site year   NLR_F_MOD NLR_FM_OLS                ANN_BR      ANN_PS      MDS            SPM            Mean   SDeviation
be1_2000          -345.33    -334.88                -345.11     -332.59    -344.85       -330.72       -338.91       6.90
be1_2001          -514.90    -521.12                -516.60     -512.07    -510.05       -514.68       -514.90       3.82
de3_2000          -516.83    -529.61                -533.83     -506.75    -518.76       -516.25       -520.34       9.84
de3_2001          -505.44    -478.84                -507.30     -501.43    -498.62       -507.29       -499.82      10.84
fi1_2001          -175.21    -170.23                -177.87     -173.09    -166.84           nan       -172.65       4.29
fi1_2002          -219.32    -215.26                -224.59     -210.88    -219.46       -193.83       -213.89      10.85
fr1_2001          -565.38    -558.17                -548.90     -560.20    -553.04       -542.40       -554.68       8.29
fr1_2002          -547.27    -550.18                -545.86     -553.62    -542.51       -542.03       -546.91       4.48
fr4_2002          -297.55    -304.54                -289.67     -323.76    -322.55       -328.88       -311.16      16.08 1)
il1_2002          -170.64    -172.56                -118.64     -147.36    -160.02           nan       -153.84     22.09 1)2)
it3_2002           -25.94     -23.37                 -35.39      -29.67     -35.55        -21.84        -28.63       5.93

                                                                                                      1) no soil temperature
                                                                                                      2) 30 day system failure



          •    Standard deviation between techniques of filling the real dataset with 
               35% gaps ≤ 16 gCm-2

                                                                                                                          24
        QuickT ime™ and a                              QuickT ime™ and a
   T IFF (LZW ) decom pressor                     T IFF (LZW ) decom  presso r
are needed to see this picture.                are n eeded to see this picture .




       QuickTime™ and a                               QuickTime™ and a
   TIFF (LZW) decompressor                        TIFF (LZW) decompressor
are needed to see this picture.                are needed to see this picture.



                                  Questions?



                                                                                   25
        QuickT ime™ and a                                    QuickT ime™ and a
   T IFF (LZW ) decom pressor                           T IFF (LZW ) decom  presso r
are needed to see this picture.                      are n eeded to see this picture .




                                 Let’s fill our
                            “knowledge gaps”
                                  and have a
                                QuickTime™ and a
                            TIFF (LZW) decompressor
                         are needed to see this picture.
                           fun and productive
                                  workshop!




                                                                                         26
                Separation of Daytime and
             QuickT ime™ and a
        T IFF (LZW ) decom pressor
     are needed to see this picture.
                                                 T IFF (L
                                              are n
                                                      QuickT im
                                                    eed
                                                               e™ and a
                                                         ZW ) decom  presso
                                                       ed to see this p
                                                                            r
                                                                       icture .




                     Nighttime Data
Keyfile: 10% artificial gaps
Fragmented Golden File: 80% daytime NEP data, 35%
  nighttime NEP data

• Real gap filling:
   20% real day gaps, 65% real night gaps
                                       1:3
• Artificial gap filling:
   8% artificial day gaps, 3.5% artificial night gaps
                                       2:1

 Important to consider daytime and nighttime data
  separately
                                                                                  27
                                 QuickT ime™ and a
                            T IFF (LZW ) decom
                                                   Bias on daily sums
                                               pressor
                         are needed to see this picture.
                                                                    T IFF (L
                                                                 are n
                                                                         QuickT im
                                                                       eed
                                                                                  e™ and a
                                                                            ZW ) decom  presso
                                                                          ed to see this p
                                                                                               r
                                                                                          icture .




                                                      Daytime data
                     • Distribution of bias error of the individual daily sums

                         Daytime data
                                                                               Daily Bias Error:
Bias Error (gCm-2)




                                                                               - up to 4 gCm-2

                                                                               ANNs leading
                                                                               ANN_BR




                                                                                                     28
                                 QuickT ime™ and a
                            T IFF (LZW ) decom
                                                  Bias on daily sums:
                                               pressor
                         are needed to see this picture.
                                                                    T IFF (L
                                                                 are n
                                                                         QuickT im
                                                                       eed
                                                                                  e™ and a
                                                                            ZW ) decom  presso
                                                                          ed to see this p
                                                                                               r
                                                                                          icture .




                                                    Nighttime data
                     • Distribution of bias error of the individual daily sums

                         Daytime data
Bias Error (gCm-2)




                                                                               Daily Bias Error:
                                                                               - up to 2 gCm-2

                                                                               ANN_PS leading




                                                                                                     29

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:22
posted:10/18/2010
language:English
pages:29