Outlier Analyses

1
3. Conduct outlier assessment of NPP and model
driver data
6. Compare model results with field measurements

• Outlier Analyses
• What is an outlier? data point unrepresentative
  of its general location or otherwise difficult
  to represent by a generalized NPP model.
• Outlier analysis tests used to assign individual
  flags
• outside of reasonable fixed limits for variables
  (e.g., NPP gt 2000 gC/m2)
• unrepresentative site for model comparisons
  (e.g., crop, pasture, plantation or wetland
  sites)
• extreme values within a biome, (e.g., outside of
  .95 percentiles)
• inconsistencies at a site, such as precipitation
  reported for the site being different than the
  precipitation derived from global climate data
• comparison of measured versus modeled ensemble
  NPP (average NPP from all models) Bias,
  Normalized Error, and Mean Absolute Error
• outliers based on regressions with driver
  variables NPP vs evapotranspiration (AET), NPP
  vs precipitation, and NPP vs temperature
• unrepresentative based on visual inspection and
  knowledge of sites
• Overall site flag
• overall flag calculated as weighted sum of
  individual flags
• minor outlier comparisons weight 1
• major outlier comparisons weight 10
• managed sites weight 100
• sites with flag gt 5 were excluded

Class ASites
Outlier Analysis Results Number of
records Steps in Outlier Analysis Class
A B C Initial NPP measurements,
sites with multiple measurements 162 2363
2768 NPP measurements passing outlier
analysis 151 1689 2624 Percent
measurements that were excluded 7 29
5 Mean NPP measurements for unique sites
excluding outliers 81 933 1637
Class B Sites

• 4. Provide model driver data for sites to
  modelers
• 5. Run global/regional models for sites without
  fine tuning models

Class C Cells