Qc2 Development 20090505
Recap from last meeting 20081119: Linear Interpolation with Triangulated Neighbours Test Results of RARR_24 Inverse Distance Squared Interpolation Above cross-validation analysis courtesy of Matthias Mohr The standard deviations of the error in precipitation amount are not very different (3.5 mm for triangulation and 3.7 for inverse distance squared)
Developments/Ongoing Recap from last meeting 20081119: Developments/Ongoing Test Tools: unit tests, automatic result plotting from algorithms and netCDF files, Interpolation options: inverse-N distance weighting, triangulation, height normalisation, General tidying and improved encapsulation, including flags, Wet/Dry discrimination, Interface to preferred neighbours station list, e.g. for triangulation, Interface to PROJ Cartographic Projections Library including geographic to UTM for linear interpolation, Expand the configuration file options, Time interpolation strategies.
Developments Test Tools: unit tests, automatic result plotting from algorithms and netCDF files, Interpolation options: inverse-N distance weighting, triangulation, height normalisation, General tidying and improved encapsulation, including flags, Wet/Dry discrimination, Interface to preferred neighbours station list, e.g. for triangulation, Interface to PROJ Cartographic Projections Library including geographic to UTM for linear interpolation, Expand the configuration file options, Time interpolation strategies. Test environment with Terje Need for filter of duplicate data arising from multiple typeids. Identification of the ”official value”. New …
Time series interpolation
Example 1: [multiple strategies]
Comparison with nearest neighbour
*NB Ole-Einer og Matthias have better methods Height correction [0.006*dHt]* *NB Ole-Einer og Matthias have better methods
Spline interpolation :: Akima Note agreement/disagreement of spatial and temporal methods Full variational analysis: combine both?
Example 2: [single points]
| 0.2|211|311|0|0|0.2| |-32767|211|311|0|0|0.7| | 1.1|211|311|0|0|1.1| 76330|2006-12-16 09:00:00 | 0.2|211|311|0|0|0.2|1111100000000010|7000000000000000| 76330|2006-12-16 10:00:00 |-32767|211|311|0|0|0.7|1000601000000007|38347000000000C1|QC1-4-211:1,hqc,hqc 76330|2006-12-16 11:00:00 | 1.1|211|311|0|0|1.1|1110100000000010|7000000000000000| | 0.2|211|311|0|0|0.2| |-32767|211|311|0|0|0.7| | 1.1|211|311|0|0|1.1|
76330|2006-12-16 09:00:00 | 0.2|211|311|0|0|0.2|1111100000000010|7000000000000000| 76330|2006-12-16 10:00:00 |-32767|211|311|0|0|0.7|1000601000000007|38347000000000C1|QC1-4-211:1,hqc,hqc 76330|2006-12-16 11:00:00 | 1.1|211|311|0|0|1.1|1110100000000010|7000000000000000|
Height corrected nearest neighbour 76330|2006-12-16 09:00:00 | 0.2|211|311|0|0|0.2|1111100000000010|7000000000000000| 76330|2006-12-16 10:00:00 |-32767|211|311|0|0|0.7|1000601000000007|38347000000000C1|QC1-4-211:1,hqc,hqc 76330|2006-12-16 11:00:00 | 1.1|211|311|0|0|1.1|1110100000000010|7000000000000000| Height corrected nearest neighbour x
Thanks to Zbigniew for identifying the application of TAN & TAX 76330| 2006-12-16 11:00:00 | 1.2 | 215 | 311 | 0 |0 | 1.2 |1110000000000010|7000000000000000| 76330| 2006-12-16 11:00:00 | 0.9 | 213 | 311 | 0 |0 | 0.9 |1110000000000010|7000000000000000| Best estimate for single missing temperature = 0.5*(TAN + TAX) ? Thanks to Zbigniew for identifying the application of TAN & TAX
Next steps Large-scale statistical analysis of results to converge on best strategies, Identification and porting (if required) of most effective spline-fitting, Akima, etc. algorithms / library, Input on algorithms welcome, Kriging!
Test environment
https://dokit.met.no/sysdok/kvalobs/driftsdokumentasjon Special thanks to quality assurance work from Terje, Bjørn, …
Test environment Operations QC2 QC1 db Achieved in vm and in earlier work. Test environment QC2 db QC1 Static test population Operations
Test environment ? ? ? Operations vm objectives QC2 QC1 db db QC0 HQC Simulator ? db QC1 Validate Output QC0 ? ? db Operations
Duplicate data filter …
Duplicate Data Filter Why How need to identify official value for each time where duplicates exist use one value for time and space interpolations all algorithms complicated by duplicates avoid more lengthy bug searches How specification of rules to identify official value specification of required configuration information define a universal flag to identify the official value in all circumstances Need for input to define these requirements
Tusen takk!