First page Back Continue Last page Overview Graphics
Data Smoothing
Data samples are intrinsically biased
- Recent samples represent short term averages
- Ancient samples represent long term averages
- Different variables have different sample periods
Integrate samples over N years
- Matches short term data to long term data
- Matches temperature to CO2 and CH4
- Isolate long and short term periodicity
- Isolate long and short term dependency
Notes:
The Nyquist theorem states that if there if there are components at a frequency N, samples must be at a rate greater than N*2 in order to avoid aliasing. The consequence of this is that there will be significant aliasing issues with data that has substantial short term, periodic variability and sample intervals greater than half the period of this variability.
The analysis program fits samples into 50 year buckets by averaging and interpolating. This compensates for samples less than 50 years apart and for missing samples. Vostok samples are always more than 50 years apart, while DomeC samples are as closely spaced as 10 years. The DomeC analysis has been performed with 10 and 25 year buckets, but other than differences in the recent, short term temperature variability, the results are the same.
The smoothing techniques used calculates an average from data points on either side of a sample, as opposed to calculating a running average of older samples.