Skip to content

Latest commit

 

History

History
146 lines (95 loc) · 4.53 KB

Notes.md

File metadata and controls

146 lines (95 loc) · 4.53 KB
Raw

Experiment Notes

2024-03-30

  • Significant slope change pre-1990. Charts changed to post-1990 to reflect that.

  • Changed calculations to use fractional year. Rates are all 10 years/decade

  • Influences are fit using both lags and exponential smoothing at the same time. A full search is performed to choose the best fit value. No smoothing and using the CMIP values are also checked.

  • Having problems with using FFT. First, it provides the value for cosines, not sines. If I expand the dataset (zero padded) other than a multiple of the total, an angle is changed so that the reconstructed cosine is correct in the center of the data set, even though the frequency isn't quite correct. Not sure how this is calculated, yet. Not using in calculations for now.

2024-03-04

  • Significant work done:

    • New annual data approach should work with all plot functions. Convert to annual at the last moment.

    • Added a Chow Test to assess breaks in data

    • still looking at how the nu value works. Clipped it to 1.0 whenever it goes below that amount. This prevents the variance from decreasing below the calculated value.

    • Added a simple exponential function to model a more gradual warming.

Next:

  • Add the ability to use the simple exponential function for all variables (find optimum value in advance for each).

  • Add the ability to just use lags (as per Foster and Rahmstorf 2011). Again find the optimum value in advance for each).

2024-02-28

  • Removed periodic signal

  • Changed fitting to only work on monthly data, then convert to annual if required.

  • Changed fitting to use lowess smoothing for removing trend, instead of a straight line. Lowess smoothing uses locally weighted slopes for each point of data.

2024-02-28

  • Added smoothing to ENSO values.

  • Consider making smoothing function an optimized exponential like Tomino

  • Check residual for any periodic signal with FFT as in Foster and Rahmstorf, 2011

2024-02-27

  • Fixed some persistent plotting problems with monthly data

  • 2002 produces a really small nu value before and after. This makes the estimated error very small. Must investigate further.

2024-02-22

  • Made the charts include annual and reduced data. Also had both slopes (before and after the breakpoint) include the breakpoint. Before, the before slope did not include it. The downside of this is that outliers will exacerbate the differences in the two slopes, bringing them in opposite directions.

2024-02-21

  • New priority: see if the temperature is accellerating.

2024-02-20

  • started a plotting function to explain how the ocean warming curve is applied.

2024-02-15

  • The pre-industrial era (PIE) offset changed

    • from 0.3117
    • to 0.366609
      This is the the offset from 1850-1899 to 1961-1990 in Hadcrut5. Other termperature sources are normalized with this value. I.e. all datasets are normalized by subracting the mean of 1961-1990, then adding 0.366609 so than now the mean of 1961-1990 = 0.366609 for all sets.

  • The plots were finalized for posting.

2024-02-14

  • Made annual version of chart. Has much better R^2 value.

  • Solar has a long-term trend which should be in the projection. But since the trend is currently -ve, it actually moves the crossing points further out. But it might not be that, so will leave the more pessimistic values, which are still more optimistic that the simple projection.

Next:

  • Make plots postable (legend, etc)

2024-02-13

  • Combining the axes onto one figure created a very squished image, even when stretched vertically to fit the entire sceen. The headings and text would all need to be adjusted. In the end, it is easier to composite the two charts onto one image manually.

  • Solar trend was removed from the fit data. Any trend would show up in the linear fit, so this prevents doubling and makes a better fit to detrended data.

  • Data with solar trend removed:

Original standard deviation was: 0.1411°C
New standard deviation is: 0.1133°C
Reduction of 19.7%
New slope is 0.191°C/decade
R² value is 0.351

Next:

  • Make a year-end annual chart

2024-02-12

  • Tried squaring ENSO (actually enso * enso.abs()) to get better matching on peaks, but the overall variance reduction went from 19.7% to only 18.0%

  • The Pacific Decadal Oscillation has no effect on the overall variance reduction.

  • MEI had a much worse fit than all the ENSO indices together

Next:

  • Move the axes onto one figure for easier posting.