Detailed Analysis of CSALT Model

The fit of the CSALT model to global temperature data is so close that it makes some sense to do a detailed analysis of the occasional glitches in the data.  This could tell us if there are other noise components that could fill the gap.

Initially what I want to do is a subjective interpretation — if one can isolate the parts of the model that don’t work so well, this may open up areas for further investigation.  For example, was the large warming spike at the end of WWII (starting in late 1943) real or was it due to calibration issues [1]?

Figure 1: Residual of the CSALT model fit to GIS data. Note the largest warming discrepancy starting in late 1943.

The following annotated fit (Figure 2) of the GISS temperature record to the CSALT model uses the GISS aerosol optical thickness data and a 12-month window to get rid of seasonal noise.

Figure 2: Peak-by-peak analysis of CSALT model to GISS temperature anomaly. Main volcanic eruptions shown as lower-case. Dark Blue is CSALT model and Green is GISS data.

To guide the eye, and as an example, the annotation labelled R indicates a period of time corresponding to the a warming residual shown in Figure 1 near 1943-1944. The individual volcanic eruptions are listed below.

label eruption year month month# intensity
a krakatoa 1883 8 42 22
b colima 1890 6 126 9.5
c calbuco 1893 1 144 5.1
d mayon 1897 6 210 4.3
e santamaria 1902 10 274 17.6
f novarupta 1912 6 390 5.7
g agung 1963 2 998 17.7
h augustine 1976 0 1152 7.3
i elchichon 1982 4 1228 18.6
j pinatubo 1991 6 1338 30

The subjective interpretation is described next for each of the points on the curve. Characterizing a portion of the fit as Excellent means that the shape, timing, and intensity of the model peak (and surrounding valley) matches the data clearly.  Labeling it Good means that the overall fit works but something is a bit off, and Marginal means that it is only showing signs of incipient matching.

A – Excellent, Krakatoa time frame
B – Excellent, shoulder dips not as deep in model
C – Excellent, model peak of SOI not as high
D – Marginal, wide peak, lots of volcanic activity may be occluding the SOI.
E – Excellent
F – Excellent
G – Marginal, cooling trough which is elevated in model
H – Good, Novarupta time frame
I – Excellent
J – Excellent
K – Poor, Small, narrow peak missing
L – Excellent
M – Excellent, two fluctuations followed by a dip
N – Good, missing valley
O – Marginal, Peak OK, but no troughs on either side
P – Marginal, Not high enough peak in 1938
Q – Excellent
R – Poor, Suppressed warming peak in 1943
S – Excellent, trough occurs
T – Excellent
U – Excellent
V – Good, follows a cooling trend
W – Excellent, recovery from Agung eruption
X – Excellent
Y – Excellent
Z – Excellent, low peak from St. Augustine eruption
1 – Excellent, narrow peak
2 – Good, Timing of peak OK but not strong enough
3 – Excellent
4 – Excellent
5 – Excellent, Pinatubo
6 – Excellent
7 – Excellent, strong El Nino of 1998, model not quite as warm as data
8 – Marginal, Shifted peak position, extra fluctuation
9 – Excellent
0 – Excellent
# – Excellent
! – Poor, Heating predicted by model not observed — until this last month.

Of the 37 total annotations the count is :
Poor – 3
Marginal – 5
Good – 4
Excellent – 26

Most of the fits are good enough that they don’t need much commentary, but we can further decompose the Marginal and Poor fits.

The Marginal peak fits are :
– D – Not as much cooling suppression in the GISS aerosol model in spite of lots of volcanic activity
– G – Deeper cooling trough than the neutral SOI shows.
– O – Neutral SOI conditions not providing enough peak-to-peak
– P – Not strong enough SOI event, note that SST bucket measurement procedures are changing during this time [1]
– 8 – SOI neutral peaks seemed to have shifted by a year or two after temperature

The Poor peak fits are (shown in Figure 3) :
– K – A very small narrow peak that may be related to a rebound from a cooling dip in 1921
– R – No SOI event during end of 1943.  Again the SST calibration may be important or war efforts may have caused some issues in collection or measurements.
– ! – The current year is not close to the degree of warming that SOI neutral conditions would imply

Timely that September’s GISS data was recently made available. The last few month’s data points is shown in the right-most panel below. For the model to continue to hold, any gap between model and data can not persist. It is indeed possible that this gap is closing with a temperature anomaly of 0.74C for September. This is something to keep an eye on.

Figure 3: Close up of poor fitting areas, K, R, ! and a marginal area P

The lag time of SOI to a globally averaged temperature anomaly is 6 months, so that we could pay special attention  on a monthly basis to how well the CSALT model is capturing the actual temperature profile . The fact that it can do so well over 130+ years (save for the occasional odd excursion) implies that it will continue to work, unless of course a different climate regime takes over.  Some believe this could happen if the circulation patterns controlled by Arctic warming start to shift the jet stream [2][3].


  1. CSALT model 
  2. Climate Variability and Inferring Global Warming


[1]  P. D. Jones and T. M. L. Wigley, “Corrections to pre-1941 SST measurements for studies of long-term changes in SSTs,” in Proc. Int. COADS Workshop, HF Diaz, K. Wolter, and SD Woodruff, Eds., Boulder, CO, NOAA Environmental Research Laboratories, 1992, pp. 227–237.
[2] G. H. Miller, S. J. Lehman, K. A. Refsnider, J. R. Southon, and Y. Zhong, “Unprecedented recent summer warmth in Arctic Canada,” Geophysical Research Letters, 2013.




8 thoughts on “Detailed Analysis of CSALT Model

  1. I learned that GISS updated their SST approach at the beginning of 2013

    “2013-01-16: Starting with the January 2013 update, NCDC’s ERSST v3b data will be used to estimate the surface air temperature anomalies over the ocean instead of a combination of Reynold’s OISST (1982 to present) and data obtained from the Hadley Center (1880-1981). “

    This may have some effect on the match between data and model after 2012 in Figure 3, where some divergence is noted.


  2. From elsewhere:

    climatereason | November 3, 2013 at 2:19 am |

    You also have the Buckets to engine water inflow question to consider.
    SSTs were effectively not measured during the war years and a collective effort wasn’t made again until the 1950′s. Even then data has been sporadically missing due to a number of political issues, for example lack of cooperation with Russia prior to and after the war.

    So we have one vote that the temperature spike of 1943-1944 could have been due to missing temperature measurements.
    During the war years I noticed that a huge chunk of SST measurements were missing south of the equator, according to the map at the bottom of this page:

    The CSALT model was not predicting a warm spike in 1943-1944 as the complementary SOI spike did not exist, yet it is possible that the global temperature signal was interpolated (or kriged) to indicate that a warming spike was there. Whether this was real or an artifact of the interpolation is up in the air.

    From GISS

    Can see that the 1943-1944 spike was there at high northern latitudes but not at equatorial and low latitudes. Also observed at higher southern latitudes, which allowed it not to suppress the spike but actually exaggerate it.


  3. Pingback: Simple models of forced warming | context/Earth

  4. Pingback: Convergence in Temperature Time Series | context/Earth

  5. Pingback: CSALT and SST corrections | context/Earth

  6. Pingback: Variational Principles in Thermodynamics | context/Earth

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s