Satellites have the potential to offer the most accurate surface temperature data. In this series I analyse the latest data records from NSSTC/UAH (Lower Troposphere) for the period Dec 1978-Feb 2024.
The HadSST4 data from 2004 on has a sinusoidal wave on top of the trend. If annual variation why not up to 2004 or in the NSSTC6 data. Looking forward to your ruminations. Regarding the last scatterplot from about 0.8 degC (HadSST4) the LOESS line seems by eyeball not to be a good fit to the data (running low). My eyes or LOESS issue?
Yep, a sinusoid has been imposed and I'll be digging into this. The LOESS fit was copied across using a default template but I'll wade in and tweak it...
The Argo system of sea temperature measuring bouys cam into effect from 2005. THe Argo system should be very accurate and have good spatial distribution. Pre 2005, SST were primarily buckets and intakes measurements from ships. So maybe the data pre and post 2005 look different because they actually are?
My very thoughts! I delve into things a bit more in part 7 but, yes, what it boils down to is two different datasets bolted together and this should not be treated as a homogeneous time series with which to do predictive modelling.
The HadSST4 data from 2004 on has a sinusoidal wave on top of the trend. If annual variation why not up to 2004 or in the NSSTC6 data. Looking forward to your ruminations. Regarding the last scatterplot from about 0.8 degC (HadSST4) the LOESS line seems by eyeball not to be a good fit to the data (running low). My eyes or LOESS issue?
Yep, a sinusoid has been imposed and I'll be digging into this. The LOESS fit was copied across using a default template but I'll wade in and tweak it...
The Argo system of sea temperature measuring bouys cam into effect from 2005. THe Argo system should be very accurate and have good spatial distribution. Pre 2005, SST were primarily buckets and intakes measurements from ships. So maybe the data pre and post 2005 look different because they actually are?
My very thoughts! I delve into things a bit more in part 7 but, yes, what it boils down to is two different datasets bolted together and this should not be treated as a homogeneous time series with which to do predictive modelling.