cca made easy. the science of seasonal climate forecasting is all about connecting forcing and...
TRANSCRIPT
CCA MADE EASY.
The Science of Seasonal Climate Forecasting is all about connecting forcing and response. Once a forcing mechanism is identified the response can be anticipated and so we can make a forecast. CCA is just the tool to connect the forcing and the response.
“CCA is about paring signal and response, the forecaster use then the signal to anticipate the response.”
Let’s start with facts in climate
During El Niño abnormally warm water (reddish) in the eastern Pacific creates enhanced convective rainfall in the Eastern Pacific.
During La Niña the warm water is located further west with precipitation confined to the Western Pacific.
FORCING AND RESPONSEThere is a strong teleconnection between the location of warm water and the occurrence of rainfall: If we can identify associated SST and rainfall patterns, we can then use such SST patterns to predict the location and timing of the corresponding rainfall. We just need a tool to identify corresponding patterns.
CCA is about finding corresponding patterns between the forcing and the
response
The basic concept is simple:1) Identify spatial patterns in two data sets, X (the
forcing) and Y (the response), that are related in time (CCA modes).
2) Use the patterns in the X data (forcing) to predict the patterns in the Y data (response).
3) Et voila.
Example of CCA : Rainfall and SST in the Pacific
Note :(1983;1998) and (1989;1999) were cases of strong associations but in opposite direction.1990 was year of weak or unrelated patterns.
83 98
89 99
90
Rainfall during DJFSST during DJF
Degree of association between those two patterns from 1981 to 2008 : positive scores means similar pattern, negative scores mean similar pattern but with opposite sign, near zero scores mean weak or unrelated pattern.
1983
1989
1990
Let’s have a closer look
La nina teleconnection
El Ninoteleconnection
Nothing special, no relation !: Call it normal
how the two patterns are actually associated
Use slide show mode (F5) to see the animation
Let’s take it further : bias correction
R=0.64
The CCA shows a strong association between GCM rainfall (X) and observed rainfall (Y) with r=0.64 showing the GCM capability to capture inter-annual rainfall variability BUT it misses the location of the rainfall in Kenya (spatial bias). If we looked at GCM rainfall in Kenya (targeted region) we would miss the signal.
conclusion
• CCA = identifying patterns that tend to coincide in time or space
• Seasonal forecasting is about capturing large scale signal (ocean, circulation, monsoon). Searching for a signal common to many stations (pattern) increases the efficiency/robustness of our forecast.
• Climate has no frontier let’s open our horizons from a single point to set of stations.
• Does it always work ? well you can quite always find good matching pair of X and
Y fields BUT the identified CCA Y field should capture lot of the variability of the raw Y field.