All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 486 179 9
PDF Downloads 275 130 3

Specification and Prediction of Global Surface Temperature and Precipitation from Global SST Using CCA

View More View Less
  • 1 Climate Prediction Center, NOAA/NWS, Washington, D.C.
Restricted access

Abstract

A reconstructed sea surface temperature (SST) dataset is used to examine relationships between SST and seasonal mean surface temperature (T) and total precipitation (P) over most of the global continents for the 1950–92 period. Both specification (i.e., simultaneous) and predictive relations are studied.

Canonical correlation analysis (CCA) is used to describe the relationships and to provide information aiding in physical interpretation. A sequence of four consecutive 3-month periods of global SST anomalies is related to T and P anomalies during the fourth period for the specification analyses, and to 3-month periods ranging from one to four seasons later for the predictive analyses. Dynamical specifications of the National Centers for Environmental Prediction (NCEP) atmospheric model, using observed SST anomalies as boundary conditions, are also examined for confirmation of and comparison with the statistical specification relationships suggested by the CCA.

Specification and predictive cross-validated skill is modest except for certain regions and/or times of the year having correlations of 0.5 and greater. Seasonal T is generally specified/predicted with greater skill than P. Some regions have seasonally in their specificability/predictability, where skill varies more strongly as a function of the target season than lead time for T, P, or both. In these cases, such as Sahel African rainfall in northern summer or northeastern Australian rainfall in May through July, the skill of specification is not substantially higher than the skills of short or even moderately long lead prediction.

Specifications and predictions are skillful in areas affected by the ENSO, including the tropical Pacific islands for all seasons, and during specific seasons in northern and eastern Australia, and parts of Africa and North and South America. Skill is lowest in Europe and midlatitude Asia where ENSO's direct influence is lacking. However, non-ENSO predictive skill sources also contribute substantially to final skill; these exist both in regions strongly and minimally influenced by ENSO. The most important of these is an interdecadal trend from the 1950s to the 1980s–90s defined by a warming in the Indian and South Atlantic Oceans paralleling a cooling in the North Pacific and Atlantic basins. Another controlling SST dipole with a less obvious trend includes mainly the tropical SST of all three ocean basins versus the extratropical (especially Northern Hemisphere) SST. Still other, more localized, SST patterns are suggested as critical.

Some of the regions that show modest but usable seasonal predictive potential have no prior specificative or predictive history because they are not directly influenced by ENSO and/or have marginal data quality or density. This is encouraging, since the statistical skill realized here should be reproducible, and hopefully surpassable, using dynamical models.

Abstract

A reconstructed sea surface temperature (SST) dataset is used to examine relationships between SST and seasonal mean surface temperature (T) and total precipitation (P) over most of the global continents for the 1950–92 period. Both specification (i.e., simultaneous) and predictive relations are studied.

Canonical correlation analysis (CCA) is used to describe the relationships and to provide information aiding in physical interpretation. A sequence of four consecutive 3-month periods of global SST anomalies is related to T and P anomalies during the fourth period for the specification analyses, and to 3-month periods ranging from one to four seasons later for the predictive analyses. Dynamical specifications of the National Centers for Environmental Prediction (NCEP) atmospheric model, using observed SST anomalies as boundary conditions, are also examined for confirmation of and comparison with the statistical specification relationships suggested by the CCA.

Specification and predictive cross-validated skill is modest except for certain regions and/or times of the year having correlations of 0.5 and greater. Seasonal T is generally specified/predicted with greater skill than P. Some regions have seasonally in their specificability/predictability, where skill varies more strongly as a function of the target season than lead time for T, P, or both. In these cases, such as Sahel African rainfall in northern summer or northeastern Australian rainfall in May through July, the skill of specification is not substantially higher than the skills of short or even moderately long lead prediction.

Specifications and predictions are skillful in areas affected by the ENSO, including the tropical Pacific islands for all seasons, and during specific seasons in northern and eastern Australia, and parts of Africa and North and South America. Skill is lowest in Europe and midlatitude Asia where ENSO's direct influence is lacking. However, non-ENSO predictive skill sources also contribute substantially to final skill; these exist both in regions strongly and minimally influenced by ENSO. The most important of these is an interdecadal trend from the 1950s to the 1980s–90s defined by a warming in the Indian and South Atlantic Oceans paralleling a cooling in the North Pacific and Atlantic basins. Another controlling SST dipole with a less obvious trend includes mainly the tropical SST of all three ocean basins versus the extratropical (especially Northern Hemisphere) SST. Still other, more localized, SST patterns are suggested as critical.

Some of the regions that show modest but usable seasonal predictive potential have no prior specificative or predictive history because they are not directly influenced by ENSO and/or have marginal data quality or density. This is encouraging, since the statistical skill realized here should be reproducible, and hopefully surpassable, using dynamical models.

Save