Global numerical weather prediction (NWP) models provide the best available framework for assessing global variability and trends for a variety of weather-climate parameters over time, including temperature. The most advanced NWP models include the Global Forecast System (GFS) model run by the US National Center for Environmental Predictions (NCEP) and the European Center for Medium-Range Weather Forecasts (ECMWF) model. These models provide output on rectangular latitude-longitude grids with horizontal grid cells that are not equal in surface area. Another approach for global weather modeling is to use icosahedral horizontal grids with grid cells that are approximately equal in surface area, such as the Finite-volume Icosahedral Model (FIM) being developed by the US National Oceanographic and Atmospheric Administration (NOAA) Earth System Research Laboratory (ESRL). The FIM approach may eventually become predominant, but for now, the older GFS and ECMWF models are in the forefront.
Reanalyses have been performed using both the GFS and ECMWF models in conjunction with additional input data not available in real-time for weather forecast model runs, plus the original input data used for the forecast runs, but with additional data quality filters. The GFS reanalysis performed by NCEP is called the Climate Forecast System Reanalysis (CFSR) and the latest ECMWF reanalysis is called the ECMWF Interim Reanalysis (ERAI). The weather models have been evolving over the years, including higher spatial resolution, and thus changes have been made to the reanalysis efforts to take advantage of more advanced model features. The oldest NOAA reanalysis effort is being maintained and updated by the National Center for Atmospheric Research (NCAR) and is called the NCEP/NCAR Reanalysis I (NCEP/NCAR R1). It has been run backward to 1948, but has a relatively low spatial resolution. However, it uses a consistent model and methodology over the entire period from 1948 to present, in contrast to changing methodologies that complicate other reanalysis efforts for comparing longer time periods. The NCEP/NCAR R1 has output horizontal grid resolution of 2.5/2.5 degrees latitude/longitude. The NCEP CFSR and more recent NCEP CFSR Version 2 (CFSV2) provide horizontal output of 0.5/0.5 degrees latitude/longitude. The CFSV2 became operational beginning April 2011. The ERAI has horizontal output resolution of 0.75/0.75 degrees latitude/longitude. The CFSR and ERAI were both run back to 1979 to provide historical analyses. Recently, the ERAI has been adjusted in an effort to compensate for possible bias differences resulting from model input changes that began in 2002.
Four different reanalysis estimates of monthly global surface temperature anomalies are compared below for the period from 1979 through 2015, with two of them extending through November 2016. The NCEP/NCAR R1, NCEP CFSR/CFSV2, and ERAI Unadjusted data sets were obtained from the University of Maine Climate Change Institute (UM CCI) and the ERAI Adjusted data set was obtained from the ECMWF Copernicus web site. Since there were changes in methodologies in the ERAI beginning 2002 and in the NCEP CFSR/CFSV2 beginning April 2011, the estimates are compared using two different reference periods, 1979-1998 and 2011-2015.
Figures 1 and 2 provide overviews of the full 1979-2016 time series of monthly global surface temperature anomaly estimates referenced to 1979-1998 and 2011-2015 respectively (click on any graph image for a larger view).
Figure 1. Overview of global surface temperature anomalies referenced to 1979-1998.
Figure 2. Overview of global surface temperature anomalies referenced to 2011-2015.
A closer look at the 20th Century estimates referenced to 1979-1998 is shown in Figure 3. All four of the estimates are in close agreement during this period, with the spread generally near or less than 0.1 degrees Celsius (C). Most of the largest outliers are from the coarser resolution NCEP/NCAR R1.
Figure 3. Global surface temperature anomalies for 1979-2000, referenced to 1979-1998.
Figure 4 provides a closer view of the estimates for the 21st Century so far, as referenced to 1979-1998. During this period, the estimates diverge substantially, with a spread closer to 0.2C most months. The NCEP CFSR/CFSV2 is generally near the top of the spread for 2001-2009, but then suddenly drops to the bottom of the spread beginning April 2010, a year before the CFSV2 became operational. The NCEP/NCAR R1 and ERAI Adjusted show the closest and most consistent match during this period.
Figure 4. Global surface temperature anomalies for 2001-2016, referenced to 1979-1998.
For an alternative perspective, the same estimates have been graphed relative to a 2011-2015 reference period, as shown in greater detail in Figures 5 and 6. From this reference perspective, there is excellent agreement among all four estimates for 2011-2015. The largest spread is in the 20th Century portion, shown in Figure 5. Oddly, the NCEP CFSR/CFSV2 separates high during the period from about 1999 through March 2010, while the other three estimates are much closer together but lower. For the period 1979-1998, the ERAI and NCEP/NCAR R1 match well, with the ERAI Unadjusted slightly higher and the NCEP CFSR/CFSV2 the highest.
Figure 5. Global surface temperature anomalies for 1979-2000, referenced to 2011-2015.
Figure 6. Global surface temperature anomalies for 2001-2016, referenced to 2011-2015.
The corresponding global temperature trends for 1979-2015 projected to 100 years range from +1.25C for ERAI Unadjusted, to +1.32C for NCEP CFSR/CFSV2, to +1.52C for NCEP/NCAR R1, to +1.67C for ERAI Adjusted. These trends are independent of any reference period and the range in these trends illustrates some of the uncertainties involved in trying to determine global temperature trends.
The four global surface temperature anomaly reanalysis data sets examined showed good agreement for 1979-1998 when referenced to that period and showed excellent agreement for 2011-2015 when referenced to that period. However, trends indicated by the four data sets differed significantly, mainly because of apparent reanalysis modifications during the period from 1999 through 2010. The differing trends indicate a small part of the uncertainty involved in trying to assess global temperature trends over long time periods.
UM CCI (NCEP/NCAR R1, NCEP CFSR/CFSV2, and ERAI Unadjusted)
ECMWF Copernicus (ERAI Adjusted)