Error message

  • Notice: Undefined offset: 1 in drupal_settings_initialize() (line 789 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: ckeditor. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: custom_breadcrumbs. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: custom_breadcrumbs_identifiers. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: fancybox. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: jquery_update. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: lexicon. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: libraries. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).
  • User warning: The following module is missing from the file system: taxonomy_sync. For information about how to fix this, see the documentation page. in _drupal_trigger_error_with_delayed_logging() (line 1143 of /var/www/html/includes/bootstrap.inc).

Historic model runs and climate variability

A common pit-fall when using scenario data

When comparing a reference period (e.g. 1981-2010) of a climate model dataset to meteorological measurements of the same period, it might be tempting to carry out a comparison of the simulated time-series to the corresponding time-series of observations. This would normally result in a rather poor agreement and thus disappointment which might lead to the conclusion that the climate scenario and/or climate model has quality problems. However, you should be aware that climate scenarios (e.g. from the CMIP5 project), also historic runs, should statistically resemble observations (neglecting biases), e.g. monthly means and variances over 30 year periods are similar to those observed, but 1 or more specific months (say Jan 1985) are not identical between model and observations. If you need such climate predictions, you are looking for seasonal to decadal predictions. If so first read the remainder of this page and the links below, then go to Seasonal to decadal predictions.

Such a conclusion based on a point-by-point comparison is unwarranted because it fails to take into account the influence of initial conditions of the simulation in relationship to slow modes of the natural climate variability. In order to achieve any reasonable point-by-point temporal agreement, climate scenarios have in the future to be replaced by climate predictions.

Explanation

The explanation to this point-by-point mismatch is as follows. The GCM simulations were initiated using initial conditions representing a possible weather situation in 1850 or 1860. But the exact state of the atmosphere and the oceans did not actually occur — it is only a situation that reasonably could have occurred in that year. In particular, the specific state of the oceans were not possible to observe at that time, so it could not be used as part of the initial condition. As the oceans are a major source for the slow natural variability it is consequently not possible to have the GCM simulation starting from a real ocean state. As the simulation progresses towards the present time, this difference between the simulated and the real world ocean state remains and results in that the simulated decadal-scale climate variability is out of phase with reality. For a more in-depth explanation see the paper by Jones and Nikulin (2009).

 


Winter surface temperature for Southern Sweden from 3 members of the ECHAM5 A1B ensemble dynamically downscaled by the RCA3 regional model. Seasonal mean temperatures are shown for each ensemble member for 3 sections of the transient integration from 1950-2100. The green line shows the ensemble mean, 20-year mean temperature for each period. Image source: Jones, C.G. & Nikulin, G. Rossby Centre Newsletter, May 2009, pp. 4-8.

The graph above illustrates the problem. For each of the three downscaled GCM simulations the interannual and decadal-scale variability is totally out of phase with each other, and none of them can be expected to reflect the observed interannual to decadal-scale variability. In effect, the model calendar cannot be expected to be in phase with reality on short time-scales. As is clearly seen in the graph, the situation changes when one focus on a slightly longer time-scale (e.g. 20-30 year periods) for which most of the natural climate variability is averaged out. However, even for a 20-30 year period the effect of natural variability — and the dependency on initital conditions — is not totally removed due to multi-decadal components of the natural variability.

What to do?

The correct way to evaluate and analyse climate scenarios is to focus on long enough periods. Usually this means 30-year periods, which can be either overlapping or consecutive. The length of the period depends on the variable, for some variables such as seasonal or annual temperatures a 20-year period may sufficient, but for other variables like monthly precipitation longer periods should be used. Within each period the analyses should focus on comparing statistics between the climate scenario and the corresponding observational record. In this way much of the natural variability is exposed as statistical measures of averages, spread, and frequencies within the different periods rather than for individual years. However, to fully remove the effect of multi-decadal climate variability the analyses have to be based on an ensemble of climate scenarios, from different GCMs and/or from several runs with different initial conditions of one GCM.

Read more

 

image: