Berkeley evaluates global model performance

Many Models Still Struggle With Overall Warming; None Replicate Regional Warming Well

The following graphics show the land surface temperature results derived from Berkeley Earth's computations versus the output of a series climate models. In the literature these models are referred to as Global Climate Models or "GCMs." Climate models are computer simulations which attempt to mathematically reproduce the Earth's climate, and are used by some scientists studying climate change to attempt to understand the future timing, magnitude, and effects of global warming.

The top portion of each graphic shows the warming trends (degrees C per century) over land, ocean, and global areas for the referenced model. The bottom portion of each graphic shows the difference between the model results and the actual historical temperature map derived from the Berkeley Earth land-based dataset. A model which perfectly replicates the historical temperature trend for all regions of the globe would show a uniform green color (the color for zero difference in temperature trend between the model and the historical record) for all land areas. The yellow/red range of the color scale indicates model temperature trends which overstate (are higher than) the actual historical trend for a region. The blue range of the color scale indicates model temperature trends which understate (are lower than) the actual historical trend for a region. NASA's "GISS-E2-H" model appears to replicate overall warming best, but is among the worst models at replicating regional trends. China's "FGOALS-g2" appears to be the best performer at replicating regional trends, but still over/under-predicts the warming trend by about 0.5C per century almost everywhere.​

Source: Berkeley Earth



Here is the best matching, GISS model for global temperature:


The nice thing about Berkley is they provide the technical details needed to evaluate models:

• Names and versions
• Run conditions
• Performance metrics
  • global mean
  • regional correlation

Of the two, the Chinese model would be the easiest to 'adjust' by multiplying by a constant. However, the GISS model may be helped with more recent data from the global CO2 observatory. The race is on!

Bob Wilson

 

I began to wonder where the 'models' are found:
CMIP5 - Data Access - Availability

It appears there are nearly half-a-hundred but as I look at them, I'm finding many are specific to goals independent of global temperature. For example, the Chinese model I looked at appears more interested in rain fall and water. With 1/3d of the human population, water for agriculture is much more important to their policy managers than global warming. So they need to make plans to deal with the food supply. Accuracy in water prediction does not need accuracy in global temperature prediction.

So the key lesson is each model needs to be evaluated for its accuracy relative to the stated goals. Identify the model and we can discuss what is going on. To be blunt, using a 'rain gauge' model for a 'thermometer' belies a critical lack of understanding.

Bob Wilson