Jan 10, 2012

The "LRC Model" Versus The GFS Model

Last year I began comparing the long range Gloabal Forecast System or GFS forecast 500mb heights to a LRC based average of actual 500mb heights. It took a few forecast periods to get the calculation ironed out but this year beginning on Nov-23 I 'officially' started the comparisons.

How is the comparison made? On a daily basis the atmospheric soundings for a location are extracted and stored into a database. Every ten days the 192-384h GFS forecast atmospheric soundings for the same location are extracted and stored into the same database. As the numbers enter the database, a calculation is run to compare the number sets and dynamically update the trend. The X factor in the calculation is the LRC duration. This number allows the averaging of the actual heights to take place. Thus far this year the cycle is recurring roughly every 44-48 days. In November I chose 48 as the constant.

The criteria that I feel are meaningful with this comparison are listed below. The station used for the data is GRB, station number 72645, Green Bay, WI. Disclaimer: I admit to my lack of statistical knowledge.

Mean - Average difference from actual
Maximum - Highest difference from actual
Minimum - Lowest difference from actual
Range - Difference between Max and Min
Standard Deviation - How widely the difference varies from actual
Count - Amount of data samples
Confidence Interval - (only used to calculate confidence)
Confidence (95%) - Likelihood the forecast lies within the range specified

The above list suggests that between the dates of Nov-23 and Jan-7 the LRC model outperformed the GFS model (192-384h). Honing in on actual with a 95% chance of the trend being within +/- 106mb. The range in spikes from actual between the two was a 99mb difference creating an argument for the LRC model being a bit more stable. Click on the image above for the entire list of results.

The greatest output of the LRC model is it can be trended months into the future. Providing the ability to see pattern fluctuations weeks and months in advance and pin pointing specific dates for when they will recur. The graphical examples below are forecast trends for 500mb in Green Bay, WI and maximum temperatures in Oshkosh, WI. I would like to note that with seasonal fluctuations in the cycle duration it is important to interpolate these trends to the current cycle duration. Click on the graphs for the forecast trends into July, 2012.

(500mb Forecast Trend For Green Bay)

(Temperature Forecast Trend For Oshkosh)

As always, if there are any questions or thoughts on my research and analysis of the LRC or how I presented the material just let me know in the comments section of the blog. Thanks for reading!