Canadian Forest Service Publications

A comparison of two approaches for generating spatial models of growing season variables for Canada. 2015. Pedlar, J.; McKenney, D.; Lawrence, K.; Papadopol, P.; Hutchinson, M.; Price, D. Journal of Applied Meteorology and Climatology 54(2):506-518.

Year: 2015

Available from: Great Lakes Forestry Centre

Catalog ID: 35871

Language: English

CFS Availability: PDF (request by e-mail)

Available from the Journal's Web site.
DOI: 10.1175/JAMC-D-14-0045.1

† This site may require a fee.

Abstract

This study produced annual spatial models (or grids) of 27 growing season variables for Canada that span two centuries (1901-2100). Temporal gaps in the availability of daily climate data – the typical and preferred source for calculating growing season variables – necessitated the use of two approaches for generating these growing season grids. The first approach, used only for the 1950-2010 period, employed a computer script to directly calculate the suite of growing season variables from existing daily climate grids. Since daily grids were not available for the remaining years, a second approach, which employed a machine learning method called boosted regression trees (BRT), was used to generate statistical models that related each growing season variable to a suite of climate and water-related predictors. These BRT models were used to generate grids of growing season variables for each year of the study period, including the 1950-2010 period to allow comparison between the two approaches. Mean absolute errors associated with the BRT-based grids were approximately 30% higher than those associated with the daily-based grids. The two approaches were also compared by calculating trends in growing season length over the 1950-2010 period. Significant increases in growing season length were obtained for nearly all ecozones across Canada and there were no significant differences in the trends obtained from the two approaches. Although the daily-based approach tended to have lower errors, the BRT approach produced comparable map products that should be valuable for periods and regions for which daily data are not available.

Plain Language Summary

This study produced spatial models of growing season variables for Canada that span two centuries (1901-2100). Understanding and representing growing season variables from the past and into the future is important for many sectors including the forest sector. Growing season grids for the 1950-2010 period were generated by processing recently developed daily temperature and precipitation grids. High quality daily grids were not available prior to 1950 or for the future, growing season variables were estimated for these periods from a suite of 57 climate and water-related predictors using a modern statistical technique called boosted regression trees (BRT). Errors associated with both approaches were reasonable; for example, the mean absolute error for growing season start date was about 6 days for the daily-based product and about 10 days for the BRT approach. We further compared the two approaches by calculating trends in growing season length for the period 1950-2010. Significant increases in growing season length were obtained for nearly all ecozones across Canada and there were no significant differences in the slopes obtained from the two approaches. This work is part of an ongoing effort to provide spatial climate data for Canada; it can be accessed directly at http://cfs.nrcan.gc.ca/projects/3 or through data requests to the corresponding author.

Funding programs

  • Clean Air Agenda - Adaptation (Forest Change)
Date modified: