Fourteen 20 m × 30 m plots were established on the Oklahoma State University Research Range 13 km west of Stillwater, Oklahoma, USA. Prior to the study, two fires burned the entire site in February 1991 and March 1996. Each plot was marked with three permanently established transects, mid-point and 5 m to each side, oriented on the long axis of the plot. Vegetation data was collected in the late growing season at 10 random points along each transect. Plant functional group composition using cover classes for tallgrasses, other grasses, forbs, legumes, and woody plants was estimated in 2003 (baseline) and then following fire treatments in 2005, 2008, 2010, 2014, and 2016. We randomly applied seven treatments to each of the two replications. The treatments were fire applied every two years during 1) January to February (Jan–Feb), 2) March to April (Mar–Apr), 3) May to June (May–Jun), 4) July to August (Jul–Aug), 5) September to October (Sep–Oct), 6) November to December (Nov–Dec), and 7) the control (No Burn; Figure 1). The plots were burned in 2004, 2006 to 2007, 2008 to 2009, 2010 to 2011, 2012 to 2013, and 2014 to 2015, wsith the same fire treatment applied to the same plot each time.
For analyses of vegetation, we used plot as the experimental unit. As an experimental study with only two replications across space and six iterations of each fire season treatment (i.e., every two years) during the 13-year study period, we consider this a long-term case study. We calculated the 2003 and 2016 means for each plant functional group and presented the mean from the two replications of each fire seasonality treatment. For woody plant cover, we used linear least squares regression to determine if woody plant cover change was correlated with time in each seasonal treatment based on the coefficient of determination (r
2), a P-value for fit linear trendlines, and the slope of significant trendlines to predict annual increases.