Data
Processing
There was an initial total of 893 tree chronologies available from the two tree ring data sources within the study area. 443 were from the Canadian Forest Service, and 450 from the ITRDB.
First, the chronology IDs, coordinates and elevation were imported from both tree ring data sources so they could be merged and used to extract monthly CMI data from ClimateNA. An example of this format can be seen in table 1 and the subsequent joined monthly CMI data can be seen in table 2.
CMI for each year was them calculated as the sum of CMI from the previous June to the current August and the monthly values were removed.
ITRDB data were imported from chronology files where the ring widths were already detrended, Alberta tree ring data were detrended using the same settings to match.
Dates were filtered to 1980 - 2010 for both RWI and CMI
CMI and RWI were converted to a wide format in separate tables where each entry represented a chronology and the variables represented each year.
146 chronologies that didn't have ring widths for half or more of the time frame were removed. The rest of the missing ring widths were imputed using randomForest (Liaw & Wiener, 2002). The percentage of NA values can be seen for each year before and after the chronologies were removed in figure 5.
Values for both CMI and RWI were standardized based on each chronology and then multiplied by negative one so PCA vectors would be directed towards lower values for each. Examples of the wide, standardized, and negative formats for CMI and RWI can be seen in table 3.
Species codes were temporarily merged to extract counts for each species. Species with a count below 10 were omitted. The count of each species considered can be seen in table 4, with the removed species represented as red text. This resulted in the removal of 27 series.
Once clusters were determined, a final table was generated that joined tree ID, Species code (SPCD), Cluster, RWI, and CMI so resilience indices could be calculated and timelines plotted for each cluster (Table 5).
The final version of data had ID, Species (SPCD), and Cluster as categorical variables; and Year, periodic CMI, and RWI as numerical variables. For the purposes of analysis, CMI was the predictor variable, and RWI was the response variable. Each chronology was considered as a sampling unit. The final number of chronologies after data processing was 720.
Exploration
The distribution of values for RWI and CMI can be seen in figure 6 and figure 7, respectively. The distribution for RWI is normal, but CMI is skewed right. Both variables were standardized by chronology, which allowed them to have the same weight in the PCA ordination.
Figure 8 shows the PCA ordinations for RWI, CMI, and both CMI and RWI. All three were considered, but using both resulted in the strongest clustering and correlation between vectors and clusters.
All trees in the original data set are displayed in figure 9, where removed trees are illustrated in red. It appears that some of the chronologies that were removed were in more spatially diverse locations and would have extended the environmental variation in the dataset. Most removals were due to shorter time series, so in future study these could be retained by analyzing a longer time frame.
The time frames of each chronology are shown in figure 10. Many of the series end close to or soon after 1985, supporting that more chronologies could be kept by altering the analyzed time frame.
