Data
Processing
Table 1. Example of data used as input for ClimateNA.
Many of the data were in separate tables and different formats that had to be merged into one table suitable for this analysis. The ITRDB site location data was in its own file, each ITRDB time series was in a separate chronology file, and the Alberta Spruce TRDB was in another table with site data and tree ring widths together, but the ring widths were in the opposite order of the ITRDB data and listed in columns instead of rows.
Table 2. Example of data received from ClimateNA. Climate variable and year fields have been added. Original fields have been multiplied by number of years so climate variables could be recorded for each series ID for each year specified.
Once the time series from each database were filtered, the site locations were merged to be used as an input in ClimateNA -- a sample of which can be seen in table 1. ClimateNA then produced the same data duplicated for each year with annual climate variables added as columns. The altered data can be seen in table 2. CMI was the only climate variable analyzed, so the others were removed. Some CMI values were listed as -9999, so these were replaced with NAs and omitted from any calculations.
Table 3. ClimateNA output data after removing climate variables other than CMI and joining the detrended tree ring widths (RW).
Separate iterators were used to read and merge the data from each database. The ITRDB iterator read each chronology file and merged it to the ClimateNA dataset based on year and ID. These were already detrended. The Spruce TRDB was converted to a comma-separated values file with only series IDs and ring widths, where the data were transposed so series IDs were in columns and years represented rows. The columns were then flipped so the earliest year was highest, making each series compatible with the detrend.series function from the dplR package in R. NA values were removed in each series so blank years did not interfere with the detrending function. These detrended ring widths were then also joined to the ClimateNA dataset based on ID and year. The data updated with detrended ring widths can be seen in table 3.
Table 4. Final dataset with added CMI, drought, and resilience index variables.
Another iterator was used to calculate the minimum CMI, mean CMI, and subsequent CMI threshold for each time series. This threshold was then compared to the CMI value in each year to determine drought years, which were listed under another column as a logical value. A nested iterator then calculated pre-drought and post-drought values for drought years with the requisite 5 years of ring widths centered on the drought event. Finally, resilience index values were calculated based on the ring width of the drought year and the calculated pre- and post drought ring width averages. These additional fields can be seen in table 4.
The final version of the data had ID, Species (SPCD), ECOZONE, and ECOZONE_NAME as categorical variables; Year, Latitude, Longitude, Elevation, CMI, detrended ring width (RW), minimum CMI (CMImin), mean CMI (CMImean), CMI threshold (CMI threshold), pre-drought detrended ring width (preRW), detrended ring width during drought (drougthRW), post-drought detrended ring width (postRW), resistance index, recovery index, and resilience index as numerical variables; and Drought as a logical variable. For the purposes of analysis, Drought was the predictor variable, and the resilience indices were the response variables. Each ring width series was considered as a sampling unit.
Exploration
Table 5. Number of ring width series by ecozone and tree species. Red values indicate totals that fell below 10 and were filtered out.
Table 5 shows the total number of ring width series for each tree species in each ecozone. Some species and ecozones had considerably more samples than others; Picea glauca in the Boreal Plains had an especially high number at 319.
Figure 5. A map showing the included ecozones and tree sample location by species.
Figure 5 shows the location of each tree sample location by species relative to Canada's ecozones. It is clear that the sample locations are not evenly distributed. For example, sample locations in the Prairies are only around the edges of the ecozone.
Figure 6. Distribution of CMI values as a percentage of range for each sample from 1900 - 2019.
Figure 7. Distribution of detrended ring width values as a percentage of range for each sample from 1900 - 2019.
The distribution of CMI and detrended ring width values as a percentage of their range throughout the analyzed period can be seen in figures 6 and 7. The values were converted to percentages based on the maximum and minimum from each series. As a result, every series had a 0% and 100% value which is why there appears to be spikes for both variables at these values. Both values have fairly normal distributions, although the detrended ring width data are flatter and slightly skewed towards lower values.
Figure 8. Scatter plot of detrended ring width and CMI as percentages of the range of each sample from 1900 - 2019.
Figure 8 is a scatterplot of the values for detrended ring widths and CMI as percentage of the range of each sample. There does not appear to be much of a correlation between the two variables. If the CMI threshold was expected to be an accurate predictor of lower ring width values, there would ideally be some kind of funnel shape where CMI and detrended ring width are more closely correlated at their lower ends.