Results & Discussion

Figure 11 shows the results of the PCA and k-means clustering, which is summarized in table 6. The spatial distributions of each cluster is also illustrated in figure 12. 13 distinct drought events were identified, and trees were separated into 10 separate clusters. There were no drought events identified in 2 of the clusters (9 and 10), while cluster 8 had no drought events indicated by CMI and RWI vectors. Drought events 1, 2, 9, and 13 were strongly indicated, while events 7 and 10 had predictor and response vectors pointing in opposite directions. The drought vectors did not point directly at affected clusters as was expected. This is likely because most drought events had effects in multiple clusters, so clusters were plotted in a space between their associated vectors. Additionally, more spatially distinct clusters also appear to have more distinct RWI and CMI variations, reflecting the spatially correlated nature of moisture regimes.

Figure 13 displays RWI and CMI over the 31-year time frame for each cluster. Years with droughts and drought responses are marked on the time series for CMI and RWI, respectively. RWI followed CMI more closely in clusters 1, 2, 3, and 7; moderately in cluster 5 and 6; and weakly in clusters 4, 8, 9, and 10. The two clusters without associated drought events both had higher CMI values, indicating that moisture requirements may not be as limiting in these areas. Cluster 8 is closer to the tree line, and appears to have other factors limiting growth, so moisture restrictions may also be less limiting.

The resilience indices for each species within each cluster can be seen in figure 14. Species with a frequency less than 5 in a cluster were omitted due to excessive confidence intervals. There is considerable variation between species and within species in different clusters, matching the expectation that drought response varies between species and climate (Serrano & Peñuelas, 2005). However, unexpectedly many of the species in each cluster did not show significant post drought decreases. Clusters that did indicate losses were 3, 4, 6, and 7. Picea glauca had similar resilience in clusters 4 and 6 despite having different levels of CMI. Pseudotsuga menziesii had low resilience in cluster 6 where CMI is low, but recovered to pre-drought levels in cluster 2 with a higher CMI despite having similar levels of resistance in both. This could indicate that lower moisture levels limit the ability of this species to recover after a drought. Pinus banksiana had very high resilience in cluster 3, increasing past pre-drought levels, but did not perform as well in cluster 4 despite similar moisture levels. In cluster 4, Picea glauca suffered less post drought than in cluster 3. This could mean that the growth increase in Pinus banksiana in cluster 4 is due to lower vitality in its competition and a relatively higher drought tolerance. The only clusters with species that suffered post drought with a confidence level greater than 95% were clusters 3, 6, and 7. All three of these clusters have locations that match previously identified vulnerable areas of the boreal forest (Hynes & Hamann, 2020). However, clusters 1 and 5 are also in these vulnerable areas and do not appear to have reduced growth post drought.

Next Steps

First, CMI is an approximation due to using potential evapotranspiration (Hogg, Barr, & Black, 2013). Root-level soil moisture is the most important factor during drought (Liu, et al., 2020), but this is difficult to measure directly due to equipment and time requirements. However, satellite soil moisture is available which could more closely mirror root soil moisture (Gruber, Scanlon, van der Schalie, Wagner, & Dorigo, 2019). Soil moisture data could also be used in conjunction with CMI to improve its accuracy.

As previously mentioned, while drought is one of the most limiting factors of tree growth, there are other environmental variables that can have a similar effect (e.g.: cold winters, late frosts, abnormally high temperatures, etc.) (Delpierre, et al., 2016). Including the effect of these events would improve the accuracy of observed drought responses.

Further, there are also many attributes that can change the effect of drought on a stand level (e.g.: density, species composition, age, etc.) (Martínez-Vilalta, López, Loepfe, & Lloret, 2012). It might be possible to analyze some of these attributes by including data from the Canadian National Forest Inventory (Beaudoin, et al., 2014).

Finally, the drought events found here could be compared to drought that has been recorded historically (Agriculture and Agri-food Canada, 2023). This would ensure that the events indicated through this analysis are accurate.