Data

Raw Data Visualization

Fig 6. Screen shot of data matrix used in the microclimate analysis. Each plot was given a unique ID, based on the site name, remnant number, forest type and transect number. Predictor variables are shown across the top row. Most of the predictor variables, particularly the percent cover variables, are continuous and zero-inflated. Forest structure variables was untransformed, though species data was Hellinger-transformed.

Outlier Detection in Microclimate Variables

Fig 6.1: Distribution of air temperature measures, relative to local ACIS station. 

Some of our microclimate sensors were disturbed by animals during the course of the field season. Data was visually assessed to determine when the sensor was disrupted and the usable range of data (if any) of disturbed sensors was determined. The period of June 26-August 16th was used for the analysis as this period had the most number of undisturbed sensors, 128. To minimize site-level variation, temperature readings were subtracted from the nearest ACIS weather station.

Outliers in the data were assessed visually using boxplots (Fig 6) and histograms (Fig 6.2). All of the microclimate variables were normally distributed, except for maximum soil temperature, which had two outliers skewing the data. These outliers were removed in order to fit the assumptions of linear models, but there is reason to believe these readings are accurate as they are from sites with high solar radiation (low canopy cover) and coverage of mineral soil. 

Fig 6.2: Histogram showing the distribution of relative temperature values. 

Influence of Forest Structure on Microclimate

Fig. 7: PCA of forest structural variables (gray) correlation with microclimate variables (black) in harvest and fire remnants. Temperature values represent the mean daily maxima and minima recorded from June-August, 2023.

Maximum daily air temperature (AirMaxT), maximum soil temperature (SoilMaxT) and minimum soil temperature (SoilMinT) are strongly correlated with each other along the component one axis (Fig 7). We see strong correlations on this axis with variables indicating shady forest interiors like canopy closure (CanopyClose), moss cover (Moss), and tall shrub cover (TS). 

Minimum air temperature (AirMinT) has a stronger correlation with the component two axis and variables associated with ground conditions and vegetation cover such as (Litter), organic layer depth (OrgLayer) and tall shrub cover (TS). 

Fig. 8: PCA of forest structural variables (gray vectors) correlation with microclimate variables (black vectors) in reference forest and remnants. Temperature values represent the mean daily maxima and minima recorded from June-August, 2023.

When comparing remnants to reference forest, our response variables (temperature) are more spread out, which is expected as the sample is more homogenous. 

Still we can observe correlations between maximum air temperature (AirMaxT) and canopy closure (CanopyClose). Soil temperatures again seem strongly correlated with ground conditions like moss cover (Moss) and organic layer depth (OrgLayer) (Fig 8.).

Air minimum temperatures are again corelated with ground conditions and percent cover of vegetation. 

Microclimate Influence on Vegetation Communities

Outlier Site Constrains Fire vs. Harvest Comparison

Fig. 9: NMDS showing microclimate variables and Hellinger-transformed vegetation data at our six disturbance sites. Ellipses represent 80% confidence intervals. Not all cover classes are shown, to minimize clutter on the graph. 

One of our fire sites, Eagle Creek (EC) contains half of our microclimate transects in fire. This site is associated with higher raw temperatures and different vegetation (Fig 9). communities due to its geography. While the rest of our sites are found in Upper Foothills and Lower Subalpine ecoregions, EC borders the Montane. Species including Festuca campestris and Androsace chamissonis are found at EC but not at the other sites. 

Including the EC sites in our species matrix will limit our ability to make inferences about species composition based on microclimate and forest structure, so they were excluded from the analysis of community composition. Since this reduces our replication of fire sites by half, we do not expect to be able to make the fire vs. harvest comparison. Fortunately our larger data set has more replication of fire sites, but does not have microclimate data.

Plant Community Varies by Plot Location

Fig. 10: Results of an direct gradient analysis on Hellinger-transformed vegetation data. Black vectors are temperature predictor variables. Ellipses represent 80% confidence intervals for different plot locations (disturbance, edge and interior). Not all cover classes are shown, to minimize clutter on the graph. 

After removing EC sites from the data, we can see more clearly how different species are correlated with microclimate and edge effects (Fig 10). Interior and disturbance plots differ in plant community compoisition, with edge plots being intermediate between the two.  We can see that moss (NONVAS) cover is associated strongly with interior plots, and litter (NLITT) is associated with disturbance plots.