Background & data used

As a practice, geneocological studies such as this one, utilize common gardens to monitor phenotypic measurements of plants from a dispersed set of populations in a common environment. By collecting seed from a variety of populations in the range of the species, and growing them up in a common garden, the effects of site to site variation are more controlled. Ultimately, the phenotypic traits of each plant observed at all of the common gardens are meant to be tied back to the seed source population so that we might learn about how population-level selective pressures translate to divergent traits (even when these plants are grown in a variety of climates). In theory, planting the same seed source population in multiple climate regimes will illuminate what traits are plastic, (i.e. change within one lifetime of a plant) and what traits change over generational time. For example, if you observe that no matter what common garden a seed source population is planted in, you always see wide leaves, and populations from other seed source populations always have narrow leaves, then it is possible to reach the conclusion that leaf width changes over generational /evolutionary time. This is important because if plants have been adapting to selective pressures over many generations and they are moved to a differing conditions, the traits they have developed might be maladapated to the new conditions. My dataset includes 39 population means (4-5 individuals) for each of sixteen gardens for several phenotypic traits such as leaf width, crown width and others.

One of the problems I hope to deal with is how to both quantify and visualize bluebunch wheatgrass phenotypic trait variability across soil gradients. One technique that may be useful is multiple linear regression. One drawback of my dataset however, is the fact that many soils variables are spatially auto correlated. This lack of independence leads to a violation of one of the assumptions of multiple linear regression. In order to deal with the spatially related variables in my dataset, it seemed appropriate to use a regression method that accounts for this spatial relationship.

Question

Is there a relationship between available water storage (AWS) in the top 25cm of the soil surface and the leaf width in bluebunch wheatgrass populations?

Approach and tools used

In order to make my results easier to interpret, I have restricted my analysis to a single common garden that contains 39 different bluebunch wheatgrass populations.

Tools:

1. geographically weighted regression analysis in ArcMap.
2. Spatial join tool in ArcMap
3. Join field tool in ArcMap.
4. Hot Spot analysis tool in ArcMap

Steps followed to complete the analysis

1. Surveyed gssurgo database for continuously variable soil metrics that might be relevant to plant growth (i.e. available water storage).
2. Joined AWS tabular data to gssurgo raster data for the entire study area.
3. Spatially joined soils tabular data to common garden source population x y coordinates, and plant phenotypic traits data from common garden monitoring.
4. Selected a common garden that had existing soils data for all seed source populations.
5. Reduced dataset to only include soils and trait data for the Wahluke common garden in seed zone 1.
6. Performed geographically weighted regression analysis.
   1. Analysis failed initially due to the wide geographic spread of populations.
   2. Set the minimum number of populations to be used in analysis to 8 (5 was too few, and 10 produced a smaller p-value).
7. Exported regression results to a .csv table
8. Joined local regression coefficients to seed source population shape file attribute table.
9. Symbolized local regression coefficients according to sign (negative = blue, positive = red)
10. Performed a hotspot analysis on mapped regression coefficients to look for significant clustering

Results

• Positive relationships between leaf width and available water storage occur in moderately wet areas
• Negative relationships between leaf width and available water storage occur in moderately wet to dry areas.
• Negative coefficients tend to be near other negative coefficients spatially
• Positive coefficients tend to be near other positive coefficients as well

Critique

• Limited to the analysis of one plant trait and one common garden at a time.
• The resolution and accuracy of the AWS data is questionable.
• The AWS values may not vary enough to be biologically significant.