Take a look at all 4 result maps on the same page.

Trends

There's a clear trend of increasing “Best” quality area from 1990 to 2050, but a less clear trend between 2050 and 2100. Between 2050 and 2100 the only visible trend is a replacement of “Moderate” quality land with “High” quality land, and not much expansion of the “Best” land. This follows from the fact that from 1990 to 2050, a huge increase in EGDD causes much of the land to be Class 2 or higher (above 1200 GDDs), and from 2050 to 2100 the edges are filled in but the main body of the Yukon simply remains in the highest class. With a further analysis that included a Class 1/Class 2 distinction, this could be clarified, and more of a shift could be seen from 2050 to 2100. However, for the purpose of comparison I lumped Class 1 and Class 2 together for the initial analysis. Further, given Class 2 is the point at which most crops can be grown successfully (with regards to GDD), there is a key point that arises from this, which is that within 50 years the majority of the Yukon will be within Class 2 and not substantially limited by Growing Degree Days.

Further, it appears that the distribution of “Best” land is rather patchy, with the highest concentration around Dawson City in the mid-Northwest, and some around and North of Whitehorse. This may be a combination of the influence of the roads layer (which goes to and from those towns) and the soil texture layer (which had loam areas concentrated in the Southwest corner). A sort of positive feedback loop is present here, in that the towns must be placed in areas that are feasible for human population, and these are often flatter areas, and given the remoteness may also have had to be somewhat correlated to agricultural activities. Roads are then built in the context of these towns, and by using proximity to roads some of the biases would be incorporated into the analysis. I believe this is a good bias, as it accounts for the fact that being close to towns is good for allocation of resources, safety, accessibility, etc. I didn't directly run a proximity analysis for towns because (1) that would adding to a bias already present in the roads layer and (2) that would bias the analysis towards already explored areas rather than looking at the potential for entirely new towns based on agriculture, in which I was more interested.

Sources of Error and Further Analysis

There is one major source of error in particular: scale. Performing this analysis at the scale of the entire Yukon involved low resolution soil data and broad area EGDD values (using ecodistrict polygons as opposed to raster based temperature data). For a more precise and locally-significant analysis, it would be crucial to develop or locate detailed soils maps and a higher resolution temperature data. It might be conducive to zoom in on the areas with clusters of “Best” designations and see how suitable the features really are, and how the factors I considered to be constant might actually change (creeks drying up, vegetation changes causing shifting soil regimes, etc).

However, using the EGDD ecodistrict polygons did lead me to another key discovery- that approximating EGDD using mean temperature values is a fairly valid estimate, and though it will somewhat underestimate the final values and will underestimate the variation between areas, the trend direction and magnitude are maintained. For a more complete and accurate analysis, it would be best to go further and perform the analysis and the calculations using a higher resolution raster data set of mean monthly temperatures.

Error graph:

The orange line is the EGDD from daily temperature values, the blue is from my approximated EGDD values from monthly temperature values, and the yellow line is the difference between them. The x-axis is simply the various polygons arranged in order from fewest EGDD to most EGDD for simplicity. The y-axis is EGDD. 

Another area in which further detail would be appropriate is within the proximity to roads and rivers layers. I was unable to locate data for the size of the rivers or roads, which would be quite useful for determining how much water various watersheds could provide. I think it's interesting that in the calculated 1990 map the only cluster of “Best” land is located right around Dawson City, and I wonder if this is because of the aforementioned road bias, as well as the location of the Yukon River causing a high concentration of drainage sources in the region. However, a more detailed analysis would incorporate the different order streams, which would limit and focus the ideal agricultural lands to areas with substantial water sources rather than just any recorded water source. For the proximity to roads criterion, rather than dividing up the roads layer into different sized roads, I think the best approach would be to obtain information about the less-used but decently maintained roads that exist throughout the Yukon.

Also, I have made the assumption that EGDD are the most important factor in Yukon agriculture, with water being the second, based on a few different sources. However, a sensitivity analysis should be performed to see how much the weighting of different criteria affect the results, and what sort of limits arise with different weightings. I think this would best to do within the context of better quality layers, with a consideration for the factors above: Higher resolution temperature data, unequally weighted river data, a more extensive road network, and better quality soil data. I regret not being able to develop some of these layers, but time was an issue and despite being highly interested in the topic myself, much more work could and should be done before solid conclusions are made.