The wind energy development suitability product is a per-pixel (30 square-meters) model representation of the predicted probability (0.00-1.00) that an area can support wind energy development. The result is represented as a percentage, such that any value greater-than 0.5 would be classified as suitable for wind energy development in model space. To model suitability for wind energy development, we used 9,399 observations of ‘windmills’ taken from the FAA Digital Obstruction File (http://bit.ly/dof_12549; accessed June, 2016). Because the extent of the LCD region is limited to the panhandle region of Texas, we excluded all windmill observations outside of Texas from consideration during model building. To generate psuedo-absences, we generated 10,000 random points within the state of Texas and randomly selected 9,399 observations that were more-than 250 meters away from a wind turbine (consistent with the k=1 nearest neighbor distance between turbines in Texas wind farms). We fit our model to a suite of variables hypothesized to be limiting factors governing wind energy development and use. These include : proximity to 10 MW, 12 MW, 22 MW, 115 MW, 138 MW, 161 MW, 230MW, and 500 MW transmission lines, wind production class (1-5), continuous wind speed (at 100 m), as well as elevation, slope, aspect, and topographic roughness (at 8100 m2 and 656,100 m2 scales). Our transmission line data for Texas were purchased from S& P Global Platts (http://www.platts.com. Our wind production class data were derived from the NREL 50 meter national wind production potential product (https://goo.gl/yT9kkM). And our continuous wind speed were derived from NREL 80 meter national wind product (https://goo.gl/oguPbz). All covariates were rasterized and gridded to a consistent 30m resolution for analysis (Albers, NAD83). To account for potential model overfit, we used a threshold-based optimization procedure that sought to minimize Random Forests’ out-of-bag (OOB) error statistic using an importance threshold (parsimony=0.3) applied to each candidate variable in an iterative model building process (Murphy et al., 2010). Variables with an OOB importance consistently in the lower (30th) percentile were excluded from the final model. Our final candidate variable series was : proximity to 12, 138, 22 and 69 MW transmission lines, elevation, and continuous wind speed (100 m). Finally, we further condensed our variable space by aggregating each individual proximity-to-transmission raster selected for inclusion in our final model as a single surface, such that wind suitability ~ f(proximity to transmission, elevation, continuous wind speed). We then projected suitability across the state of Texas as a continuous raster surface and used quantile-based contour-delineation to extract relevant suitability hotspots for the LCD.
