We calculated lines-of-sight viewscapes for all 2XXX houses using  a 10-m DEM [Source of DEM] and a computationally efficient line-sweeping algorithm in GRASS GIS (Haverkort, Toma, & Zhuang, 2009). This DEM represents the highest-resolution elevation data available since publicly-available LiDAR data do not exist for the study area. We set the observer height at 3-m above the surface to simulate a typical household viewpoint and we restricted the maximum visibility distance to 10-km in all directions. Though LiDAR data have been shown to more accurately model viewscapes in the presence of forest (Vukomanovic et al. 2018), this region is dominated by low stature desert plant communities. From the line-of-site visibility analysis , we calculated four metrics of visual quality: viewscape privacy, size, greenness, and ruggedness. 
It's a line-of-sight, so the number of people that you can see = the number of people that can see you. 
2.3 Line-of-Sight Viewscape Modeling
Here is the other paragraph. I was not sure about the terminology and missing little bit of context, so it definitely needs to be adjusted based on the text around.
Within our study area we created a regular grid of viewpoints with 30m spacing and computed viewsheds for each of these viewpoints using the 10-m DEM.The resulting 30-m resolution total viewshed was then derived by storing the number of visible cells from each view point rather than using raster algebra to add each individual viewshed one at a time (Llobera et al, 2010). Based on the computed viewsheds, we derived several variables for each viewpoint - viewshed size, the number of visible houses, greenness based on NDVI, and terrain ruggedness. By assigning each of the individual computed values on the grid, we created several 30-m resolution raster layers representing spatially continuous visual attributes of the landscape - visual prominence (Llobera 2003), privacy index, greenness, and terrain complexity, respectively.
Since deriving the total viewshed and other visual attributes for our study area required to compute 40000 individual viewsheds, we parallelized the computation in high-performance computing environment to reduce the required time from several months to several days. (I can add more precise numbers if needed). The individual viewshed computations were performed using GRASS GIS module r.viewshed with 3-m observer's height and while considering the curvature of the earth. Since the maximum visibility distance was set to 10 km in all directions, we sufficiently increased the extent of the DEM to avoid any edge effects.
Llobera, Marcos, David Wheatley, James Steele, Simon Cox, and Oz Parchment. "Calculating the inherent visual structure of a landscape (inherent viewshed) using high-throughput computing." Archaeolingua, 2010.
Llobera, Marcos. "Extending GIS-based visual analysis: the concept of visualscapes." International journal of geographical information science 17, no. 1 (2003): 25-48.

2.4 Statistical Analysis and Model Development

Using pairwise Wilcoxon rank sum tests, we compared differences in viewscape size and privacy between exurban,  suburban, rural, and randomly-located homes (built through 2010). Next, using autologistic regression, we tested which metrics of visual quality (viewscape privacy, size, greenness, and ruggedness) best predicted the probability of exurban development between 2010 and 2016. The autologistic form of ordinary logistic regression allowed our model to simultaneously account for expected spatial dependence in the housing data and avoid autocorrelated residuals (Besag, J., 1974, Gumpertz et al. 1997). We also considered the significance of a home's proximity to a primary road to account for the potential importance of accessibility in remote locations. Our analysis included the examination of all-possible models testing every subset of predictors and identifying the best model based on Akaike Information Citerion (AIC) (sensu Quinn & Keough 2002).

2.5 Model Application with spatially continuous, All-Possible Viewscapes 

For each 30-m pixel, computed the viewscape geometry and a value for each of the considered variables (privacy, greenness, terrain ruggedness, viewscape size, and primary roads). [Figure DD can show the number of neighbors visible from each pixel based on it's size and configuration]

3. Results

Our heads-up digitizing identified a total of 1,867 homes in 2010 (5X% exurban). Exurban homes built prior to 2010 have significantly larger viewscapes than randomly-distributed (Z times on average), suburban (X times on average), and rural (Y times on average) homes (Figure BBa). These same exurban homes were visible to significantly fewer neighbors compared to suburban homes, but more visible than rural homes. Exurban homes were also visible to XX fewer neighbors than what may be expected by chance (Figure BBb).
An additional 2XX homes were built between 2010 and 2016 with XX% located in low density exurban and rural settings. T-test show that exurban homes built after 2010 had significantly larger viewscapes (P < 0.00X) but possessed less privacy (P< 0.00X) as the region grew. The auto-logistic model of new growth showed that viewscape privacy (P = 0.003) and NDVI (P < 0.001) were the most significant predictors of the probability of exurban development between 2010 and 2016 (write equation here). Proximity to primary roads and the size and ruggedness of exurban viewscapes were not significant predictors after accounting for privacy and NDVI. Application of the model in the GIS produced a predictive map of the exurban development probability across the study region (Figure DD). 

4. Discussion

1. Exurban viewscapes are larger, but they see fewer neighbors 
Would expect random homes that are "pushed" to the edges of the study area to have larger viewscapes. Supports findings in Vukomanovic and Orr (2014) about the importance of visual scale as a visual quality metric. 
2. The fact that they see fewer neighbors 
scrap: Homes built through 2010 represent a period of first gradual (1970-XXXX) and then rapid (XXXX) growth, which virtually came to a stop with the great recession. This pre-2010 period of housing growth represents a wide-range of drivers, while XX% of the new homes after 2010 has been exurban.

Figures:

Figure AA: Study Area Map
Figure BB: Photos of exurban development
Figure CC: a) Viewscape Size by Density Class; b) Privacy by Density Class ) [do 3 pairwise Wilcoxon tests and include bar graph (include the statistics in the caption)]
Figure DD: Spatially continuous, all-possible viewscapes visualization: a) viewscape size, b) privacy, c) greenness, d) terrain ruggedness
Figure EE: Predictive map of the probability of exurban home development (INCLUDE EQUATION AGAIN SO FIGURE STANDS ON ITS OWN). Locations with the highest likelihood of exurban development occur in places that exhibit opportunities to A) view the scenic beauty of vegetated "green" landscapes....., B) build in highly private, low density settings, and c) the combination of green and private viewscapes.
Jelena To-Do:
-Figure BB and the wilcoxon tests
SCRAPS:
The region provides a unique opportunity for studying drivers of exurban viewscapes as the surrounding mountains provide scenic vistas with vertical visual boundaries that constrain residents’ views to the interior of the Plain and the inward-facing mountain slopes
REFERENCES
Gumpertz, M.L., Graham, J.M., Ristaino, J.B., 1997. Autologistic model of spatial pattern of phytophthora epidemic in bell pepper: effects of soil variables on disease presence. J. Agric. Biol. Environ. Stat. 2, 131–156.
Besag, J., 1974. Spatial interaction and the statistical analysis of lattice systems. J. R. Stat. Soc. B 36, 192–236.