Introduction

(About this document: This is an R notebook. As such, it is very informal and contains lots of typos and poorly constructed fragment sentences. It’s purpose is to generate converstation among staff. It’s not suitable for publishing in the Daily Camera.)

Using crowdsourced geolocation data (hereafter, cell phone data) provided by Unacast, we can explore patterns of travel on OSMP properties. For this report, we focus on two Habitat Conservation Areas (Longs Canyon and Goshawk) to see if we can detect off trail travel, and whether this “signal” changes between two different years (2019 and 2020). The data we have available is for the month of June for each year.

Load the data, count the # of records for each year and area

library(ggmap)
library(dplyr)
library(raster)
library(pander)
options(stringsAsFactors = FALSE)

setwd('C:/Users/anacb1/City of Boulder/Science Officer - Documents/Projects/2020_Unacast/Data/')
d <- read.csv('DerivedCsvs/DeviceLevelLongGosh.csv')

pander(addmargins(table(d$area, d$year)))

About 25k records in the dataset for these two areas and years. Each record is the location of a single device. There are way more records (120% increase) in 2020 than 2019. Note, for % change, 100% increase equals a doubling; in our case, 120% increase is the same as saying a 2.2 fold increase. We’ll use % increase from here on out.

In comparison, when we compare the year-over-year change in a larger dataset (i.e., an area that includes all of our trails and BCPOS’s trails), we see an increase from 3.7 million records in 2019 to 5.4 million records in 2020, for a 46% increase.

Thus, we have a backround rate of increase of 46% for the full dataset, compared to our local rate of increase of 120%.

To put it another way, if we take the 1651 records observed in Goshawk in 2019, we’d expect this to grow to 2410 in 2020 (1651 * 1.46), just following the background rate of increase in the larger dataset. Instead, we see 3498 records, above and beyond our background expectation.

What does that all mean? Well, it’s tempting to think that this data is in fact showing us a “real” increase in visitation to these areas in 2020 vs 2019.

Map 1. Cell phone records by area and year

bbox <- make_bbox(lon = d$centroid_lon, lat = d$centroid_lat, f = .1)
map <- get_map(location = bbox)

ggmap(map) + 
  geom_point(data = d, aes(x = centroid_lon, y = centroid_lat, col = factor(year)), size = 0.5) +
  facet_wrap(~year) +
  theme(legend.position = "none") +
  scale_color_manual(values=c('black', 'red'))

Even at this coarse resolution, you can see a much higher density of records in 2020 than 2019 for both areas. You can zoom in on this data here: ArcGIS Online Web Map.

Map 2. Cell phone records at Goshawk

At this scale, we can see some patterns:

1. Again, more points in 2020 (red) than 2019 (black)
2. Lots of records along the road and the trailheads, relative to the trails
3. Pretty clear rendering of where our trails are (i.e., lots of records along trails)
4. Lots of records in the middle of the loop trails, especially red dots, potentially indicating off trail travel.