Technical Application: LCLU Change in Amherst, New York

Abstract

This project examined the land cover and land use (LCLU) changes between 2010 – 2024 in Amherst, New York. Additionally, since Amherst hosts multiple academic campuses– such as the University at Buffalo’s (UB) North Campus, the study looked at any evidence of physical changes resulting from studentification. The study found that there was an increase in residential land use supporting the overall Town’s population growth. The study also identified the majority of apartment land use change occurring within a two-mile radius of the university campus, which supports the University’s increase student population along with the physical land use changes corresponding with the studentification process

Introduction

Urban growth, enhanced by increasing populations, has been a primary factor in vegetation to urban surface land cover change (Ducey et al., 2018; Nyiranshimiyimanan & Mupenzi, 2020). In towns that host academic institutions, commonly referred to as “college towns”, overall town population varies based on the academic calendar, resulting in major seasonal increase and decrease population shifts (Sage et al., 2012). College towns often face studentification, a process where the growing student population significantly affects town characteristics. The studentification process affects a range of social, cultural, economic, and physical transformations (Mohammed & Ukai, 2022; Smith, 2004). While an increase in student population does create some economic benefits, it often creates challenges for long-term residents (Jolivet et al., 2023; Mosey, 2017). An increase in student population often attracts private businesses, such as chain retailers and real estate developers, which increases urban cover and land use tailored towards students. This shift in businesses and land use can displace locally owned businesses and alter neighborhood characteristics (Savage, 2025; Zasina, 2021). The construction of student-oriented facilities also increases property values near the campus(s), making real estate unaffordable for non-student residents and increasing local residential feelings of exclusion from these areas (Baron & Diamant, 2016; Mosey, 2017; Rogaly & Taylor, 2015; Savage, 2025). Physically, increase student populations often lead to more off-campus student housing– resulting in greater noise and traffic during the academic year, while academic breaks can leave parts of the town feeling like “ghost towns” (Kinton et al., 2018; Sage et al., 2012). Despite these challenges, an increase in student population can bring benefits such as higher consumerism and a workforce, allowing students to contribute positively to the local economy during academic periods (Zasina, 2021). This project was conducted to examine land cover and land use changes between 2010 – 2024, in Amherst, New York, a college town. The study aims to assess whether these changes reflect the Town and the university’s, residing inside the boundary, population dynamics and if there is any evidence of studentification based on the land use changes.

Study Area

The project focused on the Town of Amherst (Amherst), located in Erie County, New York. With parts of SUNY University at Buffalo (UB) and Erie Community College residing within the Town’s boundary, Amherst is considered a college town (Town of Amherst, 2025). This project focuses on land cover and land use change across Amherst, and specifically near the UB’s North Campus area. According to the US Census, the Town of Amherst experienced a 6.84% increase in its total population from 2010 – 2023. The population grew from 120,945 in 2010 to 129,209 in 2023, resulting in an increase of 8,264 residents (U.S. Census Bureau, 2023).

Code 
library(tidycensus)
library(tidyverse)
library (ggplot2)
library(knitr)
library(kableExtra)

#2010 Total Population
amherst_pop_2010 <- get_acs(
  geography = "county subdivision",
  variables = c("B01003_001"),  # Total population
  state = "NY",
  year = 2010,
  survey = "acs5",
  cache_table = TRUE
)

amherst_pop_2010 <- amherst_pop_2010 %>% 
  filter(str_detect(NAME, "Amherst town, Erie County, New York"))

#amherst_pop_2010


#2015 Total Population
amherst_pop_2015 <- get_acs(
  geography = "county subdivision",
  variables = c("B01003_001"),  # Total population
  state = "NY",
  year = 2015,
  survey = "acs5",
  cache_table = TRUE
)

amherst_pop_2015 <- amherst_pop_2015%>% 
  filter(str_detect(NAME, "Amherst town, Erie County, New York"))

#amherst_pop_2015

#2020 Total Population
amherst_pop_2020 <- get_acs(
  geography = "county subdivision",
  variables = c("B01003_001"),  # Total population
  state = "NY",
  year = 2020,
  survey = "acs5",
  cache_table = TRUE
)

amherst_pop_2020 <- amherst_pop_2020%>% 
  filter(str_detect(NAME, "Amherst town, Erie County, New York"))

#amherst_pop_2015

#2023 Total Population
amherst_pop_2023 <- get_acs(
  geography = "county subdivision",
  variables = c("B01003_001"),  # Total population
  state = "NY",
  year = 2023,
  survey = "acs5",
  cache_table = TRUE
)

amherst_pop_2023 <- amherst_pop_2023%>% 
  filter(str_detect(NAME, "Amherst town, Erie County, New York"))

#amherst_pop_2023


amherst_trend <- tibble (
  year = c(2010, 2015, 2020, 2023),
  population = c(
    amherst_pop_2010$estimate,
    amherst_pop_2015$estimate,
    amherst_pop_2020$estimate,
    amherst_pop_2023$estimate
  )
)


ggplot(amherst_trend, aes(x = year, y = population)) +
  geom_line(color = "steelblue", size = 1.2) +
  geom_point(size = 3, color = "darkred") +
  scale_x_continuous(
    breaks = seq(2010, 2025, by = 5),  # Tick marks every 5 years
    limits = c(2010, 2025)             # Optional: constrain x-axis
  ) +
  labs(
    title = "Amherst, NY Population Trend (2010–2023)",
    x = "Year",
    y = "Population Estimate"
  ) +
  theme_minimal()

Code 
kable(amherst_trend, col.names = c('Year', 'Total Population'), align = 'c', caption = "Town of Amherst Total Population from 2010 - 2023") %>% 
  row_spec(row = 0, color = 'white', background = 'dodgerblue', font_size = 20) %>%
  footnote(symbol = 'Data gathered from US Census for respective years') %>%
  kable_styling(bootstrap_options = 'striped')
Town of Amherst Total Population from 2010 - 2023
Year Total Population
2010 120945
2015 124044
2020 125799
2023 130209
* Data gathered from US Census for respective years

From the University at Buffalo enrollment data, UB’s student enrollment rose by 2.9% from 2010 to 2015, followed by an 8.5% increase from 2015 – 2020. However, between 2020 – 2024, total enrollment declined by 1.4% (University at Buffalo, 2023)

Code 
UB_pop_2010 = c(19389, 9659, 29048)
UB_pop_2013 = c(19380, 10020, 29850)
UB_pop_2015 = c(19951, 9855, 29806)
UB_pop_2017 = c(21020, 9628, 30648)
UB_pop_2020 = c(22306, 10041, 32347)
UB_pop_2022 = c(20761, 11338, 32099)
UB_pop_2024 = c(20359, 11544, 31903)

UB_enrollment <- tibble(
  year = c(2010, 2013, 2015, 2017, 2020, 2022, 2024),
  undergrad = c(19389, 19380, 19951, 21020, 22306, 20761, 20359),
  grad = c(9659, 10020, 9855, 9628, 10041, 11338, 11544),
  total = c(29048, 29850, 29806, 30648, 32347, 32099, 31903)
)

kable(UB_enrollment, col.names = c('Year', 'Undergrad', 'Grad', 'Total'), align = 'c', caption = "University at Buffalo Total Student Enrollment") %>%
  row_spec(row = 0, color = 'white', background = 'dodgerblue', font_size = 20) %>%
  footnote(symbol = 'Data gathered from SUNY Univeristy at Buffalo') %>%
  kable_styling(bootstrap_options = 'striped')
University at Buffalo Total Student Enrollment
Year Undergrad Grad Total
2010 19389 9659 29048
2013 19380 10020 29850
2015 19951 9855 29806
2017 21020 9628 30648
2020 22306 10041 32347
2022 20761 11338 32099
2024 20359 11544 31903
* Data gathered from SUNY Univeristy at Buffalo
Code 
UB_long <- UB_enrollment %>%
  pivot_longer(cols = undergrad:total, names_to = "group", values_to = "population")


ggplot(UB_long, aes(x = year, y = population, color = group)) +
  geom_line(size = 1.2) +
  geom_point(size = 3) +
  scale_x_continuous(
    breaks = seq(2010, 2025, by = 5),
    limits = c(2010, 2025)
  ) +
  scale_y_continuous(labels = scales::comma) +
  labs(
    title = "University at Buffalo Enrollment Trends (2010–2024)",
    x = "Year",
    y = "Enrollment",
    color = "Population Type"
  ) +
  theme_minimal()

Code 
#rate of Change
calc_growth <- function(start, end) {
  ((end - start) / start) * 100
}

years <- UB_enrollment$year
ug <- UB_enrollment$undergrad
grad <- UB_enrollment$grad
tot <- UB_enrollment$total

#growth rates
growth_data <- tibble(
  Interval = c("2010–2015", "2015–2020", "2020–2024"),
  Undergrad = c(
    calc_growth(ug[1], ug[3]),  #10-15
    calc_growth(ug[3], ug[5]),  #15-20
    calc_growth(ug[5], ug[7])   #20-24
  ),
  Graduate = c(
    calc_growth(grad[1], grad[3]),
    calc_growth(grad[3], grad[5]),
    calc_growth(grad[5], grad[7])
  ),
  Total = c(
    calc_growth(tot[1], tot[3]),
    calc_growth(tot[3], tot[5]),
    calc_growth(tot[5], tot[7])
  )
)

growth_data %>%
  mutate(across(ends_with("_growth"), ~ round(., 2))) %>%
  knitr::kable(caption = "UB Enrollment Rates of Change (%)")
UB Enrollment Rates of Change (%)
Interval Undergrad Graduate Total
2010–2015 2.898551 2.029196 2.609474
2015–2020 11.803920 1.887367 8.525129
2020–2024 -8.728593 14.968629 -1.372616

Methods

Data

Landsat-5TM, Landsat-8 OLI, and Landsat-9 OLI-2 Surface Reflectance products (30 m²; Path 17, Row 30) were used for this project. Sentinel-2 OLI products (10 m²) were incorporated as the ground truth dataset for calculating Landsat land cover accuracy. Landsat products were sourced from USGS EarthExplorer and Sentinel products were sourced from the Copernicus Browser. All remote sensing products were acquired between August 1 to September 21 with the least visible cloud cover within the study site. Landsat products, used for classifications, were created in 2010, 2015, 2017, 2020, and 2024 while Sentinel products, used as ground truth samples, were created in 2017, 2020, and 2024. For the classification process, the blue (450-515 nm), green (525-605 nm), red (630-690 nm), and near-infrared (750-900 nm) bands were utilized (Landsat-5: B1, B2, B3, B4) (Landsat-8 and -9, and Sentinel-2: B2, B3, B4, B5).

Remote Sensing Imagery Preprocessing

Before the classification process begun, all remotely sensed images were reprojected to a common coordinate system: Universal Transverse Mercator (UTM) WGS 8436 using ArcGIS Pro. All remote sensing raster images were clipped to the Town of Amherst boundary. Areas where agricultural land use changed from bare soil to vegetation cover, due to harvesting, significantly affected multiple accuracy metrics. To address this issue, the images were further clipped to a bounding box of (easting 676515 – 687885, northing 4757295 – 4769175).

Image Classification

All images were classified using a supervised classification method in ArcGIS Pro. All Landsat and Sentinel images were initially classified into four land cover categories: Water, Building, Road, and Vegetation. Due to the large spatial resolution of Landsat products, the building and road classes were later merged into a single urban surface class to conduct a change analysis. Training samples were placed throughout the entire study area– with a minimum of 2,000 Landsat training pixels and 50,000 Sentinel training pixels. Due to the absence of accuracy assessments, higher quantities of training pixels were required for the 2010 and 2015 Landsat images with 9,776 and 5,686 total pixels, respectively. The 2017, 2020, and 2024 Landsat images were classified based on how well their results were during their accuracy assessments with the classified Sentinel images. After analyzing areas with inconsistent classifications, pixels classified as building or road were merged into one single urban surface class, to help the accuracy metrics using RStudio.

Training Polygons Utilizing ArcGIS Pro

Training Sample Pixels For Each Raster Image

Initial pixel classifications was water (0), buildings (1), roads (2), vegetation (3)

Accuracy Assessments

Accuracy assessments were performed with ArcGIS Pro on the classified 2017, 2020, and 2024 Landsat images, with the classified Sentinel 2017, 2020, and 2024 images used as the referenced ground truths. Four-class confusion matrices were created using randomly selected pixels within the study site. Each class had relatively similar randomly selected pixel counts within each year assessed. In 2017, 2020, and 2024, each class had between 400 – 450, 830 – 960, and 750 – 965 pixels, respectively. User Accuracy (UA), Producer Accuracy (PA), Overall Accuracy (OA), and Kappa Score were all derived from the resulting confusion matrices. After assessing the accuracy measurements, the building and road classes were combined into a single urban surface class, to address challenges arising from the 30m2 Landsat spatial resolution such as the poor edge detection in residential neighborhoods. The four-class confusion matrices were then updated to three-class confusion matrices and corresponding accuracies metrics were updated accordingly.

Initial four-class accuracy assessments

Three-class accuracy assessments

Land Use Change Analysis

Land use change analysis was conducted using 2010 and 2024 Landsat images and focused on the 2010 image pixels classified as vegetation and later reclassified in the 2024 image as urban surface. A local analysis was performed using a focal 5×5 pixel moving window to identify hotspots of vegetation-to-urban surface change. Hotspots were defined as areas where at least 15 out of the 25 neighboring pixels within the window indicated vegetation-to urban surface change. After each hotspot was identified, the land use was manually assigned using Google Maps. Land use analysis was carried out for the entire study area. To investigate any correlation between land use change and University at Buffalo’s student population dynamics, a two-mile buffer was placed around UB’s North Campus. Land use hotspots, within this two-mile buffer, were later compared to the overall land use classification and the land use located outside the buffer.

Load necessary packages

Code 
library(terra)
library(sf)
library(ggplot2)
library(leaflet)
library(sp)
library(raster)
library(dbscan)
library(dplyr)
library(osmdata)
library(ggspatial)
library(prettymapr)
library(kableExtra)

Import classified land raster files and Amherst, NY town boundary

Code 
r10 <- rast("GEO653_documents/Amherst_2010.tif") 
r15 <- rast("GEO653_documents/Amherst_2015.tif") 
r17 <- rast("GEO653_documents/Amherst_2017.tif") 
r20 <- rast("GEO653_documents/Amherst_2020.tif") 
r24 <- rast("GEO653_documents/Amherst_2024.tif") 

amherst <- vect("GEO653_documents/BB.shp")

Reproject variables and clean up the data

Code 
amherst <- project(amherst, crs(r17))
amherst_sf <- st_as_sf(amherst)

#crop boundaries
clip_raster <- function(r, boundary) 
                                          {mask (crop (r, boundary[2, ]), boundary)}

r10 <- clip_raster(r10, amherst)
r15 <- clip_raster(r15, amherst)
r17 <- clip_raster(r17, amherst)
r20 <- clip_raster(r20, amherst)
r24 <- clip_raster(r24, amherst)


rcl <- matrix(c(
  10, 1,
  20, 2,
  30, 3,
  60, 4
), ncol = 2, byrow = TRUE)



r10_reclass <- classify(r10, rcl)
r15_reclass <- classify(r15, rcl)
r17_reclass <- classify(r17, rcl)
r20_reclass <- classify(r20, rcl)
r24_reclass <- classify(r24, rcl)

r10_aligned <- r10_reclass
r15_aligned <- resample(r15_reclass, r10_reclass, method = "near")
r17_aligned <- resample(r17_reclass, r10_reclass, method = "near")
r20_aligned <- resample(r20_reclass, r10_reclass, method = "near")
r24_aligned <- resample(r24_reclass, r10_reclass, method = "near")

r10_aligned[r10_aligned == 255] <- NA
r15_aligned[r15_aligned == 255] <- NA
r17_aligned[r17_aligned == 255] <- NA
r20_aligned[r20_aligned == 255] <- NA
r24_aligned[r24_aligned == 255] <- NA

Plot to show areas that changed from vegetation to developed urban between different years

Code 
par(mfrow = c(1, 3), oma = c(0, 0, 4, 0)) 

veg_to_dev_10_15 <- (r10_aligned %in% c(0,3)) & (r15_aligned %in% c(1,2))
plot(veg_to_dev_10_15, main = "2010 - 2015")

veg_to_dev_15_17 <- (r15_aligned %in% c(0,3)) & (r17_aligned %in% c(1,2))
plot(veg_to_dev_15_17, main = "2015 - 2017")

veg_to_dev_17_20 <- (r17_aligned %in% c(0,3)) & (r20_aligned %in% c(1,2))
plot(veg_to_dev_17_20, main = "2017 - 2020")

mtext("Areas of Vegetation to Developed",
      outer = TRUE, cex = 1.5, line = 2)

The change analysis will use 2010-2024

Code 
par(mfrow = c(1, 1), oma = c(0, 0, 4, 0)) 
veg_to_dev_10_24 <- (r10_aligned %in% c(0,3)) & (r24_aligned %in% c(1,2))
plot(veg_to_dev_10_24, main = "2010 - 2024")

veg_to_dev_17_24 <- (r17_aligned %in% c(0,3)) & (r24_aligned %in% c(1,2))
#plot(veg_to_dev_17_24, main = "2017 - 2024")

veg_to_dev_20_24 <- (r20_aligned %in% c(0,3)) & (r24_aligned %in% c(1,2))
#plot(veg_to_dev_20_24, main = "2020 - 2024")

mtext("Areas of Vegetation to Developed",
      outer = TRUE, cex = 1.5, line = 2)

Find spatial clustering of land change (‘True’ cells)

Code 
moving_window <- matrix(1, 5, 5)
Code 
veg_loss_hotspots_10_24 <- focal(veg_to_dev_10_24, w = moving_window, fun = sum, na.policy = "omit")
plot(veg_loss_hotspots_10_24, main = "Vegetation to Development Hotspots 2010 - 2024", col = terrain.colors(170))

Code 
strong_hotspots_10_24<- veg_loss_hotspots_10_24 > 15
plot(strong_hotspots_10_24, main = "Strong Vegetation Loss Hotspots 2010 - 2024", col = c("white", "red"))

Code 
n_hotspot_pixels_10_24 <- global(strong_hotspots_10_24, sum, na.rm = TRUE)

hotspot_area_m2_10_24 <- n_hotspot_pixels_10_24 * 900

cat(sprintf("There is a total of %.2f m² of hotspots identified.\n", hotspot_area_m2_10_24))
There is a total of 953100.00 m² of hotspots identified.

After identifying the hotspot areas, convert the raster pixels into polygons and calculate the centroid of each polygon. Afterwards, determine the land use type for each centroid using OpenStreetMap data.

Code 
#convert raster hotspots to polygons
hotspot_polygons <- as.polygons(strong_hotspots_10_24, dissolve = FALSE)
hotspot_polygons <- hotspot_polygons[hotspot_polygons[[1]] == 1, ]

#find centroids and coords to polygons
centroids <- centroids(hotspot_polygons)
coords <- crds(centroids)

#find clustering of centroids 
eps_value <- 50  # adjust as needed
clust <- dbscan(coords, eps = eps_value, minPts = 1)
centroids$cluster <- clust$cluster

attr_df <- as.data.frame(centroids)

#make grouped centroids into one single centroid by average (x,y)
centroids_df <- cbind(attr_df, x = coords[,1], y = coords[,2])

cluster_summary <- centroids_df %>%
  group_by(cluster) %>%
  summarize(
    mean_x = mean(x),
    mean_y = mean(y),
    count = n()
  )


cluster_centroids <- vect(cluster_summary[, c("mean_x", "mean_y")], 
                         geom = c("mean_x", "mean_y"), crs = crs(centroids))
#areas of hotspots
plot(cluster_centroids, main = 'Identified Centroids')

Code 
clustered_centroids_sf <- st_as_sf(cluster_centroids)
clustered_centroids_wgs84 <- st_transform(clustered_centroids_sf, 4326)

buffer_radius <- 0.00025  # ~25 meters in lat/lon degrees
centroid_buffers <- st_buffer(clustered_centroids_wgs84, dist = buffer_radius)

global_bbox <- st_bbox(centroid_buffers)

osm_key <- "landuse"

query <- opq(bbox = global_bbox) %>%
  add_osm_feature(key = osm_key)

osm_data <- tryCatch(
  osmdata_sf(query),
  error = function(e) NULL
)

if (!is.null(osm_data) && nrow(osm_data$osm_polygons) > 0) {
  osm_landuse <- osm_data$osm_polygons
  

  overlaps <- st_join(centroid_buffers, osm_landuse, join = st_intersects, left = FALSE)
  
  if (nrow(overlaps) > 0) {
    
    cat("Matched landuse types:\n")
    print(table(overlaps$landuse))

   
    plot(st_geometry(centroid_buffers), col = NA, border = "gray", main = "Land Use Around Clustered Centroids")
    plot(st_geometry(osm_landuse), col = "lightgreen", border = "green", add = TRUE)
    plot(st_geometry(clustered_centroids_wgs84), col = "black", pch = 19, add = TRUE)

  } else {
    cat("no landuse polygons overlapped with point buffer.")
  }
} else {
  cat("no landuse data returned from OSM.")
}
Matched landuse types:

  brownfield construction       meadow  residential 
           1            3            1           13

Code 
#buffer the centroids to find the various land use around the area
centroid_buffers$id <- seq_len(nrow(centroid_buffers))

landuse_matches <- st_join(centroid_buffers, osm_landuse, join = st_intersects, left = FALSE)

#find the most dominant land use and assign to the centroid
dominant_landuse <- landuse_matches %>%
  st_drop_geometry() %>%
  group_by(id, landuse) %>%
  summarize(count = n(), .groups = "drop") %>%
  group_by(id) %>%
  slice_max(order_by = count, n = 1, with_ties = FALSE)


centroid_with_landuse <- left_join(centroid_buffers, dominant_landuse, by = "id")
centroid_points <- st_centroid(centroid_with_landuse)

centroid_points <- centroid_points %>%
  mutate(
    landuse = ifelse(is.na(landuse), "not found", landuse),
    count = 1)

ggplot() +
  annotation_map_tile(type = "osm") +
  geom_sf(data = centroid_points, aes(color = landuse), size = 3, shape = 19) +
  scale_color_viridis_d(option = "plasma", name = "Dominant Landuse") +
  theme_minimal() +
  labs(
    title = "Centroids Colored by Dominant Landuse"
  ) +
  theme(legend.position = "right")

Identifying land use change using OSM data creates a large amount of uncertainty due to the data manipulation required to convert 10m raster cells into polygons and points. Therefore, it was decided for this project to use physical analysis to determine land use within the hotspot regions. The original hotspot cells were overlaid onto a Leaflet map to create a visual identification of land use for these hotpot areas.

Code 
#2010 - 2024 Hotspot 
strong_hotspots_proj_10_24 <- project(strong_hotspots_10_24, "EPSG:3857")

#Convert True/False to (1/0)

strong_hotspots_proj_10_24_numeric <- as.numeric(strong_hotspots_proj_10_24)
strong_hotspots_proj_10_24_numeric_no_zeros <- mask(strong_hotspots_proj_10_24_numeric, strong_hotspots_proj_10_24_numeric, maskvalues = c(0, NA))

pal_hotspots <- colorNumeric(
  palette = c("transparent", "red"),
  domain = c(0, 1)
)

#Leafet map with osm to identify land use of hotspot areas
leaflet() %>%
  addProviderTiles(providers$OpenStreetMap) %>%
  addRasterImage(
    strong_hotspots_proj_10_24_numeric_no_zeros,
    colors = pal_hotspots,
    opacity = 0.50,  # Show hotspots with full opacity
    project = FALSE
  ) %>%
  addLegend(
    position = "bottomright",
    pal = pal_hotspots,
    values = c(0, 1),
    title = "Vegetation Loss Hotspots (2010-2024)",
    labels = c("No Hotspot", "Hotspot")
  )
Code 
hotspots <- read_sf("GEO653_documents/Hotspot_Points.shp")
UB <- read_sf("GEO653_documents/UB.shp")
amherst <- read_sf("GEO653_documents/BB.shp")


num_hotspots <- length(hotspots)


df <- as.data.frame(hotspots)

df <- df %>%
  mutate(Name = ifelse(Name == "Commerical", "Commercial", Name)) %>%
  mutate(Name = ifelse(Name == "Apartment", "Apartments", Name))%>%
  mutate(Name = ifelse(Name == "Residental", "Residential", Name)) %>%
  group_by(Name) %>%
  summarize(Count = n())

hotspots <- hotspots %>%
  mutate(Name = ifelse(Name == "Commerical", "Commercial", Name)) %>%
  mutate(Name = ifelse(Name == "Apartment", "Apartments", Name)) %>%
  mutate(Name = ifelse(Name == "Residental", "Residential", Name))

kable(df, caption = " 2024 Hotspot Land Use")%>% 
  kable_styling(bootstrap_options = 'striped')
2024 Hotspot Land Use
Name Count
Apartments 8
Commercial 2
False 2
Hospital 1
In Progress 1
Misc. 1
Parking Lot 2
Residential 11
Solar 1
Code 
amherst$Source <- "Amherst"
UB$Source <- "UB"
hotspots$Source <- "Hotspots"

ub_utm <- st_transform(UB, crs = 32618)
buffer_2mile <- st_buffer(ub_utm, dist = 3218.68)

ggplot() +
  geom_sf(data = amherst, aes(fill = Source), color = "black", alpha = 0.3) +
  geom_sf(data = UB, aes(shape = Source), color = "blue", size = 4) +  # UB points
  geom_sf(data = hotspots, aes(color = Name), size = 2, alpha = 0.8) +
  geom_sf(data = buffer_2mile, fill = "lightblue", alpha = 0.5) +
  scale_shape_manual(values = c("UB" = 16)) + 
  scale_fill_manual(values = c("Amherst" = "lightgray")) +
  scale_color_manual(values = c(
    "Residential" = "red",
    "Apartments" = "purple",
    "Commercial" = "orange",
    "Hospital" = "green",
    "Parking Lot" = "brown",
    "In Progress" = "pink",
    "Solar" = "black",
    'Other' = 'gray'
  )) +
  theme_minimal() +
  labs(
    title = "Hotspots in Amherst and UB Area",
    fill = "Polygon Layer",
    color = "Hotspot Type",
    
    shape = "UB Symbol"
  )

Code 
hotspots_utm <- st_transform(hotspots, crs = st_crs(buffer_2mile))
hotspots_within_buffer <- st_intersection(hotspots_utm, buffer_2mile)

df2 <- as.data.frame(hotspots_within_buffer)

df2 <- df2 %>%
  group_by(Name) %>%
  summarize(Count = n())

kable(df2, caption = "2024 Hotspot Land Use in 2 Mile Buffer of University at Buffalo")%>% 
  kable_styling(bootstrap_options = 'striped')
2024 Hotspot Land Use in 2 Mile Buffer of University at Buffalo
Name Count
Apartments 7
Commercial 1
False 2
Hospital 1
In Progress 1
Misc. 1
Parking Lot 2
Residential 1

Results

Land Use Change in Amherst, New York

Land use analysis, conducted between the 2010 and 2024 Landsat images, identified 29 hotspots, within the entire study region, where the vegetation class transformed into the urban surface class. Of these 29 hotspots, two hotspots were determined to be false classifications, which was likely due to the 15-pixel threshold used to define a hotspot. Among the 27 true vegetation-to-urban surface hotspots, the main land use types were classified as residential and apartments.

The increase in residential and apartment developments aligns with population growth in Amherst between 2010 and 2023. The observed increase in housing land use positively correlates with Amherst’s increasing population.

Land Use Change Near University at Buffalo’s North Campus

A proximity analysis was conducted after applying a two-mile buffer around the University at Buffalo’s North Campus. Of the 27 vegetation-to-urban surface hotspots identified within the study area, 14 hotspots were located within the two-mile buffer zone. Most of these hotspots, inside the buffer, were classified as apartments (7 out of 8 total apartment hotspots), suggesting that the apartment developments, within the buffer, are likely oriented towards university students. In contrast, the majority of residential hotspots (10 out of 11) were located outside the buffer, suggesting that these areas are oriented towards local residents. The addition of the seven student-oriented apartments, near the campus, reflects the physical changes associated with studentification (Smith, 2004). As more developments are catered towards the seasonal student population, it is likely that these areas experience fluctuations in activity- becoming quieter and empty during academic breaks and busier with increased traffic and noise during the academic season. Additionally, University at Buffalo’s enrollment data, from 2010 – 2024, showed an increase in student enrollments between 2010 – 2020, which may have contributed to the rise in apartment buildings. However, the enrollment decline from 2020 – 2024, could further enhance the sense of emptiness in these student-focused areas during academic breaks (Kinton et al., 2018; Sage et al., 2012).

Conclusion

Attempts to use OpenStreetMap (OSM) data and land-change centroids to identify areas of land-use change were not successful. This may be due to the high spatial specificity of OSM’s vector data, which limited its compatibility with raster-based change detection. Other approaches could be explored to improve detection accuracy. For example, automated land-use classification methods developed in previous studies have shown strong potential for detecting urban change (Hu & Wang, 2013).

A total of 27 vegetation-to-urban surface hotspots were identified over the study period. The major hotspots were classified as residential neighborhoods and apartment buildings. Residential hotspots were primarily located outside the two-mile radius from the University at Buffalo’s North Campus, suggesting they are oriented toward local residents and reflect the town’s overall population growth, as indicated by U.S. Census data.

In contrast, most apartment hotspots were concentrated within the two-mile buffer zone surrounding the campus (likely serving the student population). The increase in off-campus student housing indicates signs of studentification, particularly in terms of physical land change. Based on findings from other studentification studies, these student-oriented neighborhoods may experience seasonal fluctuations, making areas appearing quiet during academic breaks and significantly busier, with increased traffic and noise, during the academic year.

References

Baron, M. G., & Diamant, E. R. (2016). Real estate in studentified neighborhoods. ERSA Conference Papers, ersa16p642.

Ducey, M. J., Johnson, K. M., Belair, E. P., & Cook, B. D. (2018). The influence of human demography on land cover change in the Great Lakes States, USA. Environmental Management, 62, 1089–1107.

Hu, S., & Wang, L. (2013). Automated urban land-use classification with remote sensing. International Journal of Remote Sensing, 34(3), 790–803. https://doi.org/10.1080/01431161.2012.714510

Jolivet, V., Reiser, C., Baumann, Y., & Gonzalès, R. (2023). Before displacement: Studentification, campus-led gentrification and rental market transformation in a multiethnic neighborhood (Parc-Extension, Montréal). Urban Geography, 44(5), 983–1002.

Kinton, C., Smith, D. P., Harrison, J., & Culora, A. (2018). New frontiers of studentification: The commodification of student housing as a driver of urban change. The Geographical Journal, 184(3), 242–254.

Mohammed, A. M., & Ukai, T. (2022). University campuses as agents for urban change. Environmental & Socio-Economic Studies, 10(1), 22–37.

Mosey, M. (2017). Studentification: The impact on residents of an English city. GEOverse, 75.

Nyiranshimiyimanan, S., & Mupenzi, C. (2020). Impacts of Population Growth on Land Cover Change. Case of Nyanza District. East African Journal of Science and Technology, 11(11).

Rogaly, B., & Taylor, B. (2015). For the likes of us? Retelling the classed production of a British university campus. ACME: An International Journal for Critical Geographies, 14(1), 235–259.

Sage, J., Smith, D., & Hubbard, P. (2012). The rapidity of studentification and population change: There goes the (student) hood. Population, Space and Place, 18(5), 597–613.

Savage, C. (2025). Studentification Unveiled: Navigating Socioeconomic Shifts, Housing Pressures, and Community Cohesion in Aging Communities [PhD Thesis].

Smith, D. P. (2004). ‘Studentification’: The gentrification factory? In Gentrification in a global context (pp. 73–90). Routledge.

Town of Amherst. (2025). Community Profile. https://www.amherst.ny.us/content/departments.php?dept_id=dept_09&div_id=div_11&menu_id=menu_03

University at Buffalo. (2023). Student Enrollment [Dataset]. https://www.buffalo.edu/oia/facts-publications/factbook/student/enrollment.html

U.S. Census Bureau. (2023). Amherst town, Erie County, New York [Dataset]. https://www.census.gov/quickfacts/fact/table/amhersttowneriecountynewyork/PST045224

Zasina, J. (2021). The student urban leisure sector: Towards commercial studentification? Local Economy, 36(5), 374–390.