优化同一sf对象中多要素的凸包生成效率

Great question! When dealing with thousands of SF features, for loops can get pretty slow—especially with repeated bind_rows() calls that copy your entire data frame every iteration. Let's refactor your workflow using dplyr grouping + purrr::pmap to make it vectorized and way more efficient. Here's how to adapt your existing logic:

Step 1: Load Data & Prep Grouped Structure #

First, we'll load your data as before, then organize it so each UID's original polygon and associated points are grouped together (this avoids repeated filtering in loops):

library(sf)
library(dplyr)
library(purrr)
library(concaveman)

# Load and prepare data (same as your original code)
download.file("https://drive.google.com/uc?export=download&id=1-I4F2NYvFWkNqy7ASFNxnyrwr_wT0lGF" , destfile="ProximityAreas.zip")
unzip("ProximityAreas.zip")
Proximity_Areas <- st_read("Proximity_Areas.gpkg") 
Proximity_Areas_Points <- Proximity_Areas %>% st_cast("POINT")

# Group data by UID: each row holds one UID's original polygon and its points
uid_groups <- Proximity_Areas %>%
  left_join(
    Proximity_Areas_Points %>% 
      group_by(UID) %>% 
      nest(points = geometry),  # Nest points into a list column per UID
    by = "UID"
  ) %>%
  select(UID, geometry, points)

Step 2: Encapsulate Processing Logic in a Function #

We'll turn your loop's step-by-step logic into a reusable function. This makes the code cleaner and easier to debug:

process_uid_group <- function(uid, original_poly, points) {
  # Generate convex hull (concavity=1 = strict convex hull)
  convex_hull <- st_make_valid(concaveman(points, concavity = 1, length_threshold = 0))
  
  # Clip convex hull to original polygon boundary
  clipped_convex <- st_intersection(convex_hull, original_poly) %>%
    mutate(UID = uid) %>%
    select(UID)
  
  # Handle geometry collections (extract and union polygons)
  if (st_geometry_type(clipped_convex) == "GEOMETRYCOLLECTION") {
    clipped_convex <- clipped_convex %>%
      st_collection_extract("POLYGON") %>%
      group_by(UID) %>%
      summarize(geometry = st_union(geometry)) %>%
      ungroup()
  }
  
  # Convert non-polygon geometries (points/lines) to polygons via buffer
  if (!(st_geometry_type(clipped_convex) %in% c("POLYGON", "MULTIPOLYGON"))) {
    end_cap_style <- ifelse(st_geometry_type(clipped_convex) == "POINT", "SQUARE", "FLAT")
    clipped_convex <- clipped_convex %>%
      st_buffer(dist = 12.5, endCapStyle = end_cap_style) %>%
      st_cast("MULTIPOLYGON") %>%
      st_intersection(original_poly) %>%
      select(UID)
  }
  
  # Ensure geometry is MULTIPOLYGON for consistency
  if (st_geometry_type(clipped_convex) == "POLYGON") {
    clipped_convex <- st_cast(clipped_convex, "MULTIPOLYGON")
  }
  
  # Fix any invalid geometries
  if (!st_is_valid(clipped_convex)) {
    clipped_convex <- st_make_valid(clipped_convex)
  }
  
  return(clipped_convex)
}

Step 3: Process All Groups with purrr::pmap #

Instead of looping through each UID manually, we'll use pmap to apply our function to every group in parallel (conceptually), then combine the results:

# Process all UIDs and combine results into a single SF object
Proximity_Convex <- uid_groups %>%
  mutate(convex_result = pmap(list(UID, geometry, points), process_uid_group)) %>%
  select(convex_result) %>%
  unnest(convex_result) %>%
  st_as_sf()

Why This Is Faster Than Your Original Loop #

• No repeated data copying: Your original loop uses bind_rows() every iteration, which creates a full copy of your growing data frame each time. For thousands of features, this is a massive performance drain.
• Optimized vectorized operations: dplyr::nest() and purrr::pmap() handle iteration in a way that minimizes R's for-loop overhead, making it much faster for large datasets.
• Cleaner maintainability: Encapsulating logic in a function makes it easier to tweak edge cases (like buffer sizes or geometry fixes) without rewriting the entire loop.

内容的提问来源于stack exchange，提问作者Chris