如何加速大分辨率栅格间重叠对应单元格的提取操作

Your current approach is slow because it loops through every low-resolution cell individually, creating and projecting polygons one at a time. Instead, use vectorized operations in terra to process all low-resolution cells in bulk:

Step-by-Step Optimized Code #

library(terra)

# Create example rasters (same as your original)
highRes <- rast()
res(highRes) <- c(0.1, 0.1)
highRes[] <- sample(1:10, ncell(highRes), replace = TRUE)
lowRes <- rast(res = c(1, 1), xmin = -170, xmax = -50, ymin = 0, ymax = 80)
lowRes[] <- 1
lowRes <- shift(lowRes, dx = 0.02, dy = 0.03) # Not aligned

# 1. Ensure both rasters share the same CRS (project lowRes if needed)
lowRes_proj <- project(lowRes, crs(highRes))

# 2. Convert entire low-res raster to individual cell polygons (no dissolution)
low_polys <- as.polygons(lowRes_proj, dissolve = FALSE)
# Add a unique ID for each low-res cell to track results
low_polys$cell_id <- seq_len(nrow(low_polys))

# 3. Extract high-res values with exact coverage weights in one bulk operation
result <- extract(highRes, low_polys, weights = TRUE, exact = TRUE)

# Result structure:
# - ID: Matches cell_id in low_polys (low-res cell identifier)
# - value: Value from high-res cell
# - weight: Fraction of the high-res cell covered by the low-res cell

Why This Is Faster #

• No loops: Avoids the overhead of iterating through each low-res cell in R, which is slow for large datasets.
• Bulk processing: terra handles polygon intersections and value extraction in optimized C++ code, which is far more efficient than per-cell operations.
• Minimized redundant operations: Converts all low-res cells to polygons in one step instead of creating a new raster/polygon for each cell.

Additional Optimizations #

• Memory management: For extremely large global rasters, consider processing in chunks using terra::spatSample or splitting the low-res raster into regions if memory is constrained.
• File formats: Store rasters in efficient formats like GTiff with LZW compression to reduce I/O time when reading from disk.
• CRS alignment: If your rasters already share the same CRS, skip the project step to save time.

内容的提问来源于stack exchange，提问作者Pascal