This repo is an effort at consoloditating the current raster tools available on R and producing further optimised revisions of such tools. Currently, manipulating and analysing large geospatial data in R is both a rocky and slow process if you don't have a particularly fast machine. Further, not all methods work as intended and often use sub-optimal solutions.
To illustrate, below are the results for classifying and extracting the xyz values of a relatively large raster. Test results compare functions written here and those found in the raster package. All testing was conducted on a system with an i5-7300u and 8gb RAM, with Manjaro XFCE as the operating system on linux.
The test data used here is a digital terrain model (DTM) of Plymouth and the surrounding areas, at a resolution of ~1m. This DTM is ~317.9 Mb in GeoTiff format and contains roughly 79 million points. It was constructed from the ascii LIDAR files: SX45ne, SX45nw, SX45se, SX45sw, SX46se, SX46sw, SX55nw, SX55sw, SX56sw, which can be downloaded here.
Here the the raster was classified to four thresholds: min, 2m, 5m and max.
library(raster)
DTM <- raster("DTM.tif")
# rgeo
microbenchmark::microbenchmark(classify(DTM, at = c(2,5)), times = 10L)
# raster package
microbenchmark::microbenchmark(raster::cut(DTM, breaks = c(-10.08, 2, 5, 126.170)), times = 10L)Unit: seconds
| expr | min | lq | mean | median | uq | max | neval |
|---|---|---|---|---|---|---|---|
| classify(DTM, at = c(2, 5)) | 3.039677 | 3.054341 | 3.08061 | 3.064573 | 3.070116 | 3.269607 | 10 |
| raster::cut(DTM, breaks = c(-10.08, 2, 5, 126.17)) | 3.851293 | 7.822391 | 7.560331 | 8.01715 | 8.06603 | 8.207009 | 10 |
Here the DTM was simply converted to an xyz dataframe.
library(raster)
DTM <- raster("DTM.tif")
# rgeo
microbenchmark::microbenchmark(extract_xyz(DTM), times = 10L)
# raster package
microbenchmark::microbenchmark(raster::rasterToPoints(DTM), times = 10L)Unit: seconds
| expr | min | lq | mean | median | uq | max | neval |
|---|---|---|---|---|---|---|---|
| extract_xyz(DTM) | 3.976426 | 4.1043 | 4.095051 | 4.106489 | 4.110322 | 4.116398 | 10 |
| raster::rasterToPoints(DTM) | 4.927814 | 12.77142 | 12.00122 | 12.78671 | 12.79453 | 12.80327 | 10 |
Both results show that a substantial reduction in time taken can be found when revising current available methods. Speed-ups of up to ~3x were found, with optimisations remaining pure R based and not offloading any computation onto a faster C language.
Please note that the methods included in this repo are S3 functions and only support single raster layers. The S4 methods found in the raster package are still the best choice for handling more complex raster objects.