Preparation of input data: geomorphological analysis with whitebox

Alban de Lavenne & Antoine Casquin

2026-07-24

The hydrological modelling of the transfR package is based on a geomorphological analysis of the studied catchments. In this vignette, we give some guidance on how to perform this geomorphological analysis. More specifically, we extract the catchment delineation and hydraulic length maps from a digital elevation model (DEM). This analysis is one of the two inputs needed (together with the time series of flow observations) to build a transfR object and start using the transfR package (see the Get started with transfR vignette).

Hydraulic length is defined as the distance within the river network along an identified flow path to the outlet. It can be extracted from a DEM in many different ways, such as with the GRASS toolkits (Jasiewicz and Metz 2011), Whitebox GAT (see J. B. Lindsay (2016) or WhiteboxTools), TauDEM (D. Tarboton, Utah State University) or online services (Squividant et al. (2015) for catchment delineation only). This vignette presents one possible workflow by making use of two main R packages:

Functions in the whitebox package generally work with files on disk rather than objects in R memory. A working directory should therefore be defined for these files.

wbt_wd <- tempdir(check = TRUE)

1. Retrieve elevation data with elevatr

The elevatr package retrieves grids (rasters) of elevation data worldwide for various zoom levels and data sources. It uses the Open Topography Global Datasets API to access Shuttle Radar Topography Mission (SRTM) data.

A projected coordinate reference system must be assigned to the catchment delineation and hydraulic length maps used with the whitebox package. We use the Lambert-93 projection, the official projection for maps of metropolitan France, whose EPSG code is 2154.

library(elevatr)
library(progress) # Needed by elevatr

# Set up a projection (French Lambert-93 projection)
EPSG <- 2154 

# Define a bbox that will encompass the catchments of the study area
blavet_bbox <- st_bbox(c(xmin = -3.3, xmax = -2.7, ymax = 48.11, ymin = 47.77), 
                           crs = st_crs(4326))
blavet_loc <- st_as_sfc(blavet_bbox) |> st_sf()

# Retrieve elevation data as raster
dem_raw <- elevatr::get_elev_raster(blavet_loc, z = 10) # ~76 m resolution

# Project and define spatial resolution: 
dem_100m  <- st_warp(st_as_stars(dem_raw), cellsize = 100, crs = st_crs(EPSG))
names(dem_100m) <- "warp"

# Set negative values (ocean) to NA
dem_100m[dem_100m < 0] <- NA

# Write to file
write_stars(dem_100m["warp"], file.path(wbt_wd,"dem_100m.tif"))

2. Retrieve a known river network and burn it into the DEM (optional)

The hydrological modelling distinguishes hillslopes from the river network. They have very different transfer dynamics, and the transfR package aims to describe only the transfer function of the river network. Defining where this river network begins and which flow path it takes is a key, non-trivial issue for hydrogeomorphologists. The easiest way to draw a drainage network from a DEM is usually to define a minimum drainage area threshold, above which the drainage network is assumed to start. However, defining this threshold can be difficult because it may vary spatially, especially with the geology of the region. It may therefore be better to use a known river network and force the flow paths to follow it. Here we use the French TOPAGE river network as a reference (see the description of the Blavet dataset to download it from the “Sandre - Eau France” Web Feature Service (WFS)). We implement a stream-burning technique in the DEM with whitebox::wbt_burn_streams_at_roads(), following John B. Lindsay (2016).

library(transfR)
library(whitebox)

# Get the French Topage river network from the Blavet dataset
data(Blavet)
CoursEau_Topage2019 <- Blavet$network

# Change projection and write files
network_topage <- st_transform(CoursEau_Topage2019, EPSG)
st_write(network_topage, file.path(wbt_wd, "network_topage.shp"), 
             delete_layer = TRUE, quiet = TRUE)
whitebox::wbt_rasterize_streams("network_topage.shp", 
                                base = "dem_100m.tif", 
                                output = "network_topage.tif", 
                                nodata = 0, 
                                wd = wbt_wd)

# Burn this river network on the DEM
# We will neglect the effect of the road embankments at this DEM resolution of 100m 
# by creating an empty shapefile for roads
st_write(st_sfc(st_multilinestring(),crs = EPSG), file.path(wbt_wd,"roads.shp"), 
             delete_layer = TRUE, quiet = TRUE)
whitebox::wbt_burn_streams_at_roads(dem = "dem_100m.tif", 
                        streams = "network_topage.shp", 
                        roads = "roads.shp",
                        output = "dem_100m_burn.tif", 
                        wd = wbt_wd)

3. Delineate catchments from their outlet coordinates

The whitebox package provides all the tools required for a standard catchment delineation workflow from a DEM and outlet coordinates. It consists of six steps:

# Remove the depressions on the DEM
whitebox::wbt_fill_depressions(dem = "dem_100m_burn.tif",
                               output = "dem_fill.tif",
                               wd = wbt_wd)

# Flow direction raster
whitebox::wbt_d8_pointer(dem = "dem_fill.tif",
                         output = "d8.tif",
                         wd = wbt_wd)

# Compute flow accumulation
whitebox::wbt_d8_flow_accumulation(input = "d8.tif",
                                   pntr = TRUE,
                                   output ="facc.tif",
                                   wd = wbt_wd)

# Extract a stream network (threshold = 1 km^2) consistent with flow direction
whitebox::wbt_extract_streams(flow_accum =  "facc.tif",
                              threshold = 100, # 100 cells for 1 km^2
                              output = "network_1km2.tif",
                              zero_background = TRUE,
                              wd = wbt_wd)
whitebox::wbt_remove_short_streams(d8_pntr = "d8.tif",
                                   streams = "network_1km2.tif",
                                   output = "network_d8.tif",
                                   min_length= 200,
                                   wd = wbt_wd)

Outlet coordinates are retrieved from hydro.eaufrance.fr and snapped to the river-network pixel that is consistent with the previously defined flow directions.

# Locate the outlets of the studied catchments (with manual adjustments to facilitate snapping)
outlets_coordinates <- data.frame(id = names(Blavet$hl),
                             X = c(254010.612,255940-100,255903,237201,273672,265550),
                             Y = c(6772515.474,6776418-200,6776495,6774304-200,6762681,6783313))
outlets <- st_as_sf(outlets_coordinates, coords = c("X", "Y"), crs=2154)
st_write(outlets, dsn = file.path(wbt_wd, "outlets.shp"), 
             delete_layer = TRUE, quiet = TRUE)

# Snap the outlets on the stream raster
whitebox::wbt_jenson_snap_pour_points(pour_pts = "outlets.shp",
                                      streams = "network_d8.tif",
                                      output = "outlets_snapped.shp",
                                      snap_dist = 200,
                                      wd = wbt_wd) 
outlets_snapped <- st_read(file.path(wbt_wd, "outlets_snapped.shp"), quiet = TRUE)

# Delineate catchments 
catchments <- st_sf(id = character(), geometry = st_sfc(crs = EPSG))
for(id in outlets_snapped$id){
  st_write(outlets_snapped[outlets_snapped$id==id,],
               file.path(wbt_wd, paste0(id, "_outlet.shp")), delete_layer = TRUE, quiet = TRUE)
  whitebox::wbt_watershed(d8_pntr = "d8.tif",
                          pour_pts = paste0(id, "_outlet.shp"),
                          output = paste0(id, "_catchment.tif"),
                          wd = wbt_wd)
  # Vectorize catchments
  drainage <- read_stars(file.path(wbt_wd, paste0(id, "_catchment.tif")))
  drainage_polygon <- st_as_sf(drainage, as_points = FALSE, merge = TRUE, na.rm = TRUE)
  catchment_geometry <- st_union(st_geometry(drainage_polygon)) |> st_cast("POLYGON")
  # Remove isolated raster fragments and retain the main catchment.
  catchment_geometry <- catchment_geometry[which.max(st_area(catchment_geometry))]
  catchments <- rbind(catchments, st_sf(id = id, geometry = catchment_geometry))
}

The resulting catchment delineations can be plotted and checked.

# Compare drainage areas to the dataset provided with transfR
compare_areas <- data.frame(
  name = names(Blavet$hl),
  expected_area =  st_area(st_geometry(Blavet$obs)) |> units::set_units("km^2") |> round(1),
  computed_area = st_area(catchments) |> units::set_units("km^2") |> round(1)
)

print(compare_areas)
#>            name expected_area computed_area
#> 1      J5613010  314.8 [km^2]  314.8 [km^2]
#> 2      J5618310   15.5 [km^2]   15.6 [km^2]
#> 3      J5618320    6.8 [km^2]    6.7 [km^2]
#> 4      J5704810   45.3 [km^2]   45.8 [km^2]
#> 5      J8433020  134.5 [km^2]  134.6 [km^2]
#> 6 AgrHys_Naizin    4.9 [km^2]    5.2 [km^2]
# Plot catchment delineation
par(oma = c(0, 0, 0, 6))
plot(catchments[,"id"], main = "Catchment delineations", key.pos = 4, reset = FALSE)
plot(st_geometry(st_intersection(network_topage,catchments)), 
     col = "white", lwd = 1.5, add = TRUE)
plot(outlets_snapped, col = "black", pch = 16, add = TRUE)

4. Compute flow path length and hydraulic length for each catchment

The whitebox package does not provide a function to compute hydraulic length directly. However, it can compute the total downslope flow path length to the outlet using the whitebox::wbt_downslope_flowpath_length() function and the downslope distance to the stream using the whitebox::wbt_downslope_distance_to_stream() function. Hydraulic length does not include the flow path over hillslopes, so it can simply be calculated as the difference between these two flow distances.

Flow path lengths are computed once across the entire study area. For each catchment, this raster is then trimmed (i.e. NA values outside the catchment bounding box are removed), and the flow path lengths are adjusted so that the minimum is equal to half the pixel width or height. Finally, the hydraulic lengths are gathered in a list as expected by the as_transfr() function.

# Compute hydraulic length
whitebox::wbt_downslope_flowpath_length(d8_pntr = "d8.tif",
                                        output = "fpl.tif",
                                        wd = wbt_wd)
whitebox::wbt_downslope_distance_to_stream(dem = "dem_fill.tif",
                                           streams = "network_topage.tif",
                                           output = "d2s.tif",
                                           wd = wbt_wd)
fpl <- read_stars(file.path(wbt_wd, "fpl.tif"))
d2s <- read_stars(file.path(wbt_wd, "d2s.tif"))
hl_region <- fpl-d2s
names(hl_region) <- "hl"

# Crop hydraulic length for each catchment
hl <- list()
for(id in catchments$id){
  crop <- st_crop(hl_region, catchments[catchments$id == id, ])
  crop <- crop-min(crop$hl, na.rm = TRUE)
  crop$hl <- units::set_units(crop$hl, "m")
  hl[[id]] <- crop
}

The resulting hydraulic length maps can be plotted and checked.

i <- 1
network <- st_geometry(st_intersection(network_topage, catchments[i, ]))
plot(hl[[i]], main = paste("Hydraulic length of catchment", i,"[m]"), 
     col = hcl.colors(20, palette = "Teal"), key.pos = 1, reset = FALSE)
plot(network, col = "white", lwd = 1.5, add = TRUE)

5. Creating a transfR object and running a simulation

Catchment delineations can be used as the spatial dimension of stars objects to georeference the observed flow time series of gauged catchments and locate ungauged catchments (see vignette Preparation of input data: creation of a stars object for details). Here we will create a stars object using the observed discharge of the Blavet dataset and the delineations that we just computed with whitebox.

obs_st <- st_as_stars(list(Qobs = Blavet$obs$Qobs), 
                            dimensions = st_dimensions(
                              time = st_get_dimension_values(Blavet$obs,1),
                              space = st_geometry(catchments)))

The hydraulic length maps can then be used to create an object of class transfR with the as_transfr() function (argument hl) and to perform simulations.

obs <- as_transfr(st = obs_st, hl = hl)

Hydrographs can then be transferred from the gauged catchments to the ungauged catchments using the quick_transfr() function.

obs <- quick_transfr(obs, velocity = "brittany2013", cv = TRUE, parallel = TRUE, cores = 2)

The simulated time series will be available in its stars object as new attributes.

obs$st

References

Jasiewicz, JarosŁaw, and Markus Metz. 2011. “A New GRASS GIS Toolkit for Hortonian Analysis of Drainage Networks.” Computers & Geosciences 37 (8): 1162–73. https://doi.org/10.1016/j.cageo.2011.03.003.
Lindsay, J. B. 2016. “Whitebox GAT: A Case Study in Geomorphometric Analysis.” Computers & Geosciences 95 (October): 75–84. https://doi.org/10.1016/j.cageo.2016.07.003.
Lindsay, John B. 2016. “The Practice of DEM Stream Burning Revisited.” Earth Surface Processes and Landforms 41 (5): 658–68. https://doi.org/10.1002/esp.3888.
Squividant, H., R. Bera, P. Aurousseau, and C. Cudennec. 2015. “Online Watershed Boundary Delineation: Sharing Models Through Spatial Data Infrastructures.” Proceedings of the International Association of Hydrological Sciences 368: 144–49. https://doi.org/10.5194/piahs-368-144-2015.