Station interpolation workflow

This article documents the station-data side of the same test-area workflow. The region of interest is the São Lourenço basin area in Mato Grosso, Brazil. The station workflow uses daily precipitation observations from March 2017. It is presented as an interpolation and validation-preparation workflow, not as a scientific validation against the ERA5-Land 2025 data used in the gridded case.

Station observations

The input folder contains one precipitation file per station and a pcp.txt main file with station metadata.

station_folder <- file.path(case_root, "data/input/weather-station-data/pcp_stations")

pcp_table <- files_to_table(
  files_path = station_folder,
  files_pattern = "p-",
  start_date = "2017-03-01",
  end_date = "2017-03-31",
  na_value = -99,
  neg_to_zero = FALSE
)

The case has 14 stations and 31 daily records. A daily station table has the shape expected by the interpolation functions:

ID	NAME	LAT	LON	ELEVATION	pcp
1	p-1553003	-15.940	-53.450	597	0.0
2	p-1554006	-15.989	-54.968	256	8.4
3	p-1555005	-15.840	-55.320	783	0.0

Gap filling

count_na() identifies missing values and fill_gap() fills station gaps using the configured temporal correlation period.

count_na(pcp_table[-1], percent = TRUE)

gap_filled <- fill_gap(
  dataset = pcp_table,
  corPeriod = "daily"
)

count_na(gap_filled[-1], percent = TRUE)

The filled table can be written back to SWAT-style station files.

table_to_files(
  table = gap_filled,
  folder_path = gap_filled_output,
  first_date = "2017-03-01",
  file_extension = "txt"
)

Daily tables for interpolation

point_to_daily() changes the perspective from one file per station to one table per day. save_daily_tbl() writes those daily tables to disk.

daily_tables <- point_to_daily(
  my_folder = gap_filled_output,
  var_pattern = "p-",
  main_pattern = "pcp",
  start_date = "20170301",
  end_date = "20170331",
  interval = "day",
  na_value = -99,
  neg_to_zero = TRUE,
  prefix = "day_"
)

save_daily_tbl(
  tbl_list = daily_tables,
  path = daily_tbl_output
)

The precomputed run created 31 daily tables.

Interpolation to an area

ts_to_area() applies trend-surface interpolation to each daily station table and returns a multi-layer raster. The case wrote selected days to GeoTIFF for mapping and inspection.

raster_interpolated <- ts_to_area(
  my_folder = daily_tbl_output,
  bassin_limit_path = basin_limit,
  poly_degree = 2,
  resolution = 0.001
)

terra::writeRaster(
  terra::rast(raster_interpolated)[[18]],
  file.path(output_dir, "interpolation/raster_interpolated/interpolated_2017-03-18.tif"),
  overwrite = TRUE
)

The generated GeoTIFF for 2017-03-18 is about 11.4 MB and is treated as a case output artifact.

Interpolation to SWAT target points

ts_to_point() estimates the station-derived precipitation series at target points such as sub-basin centroids. ts_point_to_files() writes one SWAT-style file per target point.

points_interpolated <- ts_to_point(
  my_folder = daily_tbl_output,
  targeted_points_path = centroid_points,
  poly_degree = 2
)

ts_point_to_files(
  points_list = points_interpolated,
  output_folder = file.path(output_dir, "interpolation/pcp/pixel_d"),
  file_prefix = "pcp",
  start_date = "2017-03-01"
)

main_tbl <- var_main_creator(
  targeted_points_path = centroid_points,
  var_name = "pcp",
  col_elev = "Elev"
)

The precomputed run produced 258 point files plus a precipitation mainFile.csv.

Validation preparation

tbl_from_references() can extract values from a raster at station or centroid locations, producing the wide table needed by validation tools.

simulated_at_stations <- tbl_from_references(
  raster_file = raster_file,
  ref_points = file.path(gap_filled_output, "pcp.txt"),
  prefix_colname = "sim"
)

This prepares the data for packages such as hydroGOF, but it does not compute validation metrics itself. A scientific validation requires matching variable, period, spatial reference points, and time-step ordering between observed and simulated data.