NetCDF-CF Geometry and Timeseries Tools for R"
In ncdfgeom: 'NetCDF' Geometry and Time Series

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width=6, 
  fig.height=4
)
options(scipen = 9999)

Introduction

ncdfgeom is intended to write spatial geometries, their attributes, and timeseries data (that would typically be stored in two or more files) into a single file. The package provides functions to read and write NetCDF-CF Discrete Sampling Geometries point and timeseries feature types as well as NetCDF-CF spatial geometries. These utilities are meant to be general, but were designed to support working with typical geospatial feature data with linked attributes and time series in NetCDF. Supported data types include:

Variables from R data.frame tables with one row per geometry are read from or written to NetCDF variables.
data.frame tables with a time series per column are read from or written as NetCDF-CF DSG timeSeries FeatureType data.
data.frame table variables with a single-time-step observation for a given point location in each row are written to a NetCDF-CF DSG Point FeatureType.
sp and sf spatial point, line, and polygon types can be read from or written to NetCDF-CF geometry variables introduced in CF-1.8

For timeseries, two formats are supported:

a wide format data.frame with timesteps in rows and geometry "instances" in columns with required attributes of geometry "instances" provided separately. This format lends its self to data where the same timestamps are used for every row and data exists for all geometry instances for all time steps -- it is sometimes referred to as the orthogonal array, or "space-wide", encoding.
a long format data.frame where each row contains all the geometry "instance" metadata, a time stamp, and the variables to be stored for that time step. This format lends it self to data where each geometry instance has unique timesteps and/or data is not available for each geometry instance at the same timesteps (sparse arrays).

Additional read / write functions to include additional DSG feature types will be implemented in the future and contributions are welcomed. ncdfgeom is a work in progress. Please review the "issues" list to submit issues and/or see what changes are planned.

Installation

At the time of writing, installation is only available via remotes or building the package directly as one would for development purposes.

install.packages("remotes")
remotes::install_github("DOI-USGS/ncdfgeom")

When on CRAN, the package will be installed with the typical install.packages method.

install.packages("ncdfgeom")

For this demo, we'll use sf, dplyr, and ncdfgeom.

library(sf)
library(dplyr)
library(ncdfgeom)

Sample Data

We will start with two dataframes: prcp_data and climdiv_poly containing precipitation estimate timeseries and polygon geometries that describe the boundaries of climate divisions respectively. The precipitation data has one time series for each geometry.

Code to download and prep these data is shown at the bottom. More info about the data can be found at: doi:10.7289/V5M32STR

The data required for this demo has been cached in the ncdfgeom package and is available as shown below.

`prcp_data`

prcp_data <- readRDS(system.file("extdata/climdiv-pcpndv.rds", package = "ncdfgeom"))
print(prcp_data, max_extra_cols = 0)
plot(prcp_data$date, prcp_data$`0101`, col = "red", 
     xlab = "date", ylab = "monthly precip (inches)", main = "Sample Timeseries for 0101-'Northern Valley'")
lines(prcp_data$date, prcp_data$`0101`)

`climdiv_poly`

climdiv_poly <- read_sf(system.file("extdata/climdiv.gpkg", package = "ncdfgeom"))
print(climdiv_poly)
plot(st_geometry(climdiv_poly), main = "Climate Divisions with 0101-'Northern Valley' Highlighted")
plot(st_geometry(filter(climdiv_poly, CLIMDIV == "0101")), col = "red", add = TRUE)

As shown above, we have two data.frames. One has 344 columns and the other 344 rows. These 344 climate divisions will be our "instance" dimension when we write to NetCDF.

Write Timeseries and Geometry to NetCDF

The NetCDF discrete sampling geometries timeseries standard requires point lat/lon coordinate locations for timeseries data. In the code below, we calculate these values and write the timeseries data to a netcdf file.

climdiv_centroids <- climdiv_poly %>%
  st_transform(5070) %>% # Albers Equal Area
  st_set_agr("constant") %>%
  st_centroid() %>%
  st_transform(4269) %>% #NAD83 Lat/Lon
  st_coordinates() %>%
  as.data.frame()

nc_file <- "climdiv_prcp.nc"

prcp_dates <- prcp_data$date
prcp_data <- select(prcp_data, -date)
prcp_meta <- list(name = "climdiv_prcp_inches", 
                  long_name = "Estimated Monthly Precipitation (Inches)")

write_timeseries_dsg(nc_file = nc_file, 
                     instance_names = climdiv_poly$CLIMDIV, 
                     lats = climdiv_centroids$Y, 
                     lons = climdiv_centroids$X, 
                     times = prcp_dates, 
                     data = prcp_data, 
                     data_unit = rep("inches", (ncol(prcp_data) - 1)), 
                     data_prec = "float", 
                     data_metadata = prcp_meta, 
                     attributes = list(title = "Demonstation of ncdfgeom"), 
                     add_to_existing = FALSE)

climdiv_poly <- st_sf(st_cast(climdiv_poly, "MULTIPOLYGON"))

write_geometry(nc_file = "climdiv_prcp.nc", 
               geom_data = climdiv_poly,
               variables = "climdiv_prcp_inches")

Now we have a file with a structure as shown in the ncdump output below.

try({ncdump <- system(paste("ncdump -h", nc_file), intern = TRUE)
cat(ncdump, sep = "\n")}, silent = TRUE)

For more information about the polygon and timeseries data structures used here, see the NetCDF-CF standard.

Read Timeseries and Geometry from NetCDF

Now that we have all our data in a single file, we can read it back in.

# First read the timeseries.
prcp_data <- read_timeseries_dsg("climdiv_prcp.nc")

# Now read the geometry.
climdiv_poly <- read_geometry("climdiv_prcp.nc")

Here's what the data we read in looks like:

names(prcp_data)
class(prcp_data$time)
names(prcp_data$varmeta$climdiv_prcp_inches)
prcp_data$data_unit
prcp_data$data_prec
str(names(prcp_data$data_frames$climdiv_prcp_inches))
prcp_data$global_attributes
names(climdiv_poly)

# Because we've gotta have pretty colors!
p_colors <- function (n, name = c("precip_colors")) {
# Thanks! https://quantdev.ssri.psu.edu/tutorials/generating-custom-color-palette-function-r
    p_rgb <- col2rgb(c("#FAFBF3", "#F0F8E3", "#D4E9CA", 
                       "#BBE0CE", "#B7DAD0", "#B0CCD7", 
                       "#A9B8D7", "#A297C2", "#8F6F9E", 
                       "#684A77", "#41234D"))
    precip_colors = rgb(p_rgb[1,],p_rgb[2,],p_rgb[3,],maxColorValue = 255)
    name = match.arg(name)
    orig = eval(parse(text = name))
    rgb = t(col2rgb(orig))
    temp = matrix(NA, ncol = 3, nrow = n)
    x = seq(0, 1, , length(orig))
    xg = seq(0, 1, , n)
    for (k in 1:3) {
        hold = spline(x, rgb[, k], n = n)$y
        hold[hold < 0] = 0
        hold[hold > 255] = 255
        temp[, k] = round(hold)
    }
    palette = rgb(temp[, 1], temp[, 2], temp[, 3], maxColorValue = 255)
    palette
}

To better understand what is actually in this file, let's visualize the data we just read. Below we join a sum of the precipitation timeseries to the climate division polygons and plot them up.

climdiv_poly <- climdiv_poly %>%
  st_transform(3857) %>% # web mercator
  st_simplify(dTolerance = 5000)

title <- paste0("\n Sum of: ", prcp_data$varmeta$climdiv_prcp_inches$long_name, "\n", 
                format(prcp_data$time[1], 
                         "%Y-%m", tz = "UTC"), " - ", 
                format(prcp_data$time[length(prcp_data$time)], 
                         "%Y-%m", tz = "UTC"))

prcp_sum <- apply(prcp_data$data_frames$climdiv_prcp_inches, 
                  2, sum, na.rm = TRUE)

prcp <- data.frame(CLIMDIV = names(prcp_sum), 
                   prcp = as.numeric(prcp_sum), 
                   stringsAsFactors = FALSE) %>%
  right_join(climdiv_poly, by = "CLIMDIV") %>% 
  st_as_sf()

plot(prcp["prcp"], lwd = 0.1, pal = p_colors, 
     breaks = seq(0, 14000, 1000),
     main = title,
     key.pos = 3, key.length = lcm(20))

This is the code used to download and prep the precipitation and spatial data. Provided for reproducibility and is not run here.

# Description here: ftp://ftp.ncdc.noaa.gov/pub/data/cirs/climdiv/divisional-readme.txt
prcp_url <- "ftp://ftp.ncdc.noaa.gov/pub/data/cirs/climdiv/climdiv-pcpndv-v1.0.0-20190408"

prcp_file <- "prcp.txt"

download.file(url = prcp_url, destfile = prcp_file, quiet = TRUE)

division_url <- "ftp://ftp.ncdc.noaa.gov/pub/data/cirs/climdiv/CONUS_CLIMATE_DIVISIONS.shp.zip"
division_file <- "CONUS_CLIMATE_DIVISIONS.shp.zip"

download.file(url = division_url, destfile = division_file, quiet = TRUE)
unzip("CONUS_CLIMATE_DIVISIONS.shp.zip")

climdiv_poly <- read_sf("GIS.OFFICIAL_CLIM_DIVISIONS.shp") %>%
  select("CLIMDIV", CLIMDIV_NAME = "NAME") %>%
  mutate(CLIMDIV = ifelse(nchar(as.character(CLIMDIV)) == 3, 
                          paste0("0",as.character(CLIMDIV)),
                          as.character(CLIMDIV))) %>%
  st_simplify(dTolerance = 0.0125)

month_strings <- c("jan", "feb", "mar", "apr", "may", "jun", 
                   "jul", "aug", "sep", "oct", "nov", "dec")

prcp_data <- read.table(prcp_file, header = FALSE, 
                        colClasses = c("character", 
                                       rep("numeric", 12)),
                        col.names = c("meta", month_strings))

# Here we gather the data into a long format and prep it for ncdfgeom.
prcp_data <- prcp_data %>%
  gather(key = "month", value = "precip_inches", -meta) %>%
  mutate(climdiv = paste0(substr(meta, 1, 2), substr(meta, 3, 4)),
         year = substr(meta, 7, 10),
         precip_inches = ifelse(precip_inches < 0, NA, precip_inches)) %>%
  mutate(date = as.Date(paste0("01", "-", month, "-", year), 
                        format = "%d-%b-%Y")) %>%
  select(-meta, -month, -year) %>%
  filter(climdiv %in% climdiv_poly$CLIMDIV) %>%
  spread(key = "climdiv", value = "precip_inches") %>%
  filter(!is.na(`0101`))
  as_tibble()

# Now make sure things are in the same order.
climdiv_names <- names(prcp_data)[2:length(names(prcp_data))]
climdiv_row_order <- match(climdiv_names, climdiv_poly$CLIMDIV)
climdiv_poly <- climdiv_poly[climdiv_row_order, ]

sf::write_sf(climdiv_poly, "climdiv.gpkg")
saveRDS(prcp_data, "climdiv-pcpndv.rds")

unlink("GIS*")
unlink("CONUS_CLIMATE_DIVISIONS.shp.zip")
unlink("prcp.txt")

Here's the p_colors function used in plotting above.

p_colors <- function (n, name = c("precip_colors")) {
# Thanks! https://quantdev.ssri.psu.edu/tutorials/generating-custom-color-palette-function-r
    p_rgb <- col2rgb(c("#FAFBF3", "#F0F8E3", "#D4E9CA", 
                       "#BBE0CE", "#B7DAD0", "#B0CCD7", 
                       "#A9B8D7", "#A297C2", "#8F6F9E", 
                       "#684A77", "#41234D"))
    precip_colors = rgb(p_rgb[1,],p_rgb[2,],p_rgb[3,],maxColorValue = 255)
    name = match.arg(name)
    orig = eval(parse(text = name))
    rgb = t(col2rgb(orig))
    temp = matrix(NA, ncol = 3, nrow = n)
    x = seq(0, 1, , length(orig))
    xg = seq(0, 1, , n)
    for (k in 1:3) {
        hold = spline(x, rgb[, k], n = n)$y
        hold[hold < 0] = 0
        hold[hold > 255] = 255
        temp[, k] = round(hold)
    }
    palette = rgb(temp[, 1], temp[, 2], temp[, 3], maxColorValue = 255)
    palette
}