BlackMarbleR is an R package for working with Black Marble data. Black Marble is a NASA Earth Observatory project that provides global nighttime lights data. The package automates the process of downloading all relevant tiles from the NASA LAADS archive to cover a region of interest, converting the raw files (in H5 format) to georeferenced rasters, and mosaicing rasters together when needed.

• Installation
• Bearer token
• Usage
  • Setup
  • Make raster
  • Make raster stack across multiple time periods
  • Make map
  • Make figure of trends in nighttime lights
  • Workflow to update data
• Functions and arguments
  • Functions
  • Required Arguments
  • Optional Arguments
  • Argument only for bm_extract

The package can be installed via devtools.

# install.packages("devtools")
devtools::install_github("worldbank/blackmarbler")

The function requires using a Bearer Token; to obtain a token, follow the below steps:

1. Go to the NASA LAADS Archive
2. Click "Login" (bottom on top right); create an account if needed.
3. Click "See wget Download Command" (bottom near top, in the middle)
4. After clicking, you will see text that can be used to download data. The "Bearer" token will be a long string in red.

After logging in, the below will show the bearer token in red instead of INSERT_DOWNLOAD_TOKEN_HERE. Sometimes, after logging in, the NASA website will redirect to another part of the website. To obtain the bearer token, just navigate to the NASA LAADS Archive after logging in.

Before downloading and extracting Black Marble data, we first load packages, define the NASA bearer token, and define a region of interest.

#### Setup
# Load packages
library(blackmarbler)
library(geodata)
library(sf)
library(raster)
library(ggplot2)

#### Define NASA bearer token
bearer <- "BEARER-TOKEN-HERE"

### ROI
# Define region of interest (roi). The roi must be (1) an sf polygon and (2)
# in the WGS84 (epsg:4326) coordinate reference system. Here, we use the 
# getData function to load a polygon of Ghana
roi_sf <- gadm(country = "GHA", level=1, path = tempdir()) |> st_as_sf()

The below example shows making daily, monthly, and annual rasters of nighttime lights for Ghana.

### Daily data: raster for February 5, 2021
r_20210205 <- bm_raster(roi_sf = roi_sf,
                        product_id = "VNP46A2",
                        date = "2021-02-05",
                        bearer = bearer)

### Monthly data: raster for October 2021
r_202110 <- bm_raster(roi_sf = roi_sf,
                      product_id = "VNP46A3",
                      date = "2021-10-01", # The day is ignored
                      bearer = bearer)

### Annual data: raster for 2021
r_2021 <- bm_raster(roi_sf = roi_sf,
                    product_id = "VNP46A4",
                    date = 2021,
                    bearer = bearer)

To extract data for multiple time periods, add multiple time periods to date. The function will return a raster stack, where each raster band corresponds to a different date. The below code provides examples getting data across multiple days, months, and years.

#### Daily data in March 2021
r_daily <- bm_raster(roi_sf = roi_sf,
                     product_id = "VNP46A3",
                     date = seq.Date(from = ymd("2021-03-01"), to = ymd("2021-03-31"), by = "day"),
                     bearer = bearer)

#### Monthly aggregated data in 2021 and 2022
r_monthly <- bm_raster(roi_sf = roi_sf,
                       product_id = "VNP46A3",
                       date = seq.Date(from = ymd("2021-01-01"), to = ymd("2022-12-01"), by = "month"),
                       bearer = bearer)

#### Yearly aggregated data in 2012 and 2021
r_annual <- bm_raster(roi_sf = roi_sf,
                      product_id = "VNP46A4",
                      date = 2012:2021,
                      bearer = bearer)

Using one of the rasters, we can make a map of nighttime lights

#### Make raster
r <- bm_raster(roi_sf = roi_sf,
               product_id = "VNP46A3",
               date = "2021-10-01",
               bearer = bearer)
               
#### Prep data
r <- r |> mask(roi_sf) 

r_df <- rasterToPoints(r, spatial = TRUE) |> as.data.frame()
names(r_df) <- c("value", "x", "y")

## Remove very low values of NTL; can be considered noise 
r_df$value[r_df$value <= 2] <- 0

## Distribution is skewed, so log
r_df$value_adj <- log(r_df$value+1)

##### Map 
p <- ggplot() +
  geom_raster(data = r_df, 
  aes(x = x, y = y, 
  fill = value_adj)) +
  scale_fill_gradient2(low = "black",
                       mid = "yellow",
                       high = "red",
                       midpoint = 4.5) +
  labs(title = "Nighttime Lights: October 2021") +
  coord_quickmap() + 
  theme_void() +
  theme(plot.title = element_text(face = "bold", hjust = 0.5),
  legend.position = "none")

We can use the bm_extract function to observe changes in nighttime lights over time. The bm_extract function leverages the exactextractr package to aggregate nighttime lights data to polygons. Below we show trends in annual nighttime lights data across Ghana's first administrative divisions.

#### Extract annual data
ntl_df <- bm_extract(roi_sf = roi_sf,
                     product_id = "VNP46A4",
                     date = 2012:2022,
                     bearer = bearer)

#### Trends over time
ntl_df |>
  ggplot() +
  geom_col(aes(x = date,
  y = ntl_mean),
  fill = "darkorange") +
  facet_wrap(~NAME_1) +
  labs(x = NULL,
       y = "NTL Luminosity",
       title = "Ghana Admin Level 1: Annual Average Nighttime Lights") +
  theme_minimal() +
  theme(strip.text = element_text(face = "bold")) 

Some users may want to monitor near-real-time changes in nighttime lights. For example, daily Black Marble nighttime lights data is updated regularly, where data is available roughly on a week delay; same use cases may require examining trends in daily nighttime lights data as new data becomes available. Below shows example code that could be regularly run to produce an updated daily dataset of nighttime lights.

The below code produces a dataframe of nighttime lights for each date, where average nighttime lights for Ghana's 1st administrative division is produced. The code will check whether data has already been downloaded/extracted for a specific date, and only download/extract new data.

# Create directories to store data
dir.create(file.path(getwd(), "bm_files"))
dir.create(file.path(getwd(), "bm_files", "daily"))

# Extract daily-level nighttime lights data for Ghana's first administrative divisions. 
# Save a separate dataset for each date in the `"~/Desktop/bm_files/daily"` directory. 
# The code extracts data from January 1, 2023 to today. Given that daily nighttime lights 
# data is produced on roughly a week delay, the function will only extract data that exists; 
# it will skip extracting data for dates where data has not yet been produced by NASA Black Marble. 
bm_extract(roi_sf = roi_sf,
           product_id = "VNP46A2",
           date = seq.Date(from = ymd("2023-01-01"), to = Sys.Date(), by = 1),
           bearer = bearer,
           output_location_type = "file",
           file_dir = file.path(getwd(), "bm_files", "daily"))

# Append daily-level datasets into one file
file.path(getwd(), "bm_files", "daily") |>
  list.files(pattern = "*.Rds",
  full.names = T) |>
  map_df(readRDS) |>
  saveRDS(file.path(getwd(), "bm_files", "ntl_daily.Rds"))

The package provides two functions.

• bm_raster produces a raster of Black Marble nighttime lights.
• bm_extract produces a dataframe of aggregated nighttime lights to a region of interest (e.g., average nighttime lights within US States).

Both functions take the following arguments:

• roi_sf: Region of interest; sf polygon. Must be in the WGS 84 (epsg:4326) coordinate reference system. For bm_extract, aggregates nighttime lights within each polygon of roi_sf.

• product_id: One of the following:

  • "VNP46A1": Daily (raw)
  • "VNP46A2": Daily (corrected)
  • "VNP46A3": Monthly
  • "VNP46A4": Annual

• date: Date of raster data. Entering one date will produce a raster. Entering multiple dates will produce a raster stack.

  • For product_ids "VNP46A1" and "VNP46A2", a date (eg, "2021-10-03").
  • For product_id "VNP46A3", a date or year-month (e.g., "2021-10-01", where the day will be ignored, or "2021-10").
  • For product_id "VNP46A4", year or date (e.g., "2021-10-01", where the month and day will be ignored, or 2021).

• bearer: NASA bearer token. For instructions on how to create a token, see here.

• variable: Variable to used to create raster (default: NULL). For information on all variable choices, see here; for VNP46A1, see Table 3; for VNP46A2 see Table 6; for VNP46A3 and VNP46A4, see Table 9. If NULL, uses the following default variables:

  • For product_id "VNP46A1", uses DNB_At_Sensor_Radiance_500m.
  • For product_id "VNP46A2", uses Gap_Filled_DNB_BRDF-Corrected_NTL.
  • For product_ids "VNP46A3" and "VNP46A4", uses NearNadir_Composite_Snow_Free.

• quality_flag_rm: Quality flag values to use to set values to NA. Each pixel has a quality flag value, where low quality values can be removed. Values are set to NA for each value in ther quality_flag_rm vector. (Default: c(255)).

  • For VNP46A1 and VNP46A2 (daily data):

    • 0: High-quality, Persistent nighttime lights
    • 1: High-quality, Ephemeral nighttime Lights
    • 2: Poor-quality, Outlier, potential cloud contamination, or other issues
    • 255: No retrieval, Fill value (masked out on ingestion)

  • For VNP46A3 and VNP46A4 (monthly and annual data):

    • 0: Good-quality, The number of observations used for the composite is larger than 3
    • 1: Poor-quality, The number of observations used for the composite is less than or equal to 3
    • 2: Gap filled NTL based on historical data
    • 255: Fill value

• output_location_type: Where output should be stored (default: r_memory). Either:

  • r_memory where the function will return an output in R
  • file where the function will export the data as a file. For bm_raster, a .tif file will be saved; for bm_extract, a .Rds file will be saved. A file is saved for each date. Consequently, if date = c(2018, 2019, 2020), three datasets will be saved: one for each year. Saving a dataset for each date can facilitate re-running the function later and only downloading data for dates where data have not been downloaded.

If output_location_type = "file", the following arguments can be used:

• file_dir: The directory where data should be exported (default: NULL, so the working directory will be used)
• file_prefix: Prefix to add to the file to be saved. The file will be saved as the following: [file_prefix][product_id]_t[date].[tif/Rds]
• file_skip_if_exists: Whether the function should first check wither the file already exists, and to skip downloading or extracting data if the data for that date if the file already exists (default: TRUE). If the function is first run with date = c(2018, 2019, 2020), then is later run with date = c(2018, 2019, 2020, 2021), the function will only download/extract data for 2021. Skipping existing files can facilitate re-running the function at a later date to download only more recent data.

• aggregation_fun: A vector of functions to aggregate data (default: "mean"). The exact_extract function from the exactextractr package is used for aggregations; this parameter is passed to fun argument in exactextractr::exact_extract.
• add_n_pixels: Whether to add a variable indicating the number of nighttime light pixels used to compute nighttime lights statistics (eg, number of pixels used to compute average of nighttime lights). When TRUE, it adds three values: n_non_na_pixels (the number of non-NA pixels used for computing nighttime light statistics); n_pixels (the total number of pixels); and prop_non_na_pixels the proportion of the two. (Default: TRUE).