AppEEARS functionality

Koen Hufkens

2023-09-15

Setup

Before starting save the provided NASA Earth Data password to your local keychain. The package does not allow you to use your password inline in scripts to limit security issues when sharing scripts on github or otherwise.

library(appeears)

# set a key to the keychain
rs_set_key(
  user = "earth_data_user",
  password = "XXXXXXXXXXXXXXXXXXXXXX"
  )

# you can retrieve the password using
rs_get_key(user = "earth_data_user")

# the output should be the key you provided
# "XXXXXXXXXXXXXXXXXXXXXX"

Downloads are managed using a Bearer/session token. This token is valid for 48 hours, after which it will expire and you will need to request a new one. Although downloads are managed using the user (keychain) details only, you can request the current token using rs_login(), while rs_logout() will explicitly invalidate the current session token.

# request the current token
token <- rs_login(user = "earth_data_user")

# invalidate the current session
rs_logout(token)

Use

Listing available data

You can list all available products on the AppEEARS API using the rs_products() function. Similarly, you can list all available layers for a given product using rs_layers(). Note that when queryign the layers for a product you will need to use the field ProductAndVersion field, as some products are listed under various versions.

# list all product information
products <- rs_products()

# print the start of all products with their versions
head(products$ProductAndVersion)
#> [1] "GPW_DataQualityInd.411"    "GPW_UN_Adj_PopCount.411"  
#> [3] "GPW_UN_Adj_PopDensity.411" "GPW_Basic_Demog_Char.411" 
#> [5] "MCD12Q1.061"               "MCD12Q2.061"

# list all layers for a particular
# product
layers <- rs_layers(
  product = "MCD12Q2.006"
)

head(layers)
#>                  V1    Band
#> 1 Product not found message

Point based data requests

All point based queries are made by first creating a tidy data frame with the desired products and layers to query.

In this data frame task specifies the overall name of the task to run (this prefix will be used to name the final downloaded files). The subtask denotes the various locations and or products you want to query. As such, you can query multiple locations in the same larger task, avoiding multiple queries to the API.

The latitude and longitude fields specify geographic coordinates of query locations, while start and end columns define the range of the data queried. Note that the date range will cover the maximum date range across all subtasks. If date ranges vary widely it is advised to create separate tasks.

Finally the product and layer columns denote the remote sensing product and particular layer to download. A full list of products can be queried using rs_products(), while the layers of a particular product can be listed using rs_layers(). Note that the product needs to be specified using the full product name, including the version of the product (as stored in the ProductAndVersion field).

For point and area based queries all data are saved in a subdirectory of the main path as defined by the task name.

# Load the library
library(appeears)

# list all products
rs_products()

# list layers of the MOD11A2.061 product
rs_layers("MOD11A2.061")

df <- data.frame(
  task = "time_series",
  subtask = "US-Ha1",
  latitude = 42.5378,
  longitude = -72.1715,
  start = "2010-01-01",
  end = "2010-12-31",
  product = "MCD43A4.061",
  layer = c("Nadir_Reflectance_Band3","Nadir_Reflectance_Band4")
)

# build the area based request/task
# rename the task name so data will
# be saved in the "point" folder
# as defined by the task name
df$task <- "point"
task <- rs_build_task(df = df)

# request the task to be executed
rs_request(
  request = task,
  user = "earth_data_user",
  transfer = TRUE,
  path = "~/some_path",
  verbose = TRUE
)

We can now read in the downloaded data from the download path. Note that a subfolder will be created based upon the task name, and “_” will be converted internally to “-” in the downloaded file names.

# read in data
time_series <- read.table(
  "~/some_path/time_series/time-series-MCD43A4-061-results.csv",
  header = TRUE,
  sep = ","
  )

The downloaded data contains both requested bands as columns and we can do band math easily. Below the Normalized Difference Vegetation Index (NDVI) is calculated using bands 3 (red) and band 4 (NIR). We use the MODLAND quality control bands to screen out spurious values (those equal to 255). Finally, a quick plot is generated for the NDVI values for year 2010.

# convert band 3 and 4 to NDVI
time_series <- time_series |>
  mutate(
    Date = as.Date(Date),
    NDVI = (MCD43A4_061_Nadir_Reflectance_Band4 - MCD43A4_061_Nadir_Reflectance_Band3)/
      (MCD43A4_061_Nadir_Reflectance_Band4 + MCD43A4_061_Nadir_Reflectance_Band3)
  )

# screen for quality control
time_series <- time_series |>
  mutate(
    NDVI = ifelse(MCD43A4_061_BRDF_Albedo_Band_Mandatory_Quality_Band4 == 255 |
                    MCD43A4_061_BRDF_Albedo_Band_Mandatory_Quality_Band3 == 255,
                  NA, NDVI)
  )

# plot the time series
ggplot(time_series) +
  geom_point(
    aes(
      Date,
      NDVI
    )
  ) +
  theme_bw()

Area based data requests

You can select a region-of-interest (ROI) instead of point based data, using both sf polygons or the extent (bounding box) of an existing terra SpatRaster object. Both methods follow the same workflow.

{sf} polygon ROI

When using an sf object, provide it to the roi argument of the rs_build_task() function. The sf object must be of class sf not sfc when required convert sfc data using st_as_sf().

Note however that at the time only as simple polygon is supported. Multiple polygons in the same sf object might result in failure to query the data.

Furthermore, no other means will be provided to specify a region-of-interest. As such, you will always have to query a region-of-interest using an sf object. This ensures consistency across queries and allows for rapid visualization of a region of interest (in contrast to a simple list of e.g. top-left, bottom-right coordinates).

# load the required libraries
library(appeears)
library(sf)
library(dplyr)
library(ggplot2)

df <- data.frame(
  task = "time_series",
  subtask = "subtask",
  latitude = 42.5378,
  longitude = -72.1715,
  start = "2010-01-01",
  end = "2010-12-31",
  product = "MCD12Q2.006",
  layer = c("Greenup")
)

# load the north carolina demo data
# included in the {sf} package
# and only retain Camden county
roi <- st_read(system.file("gpkg/nc.gpkg", package="sf"), quiet = TRUE) |>
  filter(
    NAME == "Camden"
  )
# build the area based request/task
# rename the task name so data will
# be saved in the "polygon" folder
# as defined by the task name
df$task <- "polygon"
task <- rs_build_task(
  df = df,
  roi = roi,
  format = "geotiff"
)

# request the task to be executed
rs_request(
  request = task,
  user = "earth_data_user",
  transfer = TRUE,
  path = "~/some_path",
  verbose = TRUE
)

Once downloaded the geotiff can be read in using the terra package from the subfolder defined by the task name.

library(terra)
r_polygon <- terra::rast(
  file.path("~/some_path","polygon/MCD12Q2.006_Greenup_0_doy2010001_aid0001.tif")
  )

A quick plot of both the raster and vector polygon show the cut-out as provided by the AppEEARS API.

# convert to data frame for plotting
# with ggplot2, otherwise use the
# tidyterra package and geom_spatrast()
df_polygon <- r_polygon |>
  as.data.frame(xy=TRUE)

# convert incremental values (days since Jan 1, 1970
# to DOY)
df_polygon <- df_polygon |>
  mutate(
    DOY =  as.numeric(
      format(
        as.Date("1970-01-01") + MCD12Q2.006_Greenup_0_doy2010001_aid0001, "%j"
        )
      )
  ) |>
  filter(
    DOY < 180
  )

head(df_polygon)
#>           x        y MCD12Q2.006_Greenup_0_doy2010001_aid0001 DOY
#> 1 -76.56458 36.55625                                    14695  86
#> 2 -76.56042 36.55625                                    14692  83
#> 3 -76.55625 36.55625                                    14692  83
#> 4 -76.55208 36.55625                                    14692  83
#> 5 -76.54792 36.55625                                    14688  79
#> 6 -76.54375 36.55625                                    14690  81

ggplot() +
  geom_raster(
    data = df_polygon,
    aes(
      x,
      y,
      fill = DOY
    )
  ) +
  labs(
    x = "",
    y = ""
  ) +
  scale_fill_viridis_c() +
  geom_sf(
    data = roi,
    fill = NA,
    colour = "red",
    lwd = 2
    ) +
  theme_bw()

{terra} SpatRaster ROI

The terra based region-of-interest workflow is similar to that of sf polygon based queries. One only has to provide a SpatRaster as an roi argument in rs_build_task() to query a region of the same extent as the SpatRaster. The use case for this functionality is obvious, creating a quick way to sample new data for an existing data set (using the same coverage).

Note that unlike the sf method a bounding box is used and masked data is ignored (the full extent is downloaded and masking will have to be repeated afterwards).

# load the required libraries
library(terra)
library(ggplot2)
library(patchwork)

# create a SpatRaster ROI from the terra demo file
f <- system.file("ex/elev.tif", package="terra")
roi <- terra::rast(f)

# build the area based request/task
# rename the task name so data will
# be saved in the "raster" folder
# as defined by the task name
df$task <- "raster"
task <- rs_build_task(
  df = df,
  roi = roi,
  format = "geotiff"
)

# request the task to be executed
rs_request(
  request = task,
  user = "earth_data_user",
  transfer = TRUE,
  path = "~/some_path",
  verbose = TRUE
)

Once downloaded the geotiff can be read in using the terra package from the subfolder defined by the task name.

r_raster <- terra::rast(
  file.path("~/some_path","raster/MCD12Q2.006_Greenup_0_doy2010001_aid0001.tif")
  )

A quick plot of both the raster and vector polygon show the cut-out as provided by the AppEEARS API.

# convert to data frame for plotting
# with ggplot2, otherwise use the
# tidyterra package and geom_spatrast()

df_raster <- r_raster |>
  as.data.frame(xy=TRUE)

# convert incremental values (days since Jan 1, 1970
# to DOY)
df_raster <- df_raster |>
  mutate(
    DOY =  as.numeric(
      format(
        as.Date("1970-01-01") + MCD12Q2.006_Greenup_0_doy2010001_aid0001, "%j"
        )
      )
  ) |>
  filter(
    DOY < 180
  )

df_source <- roi |>
  as.data.frame(xy=TRUE)

p <- ggplot() +
  geom_raster(
    data = df_source,
    aes(
      x,
      y,
      fill = elevation
    )
  ) +
  labs(
    x = "",
    y = ""
  ) +
  scale_fill_viridis_c() +
  theme_bw()

p2 <- ggplot() +
  geom_raster(
    data = df_raster,
    aes(
      x,
      y,
      fill = DOY
    )
  ) +
  labs(
    x = "",
    y = ""
  ) +
  scale_fill_viridis_c() +
  theme_bw()

# patchwork side by side plot
p | p2