In cboettig/neonstore: NEON Data Store

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%",
  cache = FALSE
)
library(neonstore)
library(dplyr)
library(stringr)
Sys.setenv("NEONSTORE_HOME" = tempfile())
Sys.setenv("NEONSTORE_DB" = tempfile())

neonstore

neonstore provides quick access and persistent storage of NEON data tables. neonstore emphasizes simplicity and a clean data provenance trail, see Provenance section below.

Installation

Install the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("cboettig/neonstore")

Quickstart

library(neonstore)
library(tidyverse)

Discover data products of interest:

products <- neon_products()
products |>
  filter(str_detect(keywords, "bird")) |> 
  select(productName, productCode)

You may also prefer to explore the NEON Data Portal website interactively.

Download-based workflow

Once we have identified a data product code, we can download all associated data files, e.g. in the bird survey data. Optionally, we can restrict this download to a set of sites or date ranges of interest, (see function documentation for details).

neon_download("DP1.10003.001")

View your store of NEON products:

neon_index()

These files will persist between sessions, so you only need to download once or to retrieve updates. neon_index() can take arguments to filter by product or pattern (regular expression) in table name, e.g. neon_index(table = "brd").

Database backend

neonstore now supports a backend relation database as well. Import data from the raw downloaded files using neon_store():

neon_store(product = "DP1.10003.001")

Access an imported table using neon_table() instead of neon_read():

neon_table("brd_countdata")

Note that we need to include the product name in the table name when accessing the database, as table names alone may not be unique. RStudio users can also list and explore all tables interactively in the Connections pane in RStudio using the function neon_pane().

Larger-than-RAM data

When working across data from many sites or years simultaneously, it is easy for data to be too big for R to fit into working memory. This is especially true when working with sensor data. neonstore makes it easy to work with such data using dplyr-operations though. Just include the option lazy = TRUE, and most dplyr operations will execute quickly on disk instead (by leveraging the dbplyr backend and the power of the duckdb database).

brd <- neon_table("brd_countdata", lazy=TRUE)
# unique species per site?
brd |> 
  distinct(siteID, scientificName) |> 
  count(siteID, sort=TRUE) |> 
  collect()

Use the function collect() at the end of a chain of dplyr functions to bring the resulting data into R.

NEW: Cloud-based workflow

It is now possible to access data directly from NEON's cloud storage system without downloading. (Note: this still must ping the NEON API to obtain the most recent list of files, and this list is subject to rate limits). Like the local database approach, this strategy works for larger-than-RAM data, and can be substantially faster than downloading. However, if you work frequently with the same data products and have ample disk space available, you will find the one-time wait for downloading to be faster.

brd <- neon_cloud("brd_countdata", product="DP1.10003.001")

brd |> 
  distinct(siteID, scientificName) |> 
  count(siteID, sort=TRUE) |> 
  collect()

Note on API limits

If neon_download() exceeds the API request limit (with or without the token), neonstore will simply pause for the required amount of time to avoid rate-limit-based errors.

The NEON API now rate-limits requests.. Using a personal token will increase the number of requests you can make before encountering this delay. See link for directions on registering for a token. Then pass this token in .token argument of neon_download(), or for frequent use, add this token as an environmental variable, NEON_DATA to your local .Renviron file in your user's home directory. neon_download() must first query each the API of each NEON site which collects that product, for each month the product is collected.

(It would be much more efficient on the NEON server if the API could take queries of the from /data/<product>/<site>, and pool the results, rather than require each month of sampling separately!)

unlink("my_neondata.zip")
unlink("index.csv")
Sys.unsetenv("NEONSTORE_HOME")
Sys.unsetenv("NEONSTORE_DB")

Non-stacking files and low-level interface

At it's core, neonstore is simply a mechanism to download files from the NEON API. While the .csv files from the Observation Systems (OS, e.g. bird count surveys), and Instrument Systems (e.g. aquatic sensors) are typically stacked into large tables, other products, such as the .laz and .tif images produced by the airborne observation platform (AOP) sensors such as LIDAR and cameras still require the user to work directly with the downloaded files returned by neon_index(). Note that the local database can process Eddy Covariance data (h5 files), but at present this does not work with neon_cloud().

cboettig/neonstore documentation built on April 7, 2024, 8:31 p.m.