johnnybwell/neon_CPER

NEON CPER Data

TL;DR

This repo downloads and stores sensor data from the NSF NEON (National Ecological Observatory Network) network for the CPER site (Central Plains Experimental Range, CO). It's a sibling of neon_CLBJ, reusing the same catalog-driven download architecture. Five scripts pull three different kinds of NEON/NEON-adjacent data:

Script	Source	Data
scripts/download_neon_product.py	NEON /data API (neonutilities.load_by_product)	any single CSV-tabular sensor product
scripts/download_all_products.py	same, looped, fixed --years-back window	all 11 cataloged sensor products at once
scripts/download_full_history.py	same, looped, each product's own true full-history range	all 11 cataloged sensor products, no trailing-window guessing
scripts/neon_site_availability.py	NEON /sites metadata API (no token needed)	per-product true earliest/latest available month for a site
scripts/download_phenocam_gcc.py	PhenoCam Network (phenocam.nau.edu)	daily GCC greenness summaries, CPER's camera(s), --site-driven
scripts/download_eddy_covariance.py	NEON /data API (neonutilities.zips_by_product + stack_eddy)	bundled eddy covariance / NEE flux data

Which product maps to which folder/table/interval is defined in metadata/sensor_catalog.yaml, not hardcoded in any script -- same as neon_CLBJ. You need a free NEON API token (env var NEON_API_TOKEN) for anything going through NEON's own /data API (not needed for the PhenoCam script or neon_site_availability.py, which only hits NEON's open /sites metadata endpoint).

Current data status: all 11 sensor products, eddy covariance, and both PhenoCam series have been downloaded (~3.1 GB sensor data, 17 MB eddy covariance, ~600 KB PhenoCam -- see Data currently on hand below for the real per-product numbers, pulled 2026-08-26). processed/ is still empty. See the Detailed guide below for the full walkthrough if you need to re-run or extend a pull.

Unlike neon_CLBJ's trailing-5-year pulls, this repo's goal is full available history per product. Coverage start differs by product and by site -- see Product catalog below for CPER's actual per-product ranges, confirmed via neon_site_availability.py on 2026-08-26.

---

Repo layout

neon_CPER/
├── README.md                  # this file
├── .gitignore                 # excludes .venv/, raw/, __pycache__/, etc. (copied from neon_CLBJ)
├── .venv/                     # Python 3.12 virtualenv (create per Setup below; pandas<3, see below)
├── scripts/
│   ├── download_neon_product.py     # single-product downloader via NEON /data API (COPIED UNCHANGED from neon_CLBJ)
│   ├── download_all_products.py     # loops download_one() over every catalog product, fixed --years-back (COPIED UNCHANGED)
│   ├── download_full_history.py     # NEW: same loop, but each product's own true full-history --start/--end
│   ├── neon_site_availability.py    # NEW: looks up each product's true available date range for a site (no token)
│   ├── download_phenocam_gcc.py     # GENERALIZED: now --site-driven, catalog-lookup instead of hardcoded aliases,
│   │                                #   plus --full-history flag (was CLBJ-only, years-back-only)
│   ├── download_eddy_covariance.py  # DP4.00200.001 HDF5 download + stack_eddy() + QC filter (COPIED UNCHANGED)
│   ├── requirements.txt             # neonutilities, pyyaml, pandas<3, requests (COPIED UNCHANGED)
├── metadata/
│   ├── sensor_catalog.yaml    # product catalog: 11 core sensor products + eddy covariance COPIED UNCHANGED from
│   │                          #   neon_CLBJ; two new CPER-specific PhenoCam entries added, see notes in the file
│   └── README.md              # folder-to-product mapping + CPER-specific findings (this repo's version)
├── raw/                       # one subfolder per product, holds combined CSVs -- all 11 sensor products + eddy
│                               #   covariance + both PhenoCam series pulled (2026-08-26), see section 5 below
│   ├── air_temperature/       # DP1.00002.001
│   ├── soil_moisture/         # DP1.00094.001
│   ├── soil_temperature/      # DP1.00041.001
│   ├── precipitation/         # DP1.00044.001
│   ├── radiation/             # DP1.00023.001 + DP1.00024.001
│   ├── humidity/              # DP1.00098.001
│   ├── pressure/              # DP1.00004.001
│   ├── wind/                  # DP1.00001.001
│   ├── soil_heat_flux/        # DP1.00040.001
│   ├── soil_co2/              # DP1.00095.001
│   ├── phenocam/
│   │   ├── understory/        # NEON.D10.CPER.DP1.00042 (ROI GR_0001 -- coverage stops 2018-03-11, see notes)
│   │   └── canopy/            # NEON.D10.CPER.DP1.00033 (ROI GR_1000 -- covers full 2016-06-29 -> present)
│   └── eddy_covariance/       # DP4.00200.001 -- large; run as its own confirmed step, see below
└── processed/                 # empty; for derived/analysis outputs you create from raw/ (not written to by any script)

---

Detailed guide

1. Setup

cd ~/projects/neon_CPER
python3 -m venv .venv
source .venv/bin/activate
pip install -r scripts/requirements.txt

Dependencies (scripts/requirements.txt): neonutilities>=2.0.1, pyyaml>=6.0, pandas<3, requests>=2.0. The pandas<3 pin is required -- stack_eddy() (used only by download_eddy_covariance.py) calls a drop(columns=..., axis=1) pattern that pandas 3.0 rejects outright.

2. Getting a NEON API token (required for NEON API scripts)

Since a June 2026 NEON policy change, the /data endpoint rejects anonymous requests outright (403 Access Denied). Metadata endpoints like /products and /sites still work without a token -- that includes neon_site_availability.py -- but any actual data pull through download_neon_product.py, download_all_products.py, download_full_history.py, or download_eddy_covariance.py needs one. download_phenocam_gcc.py does not need a token either -- it pulls from phenocam.nau.edu's own open archive, a separate system from NEON's API.

1. Create a free account at https://data.neonscience.org/
2. Generate an API token from your account page.
3. Set it in your shell before running any NEON-API script:

export NEON_API_TOKEN=<your token>

(or pass --token <your token> directly to any script that needs it). Never commit a token to this repo.

3. Downloading data

Step 0 -- check per-product availability for CPER first (no token, no download):

python3 scripts/neon_site_availability.py --site CPER

This prints each of the 11 core products' + eddy covariance's true available start_ym -> end_ym, straight from NEON's /sites metadata. See Product catalog below for the values confirmed on 2026-08-26 -- re-run this yourself before a real pull, since NEON's published range can advance.

Step 1 -- single-product test pull (confirm token + site code work before anything larger):

python3 scripts/download_neon_product.py --site CPER --product air_temperature --start 2024-01 --end 2024-03

Step 2 -- all 11 sensor products, full available history per product (recommended; replaces the fixed-window download_all_products.py for this repo's full-history goal):

python3 scripts/download_full_history.py --site CPER

Looks up each product's own true --start/--end via neon_site_availability.py internally, then calls the same download_one() building block download_neon_product.py uses. Prints per-product progress and a final summary table, same format as download_all_products.py.

(download_all_products.py --site CPER --years-back N still works unchanged too, if you ever want a fixed trailing window instead of full history.)

Step 3 -- eddy covariance, as its own separate, manually-confirmed step (large -- see Known issues):

# check the size implications first
python3 scripts/neon_site_availability.py --site CPER   # note the DP4.00200.001 row: 2016-12 -> 2026-07 (confirmed 2026-08-26)

# then run with that explicit range
python3 scripts/download_eddy_covariance.py --site CPER --start 2016-12 --end 2026-07

download_eddy_covariance.py itself estimates total download size and checks free disk space before downloading anything, and aborts if the estimate exceeds 90% of free space -- but confirm the estimate it prints looks reasonable before letting it proceed, given CPER's ~116 available site-months (vs. CLBJ's partial 48-month pull that already reached ~6.2GB).

Step 4 -- PhenoCam, always run separately from the above (no token needed; full CPER archive history, not a trailing window):

python3 scripts/download_phenocam_gcc.py --site CPER --full-history

Before running this, re-confirm the ROI codes live -- see metadata/README.md for what was found on 2026-08-26 (short version: the understory-equivalent camera's ROI stopped publishing in 2018; the canopy camera's GR_1000 ROI alone already covers its full history, don't stitch in GR_0001/GR_0002/GR_0003).

Output: one combined CSV per pull, written to the product's raw/ folder, generally named:

<SITE>_<product_shortname>_<start>_<end>.csv

e.g. CPER_air_temperature_2014-02_2026-07.csv (NEON sensor products, month granularity) or CPER_phenocam_canopy_2016-06-29_2026-03-07.csv (PhenoCam, day granularity).

4. Product catalog (metadata/sensor_catalog.yaml)

Catalog structure and fields are identical to neon_CLBJ's -- see that repo's README for the full field reference. The 11 core sensor product entries and the DP4.00200.001 eddy covariance entry are copied unchanged (NEON product catalog entries aren't site-specific -- --site is passed at download time, not baked into the catalog). Two new PhenoCam Network entries were added for CPER's cameras.

CPER's confirmed available range per product (via neon_site_availability.py, checked 2026-08-26 -- re-verify before a real pull):

Alias	ID	Folder	Available range	Source
air_temperature	DP1.00002.001	raw/air_temperature	2014-02 -> 2026-07	NEON API
soil_moisture	DP1.00094.001	raw/soil_moisture	2016-07 -> 2026-07	NEON API
soil_temperature	DP1.00041.001	raw/soil_temperature	2016-07 -> 2026-07	NEON API
precipitation	DP1.00044.001	raw/precipitation	2016-07 -> 2026-07	NEON API
radiation_net	DP1.00023.001	raw/radiation	2014-02 -> 2026-07	NEON API
radiation_par	DP1.00024.001	raw/radiation	2014-02 -> 2026-07	NEON API
relative_humidity	DP1.00098.001	raw/humidity	2015-06 -> 2026-07	NEON API
barometric_pressure	DP1.00004.001	raw/pressure	2014-05 -> 2026-07	NEON API
wind	DP1.00001.001	raw/wind	2014-02 -> 2026-07	NEON API
soil_heat_flux	DP1.00040.001	raw/soil_heat_flux	2016-08 -> 2026-07	NEON API
soil_co2	DP1.00095.001	raw/soil_co2	2016-07 -> 2026-07	NEON API
phenocam_understory	NEON.D10.CPER.DP1.00042 (ROI GR_0001)	raw/phenocam/understory	2016-06-29 -> 2018-03-11 only	PhenoCam Network
phenocam_canopy	NEON.D10.CPER.DP1.00033 (ROI GR_1000)	raw/phenocam/canopy	2016-06-29 -> 2026-03-07	PhenoCam Network
eddy covariance / NEE	DP4.00200.001	raw/eddy_covariance	2016-12 -> 2026-07	NEON API

Note that within each product's availableMonths list, NEON's /sites endpoint reported no internal gaps for CPER as of this check (every month between first and last available is listed) -- unlike CLBJ, which had a documented 13-month site-wide gap. That's what NEON's metadata claims is available, not a guarantee every month has zero missing data once actually pulled -- download_one()'s per-month months_missing report in the final summary is still the authoritative check.

To add a new product, add an entry to metadata/sensor_catalog.yaml with the relevant fields before pulling it -- don't hardcode product details into a script.

5. Data currently on hand

All 11 sensor products, eddy covariance, and both PhenoCam series have been pulled (2026-08-26), via download_full_history.py, download_eddy_covariance.py, and download_phenocam_gcc.py --full-history respectively.

11 NEON sensor products (raw/*/), each requested through 2026-07 but actually stopping at 2025-06-30 -- neonutilities excludes NEON's most recent ~13 months of data as unreviewed "provisional" by default (confirmed in every single product's log: "Provisional data were excluded from available files list."). This is expected behavior, not a bug or a download gap -- "full history" here means full reviewed history:

Product	dpid	rows	size	coverage
air_temperature	DP1.00002.001	597,744	86M	2014-02-18 -> 2025-06-30
soil_moisture	DP1.00094.001	6,272,640	1.1G	2016-07-21 -> 2025-06-30
soil_temperature	DP1.00041.001	7,063,200	1011M	2016-07-18 -> 2025-06-30
relative_humidity	DP1.00098.001	345,312	75M	2015-06-04 -> 2025-06-30
barometric_pressure	DP1.00004.001	194,880	32M	2014-05-20 -> 2025-06-30
precipitation	DP1.00044.001	78,312	8.2M	2016-07-25 -> 2025-06-30
wind	DP1.00001.001	597,744	96M	2014-02-18 -> 2025-06-30
radiation_net	DP1.00023.001	398,496	85M	2014-02-18 -> 2025-06-30
radiation_par	DP1.00024.001	796,992	119M	2014-02-18 -> 2025-06-30
soil_heat_flux	DP1.00040.001	465,120	65M	2016-08-27 -> 2025-06-30
soil_co2	DP1.00095.001	2,352,240	325M	2016-07-21 -> 2025-06-30

Every product above had exactly 13 months missing (2025-07 through 2026-07) -- the provisional-data cutoff, not a per-product data gap.

Eddy covariance (raw/eddy_covariance/): DP4.00200.001, requested 2016-12 -> 2026-07 (116 site-months) but only 103 site-months were actually available for download (same 13-month provisional cutoff as above) -- 11.5 GB of raw HDF5 downloaded, reduced to a 17 MB, 36,725-row CSV after stack_eddy(level="dp04") extraction (most of an HDF5 bundle is raw high-frequency/intermediate-level data this repo doesn't keep -- the staging dir is deleted after dp04 extraction) and the NEE QC filter (qfqm.fluxCo2.nsae.qfFinl == 0), which dropped 75.6% of rows (113,707 of 150,432). Final coverage: 2017-03-03 -> 2025-06-30 (flux processing evidently starts a few months after the site's raw availableMonths start).

PhenoCam (raw/phenocam/), day granularity, no provisional-data concept (separate archive from NEON's /data API):

Camera	ROI	Rows	Coverage	Gaps
understory-equivalent	GR_0001	621	2016-06-29 -> 2018-03-11	none
canopy	GR_1000	3,539	2016-06-29 -> 2026-03-07	none

A practical note for future re-runs: the bulk 11-product pull and a couple of individual retries hit transient ConnectionError: Cannot access NEON API failures partway through -- most likely NEON rate-limiting after the heavy request volume of a full-catalog pull, not a real outage (the API responded normally to a plain curl seconds later). A ~60-90s backoff before retrying the failed product(s) individually cleared it every time. If you rerun download_full_history.py and see this, don't assume the token or site code is broken -- retry the specific failed product(s) after a short pause.

6. Known issues and gotchas

Carried forward from neon_CLBJ (still apply here, same root causes):

• neonutilities 2.0.1 vs pandas 3.0 incompatibility. stack_eddy() calls a drop(columns=..., axis=1) pattern that pandas 3.0 rejects (ValueError: Cannot specify both 'axis' and 'index'/'columns'). Fixed here (as in neon_CLBJ) by pinning pandas<3 in scripts/requirements.txt.
• Eddy covariance size. CLBJ's measured rate was ~130-150MB per site-month; download_eddy_covariance.py's built-in estimate uses 150MB/site-month padded for safety. CPER has ~116 available site-months (2016-12 -> 2026-07) -- confirm the printed size estimate before letting a full pull proceed; consider pulling a shorter test range first (as the CLBJ repo did) if the full estimate looks too large for available disk space.
• soil_co2's real stacked table name is SCO2C_30_minute, not SCO2C_30min -- inherited catalog fix, should apply at any site since it's a neonutilities table-naming quirk, not a CLBJ-specific one.
• wind (DP1.00001.001) returns speed and direction in the same stacked table (2DWSD_30min) -- no separate product/table for each.
• precipitation uses DP1.00044.001 (weighing gauge), not the deprecated DP1.00006.001 -- confirm CPER actually publishes weighing-gauge data (CLBJ did; CPER's availableMonths for DP1.00044.001 above suggests it does, but this hasn't been checked against CPER's actual site-specific product documentation the way CLBJ's was).

New, CPER-specific findings from this repo's setup (2026-08-26):

• CPER's understory-equivalent PhenoCam camera (DP1.00042, ROI GR_0001) stopped publishing ROI-level GCC statistics on 2018-03-11, even though the camera keeps taking images through the present (yearly circular-image files exist through 2026). Only one ROI has ever been defined for this camera -- there's no successor ROI to stitch in the way CLBJ's canopy camera needed. A "full history" pull for this camera will legitimately stop in March 2018; that's not a bug.
• CPER's canopy camera (DP1.00033) has four ROI codes (GR_0001, GR_0002, GR_0003, GR_1000), which could look like CLBJ's canopy camera's three sequential ROI generations -- but they are not sequential. GR_0001/0002/0003 all cover the identical short window 2016-06-29 -> 2018-03-11 (three simultaneous ROIs retired together), while GR_1000 alone spans the camera's entire history, 2016-06-29 -> 2026-03-07, essentially gap-free. The catalog uses only GR_1000 for this entry; do not add the others to a rois: stitch list, since their date ranges overlap GR_1000's rather than extend it.
• Tower/soil sensor position codes (verticalPosition/ horizontalPosition), now confirmed for CPER by checking the actual distinct combinations present in each pulled file (2026-08-26) -- do not assume these match CLBJ's (4 air-temp heights, 5x8 soil array):
  • air_temperature / wind: single tower location (horizontalPosition 000), 3 heights (verticalPosition 010/020/030) -- CPER has 3 air-temp heights, not CLBJ's 4.
  • radiation_par: same single location, 4 heights (010-040).
  • relative_humidity / radiation_net: only 2 position combos each -- (000,040) and (003,000). The second combo's meaning wasn't resolved from the data alone (possibly a second, differently-mounted sensor); check sensor_positions.csv from a raw NEON download if you need to disambiguate before position-level analysis.
  • barometric_pressure: single position (000,035).
  • soil_moisture: 5 horizontal (soil pits, 001-005) x 8 depths (501-508) = 40 positions.
  • soil_temperature: 5 horizontal x 9 depths (501-509) = 45 positions -- one more depth than soil_moisture, don't assume the two arrays share the same depth set.
  • soil_co2: 5 horizontal x 3 depths (501-503) = 15 positions.
  • soil_heat_flux: 3 plates only, at horizontal positions 001/003/005 (a subset of the 5 soil pits), single depth 501 each.
• Every other CLBJ-specific data-quality finding (the site-wide 13-month gap, the duplicate air_temperature file, etc.) was specific to that repo's pull history and does not carry forward as an assumption for CPER -- CPER's availableMonths currently shows no internal gaps (see the catalog table above), but this should be reconfirmed against actual pulled data, not just NEON's metadata claim.

7. File naming conventions

Same as neon_CLBJ:

<SITE>_<product_shortname>_<start_YYYY-MM>_<end_YYYY-MM>.csv

e.g. CPER_air_temperature_2014-02_2026-07.csv.

PhenoCam CSVs (day granularity, not month):

<SITE>_<alias>_<start_YYYY-MM-DD>_<end_YYYY-MM-DD>.csv

e.g. CPER_phenocam_canopy_2016-06-29_2026-03-07.csv.

Raw source files from NEON (before this script concatenates them) follow NEON's own convention:

NEON.DOM.SITE.DPL.PRNUM.REV.HOR.VER.TMI.NAME.yyyy-mm.PKGTYPE.GENTIME.csv

CPER is in NEON domain D10 (Central Plains), vs. CLBJ's D11 (Southern Plains) -- e.g. a PhenoCam id here is NEON.D10.CPER.DP1.00033, not NEON.D11.CLBJ.DP1.00033.

8. neon_site_availability.py and download_full_history.py

These two scripts are new in this repo (not present in neon_CLBJ), added specifically to support a full-history pull instead of a trailing --years-back window:

• neon_site_availability.py queries NEON's /sites metadata endpoint (no token) and reports each catalog product's true availableMonths range for a given site.
• download_full_history.py uses that lookup internally to call download_one() (from download_neon_product.py, imported unchanged) with each product's own explicit --start/--end, instead of one --years-back value applied to all 11 products.

Neither script modifies download_neon_product.py's existing --years-back/resolve_date_range() behavior -- that script (and download_all_products.py, download_eddy_covariance.py) are byte-for- byte copies of neon_CLBJ's versions and remain fully drop-in reusable for a trailing-window pull at any site, this repo's included.