NASA Earthdata CMR Skill

Never hardcode or log credentials. For search operations, no auth is needed.

Checking for Earthdata credentials (bash):

# Only needed for downloads, not for CMR search
if grep -q 'machine urs.earthdata.nasa.gov' ~/.netrc 2>/dev/null; then
    echo "Earthdata credentials found in ~/.netrc"
else
    echo "No Earthdata credentials. Sign up free at https://urs.earthdata.nasa.gov/users/new"
    echo "Add to ~/.netrc: machine urs.earthdata.nasa.gov login YOUR_USER password YOUR_PASS"
fi

API Structure

Base URL: https://cmr.earthdata.nasa.gov/search/

CMR provides two main search targets: collections (datasets) and granules (individual files/scenes within a collection).

Endpoints

Common Parameters (all endpoints)

Collection-Specific Parameters

Granule-Specific Parameters

Key Collections for PNGE Research

Landsat (Surface Reflectance + Thermal)

Use for: Land use change detection near well pads, vegetation stress (NDVI), thermal mapping of produced water impoundments and flare stacks.

Sentinel-2 (via LPCLOUD)

Use for: Higher temporal resolution monitoring of well pad construction, pipeline right-of-way clearing, reclamation progress.

MODIS Thermal

Use for: Gas flare detection, thermal anomaly mapping near surface facilities, produced water pond temperature monitoring.

Land Cover and Change

Workflow

Step 1 -- Resolve Intent

Map the user's question to search strategy:

Determine the bounding box from the user's area of interest. For well-site monitoring, use a small box (0.1-0.5 degrees). For basin-scale analysis, use a larger box (1-3 degrees).

Common Appalachian Basin bounding boxes:

• Monongalia County, WV: -80.1,39.5,-79.6,39.8
• Northern WV Marcellus: -80.8,39.2,-79.5,40.0
• Entire Marcellus Shale play: -82.0,37.5,-74.5,42.5

Step 2 -- Search Collections

Find relevant collections for the user's area and time of interest:

# Find Landsat collections with granules in northern WV
curl -s "https://cmr.earthdata.nasa.gov/search/collections.json?\
keyword=landsat+level-2&\
bounding_box=-80.5,39.5,-79.5,40.0&\
has_granules=true&\
page_size=10"

Parse the response to get concept_id values for granule search.

Step 3 -- Search Granules

Use the collection concept_id to find specific scenes:

# Find recent low-cloud Landsat scenes over Morgantown area
curl -s "https://cmr.earthdata.nasa.gov/search/granules.json?\
collection_concept_id=C2021957657-LPCLOUD&\
bounding_box=-80.1,39.5,-79.6,39.8&\
temporal=2024-06-01T00:00:00Z,2024-09-30T23:59:59Z&\
cloud_cover=0,20&\
sort_key=-start_date&\
page_size=10"

# Find MODIS thermal anomalies near a well pad
curl -s "https://cmr.earthdata.nasa.gov/search/granules.json?\
short_name=MOD14&\
bounding_box=-80.0,39.6,-79.8,39.7&\
temporal=2024-01-01T00:00:00Z,2024-12-31T23:59:59Z&\
page_size=10"

Step 4 -- Parse Response

Collection search response (.json format):

{
  "feed": {
    "entry": [
      {
        "id": "C2021957657-LPCLOUD",
        "title": "Landsat 8-9 OLI/TIRS Collection 2 Level-2 Science Products",
        "short_name": "LANDSAT_OT_C2_L2",
        "time_start": "2013-03-18T00:00:00.000Z",
        "time_end": null,
        "summary": "...",
        "data_center": "LPCLOUD",
        "links": [{"href": "...", "rel": "..."}]
      }
    ]
  }
}

Granule search response (.json format):

{
  "feed": {
    "entry": [
      {
        "id": "G2345678901-LPCLOUD",
        "title": "LC09_L2SP_017032_20240715_20240716_02_T1",
        "time_start": "2024-07-15T15:45:23.000Z",
        "time_end": "2024-07-15T15:45:47.000Z",
        "cloud_cover": "12.5",
        "granule_size": "423.7",
        "links": [
          {"href": "https://...", "rel": "http://esipfed.org/ns/fedsearch/1.1/data#", "type": "application/x-tar"}
        ],
        "boxes": ["39.5 -80.1 40.0 -79.5"]
      }
    ]
  }
}

Key fields to extract:

• id -- CMR concept ID
• title -- scene identifier (contains path/row, date, processing info)
• time_start/time_end -- acquisition time
• cloud_cover -- percent cloud cover (string, parse to float)
• links -- data download URLs (require Earthdata login)
• boxes -- spatial extent [S W N E]

Step 5 -- Produce Output

Format: Results Table + Narrative

Present a markdown table of the most relevant results (cap at ~15 rows for readability), then a narrative summary covering:

1. Search results -- how many collections/granules matched
2. Best candidates -- lowest cloud cover, most recent, best spatial coverage
3. Temporal coverage -- date range of available data
4. Spatial note -- confirm the results overlap the area of interest
5. Download instructions -- how to access the actual data files
6. PNGE application -- what analysis the data supports

Example output structure:

## Landsat 8/9 Scenes -- Monongalia County, WV (Summer 2024)

| Scene ID | Date | Cloud (%) | Path/Row |
|----------|------|-----------|----------|
| LC09_L2SP_017032_20240715_... | 2024-07-15 | 12.5 | 017/032 |
| LC08_L2SP_017032_20240707_... | 2024-07-07 | 8.2 | 017/032 |
| LC09_L2SP_017032_20240629_... | 2024-06-29 | 22.1 | 017/032 |

**Summary:** 7 Landsat 8/9 Level-2 scenes found over the Monongalia County
area for June-September 2024, with cloud cover under 20%. The July 7 scene
(8.2% cloud) offers the clearest view for NDVI analysis of well pad
vegetation recovery. Path 017, Row 032 covers the core Marcellus drilling
area in northern WV.

**To download:** Requires free NASA Earthdata Login
(https://urs.earthdata.nasa.gov/users/new). Use wget with .netrc auth or
the Earthdata Search portal (https://search.earthdata.nasa.gov/).

Pagination

CMR returns page_size results per request (default 10, max 2000). Use page_num to paginate:

# Page 1
curl -s "...&page_size=100&page_num=1"

# Page 2
curl -s "...&page_size=100&page_num=2"

Check the CMR-Hits response header for the total number of matching results. If not available in the JSON body, add -D - to curl to capture headers.

Error Handling

Common mistakes:

• Bounding box order is W,S,E,N (not N,W,S,E)
• Temporal range requires full ISO 8601 with timezone: 2024-01-01T00:00:00Z
• collection_concept_id is required for granule search (unless using short_name)
• Cloud cover filter is min,max -- use 0,100 to get all scenes

PNGE Applications

Well Pad Monitoring

Use Landsat/HLS time series to track:

• Initial land clearing for pad construction
• Active drilling period (equipment visible at 30m in some cases)
• Post-drilling reclamation and revegetation (NDVI recovery curves)

Produced Water Impoundment Detection

• Thermal bands (Landsat Band 10, MODIS LST) can detect warm produced water ponds against cooler surrounding landscape
• Surface reflectance changes when ponds fill or drain

Gas Flare Detection

• MODIS/VIIRS thermal anomaly products (MOD14, VNP14IMG) detect active flares at sub-pixel level
• Cross-reference with state well permit data to identify flaring wells

Environmental Baseline Studies

• Pre-drilling vegetation and land cover baseline using MCD12Q1
• Water quality proxy via turbidity in nearby streams (requires atmospherically corrected surface reflectance)

Pipeline Route Assessment

• Land cover analysis along proposed pipeline corridors
• Change detection before/after pipeline construction

Data Access Methods

CMR search provides metadata and download URLs. Actual data access methods:

1. Earthdata Search Portal -- interactive web interface at https://search.earthdata.nasa.gov/ (easiest for ad-hoc exploration)
2. Direct download -- wget/curl with ~/.netrc Earthdata credentials
3. NASA AppEEARS -- subsetting and reformatting service at https://appeears.earthdatacloud.nasa.gov/ (good for time series extracts)
4. Cloud access -- many collections are in AWS us-west-2; use S3 direct access with Earthdata token for large-scale analysis

Caveats and Data Quality

• Search vs download: CMR search is free and unauthenticated. Downloading actual data files requires a free Earthdata Login account.
• Cloud cover: The cloud_cover metadata is scene-level and may not reflect conditions at your specific point of interest. A scene reported at 15% cloud may have clouds directly over your well site.
• Spatial resolution limits: Landsat at 30m cannot resolve individual wellheads or small equipment. It can detect pad-scale clearing (typically 1-5 acres) and large impoundments. MODIS at 1km is useful only for regional thermal patterns.
• Temporal gaps: Cloud cover, especially in Appalachia, limits usable scenes. Expect only 20-40% of Landsat passes to be usable in WV during summer months, fewer in winter.
• Processing levels: Level-2 products (surface reflectance, surface temperature) are analysis-ready. Level-1 products require atmospheric correction before quantitative analysis.
• Latency: Standard Landsat products available within 12-24 hours of acquisition. MODIS near-real-time products within 3 hours.
• Coordinate system: CMR uses decimal degrees (WGS84/EPSG:4326). Bounding box order is W,S,E,N -- different from some GIS conventions.

Implementation Notes

• Use bash with curl + jq for API calls
• Response format .json returns Atom feed format; .umm_json returns full UMM metadata (more verbose but includes all fields)
• Collection concept_id values are stable and can be hardcoded for known collections (e.g., C2021957657-LPCLOUD for Landsat 8/9 L2)
• For Landsat scene IDs: LC09_L2SP_PPPRRR_YYYYMMDD_... where PPP=path, RRR=row, YYYYMMDD=acquisition date
• CMR rate limits are generous (no published limit for search) but add 100ms delay between paginated requests as courtesy
• The CMR-Hits response header gives total result count without fetching all pages -- useful for checking result set size before paginating