NRT disaster response

The first 24 hours after a wildfire / flood / hurricane / volcanic eruption are when remote-sensing data has its biggest operational value — and exactly when most NASA analysis-grade products are still days or months away. This recipe is the workflow that uses only NRT products, accepts the accuracy tradeoff, and ships an answer within hours.

The pattern in 6 steps

1. Confirm the event with NRT detection layers — FIRMS for fire, NOAA NWS for storm tracks, USGS for earthquakes, Smithsonian GVP for volcanoes. These tell you where and when.
2. Pull the most recent cloud-penetrating imagery. Sentinel-1 SAR (12-day revisit) sees through cloud cover and night. For floods + hurricanes, this is the only optical-blind sensor that works during the event.
3. Compute “before vs after” change detection — pre-event and post-event scenes, masked to the AOI, differenced. Smooth-water (flood) = darkening at VV polarization; burn-scar = darkening at HLS NBR; building damage = SAR coherence drop.
4. Add atmospheric context if relevant — IMERG rainfall accumulation, TEMPO NO₂ for smoke chemistry, EMIT for SO₂ during volcanic eruptions.
5. Quantify impact — flooded area in km², population exposed (overlay with SEDAC legacy population grid or external), buildings affected (combine with OSM or commercial imagery).
6. Ship a map within 6 hours. Caveat the products: “preliminary; NRT accuracy ~30% off final.”

Minimal worked example — flood mapping within 12 hours

import earthaccess
import datetime as dt

earthaccess.login(strategy="netrc")

# Hurricane Ian-class scenario — Florida west coast, day after landfall
storm_aoi = (-83, 26, -81, 28)
landfall = dt.datetime(2024, 9, 28)
pre = (landfall - dt.timedelta(days=15)).strftime("%Y-%m-%d")
post = (landfall + dt.timedelta(days=2)).strftime("%Y-%m-%d")

# 1. FIRMS NRT (3-hourly fires — used here for context, not the primary event)
# (REST API; no Earthdata Login needed for public NRT)

# 2. Sentinel-1 OPERA RTC — analysis-ready SAR
pre_sar = earthaccess.search_data(short_name="OPERA_L2_RTC-S1_V1",
                                  bounding_box=storm_aoi, temporal=(pre, pre))
post_sar = earthaccess.search_data(short_name="OPERA_L2_RTC-S1_V1",
                                   bounding_box=storm_aoi, temporal=(post, post))

# 3. Compute flood mask: VV < -20 dB AND not in pre-event water mask
# (handled in r01 — see Cross-DAAC Composition recipe)

# 4. IMERG Late Run rainfall for the event window (4-hr latency)
imerg_late = earthaccess.search_data(short_name="GPM_3IMERGHHL",
                                      bounding_box=storm_aoi,
                                      temporal=(pre, post))
# Sum precipitation_cal over the storm window → mm total
# Use Late Run (4-hr latency); skip Final (3.5-month latency, useless for NRT)

# 5. Ship: flood polygon + rainfall contours + storm track on basemap
# Caveat the map: "OPERA RTC NRT; Late-run IMERG; preliminary, accuracy ~20-30%
# below Final products that ship in 3+ months."

NRT product cheat sheet (latency in parentheses)

Product	Latency	NRT use
FIRMS active fires	3 hr	Fire detection
MODIS/VIIRS aerosol NRT	3 hr	Smoke transport
Sentinel-1 GRD	3-6 hr (post-orbit)	Flood + damage
OPERA RTC-S1	8-12 hr	Analysis-ready SAR
IMERG Early Run	4 hr	Real-time rainfall
IMERG Late Run	12 hr	Tactical rainfall
TEMPO NRT	2-4 hr	Air quality (North America)
HLS NRT	24-48 hr (when cloud-free)	Optical change detection
Black Marble NRT	6-12 hr	Power outages
Worldview imagery	1-3 hr	Visual confirmation, low quantitative value

Do NOT use during NRT: GPM IMERG Final (3.5-month latency), MCD64A1 burned area (monthly), MOD06 cloud properties full-resolution (multi-day reprocessing), any “Climate Data Record” — these are reprocessed for analysis quality, not for response.

Common gotchas

• Confidence flags matter more than ever. NRT products skip validation steps. Always show the QA flag distribution in your output map.
• Late-run IMERG ≠ Final-run IMERG. They can differ by 20-40% on individual storms because Late-run lacks gauge bias correction. State which you used.
• OPERA RTC-S1 first availability is post-orbit + processing. A storm that lands Tuesday morning may not have RTC until Tuesday evening. Plan accordingly.
• NRT products get retracted. A FIRMS pixel that’s “confidence 60” may be reclassified as cloud in the next pass. Don’t claim individual hotspots in messaging; claim aggregate area.
• Population exposure is the hardest claim to make safely. Static gridded population (SEDAC, WorldPop) is years out of date; dynamic exposure during a hurricane is unknowable. Phrase as “X km² affected, including [residential / agricultural / wetland] land use” rather than “Y people displaced.”

When this pattern fails

• Clear-sky events — wildfires under high pressure with no smoke advection are sometimes better tracked with HLS NRT + FIRMS, no SAR needed.
• Floods inside urban canyons — SAR backscatter from building corner reflections doesn’t darken with water. Use NRT optical + commercial high-res if cloud-free.
• Events outside Sentinel-1 acquisition — not every orbit acquires every AOI. Check the ASF Vertex acquisition calendar before assuming a scene will exist.

Coordination — who else is doing this in real-time

• NASA Disasters Mapping Portal: disasters.nasa.gov — coordinated NASA NRT response, includes products from multiple DAACs.
• NASA Earth Applied Sciences Disasters program: assigns coordinators for major events; reach out via disasters@earthdata.nasa.gov.
• NOAA NWS + USGS — when integrating outside NASA-only products.
• Commercial complements: ICEYE + Capella for sub-daily SAR; Planet for daily 3 m optical.

Sources

• NASA Disasters: https://disasters.nasa.gov/
• OPERA RTC-S1: https://www.jpl.nasa.gov/go/opera/products/rtc-product
• FIRMS API: https://firms.modaps.eosdis.nasa.gov/api/
• IMERG release notes (latency tradeoffs): https://gpm.nasa.gov/data/imerg