In the summer of 2023, a mixed-severity wildfire burned through approximately 12,000 hectares of Douglas fir and ponderosa pine forest in the eastern Cascades foothills of Oregon — a landscape we'd been tracking for other reasons, and which became an unplanned but highly instructive case study in what NDVI time series actually tells you about recovery versus what it appears to tell you.
The short version: NDVI recovers faster than ecosystems do. This is not a revelation to ecologists, but it's a trap that satellite monitoring pipelines can fall into if the metric isn't interpreted carefully. What looks like recovery from orbit is often something quite different on the ground.
The Immediate Post-Fire Spectral Signature
In the weeks immediately following the fire, the burn scar was unambiguous in every band. Sentinel-2 true-color imagery showed the characteristic charcoal-gray of complete combustion in the high-severity areas, transitioning to a mosaic of brown standing dead timber, bare mineral soil, and isolated surviving canopy patches in mixed-severity zones. NDVI across the core burn area dropped to values between -0.1 and 0.15 — close to bare soil baseline — from pre-fire values in the 0.55-0.75 range for mature Douglas fir canopy.
The NBR (Normalized Burn Ratio, (NIR - SWIR2) / (NIR + SWIR2)) was a more diagnostic index at this stage than NDVI. Post-fire mineral soil and charred material produce a specific NBR signature — low NIR reflectance combined with elevated SWIR2 due to the carbon and mineral exposure — that distinguishes fire severity classes more reliably than NDVI alone. We mapped four severity classes across the burn area using a dNBR (differenced NBR, pre- versus post-fire) approach: unburned, low, moderate, and high severity, each with characteristic spectral behavior in the months that followed.
Months 1–4: The NDVI Rebound Trap
By month 3 post-fire (October 2023), NDVI values in the moderate-severity zones had already rebounded to 0.30-0.42 — a partial greening that looks, superficially, like the beginning of meaningful forest recovery. What was actually happening, confirmed by field observation from a forest ecologist colleague, was the rapid establishment of ruderal herbaceous species: fireweed (Chamerion angustifolium), pearly everlasting, and various annual grasses that colonize mineral soil and ash substrates extremely quickly after fire.
These pioneer species have high chlorophyll content per unit ground area and produce a strong NDVI signal. They're not forest. They're not ecologically equivalent to the pre-fire Douglas fir canopy they replaced. From an NDVI standpoint in month 3, the moderate-severity burn zone looked like it was healing. From a forest structure standpoint, it was bare mineral soil covered in annuals and some seed-germinating perennials.
In the high-severity zones, where the dNBR indicated complete combustion of the surface organic layer, the NDVI recovery was slower — reaching only 0.15-0.22 by month 4. This slower rebound was actually more ecologically appropriate; it reflected the slower colonization rate on the most severely disturbed soils, where the seed bank had been consumed and residual structure was minimal.
The Spectral Distinguisher: Red-Edge and SWIR
Distinguishing pioneer herbaceous recovery from conifer seedling establishment is one of the more practically useful things Sentinel-2's red-edge bands enable. The chlorophyll red-edge position (approximately 700-730nm) shifts as canopy structure and leaf morphology change. Herbaceous communities with thin, high-chlorophyll leaves produce a different red-edge response than conifer seedlings with their needle geometry and different internal cell structure.
By months 6-8, we were able to use the red-edge chlorophyll index (Band 7 / Band 5 - 1, or similar formulations) alongside NDVI to begin distinguishing zones where conifer seedling establishment was occurring from zones dominated by herbaceous cover. The conifer-establishing areas showed higher red-edge chlorophyll index values relative to NDVI compared to purely herbaceous areas — a contrast that widened over the second growing season as seedlings developed more canopy volume.
This is not a clean separation. The signal is subtle enough that we treat it as a probability surface rather than a binary classification. But it's substantially more informative than NDVI alone for distinguishing "green is back" from "forest is recovering."
The Invasive Species Complication
By month 9 (spring 2024), we identified several patches in the moderate-severity burn area that showed anomalously high and spatially uniform NDVI — a signature inconsistent with the heterogeneous, patchy recovery typical of native pioneer communities. Field verification by a state forestry ecologist determined that a portion of the burn area had been colonized by Scotch broom (Cytisus scoparius), an aggressive invasive shrub that thrives in disturbed, mineral-soil conditions in the Pacific Northwest.
Scotch broom produces a high, spectrally uniform NDVI signal that, in satellite data, looks indistinguishable from healthy native shrub recovery unless you have ground truth. Its spectral distinctiveness emerges primarily in the flowering period (distinctive yellow) and in winter when it retains green leaves while surrounding native shrubs are dormant — a phenological signal that requires multi-temporal analysis to detect reliably. A snapshot assessment of NDVI in summer would classify this as "recovered shrub layer." A time-series analysis that captures the winter phenology would flag the anomalous greenness as potentially invasive.
This is one of the core arguments for dense time series over snapshot analysis in recovery monitoring. Single-date or bi-annual assessments will miss the phenological signatures that distinguish native from invasive recovery trajectories.
Month 10–14: Divergence Between Severity Classes
By the end of the 14-month observation window (October 2024), the NDVI trajectories had diverged clearly between severity classes:
- Low-severity zones: NDVI largely returned to pre-fire levels (0.55-0.65), driven by resprouting shrub layer and surviving overstory trees. These areas show strong spectral recovery and are likely tracking meaningful ecological recovery, though canopy structure remains simplified compared to pre-fire condition.
- Moderate-severity zones: NDVI 0.35-0.50, reflecting a mix of herbaceous, invasive shrub, and early native conifer seedling establishment. The ecological trajectory here is uncertain — conifer regeneration success depends heavily on precipitation and competition pressure from invasives over the next 3-5 years.
- High-severity zones: NDVI 0.20-0.35, slow recovery reflecting limited seed source, degraded soil conditions, and in some areas, aggressive invasive annual grass cover. These are the areas requiring active management intervention (replanting, invasive control) if native forest recovery is the management goal.
What This Means for Recovery Monitoring Design
The practical takeaway from 14 months of watching this burn area is that post-fire recovery monitoring needs to be designed around the specific questions being asked, not around NDVI as a proxy for "recovery." If the question is "has photosynthetically active vegetation returned?", NDVI answers that within months. If the question is "is native forest recovering in a trajectory that will restore pre-fire ecological function within a management-relevant timeframe?", NDVI alone is insufficient and potentially misleading.
The metrics that add interpretive power in recovery contexts are: dNBR-derived severity classification (to contextualize recovery rate), red-edge chlorophyll indices (to distinguish structural vegetation types), phenological consistency with native species assemblages (requires dense time series), and texture/structural complexity indices in the NIR band (which begin to differentiate pioneer herbaceous cover from developing shrub or seedling canopy structure by 18-24 months post-fire).
We're tracking this burn area into 2025 and 2026 as conifer seedling cohorts — where they established — move through their first few years. The spectral signal at that stage will be weak relative to herbaceous background, and detection will depend on the accumulated phenological time series rather than any single observation. That's the nature of monitoring slow ecological processes with rapid satellite revisit: the value is in the archive, not the latest image.