Forest canopy equivalent water thickness and live fuel moisture mapping

Shortwave-infrared absorption features at 970 nm, 1200 nm and 1450 nm allow satellite retrieval of canopy equivalent water thickness and live fuel moisture content, key inputs to fire-risk modelling and drought-stress assessment. Accuracy depends heavily on canopy closure, sensor spectral resolution and the inversion algorithm chosen.

Sensors

PRISMA (ASI): Hyperspectral pushbroom imager, 30 m spatial resolution, 400–2500 nm range at 10 nm spectral sampling. Resolves the 970 nm and 1200 nm water absorption features explicitly, enabling direct band-depth retrieval. Revisit approximately 29 days at nadir; tasking can improve this.
Sentinel-2 MSI: 10–20 m spatial resolution depending on band. Bands 8A (865 nm) and 11 (1610 nm) bracket the 1200 nm absorption feature; band 12 (2190 nm) adds a second SWIR reference. 5-day revisit at mid-latitudes with two satellites. Broadband nature limits retrieval precision compared with hyperspectral data.
Landsat-8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined). SWIR1 (1570–1650 nm) and SWIR2 (2110–2290 nm) bands support NDWI and NDII indices. Spectral breadth is coarser than Sentinel-2 MSI band 11, but the long archive back to 1972 (Landsat 1–9 series) is unmatched for trend analysis.
MODIS MOD09 (Terra/Aqua): 500 m surface-reflectance product. Bands 5 (1230–1250 nm) and 6 (1628–1652 nm) allow NDWI computation at continental scale. Daily revisit makes it useful for tracking rapid moisture loss during drought or heatwave events, despite coarse spatial resolution that cannot resolve individual stands.

Why water absorbs where it does

Liquid water in leaf mesophyll tissue absorbs solar radiation at overtone and combination bands of the O-H stretch vibration. The principal atmospheric windows that satellites can exploit fall near 970 nm, 1200 nm and 1450 nm. At 1450 nm, atmospheric water vapour competes strongly with canopy water, making that feature difficult to use from space without precise atmospheric correction. The 970 nm and 1200 nm features are more accessible, though both are shallow relative to the strong absorption at 1450 nm, requiring careful continuum removal.

Equivalent water thickness (EWT, in g cm⁻²) is the path length of liquid water a photon traverses through the canopy. Live fuel moisture content (LFMC, in per cent dry weight) relates water mass to dry biomass. The two quantities are physically linked but not interchangeable: a dense canopy can have high EWT with moderate LFMC if leaf dry matter is also high. Radiative-transfer models such as PROSPECT and SCOPE separate these contributions, at the cost of requiring additional inputs or assumptions about leaf structure.

Indices versus inversion: what each approach concedes

The simplest retrievals use spectral indices. The Normalised Difference Water Index (NDWI, using NIR and SWIR bands) and the Normalised Difference Infrared Index (NDII) are computable from Sentinel-2 and Landsat OLI with no additional inputs. Published studies using Landsat-derived NDII report correlations with field-measured LFMC in the range r² 0.5–0.7 across shrubland and forest sites, depending on species composition and canopy closure. That is useful but not precise enough to replace field sampling in high-stakes fire-risk decisions.

Radiative-transfer inversion, typically using PROSPECT-D or PROSAIL coupled to a canopy model, extracts EWT more directly by fitting the full spectral shape rather than a two-band ratio. PRISMA's 10 nm sampling resolves the 970 nm band depth explicitly, and published validation campaigns report root-mean-square errors for EWT retrieval of roughly 0.002–0.005 g cm⁻² over closed broadleaf canopies. Open canopies are harder: soil background contributes SWIR reflectance that mimics reduced canopy water, biasing retrievals low. Masking with a canopy-cover layer or using a soil-adjusted index reduces but does not eliminate this problem.

Saturation, soil and the open-canopy problem

At high canopy water contents, SWIR reflectance approaches a floor set by multiple scattering and the sensor's noise floor. NDWI and NDII both saturate: above roughly 0.35 g cm⁻² EWT, incremental increases in canopy water produce diminishing index response. Dense tropical rainforest canopies often sit in this saturated regime, making the method most informative at the drier end of the moisture spectrum, which is precisely where fire risk begins to rise.

Sparse canopies below about 60 per cent fractional cover present the inverse problem. Bare or litter-covered soil has its own SWIR signature, and in many ecosystems the soil reflectance in SWIR1 exceeds that of stressed vegetation. Without an accurate fractional-cover decomposition, retrieved LFMC values in open woodland or savanna can be systematically low by 20–40 percentage points relative to destructive samples. The MODIS-derived Global LFMC dataset (published via the University of Maryland) addresses this partly by calibrating against a large global field database, but the 500 m pixel still mixes cover types.

Validation: what destructive sampling actually tells you

Ground truth for LFMC is destructive. Technicians clip fresh foliage, weigh it, oven-dry it at 70–105 °C for 24–48 hours, and weigh again. The ratio of water lost to dry weight, expressed as a percentage, is the reference value. Sampling is laborious, spatially sparse and captures only the top of the canopy accessible from the ground. Comparing a 20 m Sentinel-2 pixel to a handful of clipped shoots from one tree introduces mismatch that is structural, not just measurement error.

Published validation campaigns for PRISMA-derived EWT over Mediterranean shrubland and Italian forest sites report mean absolute errors of 0.003–0.008 g cm⁻² when inversion is constrained by leaf-area index from ancillary data. For Sentinel-2-derived NDWI over Australian eucalyptus woodland, published RMSE values for LFMC are typically 20–35 percentage points, which is honest: the method identifies relative spatial and temporal gradients reliably, but absolute values carry uncertainty that matters when thresholds are used to trigger operational fire-danger ratings.

Operational use: where the data fits into fire-risk workflows

National fire agencies in Australia, the United States and several Mediterranean countries already ingest satellite-derived LFMC into fire-danger rating systems alongside weather indices such as the McArthur Forest Fire Danger Index and the Canadian Fire Weather Index. The satellite layer adds spatial resolution that a network of weather stations cannot match, particularly in remote or complex terrain. The limitation is latency: surface-reflectance products from Sentinel-2 and Landsat require atmospheric correction, cloud masking and compositing before indices are stable, adding one to several days to the operational chain.

MODIS daily composites reduce latency at the cost of resolution. For landscape-scale fire-risk mapping over large regions, 500 m is often adequate. For identifying individual high-risk stands or planning prescribed-burn priorities at the stand level, 10–30 m data from Sentinel-2 or PRISMA is necessary. Satellize applies radiative-transfer inversion pipelines to open-constellation SWIR data and can incorporate PRISMA tasking for clients requiring hyperspectral retrieval, building on the same analytics infrastructure used in the Tonga crop-estimation programme.

Cloud cover is the practical ceiling on revisit frequency in humid forest zones. A nominal 5-day Sentinel-2 revisit may yield only two or three cloud-free observations per month over tropical or temperate maritime regions in summer. Compositing over 16–30 day windows smooths noise but reduces the ability to track rapid moisture loss during a heatwave, precisely when the information is most needed.

Choosing the right sensor for the question

If the question is continental trend monitoring at low cost, MODIS MOD09 with a simple NDWI time series is defensible and well-validated. If the question is stand-level fire-risk mapping in a fire-prone region with heterogeneous vegetation, Sentinel-2 NDII or a two-band PROSAIL inversion is a reasonable operational choice, accepting the 20–35 percentage-point LFMC uncertainty. If the question demands the highest retrieval accuracy over a specific area, PRISMA tasking with full radiative-transfer inversion is the current best available from open or semi-open sources, though its 29-day baseline revisit requires supplementation with Sentinel-2 between acquisitions.

No single sensor solves all three requirements simultaneously. The honest answer to a buyer asking for a single-source solution is that the architecture should be multi-sensor: MODIS for daily situational awareness, Sentinel-2 for weekly spatial detail and PRISMA for periodic calibration and high-accuracy retrieval over priority zones.

Typical figures

Spatial resolution (operational)	10–20 m (Sentinel-2 MSI), 30 m (Landsat-8/9 OLI, PRISMA), 500 m (MODIS MOD09)
Revisit frequency	5 days (Sentinel-2 two-satellite), 8 days (Landsat combined), ~29 days nadir (PRISMA, improvable with tasking), daily (MODIS)
Key spectral bands	970 nm, 1200 nm, 1450 nm water absorption features; SWIR1 ~1600 nm and SWIR2 ~2200 nm for broadband sensors
Spectral sampling (hyperspectral)	~10 nm (PRISMA 400–2500 nm range)
EWT retrieval accuracy (closed canopy, PRISMA)	RMSE ~0.003–0.008 g cm⁻² (published validation, Mediterranean/Italian forest sites)
LFMC retrieval accuracy (Sentinel-2 broadband)	RMSE typically 20–35 percentage points vs destructive sampling
Saturation threshold (EWT)	Index response diminishes above ~0.35 g cm⁻²; dense tropical canopies frequently exceed this
Minimum canopy cover for reliable retrieval	~60% fractional cover; below this, soil background correction is mandatory
Archive depth	Landsat series from 1972; Sentinel-2 from 2015; MODIS from 2000; PRISMA from 2019
Delivery formats	GeoTIFF (EWT/LFMC raster), NetCDF time series, GIS-ready polygon summaries by stand or administrative unit

Analytics Satellize can run

NDWI / NDII time-series stack	Two-band ratio applied to atmospherically corrected Sentinel-2 or Landsat surface reflectance	Monthly GeoTIFF stack with per-pixel trend and anomaly flags; suitable for integration into GIS or fire-danger dashboards
EWT map (radiative-transfer inversion)	PROSPECT-D or PROSAIL inversion constrained by ancillary LAI; applied to PRISMA or Sentinel-2 SWIR bands	Single-date or composited EWT raster (g cm⁻²) with per-pixel uncertainty estimate
Live fuel moisture content map	Empirical regression or machine-learning model trained on published field-sample databases; input features from SWIR indices	LFMC raster (% dry weight) at 10–500 m resolution depending on sensor, with confidence interval layer
Fire-risk moisture threshold alert	LFMC or NDWI pixel values compared against published species-specific critical thresholds (e.g. 80–100% for chaparral ignition risk)	Polygon alert layer identifying stands below threshold, delivered as GeoJSON or shapefile within agreed latency window
Seasonal moisture anomaly report	Z-score or percentile ranking of current NDWI against multi-year MODIS or Sentinel-2 climatology	PDF or interactive report showing spatial distribution of anomalies relative to historical range, updated at user-defined frequency
Open-canopy corrected LFMC	Fractional-cover decomposition (soil, green vegetation, non-photosynthetic vegetation) applied before LFMC retrieval to reduce soil-background bias	Corrected LFMC raster with fractional-cover layers as ancillary outputs; recommended for savanna, open woodland and shrubland clients

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.