Measuring Leaf Area in the Sundarbans: How Salinity Is Quietly Erasing Mangrove Functional Diversity

The study in one paragraph

The paper is "Trait-based evidence of salinity-induced functional diversity loss in mangroves: Implications for ecosystem resilience" by Md Rezaul Karim and colleagues, published open access in Ecological Applications (2026, volume 36, issue 1, article e70191; doi:10.1002/eap.70191). It is a genuinely international effort — the author list spans institutions in Bangladesh, Canada, China, the United States, the United Kingdom and Australia, with funding from the SUST Research Center, the National Geographic Society, the Asia-Pacific Network for Global Change Research and the British Ecological Society. The question it asks is deceptively simple: what happens to mangrove functional diversity when the soil gets saltier?

Note the phrasing of the question. It is not "which species are present" — that is a species inventory, and the Sundarbans have been inventoried many times. It is a harder question: what do the plants in this place actually do? What traits do their leaves carry, how do those traits adjust under salt stress, and does the community as a whole become functionally richer or functionally narrower?

Why the Sundarbans

The Sundarbans is the largest continuous mangrove forest on Earth, straddling the India–Bangladesh border at the northern edge of the Bay of Bengal. The Bangladeshi section covers roughly 599,330 hectares — about 62% of the forest — with the remaining 426,300 hectares in West Bengal, India. It is threaded by around 200 islands and more than 400 tidal rivers and canals, which is precisely what makes it useful for this kind of study: that hydrological maze generates a real, continuous salinity gradient rather than a handful of artificial treatment levels.

Ecologically, the stakes are hard to overstate. The forest hosts 334 plant species, including 50 mangrove species, and supports 50 mammal species, 320 bird species, 53 reptile species, 11 amphibian species, 177 fish species and 873 invertebrate species. The Bangladeshi Sundarbans alone account for 45% of the country's mammal species and 42% of its birds.

It is also infrastructure. The Sundarbans absorbed the first blow of cyclones Sidr (2007), Aila (2009) and Amphan (2020) before those storms reached people. Coastal protection, carbon storage and fisheries all rest on the forest continuing to function — not merely continuing to exist as a green shape on a satellite image.

Salinity as an ecological filter

Salt does not kill a mangrove forest the way a chainsaw does. It filters it. Soil salinity shapes plant physiology by altering osmotic potential, interfering with nutrient uptake and introducing ion toxicity. Species that cannot manage those pressures thin out; species that can, spread. Ecologists call this abiotic filtering — the environment selecting which trait combinations are viable.

Salinity in the Sundarbans is intensifying for reasons that are largely upstream and offshore: reduced freshwater inflow, altered hydrological regimes and sea-level rise. The researchers sampled 62 georeferenced plots of 20 × 20 m across all 54 management compartments of the Bangladeshi Sundarbans, keeping every plot at least 120 m from any water channel to avoid edge effects. Fieldwork ran from October to March in the 2021–2022 dry season — stable hydrology, mature foliage, fewer cyclones and insects to confound the signal.

Salinity itself was measured as electrical conductivity (EC), not estimated from a map. Soil came from two subplots per plot, down to 100 cm, split into five depth layers (0–10, 11–20, 21–30, 31–50 and 51–100 cm), composited per layer and read on a portable pH/EC/TDS meter at a 1:2 soil-to-distilled-water ratio. The five layer values were averaged into one EC figure per plot, in millisiemens per centimetre. Elevation, soil pH, bulk density and total nitrogen were recorded alongside — and, as it turned out, none of them explained much.

The eight foliar traits — and what each one tells you

Trait-based ecology replaces the question "who lives here?" with "what strategies live here?" A leaf is a good place to ask, because leaves sit exactly where the plant negotiates with its environment: they trade water for carbon, and salinity makes that trade brutal. The study measured eight traits, from 3–10 mature leaves per species, collected from sun-exposed positions in the mid- to lower canopy.

• Leaf area (LA, cm²) — the total surface a leaf presents. Bigger leaves capture more light but lose more water.
• Specific leaf area (SLA, cm²/g) — area per unit dry mass. High SLA means thin, cheap, fast-return leaves; low SLA means dense, expensive, durable ones.
• Leaf dry matter content (LDMC, mg/mg) — how much of the leaf is structural dry matter rather than water. A proxy for tissue density and construction cost.
• Total chlorophyll (mg/g) — photosynthetic machinery, and a direct readout of stress.
• Stomatal density (SD, stomata/mm²) — the number of gas-exchange pores. Fewer stomata means less transpiration, and less carbon gain.
• Leaf shape index (LSI) — dimensionless; how elongated versus rounded the blade is.
• Leaf succulence (LS, g/cm²) — water stored per unit area. The classic salt-dilution strategy.
• Leaf carbon content (LC, % of dry mass) — investment in structural compounds such as lignin.

Seventeen tree species were identified across the plots, each associated with a low-, mid- or high-saline zone. Some are familiar names in Bengali forestry: Sundari (Heritiera fomes), Gewa (Excoecaria agallocha), Goran (Ceriops decandra), Keora (Sonneratia apetala), Kala Bean (Avicennia officinalis).

Traits were not left at the species level. Each was converted into a community-weighted mean (CWM) per plot: the species-level trait value weighted by that species' relative abundance in the plot. A CWM answers "what is the average leaf strategy in this patch of forest," which is a different — and more ecosystem-relevant — question than "what is the average leaf of this species."

Why leaf area is the workhorse trait

Of the eight traits, leaf area is the one that does the most quiet work. It is a trait in its own right — smaller leaves are a documented response to water and salt stress. But it is also an input to other traits in the list. Specific leaf area is leaf area divided by dry mass. Leaf succulence is water content per unit leaf area. Leaf shape index depends on the blade's dimensions. Get leaf area wrong and the error propagates into three of the eight traits, quietly, without ever announcing itself.

Leaf area is not just one of the eight traits — it is the denominator of several. Its measurement error sets a floor on the quality of the entire trait matrix.

That is a demanding requirement for a forest reached by boat, where plots sit 120 m inside the treeline and the nearest laboratory is a refrigerated launch and an ice-pack journey away. Fresh area has to be measured before the leaf loses water, which is exactly the constraint that makes classical desktop scanning awkward here.

How leaf area was measured: the Petiole app in the field

The methods section is refreshingly blunt about it. Leaves were sealed in sterile zip-lock bags immediately after collection, chilled on the launch-boat, and — in the authors' own words — "scanned or measured with the Petiole mobile app to determine fresh area." Green mass was then weighed, the leaves were oven-dried at 80 °C for 48 hours for dry mass, and LDMC followed from the fresh-to-dry ratio.

What the app does is straightforward computer vision: segment the leaf from its background, then convert pixels to real square centimetres using a calibration reference in the frame. That second half is what separates a measurement from a picture. Without a known-size object in shot, an image gives you proportions; with one, it gives you cm² you can put in a table. We have written separately on whether you need a calibration plate and on how to photograph leaves for accurate leaf area measurement in the field — both questions that decide whether a trait dataset is comparable across 59 plots or merely internally consistent within one.

The other reason this matters at scale is arithmetic. Eight traits × 17 species × 3–10 leaves per species per plot, across 62 plots, is a large number of leaves to move through a measurement step one at a time. Batch measurement — many leaves in a frame, each segmented and reported separately — is the difference between a trait study that finishes and one that quietly shrinks its sample size.

Petiole Pro has now been referenced as a data-collection tool in more than 100 research papers by researchers in 30 countries, and this Sundarbans study is a good illustration of why: the app earns its place not by being clever but by working in conditions where a benchtop planimeter cannot go. You can read more about the tool on the Petiole Pro product page.

What the data showed

Functional diversity was quantified with the dbFD() function from the FD package in R, using Gower distances across the eight standardised traits with a Cailliez correction. Plots holding fewer than three functionally distinct species were dropped, leaving 59 plots for the diversity indices. Analysis ran in R 4.4.1 and Python 3.11.

Functional diversity: one index moved

Four functional diversity indices were tested against salinity, and only one responded. Rao's quadratic entropy — which measures the average trait dissimilarity between pairs of species, weighted by their abundance — declined significantly as EC rose (slope = −0.45, R² = 0.09, p < 0.05). Functional richness, evenness and divergence showed essentially nothing (all R² < 0.01; p = 0.92, 0.70 and 0.64 respectively).

That split is the interesting part, not a disappointment. Functional richness measures the volume of trait space a community occupies; Rao's Q measures how different the species within it actually are, accounting for which ones dominate. The volume stayed put while the internal dissimilarity collapsed — which is precisely the signature you would expect if salt-tolerant species with near-identical strategies were taking over without shrinking the outer envelope of the community.

Spatially, the lowest Rao's Q values clustered in the western Sundarbans, where reduced freshwater inflow drives EC highest, and where salt-tolerant generalists such as E. agallocha and C. decandra dominate.

Foliar traits: all eight shifted

Where the diversity indices were selective, the traits were unanimous. Every one of the eight community-weighted means changed significantly along the salinity gradient, in a directionally consistent way.

Read as a story rather than a table, this is a forest shifting toward water storage and osmotic balance at the cost of everything else. Fewer stomata means less transpiration — and less gas exchange. Less chlorophyll means reduced photosynthetic potential. Smaller, rounder leaves cut the surface area losing water. Rising succulence dilutes internal salt.

The apparent paradox — specific leaf area rising while leaf dry matter content falls — resolves once you see it as succulence. The leaves are not becoming cheap and fast-growing; they are becoming waterlogged. Water is displacing structural dry matter, so area per unit dry mass climbs even as the leaf becomes less, not more, acquisitive.

The multivariate picture

A principal components analysis of the community-weighted traits confirmed that these are not eight independent responses but one coordinated syndrome. The first two components explained 83.6% of total variance, with PC1 alone accounting for 70.1% — and PC1 aligned with the salinity gradient. Leaf dry matter content (loading 0.41), leaf carbon (0.37), leaf shape index (0.37) and stomatal density (0.36) pulled one way; specific leaf area (−0.37) and succulence (−0.33) pulled the other. Separately, an NMDS ordination of species composition (stress = 0.293) showed high-salinity plots (EC > 6 mS/cm) and low-salinity plots (EC < 3 mS/cm) separating along the first axis, with no discrete clusters — species turnover happens along a continuum, not at a boundary.

Trait convergence: survival is not resilience

Trait convergence is what happens when stress narrows the range of workable strategies. Under pressure, nature selects. Plants carrying similar salt-tolerant traits come to dominate, and the community's functional variety contracts around them. From the outside nothing looks wrong — there is still a canopy, still green, still a forest.

The problem is what convergence costs in options. Functional diversity is, in effect, an ecosystem's portfolio of responses to a future it has not encountered yet. When many species converge on the same strategy, the portfolio concentrates. A community optimised entirely for salt tolerance has fewer ways to answer the next stress — accelerating sea-level rise, further freshwater reduction, a new pathogen, a shifted cyclone regime.

Survival is not the same as resilience. A converged community may persist under today's salinity and still be less able to absorb tomorrow's shock, because the traits that would have answered it are no longer in the room.

There is a live example already. Heritiera fomes — Sundari, the tree the forest is named for — is a low-salinity species, and it is already declining in areas where salinity has risen, compounded by top-dying disease. Losing it does not just subtract a name from a species list; it subtracts a distinct trait combination from the community.

The overlooked factor: species abundance

One of the study's more useful contributions is methodological, and it is a caution to anyone doing trait-based work. Functional diversity metrics are sensitive to community structure, not just to the environment. Species richness can inflate functional richness simply by adding trait combinations; dominance can skew community-weighted means. Attribute those movements to salinity without controlling for them and you have misread your own data.

So the authors controlled for them — and found that biotic structure was doing real work independently of salt:

• Species abundance reduced Rao's Q (slope = −0.05, R² = 0.20, p = 0.001) — a stronger effect than salinity's own.
• Abundance reduced functional evenness (slope = −0.005, R² = 0.16, p = 0.004).
• Abundance raised functional divergence (slope = 0.002, R² = 0.12, p = 0.014).
• Species richness predicted functional richness (slope = 1.09, R² = 0.29, p < 0.001) — trait space expands with taxonomic diversity, as expected.

The interpretation: the problem is not only salt, but how salt reshapes dominance. High abundance of a few stress-tolerant species compresses trait dissimilarity and evenness on its own. Meanwhile, edaphic variables that might plausibly have mattered — soil pH, bulk density, total nitrogen, elevation — explained almost nothing. Salinity overrides them.

What this means for conservation

The management implication follows directly from the mechanism, and it is unusually concrete for a trait paper. If salinity is the dominant filter and freshwater inflow is what holds it back, then conservation priorities are hydrological before they are botanical:

1. Protect and restore low-salinity zones. These retain the higher functional and structural diversity. They are where the portfolio still exists.
2. Manage freshwater inflow. Regulated inflow and sediment control address the driver rather than the symptom.
3. Target species with distinct trait combinations. Assisted regeneration and planting of H. fomes and other freshwater-associated species in low-salinity zones — where their trait profiles actually fit — rather than planting whatever survives everywhere.
4. Use trait-based indicators as an early-warning system. Trait convergence shows up in leaf measurements before it shows up in a species inventory, and long before it shows up in canopy cover.

That last point is the one worth dwelling on. A species inventory tells you what you have already lost. Trait data tells you what you are in the process of losing — which is the only kind of information that management can still act on.

Scaling trait ecology with technology

The authors close by pointing at the obvious bottleneck. Plot-based trait surveys are rigorous and slow: 62 plots took a full dry season, by boat, in a forest of over a million hectares. They call for pairing hydrological management with remote sensing of key foliar traits — specific leaf area and chlorophyll in particular — to track mangrove health at landscape scale.

This is where computer vision, spectral imaging and AI-assisted trait monitoring have something real to offer: detecting shifts in canopy traits, greenness, chlorophyll and stress signals over areas no field team can walk. But the framing matters. These tools are not a replacement for field ecology — they are a way to scale it. Every remote-sensing model of a foliar trait is trained and validated against ground measurements of that trait, which is exactly what a boat, a zip-lock bag and a calibrated leaf photo produce.

The same logic runs through the rest of our work: quantification beats detection. It is why we argue for measuring necrotic leaf damage rather than merely detecting disease presence, and why on-device processing matters when the study site has no signal — a theme we explore in why the future of AgTech is edge AI. A mangrove plot 120 m inside the treeline is the purest possible statement of that requirement.