How much open water the capital has gained and lost across nearly four decades — counted, pixel by pixel, from the JRC satellite record rather than estimated.
The finding
Across 1984–2021, the balance runs positive: this area gained 686 hectares of surface water and lost 375 hectares, for a net increase of 311 hectares. New water on this scale usually means reservoirs, tanks or quarry ponds filling faster than older water disappears — a change worth reading alongside what was built in the same window.
These are not rounded guesses. Each hectare is a tally of 30-metre pixels that the JRC Global Surface Water record classifies by how their water state changed between 1984 and 2021, counted only where they fall inside this box. In total the analysis touched 2,036 hectares of water-affected land across an area of about 1,038 km².
- Water lost or drained375 ha
- New water686 ha
- Unchanged water356 ha
- Ephemeral water619 ha
The two frames above are the whole argument in miniature. On the left is the ground itself; on the right is the same ground repainted so that every pixel which changed shows its verdict — red and amber where water was lost, blue where new water appeared, slate where it held steady, teal where it comes and goes. You are looking at four decades of hydrological history compressed into one picture, and none of it is illustrative: each coloured cell is a pixel the engine actually classified.
The change, class by class
Surface water does not simply exist or vanish; it shifts between states, and the JRC record captures those shifts in ten transition classes. Reading them in order is what turns a single net number into an actual story.
Where water was lost: lost seasonal accounts for 338 hectares — that is 16.6% of all the water-affected land in this box, counted from 5,005 individual 30-metre pixels; permanent to seasonal accounts for 25 hectares — that is 1.2% of all the water-affected land in this box, counted from 370 individual 30-metre pixels; lost permanent accounts for 12 hectares — that is 0.6% of all the water-affected land in this box, counted from 183 individual 30-metre pixels.
Where new water appeared: new seasonal accounts for 681 hectares — that is 33.5% of all the water-affected land in this box, counted from 10,090 individual 30-metre pixels; seasonal to permanent accounts for 5 hectares — that is 0.2% of all the water-affected land in this box, counted from 73 individual 30-metre pixels; new permanent accounts for 0 hectares — that is 0% of all the water-affected land in this box, counted from 4 individual 30-metre pixels.
Water that held steady: seasonal accounts for 356 hectares — that is 17.5% of all the water-affected land in this box, counted from 5,266 individual 30-metre pixels; permanent accounts for 1 hectares — that is 0% of all the water-affected land in this box, counted from 9 individual 30-metre pixels.
Ephemeral water (present only in exceptional years): ephemeral seasonal accounts for 615 hectares — that is 30.2% of all the water-affected land in this box, counted from 9,115 individual 30-metre pixels; ephemeral permanent accounts for 3 hectares — that is 0.2% of all the water-affected land in this box, counted from 48 individual 30-metre pixels.
Losses against gains: reading the balance
For every hectare of new water in this area, roughly 0.55 hectares were lost. A ratio below one means gains are outpacing losses here, though where that new water sits matters as much as how much of it there is.
The net figure is deliberately conservative: it counts a hectare of drained lake and a hectare of new pond as cancelling out, even though they are rarely equivalent on the ground — a shaded natural wetland and an open concrete reservoir do very different things for groundwater, temperature and wildlife. That is why Dyaan reports the gains and losses separately as well as netted: the components carry information the single number throws away.
The largest single loss class here is lost seasonal, at 338 hectares. Loss classes matter out of proportion to their size, because water that disappears rarely comes back on its own — once a wetland is filled or a tank is built over, the pixel is retired from the water budget for good, and the groundwater and flood buffering it provided go with it.
The largest gain class is new seasonal, at 681 hectares. New water is worth interrogating rather than celebrating: a fresh reservoir is genuine open water, but it is engineered, concentrated and often far from where the lost water was, so it does not automatically replace what a distributed network of small tanks and wetlands used to do.
What to watch next
A single read is a snapshot of a long history; the value compounds when you watch the same box over time. The transitions record ends in 2021, so the natural next step is to pair it with a recent land-cover and vegetation read to see what has happened on the ground since — whether the losing trend has continued, stalled or reversed. If this were your city, the move would be to save this area as a monitored watch, so that the next shift shows up as an alert rather than a surprise years later.
That is the difference an accessible, reproducible measurement makes. The satellites have been recording this all along; what changes is whether anyone is reading the record while there is still time to act on it. This article is one reading, published in the open — and the same tool that produced it is waiting to run the next one, over any place you choose.
How to read surface-water change
A single satellite snapshot of a lake tells you almost nothing: catch it in the monsoon and it looks healthy, catch it in May and it looks doomed. What actually matters is the transition — how a patch of ground moved between water and not-water across many years — and that is precisely what a change record measures. It watches every pixel across nearly four decades and asks a simple question of each one: did it used to hold water, and does it still?
The answers fall into a few families. Permanent water that stayed permanent is stability. Permanent water that became seasonal is a warning: the pixel is still wet part of the year, so a naive snapshot would miss it, yet a perennial water body has quietly become an intermittent one. Lost water — permanent or seasonal ground that is now dry — is the sharp end of the story, the drained tank and the filled wetland. New water can be genuinely good news or simply a new reservoir; the class cannot tell you which, only that open water now sits where none did before.
This is why Dyaan refuses to collapse everything into one cheerful or alarming number. The net figure is the headline, but the transition classes are the paragraph underneath it — and reading them in order is how you tell a city that is banking water behind new dams from one that is losing its living wetlands while the reservoir total papers over the gap.
One more habit worth keeping: always check the denominator. A loss of a few hundred hectares means one thing in a floodplain criss-crossed with water and something very different in a dry upland where every pond counts. That is why the article states the total water-affected area — so the change can be read as a share of what was there, not a number floating free of context.
Why a city’s surface water is worth counting
Open water is not scenery; it is infrastructure that a city usually does not know it owns. Lakes and wetlands recharge the groundwater that wells and borewells draw on, so when surface water shrinks, the water table underneath tends to follow — and the cost reappears as deeper wells and drier taps. The same water bodies act as buffers in a downpour, holding runoff that would otherwise arrive all at once in the streets; lose them, and ordinary rain starts producing extraordinary floods.
There is a temperature dividend too. A body of water moderates the air around it, and a belt of wetlands is a measurable brake on the urban heat island. Drain them and the neighbourhoods that lost them run hotter, year after year, in a warming climate that is already pushing the baseline up. And there is the living layer — the birds, fish and plants for which a wetland is not amenity but habitat — which does not migrate to a concrete tank when the marsh is filled.
None of this is a claim about the specific causes in this particular box; the satellite record measures change, not motive, and honesty demands that distinction. But the general physics is not in dispute. When a growing city loses surface water on the scale this analysis can measure, the consequences show up in the water table, the flood map, the thermometer and the biodiversity register — which is exactly why it is worth measuring in the first place, early and repeatedly, rather than discovering the loss only after the well runs dry.
That is the case for treating surface-water accounting as a routine civic instrument rather than a one-off study. The data is public, the method is cheap, and the change is already happening whether or not anyone is counting. A city that watches this number can act on a trend; a city that does not, learns about it from the flood.
About the record behind these numbers
The measurements here come from the Joint Research Centre’s Global Surface Water dataset, one of the more remarkable achievements in open Earth observation. It was built by re-examining more than three decades of Landsat imagery — millions of scenes — and classifying, for every 30-metre pixel of land on the planet, when it held water and how that changed over time. The method was published in Nature in 2016 and the data is free for anyone to use, which is precisely why Dyaan builds on it: a public, peer-reviewed, global record is the opposite of a proprietary black box.
What makes the transitions layer used here so useful is that it does the hard temporal reasoning for you. Rather than handing over a stack of yearly water masks and leaving you to difference them, it distils the whole 1984–2021 history of each pixel into a single verdict — permanent, seasonal, new, lost, and the shifts between them. Dyaan’s only job is to count those verdicts honestly inside your area and to render them faithfully, which is exactly what the map above does.
Like every dataset it has limits, stated plainly: it sees open water, not water hidden under canopy or in narrow channels below its 30-metre resolution; it ends in 2021, so the most recent changes are not yet in it; and cloud, ice and terrain shadow can complicate the classification in specific places. None of that undermines the headline — it simply defines the edges of what the headline can claim, which is a distinction Dyaan is built to respect.
Reading this alongside the other layers
No single analysis is the whole truth of a place, and surface water is most revealing when it is read next to its neighbours. Lay this change map beside a land-cover read of the same box and the losses acquire a cause you can see: water that became built-up, cropland that crept to the shoreline, green that turned grey. Add a flood-exposure screen and the picture sharpens further — because the low ground that used to hold water is exactly the ground that will flood once the water is gone and the rain still comes.
This is the logic of the whole Dyaan toolkit: each engine answers one question honestly, and the answers compound. Surface water tells you what changed; land cover tells you what replaced it; elevation tells you what the consequence will be. Run in sequence over the same area, they turn a single striking number into an argument you can actually defend — and, just as importantly, one that anyone can re-run and check.
How Dyaan measured this
This article was not written from a press release or a secondary summary. It was generated directly from a live read of the underlying satellite archive, performed at publication time by the same engine that powers the Dyaan workspace. There is no human in the loop inventing a number to fit a narrative — the narrative is assembled around numbers the machine measured.
The data source is JRC Global Surface Water v1.4 — water transitions (Pekel et al., Nature 2016). Dyaan does not download the whole archive; it issues ranged HTTP requests that pull only the bytes covering this bounding box, decodes the relevant raster tiles in memory, and counts the pixels that fall inside the area — 1,537,280 of them for this read, moving roughly 43 KB over the wire. No API key is used, and nothing is cached and then re-presented as fresh: a read either succeeds against the live archive or the article records that it could not.
Because the input is a fixed, published dataset and the area is a fixed bounding box, this analysis is reproducible. Anyone can open the same area in the Dyaan workspace, run the same analysis, and get the same figures — the definition of a measurement rather than an opinion. That reproducibility is the whole point: it is what separates an intelligence product you can audit from a chart you have to trust.
What this analysis is — and what it is not
Surface-water change records where open water appeared and disappeared between 1984 and 2021 — it does not explain why, cannot tell a new reservoir from a flood, and at 30 m resolution it will miss a pond narrower than a cricket pitch. Water that shifted from permanent to seasonal is a real change worth watching even though the pixel is still "wet" for part of the year.
Dyaan’s design rule is to state that footprint plainly rather than paper over it. A number here is only ever as good as the sensor and the method behind it, and both are named. Where the data cannot answer a question, the honest answer is that it cannot — and the tool will say so rather than fill the gap with something that merely looks authoritative.
Reproduce this yourself in about a minute
Every figure above is checkable, and checking it is the fastest way to trust it:
- Open the Dyaan workspace and sign in (the free tier is enough for this).
- Search for Delhi, India, or draw a rectangle over the same area on the map.
- Choose the Surface water analysis from the panel and press Run.
- Read the headline numbers, open the table for the full breakdown, and export a CSV or PDF if you want to keep it.
You are not limited to the places Dyaan writes about. The same engine runs over any lake, city, forest or coastline on Earth — so if there is a place you actually care about, that is the one to point it at.
A short glossary
- Permanent water — a pixel that held water essentially all year, every year, across the observation period.
- Seasonal water — a pixel wet for only part of the year — a monsoon tank, a floodplain, an irrigation pond.
- Transition — the change in a pixel’s water state between the start and end of the record — the thing this analysis actually counts.
- Lost water — a pixel that was water and no longer is: drained, filled, built over or dried out.
- Hectare — 10,000 square metres — about the playing area of a large sports stadium.
Why this is written by a machine — and why that helps
This article was assembled automatically, and that is a feature rather than an apology. A person writing to a deadline is tempted to round a number toward the story they already have in mind; an engine has no story to protect. It reads the archive, counts what is there, and lays the figures out in the same structure every time, for every place, whether the result is dramatic or dull. The consistency is the point: you can compare this week’s reading with last month’s, or this city with the next, and know the only thing that changed is the ground, not the method or the mood of the author.
It also means the work scales in a way human analysis never could. There are millions of lakes, forests and floodplains on Earth and a handful of people who study any given one. An automated pipeline that reads public data honestly can, in principle, keep an eye on all of them — and surface the ones that are changing fastest to the people who would otherwise never hear about them until it was too late. That is the ambition behind these field notes: not to replace the expert, but to make sure the expert, the official and the ordinary resident all get to see the same measured record, at the same time, for free.
Every safeguard that makes the tool trustworthy is applied here too. The numbers are counted, never estimated; the limits are stated, never hidden; and if a read fails, the failure is recorded rather than papered over with a plausible-looking figure. An automated writer is only as honest as the rules it follows, and these are the rules.
The bottom line
This is an original Dyaan analysis. The words, the framing and the pictures are Dyaan’s own; the only thing borrowed is the raw public dataset it reads — JRC Global Surface Water v1.4 — water transitions (Pekel et al., Nature 2016) — which is an input, not an article. Nothing here is copied from another publication, and every number can be reproduced in the tool.
If a place matters to you — a lake near your home, a forest you grew up beside, a valley that floods every year — you do not have to wait for Dyaan to write about it. Point the tool at it yourself and read the record for that exact ground. Open the workspace → · How every number is measured →