What the atlas shows

Three views, tracked month by month for every region on Earth, from January 2015 onward.

Water Deficit — where is water short? How dry a region is now, against its own history: rainfall balance, soil water, vegetation health, groundwater, plus how often drought recurs.
Actionable Shortage — how much of that shortage could be closed? The rain missing from the last six months, set against the water overhead in clouds that could plausibly respond to rain enhancement, counted in the hours a program could act, at a conversion rate you choose.
Water Resource — how much water is passing overhead? The atmospheric water arriving above a region, the part that leaves without raining, and the rain that fell.

A region is a country, a river basin, or a ~45 km hexagon, switchable at any time and each computed separately (Regions).

How to read a value

Deficit layers say “unusual for here.” Every value is a rank against that region’s own 30-year record, never against another region. 90 means drier than 90 % of the months in its history.
Resource layers say “this much water.” Millimetres of depth over the region, comparable regardless of size. A panel toggle converts it to a whole-region total in billions of m³ (1 bn m³ typically sustains ~2 million inhabitants for a year, though consumption varies a lot by region).

Vocabulary

Anomaly, standard deviation (sd) — distance from normal, in the region’s own year-to-year swing. About one year in six is drier than −1 sd; −2 sd is genuinely rare.
Percentile — rank against a region’s own history, 0–100. On the dryness layers, 0 is the wettest and 100 the driest.
Atmospheric water — water vapour, cloud droplets and ice moving above the region. Most of it never falls here.
Enhancement-addressable atmospheric water — the share of the atmospheric water overhead that would plausibly respond to rain enhancement: what the technique can act on, not what it would yield.
Enhancement-actionable atmospheric water — the addressable water counted only in the hours a mechanism was present (convection, or forced lift over a barrier) and a drone could fly. The Δ conversion rate you set applies to this water.
Trailing six months — droughts accumulate; the shortfall and the rainfall balance read the six months ending in the selected month.
mm and bn m³ — 1 mm of depth over 1 km² is 1,000 m³ of water. The unit toggle switches between the two (1 bn m³ is a cube of 1x1x1 km).

What it does not claim

Not a forecast. Every value describes a month that has already happened.
The Actionable view is a scenario, not a yield. Δ is set by you, not measured.
No data is interpolated. Where a source has not measured, the region is marked no-data (Availability gating).

The three bands

Both shortfall views sort every measured region into one of three exclusive bands, each on its own scale.

Band	What it means	Water Deficit → Shortfalls	Actionable Shortage
In drought	The last six months came in below the drought line — one standard deviation under the 1991–2020 average.	Deep red, by the rain missing to clear that line.	One green scale for both bands: the share of the gap to normal that enhancement could add, so any two regions compare directly.
Below normal	Above the drought line, but short of an average six months.	Light grey, by the rain missing to reach normal.
At or above normal	The six-month total is at the 1991–2020 average or better.	Labelled “no shortfall” — a finding, not a gap in the data.

Live cloud overlay. The one non-monthly layer: NASA GIBS true-colour imagery, one global pass per day, available within about three hours. Decorative, not an input: today’s clouds.

Prevailing wind overlay. One arrow per hexagon for where the air at ~3 km (700 hPa, the level seeding material travels at) is going, averaged over 2015–2025; longer is faster, fainter means the direction reverses between seasons.

Water Deficit

Land only. Each component is a dryness percentile or a standardised anomaly against a fixed 1991–2020 baseline (Baselines), so a wet country and a dry one can both read “extreme”.

Drought severity headline

severity = max(rainfall balance, soil dryness, vegetation stress, groundwater) → 0–100

The worst of the four dryness percentiles, taken at region level.
Reading. 70 → at least one dimension is drier than 70 % of this region’s history.
Unit. Percentile 0–100, higher = drier.

Layer	Captures · one reading	Source	Computation	Unit · orientation
6-month rainfall balance	Six-month precipitation minus evaporative demand (P−PET) anomaly, full-sign. `−1.5` → 1.5 sd drier than normal here; `+1.0` → 1 sd wetter.	Copernicus GDO SPEI-6 (ERA5-based), var `spe06`, 0.25°, 1991 →	SPEI-6 z-score → area-weighted region mean, published as the raw full-sign z-score; dryness-percentiled only where it enters the severity max.	std deviations; negative = dry (red), positive = wet (blue)
Soil dryness	Root-zone (0–100 cm) soil-water dryness ranked against the region’s history. `90` → drier than 90 % of history.	ERA5-Land monthly means, `swvl1–3`, 0.1°, 1950 → (fallback: GDO Ensemble SMA, record ends 2024-07)	Thickness-weighted soil water (7/21/72 cm) → standardised anomaly vs the same calendar month 1991–2020 → region mean → dryness percentile.	percentile, higher = drier
Vegetation stress	How much less sunlight vegetation absorbs than normal. `85` → greenness deficit worse than 85 % of history.	Copernicus GDO fAPAR anomaly (VIIRS, MODIS before), var `fpanv`, 0.1°, ~2001 →	Source-deseasonalised anomaly, dekadal → monthly → region mean → dryness percentile vs the ~2001–2020 baseline.	percentile, higher = drier
Groundwater depletion	Groundwater-storage drought, from satellite gravity assimilated into a land-surface model. `80` → drier than 80 % of the 1948–2012 reference.	NASA GRACE-DA `GRACEDADM_CLSM025GL_7D` v3.0, var `gws_inst`, 0.25°, weekly 2003-02 → (2017-07–2018-05 missing; fallback: GDO GRACE TWS, 1°)	Weekly wetness percentile → monthly mean → region mean → `100 − wetness`, used directly (no re-percentile). The pre-2002 part of its reference is a model reconstruction, not gravity observation.	percentile, higher = drier
Drought frequency	Chronic backdrop. `0.15` → 15 % of the last 10 years were in drought here.	derived from rainfall balance (SPEI-6)	Share of the trailing 120 months up to the viewed month with SPEI-6 ≤ −1, from a continuous series that advances with the month.	share of months; context, not in severity
Baseline water stress	Structural backdrop: how heavily human demand draws on the renewable supply in a normal year. `3.4` → high stress.	WRI Aqueduct 4.0 (released Aug. 2023), layer `bws`, 1979–2019, HydroBASINS level-6 polygons	Withdrawals as a share of renewable supply, area-weighted onto both region families. Static.	score 0–5; low / low-medium / medium-high / high / extremely high, breaking at 1 / 2 / 3 / 4

Two rows are backdrop, not weather. Drought frequency and Baseline water stress sit outside the severity score and do not move with the month.

Actionable Shortage

The one view where Deficit and Resource combine (they are measured independently): a shortage on the ground divided by a supply in the air, both in millimetres, over the same six months.

coverage = Δ × Σ over the same six months of (enhancement-addressable water × workable share of that calendar month) ÷ the rain that is missing Δ from the slider, 0–8 %, default 4 %

Reading at Δ = 4 %. 0.20 → a program acting through those six months, in the hours it could, would have closed 20 % of the gap. At Δ = 0 nothing lights up.
Actionable, not merely addressable. Each month’s addressable water is counted in its workable share (Workable hours) before the six months are summed, giving the enhancement-actionable water. A build without the workable climatology counts every hour and says so.
One bar at a time, and you choose it. “Measured against” in the panel sets “the rain that is missing”: the gap to the drought line (the default) or to normal rainfall. Every share on the view, up to the worldwide figures, is on that bar and names it. Under the drought bar, regions short of normal but not in drought are grey by their gap to normal, and past 100 % the hover quotes the other bar (“clears the drought line, and covers 40 % of the way to normal”).
Δ scales the result linearly: magnitude, not ranking.
Nothing is carried from one place, or one month, to another. Each region and month is first held to its own gap. So a country can never read higher than the regions it is made of: if each could have closed 30 % of its gap, the country reads 30 %, not 100 %. Likewise month to period and country to continent to world.
Unit. Ratio, shown as %.

Indicator	Captures · one reading	Source	Computation	Unit · orientation
6-month shortfalls	The rain missing from the past six months to have kept the region out of drought. `40` → 40 mm more rain would have cleared the drought line.	ERA5-Land monthly precipitation, 0.1°, 1950 →	`D = max(0, (μ − σ) − P6)`, where P6 is the trailing 6-month regional total and μ / σ are its 1991–2020 mean and standard deviation for windows ending in the same calendar month.	mm, higher = worse
Gap to normal rainfall	The larger, milder gap: rain missing to reach an average six months. Always ≥ the shortfall.	same as above	`D_normal = max(0, μ − P6)` — the same window and baseline, without the `− σ` drought-line term.	mm; the map scale for regions below normal, and the dashed curve in the region panel
Enhancement-addressable atmospheric water	The addressable share as a depth of water, for one month. The panel prints the share underneath it.	derived: outflow depth × addressable share	Headline resource layer × the monthly addressable share. The coverage headline uses this row summed over six months.	equiv. mm/month
Enhancement-actionable atmospheric water region panel	The addressable water counted only in the hours a mechanism was present and a drone could fly. The panel prints its share of the addressable water (“4 % of addressable”) and two pills: the mechanism the month rests on — convective mode, orographic mode, or no mechanism — and the share of the water that passed in daylight (“55 % in daylight”). The mechanism pill’s title carries the three hour-shares (convective presence, orographic presence, flyable).	ECMWF ERA5 3-hourly fields, 1°, 2015–2024: CAPE and CIN, low and medium cloud cover, convective precipitation (hourly), 10 m wind gust, 2 m temperature, precipitation, the 700 hPa wind, and the vertically integrated water-vapour flux. Copernicus DEM GLO-30 for the terrain.	Per 3-hourly sample and grid cell, three tests. Convective presence, either of two: unstable air under a moderate lid with some cloud — CAPE ≥ 150 J/kg, CIN ≤ 200 J/kg and low or medium cloud cover ≥ 10 % (no CIN reading fails this one) — or a light convective shower — at least 0.1 mm of convective precipitation in the three hours around the sample, with CAPE ≥ 50 J/kg. Orographic presence: the 700 hPa wind’s component toward the crest ≥ 5 m/s, only where the terrain gate passes — at least 800 m of forced lift, the crest (95th percentile of elevation within 24 km) minus the lowest ground 10–70 km upwind of the cold-season 700 hPa mean flow; zero elsewhere. Flyable: gust below 12 m/s and outside the icing envelope (precipitating air between −10 °C and +2 °C). A sample is workable when either mechanism is present and the drone can fly; the month’s workable share is the share of the month’s passing water vapour those samples carried (weighted by vapour flux, not clock time), per grid cell, then averaged over the region. Addressable water × its calendar month’s workable share = actionable water. The daylight share counts only the workable samples between 06:00 and 18:00 local solar time, over the same passing water; drones are assumed to fly day and night.	equiv. mm/month; the share underneath as %
Enhancement-addressable atmospheric water — share	The share of the month’s passing atmospheric water in cloud regimes rain enhancement can act on. `0.30` → 30 % was enhancement-addressable. Not a map layer: it reads under the addressable depth.	ECMWF ERA5 monthly means — cloud liquid water and temperature on 18 pressure levels from 1000 to 400 hPa, with column liquid totals and surface pressure; 0.25°, 2015 →	The share of the region’s cloud liquid water at a temperature ice-nucleating seeding can act on, discounted where the column is too thin to target: `f = φ × min(1, tclw / 12 g/m²)`, with `φ` the liquid-mass-weighted membership `Σ clwc·w(T)·Δp/g ÷ tclw` through the profile, area-meaned over the region. Months ERA5 has not published fall back to the calendar-month climatology. Eligibility is glaciogenic: `w(T)` is a membership of the supercooled-liquid regime — zero outside −25 °C…0 °C, full between −20 °C and −10 °C — so warm-cloud hygroscopic seeding, with a materially weaker evidence base, is not counted. Empty where the column carries no liquid.	ratio, shown as %

Shortfall, and the gap “to normal”

The two rows differ by exactly one standard deviation of the region’s six-month rainfall. The gap between the solid and dashed curves is the safety margin; narrow means a low-variability climate where a small rainfall loss is already an emergency.

Why six months, and what the ratio assumes

The window is matched on both sides. The numerator is the enhancement-actionable water of the same trailing six months as the shortfall; the panel row shows one month (equiv. mm/month), roughly a sixth of it.
Seasons are compared to like seasons: μ and σ are for windows ending in the same calendar month, so a monsoon dry season is measured against other dry seasons.
Incomplete windows show nothing rather than a misleadingly low number.
The 0–8 % range is an assumption, anchored on what trials report. The implied enhancement, Δ × actionable water ÷ the rain that fell, is set to land where randomised programs typically land: at Δ = 4 % the median region-month reads about +5 % of its natural rainfall and the upper quartile about +13 %, inside the 5–15 % band those programs report; at the top of the range the median is near +10 %. The region panel prints the implied ”% of natural rainfall”.
Two departures from a full “back to normal” accounting. The shortfall is precipitation-only (extra rain does not replace a hot drought’s inflated evaporative demand mm-for-mm), and the bar is exiting drought, not restoration to the mean.
Caveat on the addressable share. A monthly cloud-property fraction is applied to the month’s flux, assuming the flux is spread evenly across cloudy and clear hours. The workable share is counted sample by sample on the 3-hourly record, but is a ten-year climatology per calendar month.
Convective presence is a model reading. Instability, the lid, cloud cover and showers all come from the same reanalysis at 1°; the test has not been checked against satellite or lightning observations. In the wet tropics it is met in most flyable hours of the wet season.
The eligible share is modelled, and the amount of cloud water is uncertain to a factor of two. Where the liquid sits and how cold it is comes from a weather model, not an instrument; reanalyses turn supercooled liquid into ice more readily than the atmosphere does, so this share is a low estimate in mixed-phase conditions, lowest over the Southern Ocean. Independently, the model and the MODIS satellite retrieval disagree on how much cloud liquid water there is: within about 15 % across the tropics, three to four times across the northern mid-latitudes in winter, global median ratio near 2. That envelope sits under every figure derived from it.

The panel lists the ratio’s ingredients as rows above the headline, so the share can be checked against its inputs.

Seasonal timing — when in the year the rain is missing, and how much of it was actionable then

A per-region profile, not a layer: in which months is the rain missing, and how much could a programme have added then. Twelve bars, one per calendar month:

bar height = the rain typically missing over the six months ending in that calendar month (mean over every year of the record, zeros included), against a millimetre scale · bar colour = Σ_years min(Δ × the actionable water of those six months, that year’s gap) ÷ Σ_years that year’s gap

Why the height is an amount and not a frequency. The drought line is defined per calendar month against that month’s own history, so the frequency of being below it is about one year in six in every month, everywhere, by construction. The rain missing does carry a season: largest where the six-month total is most variable, the end of the wet season. The strip uses the bar the panel is set to.
The colour is the map’s own share, per calendar month. Δ moves the colours, never the heights. Without workable hours in the build the colour is the addressable water alone, and the legend says so.
Each year is held to its own gap. A year that had more actionable water than gap closes that year and no more, so the share is a sum of capped years rather than a ratio of two averages — the same rule the map applies across a region’s months and a country’s regions. 100 % therefore means every year of the record was closed, and it is a value the arithmetic can reach rather than a ceiling imposed on it. A year whose six-month window is incomplete counts no water and keeps its gap.
The hover gives the numbers behind a bar: the missing rain in millimetres, how many years in ten that month came in short, and the millimetres a programme could have added.
The mechanism sits on the actionable water row (above), not on the strip.
Every family, on the full record. Hexagons are profiled from a full 2015-onward record kept beside the store, not the twelve months the map scrubs.
Eleven or twelve years is thin. A calendar month with fewer than eight years is left out, and a region under the no-resource floor gets no profile.

Water Resource

Observed and reanalysis quantities only: no percentiles, no baselines, no ranking against history.

Layer	Captures · one reading	Source	Computation	Unit · orientation
Leaving without raining headline	All atmospheric water crossing back out over the boundary during the month. `15` → as liquid, it would cover the region 15 mm deep.	ERA5 total-water flux — vapour (`viwve`, `viwvn`), cloud liquid (`vilwe`, `vilwn`), cloud ice (`viiwe`, `viiwn`), 0.25°, 1940 →	`∮ max(+F·n̂, 0) dl ÷ area` — the outward part of the flux integrated around the region boundary, divided by area.	equiv. mm/month (bn m³ under the unit toggle)
Arriving overhead	All atmospheric water entering the region across its boundary.	same ERA5 flux field	`∮ max(−F·n̂, 0) dl ÷ area` — the inward part of the same boundary integral.	equiv. mm/month (bn m³ under the unit toggle)
Monthly rainfall	Precipitation that actually reached the ground. Context for the two flux layers.	NASA GPM IMERG monthly, `GPM_3IMERGM` v07, var `precipitation`, 0.1°, 2000-06 →	Rate (mm/hr) × 24 × days in month → area-weighted region mean.	mm/month

How the flux layers are measured

Boundary integrals, not cell sums. Line integrals of the ERA5 total-water flux around the region’s outline, divided by its area: comparable across regions of any size, no water counted twice.
All three phases count: vapour, cloud liquid and cloud ice. Vapour dominates, roughly two orders of magnitude larger than cloud liquid.
Inflow and outflow are independent measurements. Outflow is not inflow minus rainfall; both are integrated from the same flux field.
This is the water passing through, not water available for use (Actionable Shortage).

The worldwide figures

The Worldwide button in the top bar opens the whole planet for one six-month window, computed once by the pipeline, one row per month, independent of the camera. The months on offer are the trailing twelve the high-resolution cells carry.

It does not follow the region family you are viewing. Switching between countries, basins and hexagons changes the actionable ratio by a factor of three for the same month (a smaller polygon reports a larger boundary-integral throughput). The world figures are pinned to the equal-area hexagons, except countries-touched, counted on the first-level regions because a hexagon has no country.

“Measured land” is the denominator, within-family. Every share divides by the regions with a shortfall reading that month, never by a fixed figure for the Earth’s land area. Areas are each region’s buffered supply extent (land plus a 5 km coastal margin, Masks), so a family’s areas add up to more than the real land surface.
A region reading zero is counted, not skipped. Only a region with no reading at all (masked, unobserved, or outside the record) leaves the denominator.
“Drought touches N countries” counts distinct countries with at least one first-level region in drought. No threshold.
The actionable share is a ratio of sums, weighted by area, so a single near-zero-shortfall region cannot distort the headline. Regions with an incomplete six-month supply window are left out of both sides.
Two actionable readings, against two needs. One divides by the rain missing to reach normal across every region that is short; the other by what the regions in drought need to clear the drought line, over those regions alone.
Volumes lead, the percentage follows. Where the addressable water exceeds the gap, the figures say so in words rather than as a share above 100 %.
By continent, from the same rows. The scope picker re-reads the table at a continent: a grouping, not a second measurement. Every region sits in exactly one continent or in none; the nineteen disputed areas geoBoundaries files under an internal number appear in the world figures and in no continent. River basins have no continent rows, as a basin crosses borders.

Shared components

Baselines

Nominal window 1991–2020, per-region and source-pinned (the same product the month is scored against), clipped to each record (minimum 15 years / 60 months):

Layer	Winning source	Window actually used
6-month rainfall balance	GDO SPEI-6	1991–2020
Soil dryness	ERA5-Land	1991–2020 (GDO fallback ~2001–2020)
Vegetation stress	GDO fAPAR	~2001–2020
Groundwater	GRACE-DA	1948–2012, the product’s own published reference (GDO fallback re-percentiled ~2003–2020)

The windows are not uniform, so the four severity components read “unusual for here” on slightly different definitions of normal.

Zonal aggregation

Every layer is aggregated to the region as an area-weighted mean (exactextract coverage × cos-latitude weights; a region needs ≥ 10 % valid coverage to score), except the two flux layers, which are boundary line-integrals divided by area. Coarse sources stay at their native grid; a region smaller than the source cell inherits that cell’s value.

Percentiles

Dryness percentiles pool all baseline months into one distribution (midpoint ties), not one per calendar month. De-seasonalisation happens upstream (the SPEI z-score, the GDO anomalies, GRACE’s own percentile, soil moisture standardised per calendar month), so a score ranks this month’s anomaly against all monthly anomalies.

Units: per area or total

Water quantities are stored as depths in mm. A panel control switches the Water Resource flux layers and the Addressable mm layers (Shortfalls, enhancement-addressable monthly total) to a whole-region total: depth (mm) × area (km²) × 10⁻⁶ = billion m³, shown as “bn m³”. A large region reads high because it is large.

Availability gating

A month a source has not yet published is marked unavailable in the panel, and nothing is interpolated. One exception: for a single month past a layer’s record it falls back to its most recent available month, always dated in the panel.

Masks

Water Deficit — land only.
Water Resource & Addressable — land plus a 5 km coastal buffer; open ocean masked.

The no-resource floor

A grid cell counts as no-resource only when both tests agree: almost no rain reaches it (mean annual precipitation, CHIRPS) and almost no cloud water passes over it (MODIS liquid water path weighted by cloud cover). Both thresholds catch essentially all of the Sahara core and the Arabian Empty Quarter.

Why two tests. Cloud water alone cannot tell a hyper-arid core from a farmed semi-arid highland: the deep Sahara’s readings, inflated by dust, land higher than the Sanʿaʾ highlands or Sudan’s Gezira. Rainfall separates them: under about 110 mm a year for the two cores, 278 mm or more everywhere that must be kept.
The cloud test can only remove a place from the floor, never add one. It hands back deserts dry on the ground but cloudy overhead — the Lima garúa, the Galápagos, Cape Verde, and the Gulf, home to the largest operational rain enhancement programme in the world.
It is graded. What is stored is the share of the region that is no-resource — Abu Dhabi’s south is erg, its coast is not.
Values are only withheld at the extreme. Below a 90 % share the region keeps every reading; at or above it the resource layers and the shortfall are blanked.

Permanent ice

Permanent ice has no ground readings. The Greenland and Antarctica ice sheets, and any hexagon at least half glacier (Randolph Glacier Inventory 7.0), carry no drought severity, soil dryness or groundwater; rainfall balance, rainfall and the atmospheric-water layers still apply.

Regions

Family	What it is	Count
Regions	First-level subdivisions — states, provinces, regions (geoBoundaries CGAZ). Its panel can roll the whole country up (Region and country).	3,223
Hydro	River basins at HydroBASINS level 5, named for the highest-order river crossing them, else for overlapping administrative areas.	4,734
High-Resolution	Uniform hexagons — H3 resolution 4, ~45 km across and ~1,770 km² each — covering the same land as the other two families, coastal strips included.	77,972

Region and country

A region’s panel carries a scope switch: the same rows for the wider area, assembled on the fly from the first-level regions inside it.

Water quantities are area-weighted means over the regions measured that month, so a country reports its measured half rather than a half-strength whole.
The ranked indicators are left out, not averaged. Drought severity, soil dryness, vegetation stress, groundwater, rainfall balance and baseline water stress rank a region against its own history; an average of ranks means nothing.
What replaces them is a count. The Regions short of rain chart shows, month by month, how many regions were in drought and how many below normal; hovering names them, and each name opens that region.
Static facts are counts too — regions below the no-resource floor, regions at high or extremely high baseline water stress.
A hexagon can climb two rungs: Cell | Region | Country, without changing what the map draws. A first-level region offers Region | Country.
No wider scope for river basins (they cross borders) or for the handful of disputed territories filed under themselves.

The address bar mirrors what is on screen — view, indicator, region family, month or range, conversion rate, units, theme, camera, open region panel — so copying it shares exactly that. Nothing is added to browser history, so Back still leaves the page.

Note on high resolution hexagons

Equal-area cells make the map honest: the contrast is variation in the water, not in polygon size.
Recomputed, not downscaled: same sources, same aggregation, own 1991–2020 baselines. Nothing is interpolated from countries or basins.
A cell exists wherever it touches land, not only where its centre does; cells with under 5 % land are dropped.
A coastal cell is measured over its land: values and volumes cover the land part of the hexagon.
~45 km is near the floor the sources support (coarsest input 0.25°): neighbours inside one source cell can share a value.
Trailing 12 months only, single months, no period mode or playback, hence no seasonal profile (Seasonal timing).
Glaciated cells carry no ground readings (Masks).

Months, lags and periods

The globe opens on the most recent month available, and each layer carries the date its source published that month.
Typical lags. 6-month rainfall balance, vegetation, groundwater and atmospheric water arrive within days to a week, so they are present in the newest month; soil moisture runs 2–3 months behind and rainfall ~3.5 months.
Period replaces the single month with a range and shows the mean across it — except in Actionable Shortage, where the map divides the sum of the six-month actionable-water totals by the sum of the shortfalls across the months where a shortfall exists.
High-Resolution is trailing 12 months only (Regions).

Methodology