middle estimate of the seasonal low · end of Apr 2027
Saved City reading · 5 October 2026 · Older report
Middle estimate50% range80% range
Starting from 75.3% on 5 October 2026 · 24 past seasons starting 2000–2024
The middle estimate falls to about 39% by the end of April 2027, with an 80% range of 29%–51%.
In tests where the range settings were fixed using only earlier years (2010–2025), 80% ranges held the actual level 78–84% of the time, but much less often in some years. Outcomes were more often above the range than below it.
Storage outlook by month end, percent of combined capacity. Low and high are the 10th and 90th percentiles.
Month end
Low
25%
Middle
75%
High
Reading 05 Oct
75.3%
Oct 2026
72%
73%
75%
76%
77%
Nov 2026
67%
69%
72%
74%
77%
Dec 2026
59%
61%
65%
69%
71%
Jan 2027
50%
52%
56%
62%
65%
Feb 2027
42%
45%
48%
55%
60%
Mar 2027
34%
37%
41%
49%
54%
Apr 2027
29%
34%
39%
46%
51%
May 2027
31%
33%
41%
50%
56%
Jun 2027
32%
41%
54%
64%
76%
Jul 2027
38%
56%
70%
77%
83%
Aug 2027
45%
63%
78%
84%
96%
Sept 2027
51%
64%
80%
88%
100%
An independent estimate by Cape Water, not a City of Cape Town forecast. It updates when the City publishes a new reading.
In plain terms
Cape Town’s dams usually fall through summer and refill with winter rain. We start from the latest City reading and ask where storage would go if the next 12 months behaved like each past season since 2000. The middle line is the central path. The shaded areas are the 50% and 80% ranges of those paths, widened slightly to match how past outlooks performed.
The ranges are wide because winter rainfall varies a great deal and cannot be forecast months ahead. The outlook does not use weather forecasts, current demand, restrictions or new water sources.
How accurate has it been?
We produced outlooks for 486 past dates between 16 January 2005 and 16 April 2025, each using only seasons that had already finished, and compared them with the level the City recorded. The tables below cover the 368 dates from 2010–2025. For these, the range settings were refitted each January using only outcomes already known at the time, so the results show how the method would have performed in real use.
Tables scroll sideways on narrow screens.
Average error of the middle line, in percentage points (lower is better)
Ahead
1 month
2 months
3 months
6 months
9 months
12 months
Cape Water outlook
2.6
4.2
5.2
8.1
10.6
12.3
Assume no change
5.9
11.3
16.0
23.3
18.2
13.2
Add the average seasonal change
2.6
4.0
5.1
8.3
11.3
13.5
Typical level for the date
14.7
14.8
14.8
14.4
14.3
14.5
Improvement in probability score over simpler methods (higher is better)
Ahead
1 month
2 months
3 months
6 months
9 months
12 months
Over assuming no change
67%
74%
76%
75%
60%
36%
Over adding the average seasonal change
26%
26%
25%
31%
36%
38%
Over the typical level for the date
81%
71%
63%
44%
28%
18%
How the ranges matched what happened
Ahead
1 month
2 months
3 months
6 months
9 months
12 months
80% range held the actual level
84%
84%
83%
78%
79%
79%
Same, worst and best single year
67%–100%
63%–100%
63%–100%
38%–100%
33%–100%
46%–100%
Actual level below the 80% range
5%
5%
5%
11%
10%
10%
Actual level above the 80% range
11%
10%
11%
11%
11%
12%
50% range held the actual level
58%
56%
55%
50%
48%
47%
Width of the 80% range (points)
9.3
14.2
18.8
25.0
31.7
37.6
Middle line bias (points)
−0.7
−0.6
−0.7
−1.0
−0.9
−0.8
Bias when starting below 50% (points; 56 tests, 2012–2019)
−0.7
−0.3
−0.9
−3.0
−8.8
−12.3
80% range held the level, starting below 50%
84%
84%
82%
63%
66%
63%
The probability score is the continuous ranked probability score, which rewards ranges that are both accurate and narrow. Tests overlap in time and come from 16 test years and about 25 past seasons, so each figure is uncertain by several points.
Up to nine months ahead the outlook scored 60%–76% better than assuming no change and 25%–36% better than adding the average seasonal change. Its middle line, however, is about as accurate as adding the average seasonal change: the value of the method is in the range. At twelve months the middle line is no better than assuming no change.
The 80% range held the actual level 78%–84% of the time, but this varied widely from year to year. Outcomes were above the range more often than below it, so the ranges lean pessimistic. When storage started below 50%, during the 2015–2018 drought, the middle line was much too low, partly because restrictions cut demand; the outlook shows a caution whenever storage is below 50%.
Method
Each complete 12-month season since January 2000 contributes one path. For a reading on a given date, a path follows the changes in combined storage recorded from the same calendar date in that past year. Storage is capped at 100% and cannot fall below zero. When a past season was already full and spilling, rises in its readings above 100% are counted as water flowing in (falls back towards full are spill, not use), so wet seasons are not cut short just because the dams were full at the time. Seasons from before August 2008, when Berg River Dam was not yet part of the reported system, use the five-dam total as a percentage of five-dam capacity; seasons spanning that change are left out.
The middle line is the median of the paths. The spread around it is widened by a factor of 1.05–1.15, depending on how far ahead (interpolated between the tested horizons), chosen so that 80% ranges held the outcome at least 80% of the time both over the whole record and over the last ten years. We also tested correcting the middle line for its past bias, rescaling pre-2008 seasons to six-dam volumes, widening each side separately and counting rises above full during spills. Version 1.2 counts those spill rises: we said in advance that we would adopt it only if it also won a leave-one-year-out test, and it did, scoring 1–4% better at every horizon (about 3% on average) and roughly halving the 12-month bias.
Range settings compared on the same tests (average probability score, lower is better)
Option
Score
Widening only; five-dam seasons in five-dam percentage points (version 1.0)
5.08
Widening only; five-dam seasons rescaled to six-dam volume
5.17
Bias shift and widening; five-dam seasons in five-dam points
5.35
Bias shift and widening; five-dam seasons rescaled
5.49
Separate widening below and above the middle; five-dam seasons in five-dam points
5.06
Separate widening below and above the middle; five-dam seasons rescaled
5.24
Widening only; rises above full during spills count as gains
5.04 · used
Separate widening below and above; rises above full count as gains
5.04
Model version analogue-ensemble-1.2; tests run 5 October 2026. Full test results.
Rain scenarios
On the homepage you can ask what happens if next winter is dry, near normal or wet. These scenarios use only the past seasons whose April–September rain, near the five dam areas, was below 85%, between 85% and 115%, or above 115% of the 1991–2020 average, measured with the ERA5 reanalysis (which tracked the dams’ own rain gauges closely in 2008–2020). Each group has only a handful of seasons, so the scenario shows their middle path and full range, not a tested 80% range. They answer “what happened in years like that”, not “what will happen”: winter rain cannot be forecast months ahead.
Data
Historical storage comes from two City of Cape Town dam-level archives on the City’s Open Data Portal: “Dam levels from 2000” and “Dam levels Jan 2012 to April 2026”. We add the stored volumes of the six major dams for each day (898,221 megalitres combined capacity). The series runs from 1 January 2000 to 29 April 2026, so the most recent season is not yet part of the outlook. Isolated one-day errors at individual dams were replaced with a local median, and rows filed under the wrong year were removed; readings above full capacity were kept. Archive values may include later corrections.
The outlook assumes the coming year will resemble one of roughly 25 past seasons. A drier winter than any since 2000, a change in demand, new supply from groundwater or desalination, or changes in how the system is operated can all produce outcomes outside the ranges. Demand has been lower since the 2018 drought than in many earlier years, so older seasons may overstate today’s summer drawdown. Climate change may make dry winters more common than in the record.
The 10% line is a historical reference from the City’s 2019 Water Strategy for progressively harder extraction. It is not a point at which taps stop, and this page does not estimate a “Day Zero” date.
Sources and updates
Readings come from the City of Cape Town’s published dam tables. This page checks the source automatically; responses may be cached for up to 30 minutes. The City publishes on its own schedule. These are dated reports, not live measurements.
The live source was unavailable when this page was prepared. Saved figures retain their original report dates. Reports more than nine days old are labelled as older readings.
The latest reading and weekly report can have different dates. Weekly comparisons use the same weekly series. A percentage point is the difference between two percentages; it is not a percentage change in volume. Levels above 100% can occur when dams spill.