A reservoir and a farm the plan explorer

Before you play: what this explorer shows, and why

What you see. The smallest water system that shows a real trade-off: a river, a reservoir of 40 Mm³ that fills in winter, a farm that needs 45 Mm³ in summer and 5 in winter, and the outlet where unused water leaves. Four TaqSim blocks, three edges, 100 model years of a made-up river built from the averages of a classic study (Hashimoto, Stedinger and Loucks, 1982), with the seasons swapped so that the river peaks in summer, as rivers in Central Asia do. Every one of the 441 plans was run in TaqSim over the 100 years; nothing is recalculated in the browser.

What you can change. The two knobs of the reservoir's release rule. The trigger: start rationing when the reservoir holds less than this. The ration: then deliver only this share of the summer need. A trigger of 0, or a ration of 100 %, means no rationing: deliver in full until the reservoir is empty.

What you watch. Three scores over the 100 years. Average shortage: water not delivered per year. Worst month: the largest share of a month's need that was not delivered. Years short: how many of the 100 years had a shortage at all. A plan is unbeaten when no other plan is at least as good on the first two scores and better on one; those plans are the filled dots and the stepped line in the chart.

Why this is interesting. Delivering in full until the reservoir is empty wastes nothing: it has the lowest average shortage of all 441 plans, 1.11 Mm³ a year. But when the water runs out, the last weeks of summer get almost nothing: its worst month is 72 % short. Rationing early spreads the shortage over more months and more years: the worst month gets milder, the average shortage grows. The line of unbeaten plans bends: from 72 % to 50 % in the worst month costs 0.3 Mm³ a year; going on to 35 % costs another 2.3 Mm³ a year. The first steps are cheap, the last ones expensive. And 336 of the 441 plans are beaten by some other plan, including the reasonable-sounding “ration lightly”.

What to try. Drag the two knobs, or click a dot in the trade-off chart. Load the three example plans and read the three tiles. Find the bend of the line: the plan from which a milder worst month starts to cost much more water. Then look at the two lower charts: the worst month of each year, and the reservoir's storage around the worst drought, with the trigger line that starts the rationing.

What a water manager takes away. What you count decides the answer: with one goal there is one best plan; with two goals there is a line of compromises, and choosing among them is a decision, not a calculation. Choosing the goals is the most important modelling decision, and it is not a technical one. Intuition is a poor guide to which plans are beaten; the search is not. Real reservoir rules have these two knobs many times over, one pair for each month, which is why the full basin model hands the search to an optimiser. The notebook A reservoir and a farm builds this system block by block and goes on to a winter user and a seasonal forecast.

SOURCE River 100 synthetic years STORAGE Reservoir 40 Mm³, two knobs release DEMAND Farm 45 Mm³ summer · 5 winter SINK Downstream unused water
Average shortage
Mm³ per year not delivered
Worst month
% of need not delivered
Years short
of 100

Every plan is a dot. Left is better (less water undelivered), down is better (milder worst month). Click a dot to load that plan.
your plan unbeaten plan beaten plan
The worst month of each of the 100 years (% of that month's need not delivered)
your plan no rationing
Water in the reservoir (Mm³),
your plan no rationing trigger