Evidence & method

The numbers behind AmuFlow — every one reproducible.

This page is the deep dive: how irrigation is scheduled in Uzbekistan today, a live demonstration on real field data from Khorezm, and the full benchmark against the mathematically best possible schedules.

← Back to the main page

0%avoidable crop-water stress left by today's schedule at district scale — measured against the mathematically proven optimum
0 sfor an exact solver to prove the best possible schedule for a 12-zone, 28-day district instance
0of irrigated land Uzbekistan has pledged to cover with water-saving technology by 2030 under the World Bank's Water Forward initiative

Today's system, precisely

Three layers of planning. Zero measurements in the loop.

This is not a caricature — it is the documented mechanism, from the state's own sector review and the scientific literature. Full citations at the bottom of the page.

Layer 1 · since 1977

Hydromodule norm tables

Every hectare is assigned to a soil class, and each class × crop gets a fixed seasonal norm: how many irrigations, how much water, roughly when. The tables were computed by SANIIRI in the Soviet era and remain the legal basis of seasonal plans.

Layer 2 · allocation

Limits, passed down by hand

The Ministry allocates annual water limits down a chain — basin, canal administration, water consumers' association, farm — planned in ten-day blocks. The 2025 Water Code fines over-use; nothing measures usefulness of the allocation.

Layer 3 · delivery

Rigid canal rotation

Water physically arrives in fixed rotation windows — in Khorezm, about three days per week per zone. A field that misses its window waits for the next one, whatever the crop looks like.

"Design of irrigation systems is outdated, with inflexible, supply-driven delivery of irrigation water… Delivery of water to irrigators is not responsive to their needs and climatic conditions."

— World Bank, Uzbekistan: Policy Perspectives for Irrigation and Drainage Sector Reform (2024), written with the Ministry of Water Resources

Live demonstration · real data

Three fields in Khorezm. Fourteen days. Four doses of water.

Everything below is computed from real 2024 measurements: NASA satellite weather near Urgench, ISRIC soil surveys of the same fields, FAO cotton physiology, and the documented canal rotation. Choose who decides the schedule — the water budget never changes.

+35%avoidable crop stress vs the proven optimum
4 / 4water doses used of the fixed budget
root-zone water (mm) comfort target stress threshold canal window open ▼irrigation dose (80 mm)

The fields on the map

Same fields, same fortnight — two schedules, side by side.

Three real adjacent cotton parcels on a tree-lined distributary north of Urgench, Khorezm. Press play and watch the water: each field is tinted by how much its root zone holds — deep blue just after irrigation, draining through green to amber and red as the crop goes thirsty. Left is today's fixed state calendar; right is the AmuFlow schedule. White dashed links mark neighbouring fields that share an offtake and cannot be watered on the same day; a field glows cyan while its canal window is open.

Day 1 / 14 ·

State calendar plan

+35% crop stress wasted

AmuFlow optimized

optimal · 0%
just irrigated comfortable drying stressed crop damage canal window open 💧dose delivered

Parcel outlines are traced from Esri satellite imagery of real cotton fields near Urgench (≈41.63°N, 60.63°E); the day-by-day water levels are the model's real computed trajectories for the 3-zone, 14-day instance. Background imagery © Esri; maps by Leaflet.

Does it hold at district scale?

The bigger the district, the more the calendar costs.

Avoidable stress above the best possible schedule, for growing district sizes — from 3 zones × 1 week up to 24 zones × 8 weeks (584 gate decisions). The optimizer's line barely leaves zero.

† largest two sizes measured against the best-known solution (exact proof exceeds the open-source solver's time cap); all others against the proven optimum. Simulated annealing at matched budget is in the evidence table.

Evidence & reproducibility

Every number here can be recomputed from public data.

The benchmark comes from a complete research study — problem formulation, exact solvers, metaheuristics, and (honestly reported) quantum-hardware experiments — built exclusively on public, real measurements. No synthetic convenience data, no tuned-to-win baselines: the replica of state practice follows the documented rules, and the generous "diligent farmer" variant is given perfect soil knowledge the real system doesn't have.

District instanceState calendarDiligent triggerSAAmuFlow (GA)Optimum

Gap = extra stress-weighted crop-water deficit + water cost above the optimum. SA/GA at 20,000 matched evaluations (GA column shows the 10× budget: 1–73 s per run). † best-known reference.

$ git clone https://github.com/mcpeblocker/amuflow && cd amuflow $ pip install -r code/requirements.txt $ python code/fetch_data.py # NASA POWER + SoilGrids → real inputs $ python code/classical.py # exact / greedy / SA / GA benchmark $ python amuflow/benchmark/government_replica.py # the state-plan replica → this page's numbers

Why now

The state is building the body. We are proposing the brain.

Billions are flowing into canals, pumps, and flowmeters. Every one of those dollars raises the return on the one layer no current project addresses: the decision of where the water goes.

2020

Water Sector Development Concept 2020–2030

Presidential Decree DP-6024 targets raising irrigation-network efficiency from 0.63 to 0.73 and water-saving technology on 2 million hectares.

2025

US$200M World Bank irrigation modernization

Flowmeters, gated outlets and SCADA across five regions; 540 million m³ of water losses to be cut per year. The sensing layer a scheduler needs is being installed right now.

2026

Uzbekistan joins Water Forward

A public, dated pledge: −25% irrigation losses, water-saving technology on 100% of 4.1M ha, 200,000 digitalized intake points by 2030. Somebody has to deliver measurable percentage points. That is exactly what a scheduling layer produces.

2030

Targets come due

Physical works address water lost in transit. AmuFlow addresses water delivered on time to the wrong place — the loss no canal lining can touch.

What we are building

A district-in-a-box scheduler that speaks the planner's language.

AmuFlow does not replace the Ministry's authority, the norms, or the limits. It optimizes within them — and its output is the same ten-day plan document districts already file, with better numbers and an auditable reason for every dose.

Inputs — already available

  • Approved limits & crop maps
  • Canal rotation calendars
  • Free satellite weather & soil data
  • Flowmeter/SCADA feeds where installed
→

The AmuFlow engine

  • Physically faithful district model (moisture memory, crop stages, shallow groundwater)
  • Exact optimization at district scale — proven, in seconds, offline
  • Runs on one office computer, in Uzbek, Russian, and English
→

Outputs — the same paperwork

  • Optimized ten-day delivery plan per WCA
  • A "why" for every dose
  • Planned-vs-optimal loss report — the Water Forward metric, computed
Done · 2025–26

Evidence

Research study on real Khorezm data; replica benchmark; the public website.

2026–27 season

Shadow pilot

One Khorezm district. AmuFlow schedules run alongside official ones all season. Zero operational risk; one certified gap report.

2027

Advisory mode

Planners use AmuFlow for the decade plans; A/B across willing WCAs; hardened Uzbek/Russian/English UI.

2028–30

Scale

District-by-district rollout riding the national flowmeter deployment; Water Forward reporting module.

The ask

One district. One season. One honest report — co-signed.

We are not asking anyone to change how water is managed. We are asking for permission to measure, next to the real thing, for one season.

  1. A pilot district in Khorezm — cooperation with one canal administration and 3–10 water consumers' associations, where our model is already calibrated.
  2. Data access — the district's seasonal plan, limits, rotation calendars and delivery journals; flowmeter feeds where they exist.
  3. An institutional sponsor — a partner inside a water agency, a development program, or an accelerator for a shadow-mode season: official schedules stay in force, we measure the difference.
  4. US$30–60k pilot funding — a grant, a technical-assistance window of the ongoing water projects, or a pre-seed investment.
AO
Alisher OrtikovOptimization & engineering — senior CS & EE student, KAIST. Research stack behind AmuFlow, from the physical model up to the benchmarks.
AI
Alisher IlhamovFinance & economics — Westminster International University in Tashkent. Economic model, institutional mapping, government relations.

Presidential Schools alumni — to give what was entrusted back to the nation.

Talk to us