Yellow River · smart application platform
A smart application platform for a Yellow River authority: space-air-ground IoT sensing of hydrology...
View Details →The managed asset is a river hundreds of kilometres long, with works, sluices and floodplains along it. The business is not complex; what is complex is that several units, several lines and several generations of systems coexist. Those three breaks decide that integration must come before features.
One decision shaped the platform: the hard part of river-authority digitalisation is integration itself. So the chassis comes first — unified login, unit switching, the wall dashboard and device operations — and thirteen modules hang off the same data chain. Modules can ship in phases; the chassis cannot be split.
The manual reads as thirteen parallel entries, visited in turn. Grouped by user, they are three kinds: flood and water business, sensing and alerts, governance and operations — one login, one chassis.
| Module | What is inside | Who uses it |
|---|---|---|
| Flood control | Units, flood mitigation, key flood-defence sites, rainfall statistics, flood information, response forces | Flood office: the screen opened daily before and during the season |
| Works | Classified works register per unit: embankments, sluices and danger points, each archived | Works management: who owns this reach, of what class |
| Water resources | One-map with seven layers: works, video, running state, water quality, intake records, patrols, e-government | Water resources and dispatch: seven dimensions at one pin |
| Water admin | Classified water-administration regulations per unit | Enforcement: legal basis by unit and category |
| Detention area | Quick entries to flood mitigation, works, water resources, enforcement, culture, the four-forecast platform and the one-map | Detention-area management: seven lines, one entrance |
| E-government | Independent public website for information release | Public and communications: disclosure lives apart from the platform |
| Party building | Member records, videos, the three-meeting one-lesson cycle, activities, analytics | Party affairs: organisation life and member data in one book |
| Culture | Organisational culture and integrity-building content | Communications: culture results in one place |
| Patrol app | Patrol records, engineering alert details, AI alert details | Field stations: what was patrolled, what is alerting, what the AI saw |
| Sluice safety | Water-intake sluice monitoring: statistics, curves, layout diagram, overview, water level, weather | Sluice keepers: level curve and layout on one screen |
| Wall dashboard | Live weather, alerts, flood information, real-time hydrology | Command hall: the first screen of every flood shift |
| Unit switch | Quick entry to switch between subordinate units | Everyone: data scoped by unit, one tap to change view |
| Device ops | Device counts, alarm and fault analysis, daily statistics, video | Operations: the register decides which device gets replaced |
Two easy-to-miss details: the privacy-policy checkbox at login and the unit-switch entry on the portal are not features — they are the chassis. One puts compliance into the front door; the other lets one system serve many units without data leaking across.
The value chain of flood work runs pre-flood warning, in-flood dispatch, post-flood review. Unpacked, it is six steps, each owned by a module — and crucially, none of the six changes systems: from rainfall to review, it is one flood dataset.
Why rainfall statistics own the first screen. The first input to every flood decision is rain. Rainfall is never reported by hand; it aggregates from stations and refreshes hour by hour, setting the warning level and deciding whether response forces move forward. So rainfall statistics, key sites and response forces live in the same module: whoever reads the rain can see the sites and the crews without opening a second page. Every minute of a flood season is worth saving.
Why the four-forecast platform is an entry, not a system. Forecast, warning, rehearsal and plan are the standard agenda of water digitalisation, yet every one of them feeds on existing data: forecasts on rainfall and hydrology, rehearsals on works layout, plans on response forces. So the detention-area module presents the four-forecast platform as a quick entry alongside the rest. Data does not move and definitions are not re-invented — the capability grows on existing data instead of becoming another island.
The water-resources module is the platform's map base: works, video, running state, water quality, intake records, patrols and e-government — seven layers sharing one set of coordinates, so opening any pin surfaces all seven dimensions at once.
A trade-off: aggregate the map, or rebuild the data. The thorough path is a unified warehouse with works, quality and intake migrated in. But each line has its own systems under construction and its own maintenance rhythm; migration turns a data problem into an organisational one. We shipped map aggregation without moving data: the map references coordinates and key states, and each line keeps owning its data. The cost is that the map is not a live database and can go stale if a module stops updating — a visible, attributable cost, and cheaper than the organisational risk of migration.
Why video appears in three places. Video is a layer on the one-map, an operations object in device ops, and a review tool during flood duty — one capability, three uses: dispatch looks at the site, operations looks at uptime, alerts look at the replay. The implementation has one stream and one register; the three places are references. Add or remove a camera once and all three follow — the most direct dividend of building the chassis first.
Field routine is patrol, report, repair. The patrol app splits it into three channels: patrol records produced by people, engineering alerts raised by thresholds, and AI alerts raised by video analytics — recorded separately, disposed together.
A trade-off: should AI replace patrols, or run alongside? AI watches tirelessly and could absorb much of the patrol load. But a river produces false positives — birds, glare, drifting debris. If AI dispatched crews directly, the field would drown in false alarms and then ignore the channel. So the two run in parallel: AI alerts form their own list for duty staff to screen and dispose; patrols continue by the book. AI widens the field of view; people confirm and take responsibility — the two lists sit side by side, and missed detections versus false alarms are visible for what they are.
Why disposal time is recorded. Anyone can build an alert list; the hard question is how long an alert lay unhandled. The closing time is system-written and cannot be edited by hand — the least glamorous and most valuable field in the system. With it, monthly reports can compute average closure time per alert class, which is the raw material of fault analysis. Without it, an alert system is just a louder phone.
The sluice module watches one structure (the water-intake sluice); the dashboard watches one hall. One manages a point, the other a surface — and they share the same data.
| Module | What it shows | What it solves |
|---|---|---|
| Sluice safety | Platform statistics, curves, layout diagram, overview, water level, weather | A health sheet for one sluice: level curve and layout on one screen, adverse trends visible at a glance, weather explaining why the level looks the way it does |
| Wall dashboard | Live weather, alerts, flood information, real-time hydrology | The hall's first screen: no interaction required, the overview is always up; alerts pinned at the top where nobody can pretend not to see them |
Why build a whole module around one sluice? Because a cross-province water-intake sluice carries a risk class far above ordinary works. Ordinary works need to be findable; a key sluice must be watchable — curves, layout, overview, level and weather organised around a single structure. The allocation principle behind it: sensing investment follows risk, not fairness. Six kinds of sensing on one critical sluice beat one kind spread across twenty.
The dashboard's value is not decoration but shared fact. Its four elements — live weather, alerts, flood information, real-time hydrology — come from the business systems in real time, not a slide show. During flood duty, command, duty staff and dispatch all read the same numbers; after the event, the screen doubles as the decision record.
A river authority is organised as bureau, subordinate units and field stations. One platform serves them all, and two things decide whether it survives contact with reality: how well unit-level data is isolated, and how honestly devices are registered.
After account login and the privacy-policy confirmation, the portal offers a switch entry for subordinate units. Each unit sees only its own works, patrols and alerts; the bureau sees everything.
Because the switch lives in the portal rather than inside each module, the permission model is built once: thirteen modules share the same unit-role-data-scope model, new modules inherit it automatically, and "visible here, invisible there" cracks never appear.
Device ops maintains four things: device counts, alarm and fault analysis, daily statistics, video. Together they answer two questions: how many devices are working right now, and which ones fail repeatedly.
Fault analysis aggregates history per device: a level gauge that alarmed twenty times in a month is the one to replace, not whatever broke last week. Sensing hardware will fail; the module's job is to turn "it broke" into a planned, expected repair.
Three chassis pieces — unified login, unit isolation, device operations — are the least impressive in a demo and the most appreciated in year two. Business modules decide what the system looks like at launch; the chassis decides what it looks like three years later.
As agreed with the client, this page shows no unit names and no screenshots of production screens.
Integration platforms for water authorities, government and campus estates rarely fail on a single feature. They fail on three questions: the portal chassis, agreed data definitions, and the permission model. We have ground these down across public-sector and IoT projects — bring your structure and we will tell you which of the three is your bottleneck. Related: smart factory & IoT and how we build.
MORE CASES
A smart application platform for a Yellow River authority: space-air-ground IoT sensing of hydrology...
View Details →An IoT management platform for river sand extraction and aggregate production: eight modules and 24 ...
View Details →Start with our pricing logic, then let's talk requirements — cost structure, technology comparison and acceptance criteria, all laid out in the open.