Yellow River ice-flood AI forecasting & monitoring: from river-watching to data

An ice-flood AI forecasting and monitoring system for a river authority: cameras, drone patrols, weather radar, hydro stations and remote sensing monitor stage and discharge in real time while the AI model turns those inputs into forecast hazards and tiered alerts; one map carries the battle map, the risk map and the deployment map, and field reporting runs through to duty dispatch in one loop.

Industry
River authority / flood control and disaster reduction
System type
AI forecasting + IoT monitoring + one-map GIS + field reporting
Key technology
Cameras · drone patrols · weather radar · hydro stations · remote sensing · three wall maps · four basemaps · reporting with export
Jiding scope
Requirement shaping, multi-source sensing integration, one-map and reporting modules, launch support
5sensing sources: cameras, drones, weather radar, hydro stations, remote sensing — each covers the others' blind spots
3wall maps: operation map, risk map and deployment map, one click from the home page
4basemaps: imagery, terrain, water-conservancy and vector, switched per scenario
5loop steps: report, consolidate, alert, plan, dispatch — one ice dataset end to end

About Ice Floods

First, what an ice flood is: three stages every winter

Unlike summer floods, the threat is ice, not rain. Understanding the three stages explains the design — every stage has its own module on watch.

Ice flood = ice + flood: the driver is ice, not rain — so temperature joins stage and discharge as core inputs 1 · Ice run Ice floes drift downstream Run density signals freeze-up On watch: drone patrols by list 2 · Freeze-up River freezes; ice raises stage Watch stage, discharge and ice On watch: live stage & discharge 3 · Break-up Ice melts and drifts; jams can form Highest risk: sudden hazards On watch: AI forecast + alerts + plans Risk rises stage by stage: see it in the ice run, track it through freeze-up, predict it early at break-up — all three wall maps stay one screen away The features map to stages, not departments: drone patrols and field reports during the ice run, hydrological curves through freeze-up, AI forecasting and tiered alerts at break-up. Live weather, air and water temperature stay on the home page — they are what drive the ice. The stages repeat every winter; the system turns “judging it again from scratch” into “reading it again from data”.
Three stages: each stage has a different module on watch — but stage, discharge, air and water temperature run through all of them as the raw material of the AI forecast

Sensing Matrix

Five sensing sources, each covering the others' blind spots

Five technical means behind the scenes are not the same job done five times — they are five given conditions of the same equation.

SourceWhat it seesWhy it cannot be replaced
CamerasLive footage of key reaches, control works and bridgesThe only always-on way to actually see the site: ice colour, drift speed and shore-ice extent cannot be read by a sensor
Drone patrolsMobile patrol flights over priority reaches, managed as an address listFills the blind spots of fixed points: ice moves, so fly to where the jam is; footage lands straight on the one-map
Weather radarRegional weather processes and temperature trendsTemperature is the first driver of ice: cooling freezes, warming breaks; the forecast needs to know how cold it will be next
Hydro stationsContinuous automatic series of stage and dischargeThe quantitative baseline: why stage rises at freeze-up and how discharge evolves at break-up is answered by the series
Remote sensingMacro view of ice extent and channel morphologyThe view from above: freeze extent across the whole reach cross-checks the ground points

Why five sources cannot be merged into two

The frequent question is “can we connect fewer?”. The five sources cover each other's blind spots: cameras see points but not the line, remote sensing sees the line but not the detail, hydro stations give numbers but no picture.

So the integration is all five, each in its place: footage goes to monitor lists and the one-map, series go to live monitoring and the model, macro views go to risk assessment. Remove any one and the forecast loses a given condition.

Why live weather, air and water temperature stay on the home page

Many systems treat weather as decoration. Here it is a core variable: temperature decides whether ice forms and whether it melts.

Together with temperature-trend records at control works, the duty officer starts every shift by answering “how much colder than yesterday, and which way is the ice heading”. Forecasting starts from that sentence.

Wall Maps

Three wall maps and four basemaps: the command first screen

Login lands on the home page where three ice-flood wall maps are one click away: the operation map answers “how to fight”, the risk map answers “where it hurts”, the deployment map answers “who is where”. The basemap underneath switches four ways.

Ice-flood battle map

Pops up the flood-ice emergency operation map: hazard sections, rescue routes and force deployment on one sheet. When a hazard hits, the commander's first screen is it — from “there is a hazard” to “how to fight it” without a meeting.

Ice-flood risk map

Pops up the risk map: jam-prone reaches and historically hazardous works, coloured by risk level. Duty rosters follow the risk map — higher-risk reaches get denser watches and more frequent patrols.

Deployment map & duty roster

Pops up the defence-force deployment map; the duty module lists works names with contact phones. The risk map says where to defend, the deployment map says who goes and how to call — read together, dispatch lands.

Four basemaps, one dataset

Imagery for the site as it is, terrain for the ground, the water-conservancy layer for the works system, vector for annotations and links. One dataset, four readings: assess ice on imagery, deploy forces on terrain, look up works on the water layer.

Map Toolbox

Map toolbox: answer every question on one map

Every icon and button on the map is an entry into one class of works data — the map is the entrance, the data is the point.

ToolWhat it doesQuestion it answers
Works pop-upClick any ICON on the map for flood-defence works info of that areaWhat is this works, who owns it, what class
Temperature trendCurve of temperature history at control works of river nodesHow cold has it been, which way is the ice heading
Node registryShow the node data of all flood-defence areas at onceThe full river inventory, section by section
Bridge infoBridges on priority reachesWhere the jam-prone bridges are and what a jam would affect
Custom ringDraw an area of interest on the mapMark a reach and hand it to patrol and assessment
Distance toolMeasure between any two pointsReal distance from hazard to resources, for dispatch timing

The ring and distance tools deserve a special note: they connect to no data at all, yet they are the two most used buttons in assessment meetings — turning verbal descriptions into circles and lines on the map cuts communication cost by a notch.

AI Pipeline

The AI pipeline: five data sources in, tiered alerts out

AI forecasting is not a score in a black box. It is a pipeline: five sensing streams feed one data foundation, the model produces forecasts, forecasts become tiered alerts that reach the flood-control units — every step stays inside the system.

Prediction channel: four steps from sensing to alert Cameras Drone patrols Weather radar Hydro stations Remote sensing Data fusion time & space aligned single source of truth AI forecast model stage & discharge evolution hazard risk assessment Tiered alerts level and owner alert time logged Flood-control units alerts delivered decision support The monitoring channel runs in parallel: live stage and discharge keep checking the forecast — prediction covers the future, monitoring covers now
AI pipeline: every forecast must stand up to the monitoring data — prediction and monitoring cross-check each other, and that is why “forecasting and monitoring” share one system name

Why alerts must be tiered

The value of an alert is not “it was issued” but “it reached the person who must act”. Tiers translate a forecast into actions: each level maps to watch density, patrol scope and readiness, triggered down to specific nodes and crews.

With ice alerts logging the alert time, every warning carries a complete chain — when issued, to whom, when answered. An alert stops being a sentence and becomes a traceable record.

Why forecasting and monitoring live in one system

The industry often splits them: a forecast model here, a monitoring platform there. The result: a wrong forecast cannot be diagnosed, and an anomalous reading cannot be compared with what the forecast said.

This system keeps them together: monitoring data is the forecast's raw material and its referee. The model says stage will rise; the hydro curve confirms or refutes it within hours; the deviation stays in the system as input for the next iteration.

Reporting Loop

One ice-flood information loop: from “report now” to duty dispatch

Sensors and AI cover what machines can see; ice observed by human eyes enters the system through field reporting. Five steps turn a field observation into a traceable response.

123 45 Field · capture HQ · assessment Field · response Field reporting “report now” form Consolidate & export records exportable Ice alerts alert time logged Response plans timeline + documents Duty & dispatch rosters · monitors Field eyes, structured anytime, anywhere HQ sees the whole river records stay auditable side by side with AI first reporter credited plan matches the node actions follow the book monitors one click away phones on the roster Five steps share one ice dataset: what the field reports, HQ sees; what HQ orders, the field receives The loop works both ways: reporting with export covers upward flow (records leave no room for memory); rosters with phones and monitor lists for priority reaches cover downward flow. Machine AI alerts and human field reports merge into one ice ledger — two sources, one outlet; assessment never stops to ask who reported.
The loop: upward flow relies on reporting, downward flow on plans and rosters — AI alerts and human reports merge into one ledger, two sources, one outlet

Why human reporting did not go away

With five sensing streams covering stage, discharge, footage and macro ice, field reporting can look redundant. But many key signals sit exactly in the sensors' gaps: shore-ice shapes, oddities in small channels, reports from residents — the human eye is the most flexible sensing source.

The “report now” form turns field observations into structured records that sit beside AI alerts in the same consolidation. The more automated the sensing, the more valuable the human report — it covers exactly what machines cannot see.

Why plans hang on works nodes

Ice-flood response plans are organised by flood-defence node: each works has its own plan timeline and guidance documents. When a hazard hits a reach, open that works and the plan is at hand — not one generic document for the whole river.

That is the value of “documents follow the works”: the more urgent the moment, the less time there is to search for documents. Plans granular to the node make response procedural.

FAQ

Four common questions about ice-flood monitoring

How is an ice flood different from a summer flood?

Different driver: summer floods are driven by rain; ice floods are driven by ice — the river runs with floes, freezes over, and rising water at break-up comes from ice jams, with the risk concentrated at break-up. So the forecast inputs are not only stage and discharge but also air temperature, water temperature and ice observations, and the sensing must additionally cover the ice surface and channel morphology.

Why does an ice flood suit AI forecasting?

The evolution of an ice flood couples air temperature, discharge and channel morphology — hard to quantify by experience alone. AI forecasting puts multi-source data (cameras, radar, hydro stations, remote sensing) into one model that keeps analysing and produces checkable outputs, cross-checked against live monitoring.

Why drones when there are already fixed cameras?

Ice is a moving target: run density, freeze position and jam shape all change, and fixed points cannot cover the whole reach. Drones patrol priority reaches by an address list, fill the unseen segments, and their footage lands straight on the one-map.

Why record even the temperature trend of control works?

Air and water temperature directly drive ice formation and melt. Temperature-trend records at control works let duty officers read freezing and melting trends against stage and discharge — a key model input and the first-hand material for break-up assessment.

Deliverables & Results

Deliverables and effect

Deliverables

  • Sensing integration: cameras, drone address-list patrols, weather radar, hydro stations, remote sensing
  • Live monitoring: real-time stage and discharge with forecasting; live weather, air and water temperature on the home page
  • Three wall maps: operation map, risk map, deployment map — one click from the home page
  • One map: four basemaps plus six tools — works pop-ups, temperature trends, node registry, bridge info, custom rings, distance
  • Reporting loop: field reporting with consolidation and export, ice alerts with logged times, plans with timelines and documents
  • Duty support: works rosters with contact phones, monitor lists for priority reaches

Effect & de-identification

The most direct change: judgements now stand on data. Stage and discharge no longer travel by phone; ice conditions are covered by five sensing streams; the forecast gives a checkable view of the future; tiered alerts trigger down to nodes and crews.

The second change is at the front line: field reporting replaced paper, records export in one click; phones hang on the works, priority reaches link straight to monitors — dispatch no longer starts with “find the person, then find the number”.

The third change compounds: each winter's observations, alerts and responses stay in the same system, so every year's forecast stands on more history than the last. As agreed, this page shows no client name and no production screenshots.

Building a sensing-plus-forecasting flood monitoring system? The hard parts are how the sources cover each other's blind spots, how forecasting and monitoring cross-check, and how plans hook onto duty rosters.

MORE CASES

More Related Cases

Yellow River dynamic monitoring & early-warning system
Government System

Yellow River dynamic monitoring & early-warning system

A dynamic monitoring and early-warning system for a river authority: space-air-ground IoT devices se...

View Details →
Yellow River ice-flood AI forecasting & monitoring system
Government SystemSmart flood prevention, smart ice prevention

Yellow River ice-flood AI forecasting & monitoring system

An ice-flood AI forecasting and monitoring system for a river authority: cameras, drone patrols, wea...

View Details →
Yellow River · smart application platform
Government SystemDigital Twin,Smart Yellow River

Yellow River · smart application platform

A smart application platform for a Yellow River authority: space-air-ground IoT sensing of hydrology...

View Details →

View All Cases →

Planning a similar production line or enterprise system?

Start with our pricing logic, then let's talk requirements — cost structure, technology comparison and acceptance criteria, all laid out in the open.