Simulation — synthetic / historical data, not a live sensor feed
How to use — the four wizard steps in detail
Step 1 · Pick a site — search an address (fills the coordinates for you) or type a latitude/longitude, give it a name, then click Create simulated site. Saved sites stay in the dropdown so you can come back to them: select one to rename or move it (Save changes), or remove it (Delete site — deleting a promoted site also removes it from the live roster).
Step 2 · Place the virtual sensor — the 3D terrain flies to your site. Click Place sensor at site centre, or arm Place by clicking the map and then click anywhere on the terrain to drop the pin. This is the patent Claim-12 “water level detection arrangement”; optionally set a monitoring elevation.
Step 3 · Configure & run the real pipeline — choose a forecast horizon (24 h up to 28 days), the error metric that picks the winning model (MAPE / MAE / RMSE) and how much history to train on (10–90 days), then click Run simulation. The exact ARIMA · ETS · RNN bake-off and Kappa-selected hazard classifier used on the live dams run against a rainfall-driven synthetic series generated for your site.
Step 4 · Read the results — current vs predicted level, the selected model with its replay error, the hazard level, the forecast fan chart, the model bake-off, predicted-vs-actual holdout replay, the inputs the pipeline consumed, data-quality checks, the inundation slider that animates the water plane across the 3D terrain, and alerts & action points. Want to track it for real? Promote to live monitoring roster adds the site to the control room — it stays on the synthetic feed until real sensor hardware and telemetry exist there.
Configuration lights sit top-right (maps3d · earth engine · rainfall inputs) and the terrain source & accuracy appear above the map. Everything on this page is simulated — no real dam is affected — and every result is labelled as such.
SANSA live feed — South African National Space Agency
connecting to the SANSA stream…
Streaming collection from SANSA's open SAEOSS Earth-observation catalogue (newest satellite scenes over Southern Africa), collected continuously in the background. Context only — this is not a dam sensor reading.
Step 1 · Pick a site — an address, a pin, or coordinates
the coordinates drive the synthetic rainfall series & terrain
Tell the simulation where your future dam sits. Search an address to auto-fill the coordinates (free Open-Meteo geocoder — no API key), or type a latitude/longitude yourself. Everything downstream — terrain, rainfall, the synthetic water-level series — is generated from these coordinates.
Save renames / moves the selected site · deleting a promoted site also removes it from the live roster
Create or select a site to unlock step 2.
Step 2 · Terrain & virtual sensor placement
3D terrain loads here — the free, open-source MapLibre viewer (real elevation relief + satellite imagery, no API key). If open terrain is switched off it uses Google Maps 3D instead; the chart + pipeline still run either way.
This is the patent Claim-12 “water level detection arrangement”: a virtual sensor at or proximate the dam. Drop the pin at the site centre, or click Place by clicking the map and then click anywhere on the terrain. Optionally set a monitoring elevation.
Virtual water-level sensor (patent Claim 12 — “water level detection arrangement”): place at or proximate the dam, and pick a monitoring elevation.
Sensor placement is optional — you can continue without it.
Step 3 · Configure & run the real prediction pipeline
same ARIMA · ETS · RNN bake-off & hazard Kappa selection as the live dams
The simulation generates an hourly water-level series for your site (driven by the site’s real historical catchment rainfall when the hydrology flag is on), then runs the exact same three-model bake-off and hazard classifier used on the live dams. Choose how far ahead to forecast, how the winner is picked, and how much history to train on.
Preparing…0%
Takes a few seconds — three models are trained, scored on a withheld window, and the winner forecasts the horizon.
Step 4 · Simulation result — what the pipeline produced
Current level
—
Predicted +24 h
—
Selected model
—
Hazard potential
—
Training window
—
Holdout replay error
—
Model agreement (24 h)
—
Input data quality
—
Every number on this page is simulated: the pipeline ran on a synthetic series for your site, so treat results as a demonstration of the method — not a forecast for a real dam.
Forecast fan — all three models · the selected model is bold
thin lines = the models that lost the bake-off
Model bake-off — replay error by model
winner = lowest metric on the withheld window
Predicted vs actual — holdout replay
a trailing window was withheld from training; each model tried to reproduce it
actual (withheld)selected modelARIMAETSRNN
Inputs the simulation used — training series & rainfall
exactly what was fed to the three models
level (ft) — synthetic series, left axisrainfall (mm) — scaled to its own max
Data quality — validation checks & cross-model vote
the same gatekeeper checks the live dams pass before a forecast ships
● info
Run a simulation to see the validation report.
Hydrology picture — rise, rainfall & flow around the site
Inundation overlay on the 3D terrain
circular water-plane approximation — indicative, not an engineering flood map
Run the simulation to animate the forecast water level across the horizon.
Alerts & action points (simulation context)
● info
No actions — run a simulation.
Stays on the synthetic feed until real sensor hardware + telemetry exist at this location.