WayCast identifies the locations on your route where water accumulates and informs you of the rainfall required to submerge them. It achieves this by analyzing the road’s topography, including its dips and the elevation of the surrounding ground, which determines its ability to trap water. This information is then combined with the rainfall forecast and the drainage and culvert maps along the route.

Unlike commercial navigation apps, WayCast provides this level of detail. In the context of an Indian monsoon, a driver is often stranded not by a river in flood but by an underpass or a low junction that becomes submerged after receiving 60 mm of rainfall within two hours. This is precisely the problem WayCast was designed to address.

The weather data displayed by WayCast is live satellite imagery, not a forecast. It utilizes Meteosat-9, a satellite positioned over the Indian Ocean at 45.5°E, which captures the entire subcontinent every 15 minutes. The data includes rain rate measurements obtained through infrared and microwave sensors, storm tops, cloud height, and air-mass instability. A forecast predicts future weather conditions, while WayCast provides real-time observations. Each image is accompanied by the timestamp, allowing users to determine its recency. The service operates approximately 30 minutes behind real-time due to the time required for data transmission and processing from orbit.

Furthermore, WayCast integrates weather data with your journey. Navigation apps display routes, weather apps provide forecasts, and neither records the interaction between the two. WayCast compiles a travel log for each drive, capturing the road’s elevation changes, the rainfall experienced, the temperature and pressure at the time, and the low points encountered. This comprehensive data allows users to analyze their journey against the conditions that influenced it. Consequently, a flooded junction can be explained rather than merely remembered, and the same drive can be recorded twice for further analysis.

It is important to note that WayCast’s results are based on modeling rather than direct measurement. The app does not have access to real-time data from the road itself. Instead, it calculates the expected outcomes based on the terrain and the forecast, which is a valuable tool for decision-making. However, it should be used as a supplementary source of information alongside your own judgment and official guidance.

Components and quality assessment

• Rainfall — a forecast on a 9 km grid. It effectively resolves monsoon systems but poorly handles cloudbursts over one junction.
• Ground height — a 90 m elevation model. Below approximately 2 m of relief, it cannot distinguish a dip from noise, and an underpass is smaller than one of its pixels. This is the limitation the barometer below exists to remove.
• Crossings and drains — OpenStreetMap, which is generous in cities and sparse in rural areas. An unmapped ford is simply absent; the app cannot warn users about unrecorded features.
• Water depth — a mass balance over a guessed catchment. Treat it as an order of magnitude, not a precise number.
• Dynamic environment — the model is subject to real-world changes. A culvert may block with debris, a pump may lose power during a storm, a new wall may alter water flow, and roadworks may lift a carriageway. None of these changes are reflected in the model, and any of them can transform a safe prediction into an inaccurate one within an hour.

The barometer in your phone is what resolves a causeway. A 90 m elevation model cannot see a dip of one or two feet, and one or two feet is exactly the depth that strands a car. So WayCast does not rely on the terrain model alone. Every phone running it logs air pressure as you drive, and pressure falls and rises with height: the app converts that into the rise and fall of the carriageway itself, in 20 m steps along the road.

A single drive is not a measurement. Pressure also moves with the weather, the phone warms in the sun, and a window opening changes it — so one pass carries an offset and a drift that belong to that pass alone. What removes them is repetition. The offset and drift differ on every drive, so averaging several independent drives over the same 20 m of road cancels them, while the real dip stays put. WayCast will not publish a height until at least three separate vehicles have driven that bin.

Where the one-to-two-foot figure comes from. It is not a claim, it is arithmetic, and it is worth following because it also shows what would break it.

Step 1 — how much pressure is a foot of height? Air pressure falls as you climb. How fast depends on the temperature of the air column, through a quantity called the scale height, which on a 32 °C Indian road works out at about 8,930 m. Pressure is measured in hectopascals (hPa, the millibar of old weather reports; sea level is about 1013). Dividing through:

• 1 foot (0.305 m) of height = 0.034 hPa of pressure
• 2 feet (0.610 m) = 0.068 hPa
• 1 metre = 0.112 hPa

Step 2 — can the phone even feel that? Yes, comfortably. The pressure sensor in a modern phone reads in steps of about 0.01 hPa, and by the figures above one such step is 0.089 m — three and a half inches. So the sensor is roughly four times finer than the one-foot feature we are trying to see. The sensor is not the limit.

Step 3 — the equation has to be the right one. Most phone code converts pressure to height with the International Standard Atmosphere formula, which assumes the air is 15 °C. An Indian monsoon road is 30–35 °C, and the height you recover scales with the absolute temperature of the column, so that assumption reads every dip about 6% too shallow. On a real 2 m dip measured at 32 °C that is a 141 mm error — 31% of the 1.5 ft we are aiming at, spent before any measurement is taken. WayCast therefore uses the hypsometric equation with the actual air temperature and humidity, which recovered the same 2.00 m dip exactly.

This distinction matters more than it looks. A 6% scale error is systematic — the same size and the same direction on every single pass — so a hundred drives reproduce it perfectly and averaging never touches it. It is the one error repetition cannot fix, which is why it has to be right in the equation.

Step 4 — what repetition can fix. Each drive also carries an offset and a drift of its own: the weather changes during the pass, the phone warms in the sun, a window opens. These differ from drive to drive, so they behave like random noise and they shrink when drives are averaged — by the square root of the number of passes, which is the standard result for averaging independent measurements. Measured on the library as it stands, the spread between drives over the same 20 m of road is 0.23 m. So:

That is the whole argument. Three independent drives put the uncertainty at about five inches, and a one-foot dip stands more than twice clear of it. The same dip sits at one-seventh of the terrain model's noise, where nothing can be said about it at all. Three drives is also exactly why WayCast refuses to publish a height until three separate vehicles have crossed that bin — below that the number would not have earned the right to be believed.

The precision is therefore already proven; what is still missing is coverage. Of 4,135 distinct 20 m bins contributed so far, 3,905 have been driven exactly once and only 11 have been driven three or more times. A published road profile also needs 25 neighbouring bins — half a kilometre of continuous road — and the longest unbroken stretch is currently four. This is not a flaw in the method. It is simply the number of drives, and it is the one thing only you can change: every time you drive a road with WayCast open, you are surveying it.

Every model is an approximation, including this one. This is not a defect unique to WayCast; it is a fundamental characteristic of models. Weather forecasts, flood forecasts, and bridge designs are all approximations, and all of them are valuable because their errors are known and bounded. The current task is to establish these bounds for this specific model.

This season serves as a testing ground, and your reports are crucial. Each time you confirm that a junction flooded as predicted or report that it remained dry, that site undergoes calibration. The trigger for the prediction becomes a measured figure rather than a computed one. Confirmations are as valuable as corrections. A model tested against sufficient real-world data ceases to be a guess and becomes evidence. That is precisely what we intend to demonstrate.

What WayCast does not replace. WayCast does not replace the India Meteorological Department, the Central Water Commission, or local police and authorities. If any of them advises against traveling or using a particular road, they are correct, and this app should not contradict their guidance. Never enter flowing water based solely on a screen’s indication of survivable depth, as 15 cm of moving water can easily float a small car.

Supported by W.I.F.E., Vadodara

Apache Licence 2.0 · provided without warranty of any kind · Dr. Dhananjay A. Sant, National Institute of Advanced Studies, Bengaluru