How to specify a river water level station: range, accuracy, mounting, power and data quality

A water level station fails in ways that are rarely visible on a datasheet: the beam clips a pier, the panel is sized for summer, the modem drains the battery, the reading is right in the lab and wrong in a flood. This guide treats the specification as an engineering exercise. Each step is a calculation or a test you can do before you order. It applies to any make of sensor and logger.

Set the measuring range from site data

Start from the river, not the sensor. You need three numbers:

• Lowest stage you want to record (dry-season and drought levels).
• Highest flood level on record, and the design flood you are planning for.
• Mounting height of the sensor above a fixed datum.

The sensor must measure the distance from the mounting point to the water across the whole swing:

d_min = mounting height − highest expected stage     (water close to the sensor)
d_max = mounting height − lowest stage               (water far from the sensor)

Two constraints follow:

1. d_min must be greater than the sensor's blocking distance (the dead band close to the antenna where it cannot measure), with a margin. The margin has to cover floods higher than any on record, afflux at a bridge, and debris piling against the structure.
2. d_max must be inside the sensor's maximum range with a signal margin, since the echo weakens with distance.

Use the highest flood mark you can find on the ground and in local records, not a figure from a brochure. Hill rivers can rise by metres within hours, and a sensor that cannot see the peak is blind exactly when it matters.

Derive the accuracy you need from the discharge you want

Accuracy is not a number to maximise; it is a number to match to a purpose. If the station supports a rating curve Q = C(h − h0)^b, the discharge error from a stage error is:

ΔQ / Q  ≈  b · Δh / (h − h0)

Rearranged, the stage accuracy you need for a target discharge accuracy is:

Δh  ≈  (ΔQ/Q) · (h − h0) / b

For a 2 % discharge target and b = 2, the stage must be good to 1 % of the depth over the control: 3 mm at 0.3 m, 10 mm at 1 m, 50 mm at 5 m. The WMO Manual on Stream Gauging gives the same reasoning, using the sensitivity of the control: a 5 mm stage error costs 1 % where a 1 % change in discharge moves the stage 5 mm, and 10 % where it moves the stage only 0.5 mm [1]. The Manual also states that gauge readings should normally be made to the nearest 0.003 m, and that the acceptable error for a pressure transducer is 0.003 m or 0.10 % of the effective stage, whichever is greater [1].

So decide first which of three uses applies:

Check the geometry of the beam

For non-contact radar or ultrasonic sensors, the beam footprint at the water surface is:

footprint diameter = 2 · d · tan(θ / 2)

At 10 m a 3° beam covers 0.5 m, an 8° beam 1.4 m and a 12° beam 2.1 m. Walk the site with these numbers and check:

• No pier, abutment, rail, cable or branch lies inside the footprint, or in the cone around it where side lobes spill energy.
• The sensor is mounted clear of the structure it hangs from. A bracket extending 0.5 m beyond a bridge parapet, with a rigid mount, avoids the structure and the echoes it returns.
• The beam points vertically. A tilt of θt shifts the measured distance by a factor of 1/cos θt; a 2° tilt over 10 m adds about 6 mm. A mounting with a level reference and a lock is cheap insurance.
• The mount is stiff. A mount that flexes with wind or with the sun's heat moves the zero of your record by the same amount.

Decide how the station will sample

Sampling is a design decision with consequences. The Central Water Commission's field manual shows the underlying concern for manual gauging: during the monsoon on small catchments the greatest frequency of observation is needed, and in rapidly changing flow the true maximum can be higher than the highest hourly reading, because it falls between the observations [2]. An automatic station can do much better, but only if you set it up to.

A station that logs every 15 minutes will see a 1 m per 10 minute rise as 1.5 m between readings, which is enough to alarm, but it will report the peak up to 15 minutes late. Good practice:

• Normal interval (for example every 5 to 15 minutes) to save power.
• Adaptive interval: drop to every minute or faster when the level exceeds a threshold or the rate of rise exceeds a limit.
• Burst averaging within each reading (for example 10 to 30 seconds of samples, reported as a median or mean) to suppress waves, the same practice the CWC manual describes for a human observer taking the mean over about 30 seconds [2].
• On-device alarm. The decision to send an urgent message must not depend on the server.

Size the power system

Most river sites have no mains supply, so energy is a design problem. Do the arithmetic for your own components, with this structure. The figures below are illustrative, to show the method; replace them with datasheet values.

Note how the modem dominates. Transmitting less often, or batching several readings into one connection, saves more energy than any sensor change. Then:

Panel (W)   =  daily energy (Wh) / (peak sun hours × system efficiency)
Battery (Wh) = daily energy × days of autonomy / usable depth of discharge

With 2 peak sun hours (a cloudy monsoon month at a hill site, not the annual average), 70 % system efficiency, 5 days of autonomy and 50 % usable depth: panel 3.4 / (2 × 0.7) ≈ 2.4 W, battery 3.4 × 5 / 0.5 = 34 Wh. Then add margin for temperature, because cold cuts battery capacity, and for dust or snow cover on the panel. Size for the worst month, not the average.

Plan the telemetry

• Link. GSM (2G / 3G / 4G) covers most sites. Measure signal strength on site at the antenna height, not at the road. For valleys without coverage, LoRa to a gateway or repeater, or a satellite link, are the usual alternatives. A hybrid design with a local LoRa network and a cellular gateway is common in remote catchments [4].
• Store and forward. The logger must keep readings during an outage and send them when the link returns. At 5-minute readings, 30 days is 8,640 records, which is trivial memory, so insist on at least that.
• Message size and protocol. Each reading is a few tens of bytes. A lightweight protocol such as MQTT with acknowledgements suits intermittent links, and Modbus or SDI-12 on the sensor side keeps integration simple.
• Clock. Every reading carries a timestamp from a clock that is synchronised, because a flood record with a wandering clock is hard to trust.
• Health data. Battery voltage, panel current, signal strength and logger temperature travel with every message, so a failing station is found before it fails.

Write the data quality rules

Raw readings include spikes, dropouts and stuck values. Put rules in the logger and again on the server, and keep the flag with the value:

Do not delete suspect values. Keep them with a flag, because you cannot recover a value you discarded.

Write the acceptance test

A station is accepted when it has been tested against a reference, at your site, across the range. A reasonable acceptance procedure:

1. Bench and temperature test. Distance accuracy at several ranges and temperatures, as done in laboratory evaluations of radar sensors by the USGS Hydrologic Instrumentation Facility [3].
2. Zero and range check on site. Compare the sensor with a surveyed staff gauge or tape-down at several stages, reading the reference to the nearest 3 mm [1].
3. High-flow check. A reading during a significant rise, comparing against a reference when waves and mounting movement are greatest.
4. Communication test. Pull the SIM, wait, and check that the records are back-filled in order when the link returns.
5. Power test. Measure the actual current and check the energy budget against the sizing above.
6. Repeat each season and keep a log of every adjustment.

Maintenance and uptime

Telemetry networks fail more from neglect than from design. A Comptroller and Auditor General review of India's flood forecasting scheme, summarised by PRS Legislative Research, found that as of August 2016, 59 % of the existing telemetry stations were non-functional, leaving no real-time data for the corresponding periods [5]. The lesson is to specify maintenance as part of the system:

• Choose devices with no moving parts and no contact with the water, so site visits can be rare.
• Make the station report its own health so the server, not a visit, finds the fault.
• Set an availability target. 95 % uptime is about 18 days of missing data a year, 99 % is about 3.7 days.
• Plan spares and the response time of whoever services the site.

Where ClimMaTech fits

Tarang is our non-contact radar level sensor, specified at ±1 mm accuracy and 0.5 mm resolution, with a measuring range up to 50 m, customisable to the site. It works from -40 °C to +70 °C, reports over GSM (2G / 3G / 4G) with LoRa and other options by site, and runs on solar power. Pravaah adds surface velocity for AWVLR stations, and DeviceCloud brings the data together. For the physics and error sources of each sensor type, see radar, ultrasonic, float and pressure sensors compared. To discuss a layout for your river, talk to us.

References

1. World Meteorological Organization, Manual on Stream Gauging, Volume I: Fieldwork, WMO-No. 1044 (2010). WMO publication page.
2. Central Water Commission, India, Gauge and Discharge Measurement Field Manual. cwc.gov.in.
3. U.S. Geological Survey, Hydrologic Instrumentation Facility, Laboratory evaluation of a radar water-level sensor. usgs.gov.
4. Water level monitoring using LoRa and cellular gateways in remote areas (conference paper). Atlantis Press.
5. PRS Legislative Research, CAG report summary on schemes for flood control and flood forecasting. PRS India.

Calculations are illustrative and use stated assumptions. Replace them with the datasheet values of the equipment you are considering.