Choosing a Site for a Rain Gauge Network
A regional water authority has budget for only four automatic rain gauges to represent a 3,200 km² catchment. Existing records show three zones: a coastal plain (0–200 m, 40% of area, mean annual rainfall 640 mm), a mid-slope belt (200–700 m, 35% of area, 1,150 mm), and a high ridge (700–1,400 m, 25% of area, 1,850 mm). Two candidate sites are proposed: Site A sits on the coastal plain near a town, is easy to access, and has 30 years of manual records. Site B is on the high ridge, requires a 4-hour hike, but lies in the wettest zone. A colleague argues that because the plain covers the largest share of area, most gauges should go there. A farmer downstream complains that floods are under-predicted, and the authority suspects the ridge is under-sampled.
1. Evaluate the colleague's argument that gauge allocation should follow the largest area share. What principle should guide allocation instead?
Read the worked answer
The colleague is applying an area-proportional rule, which is defensible only if rainfall varied little across the catchment. Here it varies almost threefold (640 to 1,850 mm), so equal-area allocation would systematically under-represent the ridge, which generates much of the runoff. The guiding principle is stratified sampling weighted by variability and hydrological influence, not by area alone. A sound approach: (1) treat the three elevation bands as strata; (2) allocate at least one gauge to each stratum so no zone is unmeasured; (3) give extra weight to the ridge because it is both the wettest and the most variable, and because it drives downstream flood peaks; (4) use the plain gauge mainly to anchor the long record and calibrate. With four gauges, a reasonable split is one on the plain, one in mid-slope, and two on the ridge. The colleague's rule would leave the ridge with one gauge at most, which cannot capture its spatial spread.
2. Weigh Site A against Site B using the criteria of representativeness, record continuity, and practical reliability. Recommend how each site should be used.
Read the worked answer
The two sites serve different purposes and need not compete.
- Representativeness: Site B is far more valuable for spatial coverage because the ridge is under-sampled and contributes disproportionately to runoff. Site A represents a zone that is already comparatively well understood and is the least variable.
- Record continuity: Site A's 30-year manual record is a major asset. Breaking it would destroy a benchmark series used for trend analysis and for calibrating the automatic gauge.
- Practical reliability: Site A is easy to service, so data gaps are unlikely. Site B's 4-hour hike raises the risk of delayed maintenance, battery failure, and lost data, especially in winter.
Recommendation: Keep Site A as a continuing reference station, ideally co-locating the new automatic gauge with the manual one so the record continues unbroken. Install Site B as a new automatic station but pair it with remote telemetry, solar power, and a maintenance schedule that accounts for access time. Do not treat the two as alternatives; they answer different questions.
3. The farmer says floods are under-predicted. Explain how the sampling design could cause this, and state one check the authority should run before blaming the model.
Read the worked answer
If the ridge is under-sampled, the rainfall input to the flood model is biased low exactly where the heaviest rain falls. A model calibrated on plain-dominated data will learn a rainfall–runoff relationship that understates how much water the ridge delivers, so predicted peaks will be too low. The bias compounds because the ridge is both wetter and more variable, meaning extreme events there are the ones most likely to be missed.
Check to run first: compare observed and modelled peak flows for several past events and inspect whether the errors grow with the share of catchment rainfall falling on the ridge. If residuals are systematically negative when ridge rainfall is high, the problem is input sampling, not the model structure. Only after ruling that out should the authority revise the model itself. This ordering avoids recalibrating a model to compensate for a data gap, which would embed the bias rather than remove it.