Borehole siting: how to raise drilling success before you pay the driller

Good borehole siting means knowing, before you sign the drilling contract, how likely a site is to give enough water, how deep the first water-bearing layer sits and how deep to drill. That answer comes from the drilling records already held around the site, read together with the ground surface and a field survey, and it should come with odds you can check, not a single "drill here" mark.

Why boreholes come up dry or weak

A dry or low-yield borehole costs nearly the full drilling price, and the money is spent before anyone knows the result. Where groundwater is plentiful, a site can be chosen mainly on social grounds; where it is not, a British Geological Survey review warns that an exclusively social approach with minimal technical input "can lead to many dry wells and boreholes", and that the same costly mistakes recur because drilling data are rarely collected [1].

The Rural Water Supply Network's code of practice makes the same point from the money side: a borehole is meant to last 20 to 50 years, so the cheapest drilling is not always the most cost-effective, and a borehole that users abandon is not cost-effective at all [2]. Siting is the first place where that long-term value is won or lost.

Where to drill a borehole: what the ground is doing

Weathered basement aquifers

Across much of East Africa, including Uganda, boreholes draw from a weathered basement aquifer: crystalline rock rotted near the surface into a clay-rich regolith, with more permeable weathered and fractured rock just above fresh bedrock. Groundwater is found "where the rocks have been significantly weathered or in underlying fracture zones", weathering ranges from a few metres in arid areas to more than 90 m in the humid tropics, and the deeper the weathering, the more sustainable the supply [1]. A good site and a poor one can therefore sit close together.

The ambition also matters. A continental assessment found that appropriately sited and constructed boreholes can support hand-pump abstraction of 0.1 to 0.3 L/s across much of Africa, but that yields above 5 L/s are far less widely possible [3]. A hand-pump target and a motorised-pump target need different questions asked of the same records.

The hydrogeological survey still matters

A record-based siting study does not replace the field. BGS work for a UNICEF programme in northern Ghana recommends starting with a detailed reconnaissance: geological and hydrogeological maps, lineament maps and records of earlier siting and drilling, then field checks of existing water sources, then geophysical profiling (EM34 or 2D resistivity) along long lines across likely water-bearing features [4]. The same report notes that geophysics is often applied or interpreted poorly when the local hydrogeology is not understood [4]. Records are what give that understanding.

Siting from records: a five-step routine

1. Set the target yield. A hand pump and a motorised pump are different tests. Our drilling odds use 0.5 m³/h (about 0.14 L/s) for a hand pump and 3 m³/h for a motorised pump.
2. Gather every nearby record. Drilled depth, depth to water, depth to bedrock, the driller's log of materials and the yield at completion, including the failures. Dry holes are the most informative records and the most often lost.
3. Put them under the real terrain. Depth to the water table only makes sense against ground level, so a digital elevation model sets the surface and each record is placed on it.
4. Estimate, with uncertainty, at the site. The chance of reaching the target yield, the depth to the first water-bearing layer with a range, and a recommended drilling depth. A single number without a range hides the risk.
5. Check how far the nearest records are. An estimate 50 m from ten good records is not the same as one 400 m from two. Make the distance part of the answer.

The result is a drilling odds map: for every point, the chance that a borehole there would reach the target. It tells you where to spend the geophysics budget, where to drill first and where the records are too thin to say.

How do you know the odds are honest? A blind test

Any model can draw a convincing map. The test that matters is a blind test: hide some wells, predict them from the rest, then compare what the model said with what happened. We ran that test for the KAI drilling odds in the open, on public records from the Township of Langley, British Columbia, Canada. The data come from GWELLS, the province's registered-well database, used under the Open Government Licence – British Columbia [5].

The setting is glacial and glaciomarine sediments: sand and gravel aquifers between clay and till, over bedrock at depth. Of 4,288 wells, 4,129 had a log of materials, 3,771 a yield and 3,252 a water level. We held out 595 wells (15%) and predicted them from the rest; 536 of them had a yield to score. The median distance from a held-out well to its nearest training well was 87 m.

The KAI map answer ranked wet and dry sites with an AUC of 0.83, or 0.84 when read at each well's exact registered location, against 0.81 for the nearest-8-wells rule. At this target 83% of held-out wells succeeded, so even the area rate is a fair guess; the value is in telling the 17% apart.

The second check is the "said against happened" table. Where the model said about 90%, 93% of those 382 wells succeeded; where it said about 74%, 78% of 85 wells did; where it said about 50%, 53% of 30 wells did. The low bands are small and less tidy (about 32% said, 23% happened, 35 wells), which is exactly why the table is published in full.

Three further results are worth knowing before you trust any drilling odds map, ours included:

• Distance decides skill. For the map answer, on held-out wells within 150 m of a training well (395 wells) the AUC was 0.86. Between 150 m and 400 m (127 wells) it fell to 0.69, and the Brier score was no better than the area rate. Thin records mean wide odds.
• Higher targets are harder. For a 3 m³/h motorised pump, 44% of held-out wells succeeded and the AUC fell to 0.67. The odds were still honest (about 50% said, 50% happened over 275 wells) but they separate sites less sharply.
• Depth comes with a range. The depth to the first water-bearing unit was within its stated range for 80.5% of wells, with a typical error of 10.8 m. The recommended drilling depth was deep enough for 78% of successful wells, at a median of 18 m of extra drilling, against 70% for the 80th-percentile depth of the nearest 8 wells.

Glacial sediments are not weathered basement

The blind test is in Canada, in glacial sediments, where water sits in sand and gravel bodies between clay and till. In a weathered basement aquifer the water sits in the weathered zone and fractures above fresh rock, and yields depend on how thick that zone is and whether fractures were struck.

The method can be applied there, because it learns from the materials logged by drillers and from yields at completion, whatever the rock. Its skill does not transfer with it. In weathered basement a borehole's yield depends mostly on how many water-bearing fractures it strikes, and studies on several continents find large variability in the permeability and porosity of the weathered layers [6]; drilling records are also often incomplete [1]. So the skill has to be measured again on local records with the same blind test before anyone budgets on it. Until then, treat a basement drilling odds map as a guide to where to survey, not as a substitute for the survey.

For the Ugandan cattle corridor, the open chapter Groundwater intelligence for Uganda's cattle corridor (Nafula and Kakooza, 2026) describes how borehole records there were turned into a 3D aquifer model.

What KAI offers, and what it does not

KAI turns borehole records into evidence for siting decisions:

• GroundWater Analyst builds a 3D model of the aquifer from your records under the real terrain: water table, top of bedrock, layers and every borehole, shared with a link.
• The drilling odds demo in the Heterogeneity Lab shows the chance of reaching 0.5 or 3 m³/h, the depth to the first water-bearing layer and how deep to drill, with the blind test above. Signed in, you can run it on your own records.
• A borehole siting Site Check gives depth to water and bedrock for one site, a cross-section, the nearby records and a plain recommendation; district aquifer mapping does the same for a whole area.

KAI does not drill, hire out rigs or other equipment, or run field acquisition such as geophysical surveys or pumping tests. Those stay with your hydrogeologist and driller. What we add is the record-based picture before they start, and a check afterwards: every new borehole, wet or dry, is a new record that sharpens the next estimate.

Next step

Open the drilling odds demo and tap a few sites to see the answer and its range. If you hold borehole records for an area you plan to drill, ask us to run the same blind test on them first, so you know how far to trust the map before you pay the driller.