Shallow groundwater characterisation and hydrograph classification in the coastal city of Ōtautahi/Christchurch, New Zealand

The problem

Groundwater lies at shallow depth beneath many low-lying coastal cities. Sea-level rise can raise this groundwater (Habel et al., 2023; Knott et al., 2019). Rising groundwater can damage pipes and basements below the ground surface and roads and buildings above it (Befus et al., 2020; May, 2020; Su et al., 2022; Thorndahl et al., 2016).

Christchurch is a coastal city of more than 400,000 people, built on flat, low-lying sediments close to rivers and the sea. Following White et al. (2007), vegetation recorded on maps compiled in 1856 suggests that wetland or near-surface groundwater covered 5,215 ha of the Avon-Heathcote area before the city developed. That is about 52 km², an area roughly 30% the size of the 173 km² that we mapped (our arithmetic). By 2015 to 2016, less than 6% of that wetland remained. Our goal in this study was to characterise this shallow groundwater, including how its depth varies in space and time and how it relates to natural and human stressors.

What we did

Most of our records came from the Automatic Piezometer Project network. The Christchurch City Council acquired it from the Earthquake Commission in 2016, after the 2010–2011 Canterbury Earthquake Sequence (CES). Its 247 sites have loggers that record the water table every 10 minutes. We added long-term manual wells and regional council sites, and selected sites with records from 2016 to 2020. Because land subsided during the earthquakes, ground elevation models are uncertain. We therefore worked with depth to groundwater, which is likely to be more accurate than a water-table elevation.

First, we interpolated median depths between wells onto a 200 m grid. We then used Moran's I, a statistic of spatial similarity, to check whether the size of water-level fluctuations was grouped in space. We tested each record for a trend with the seasonal Kendall test, which allows for seasonal variation. Next, we sorted 210 records, each 39.5 months long and sampled every 4 hours, into groups of similar shape with k-means clustering. Finally, we used cross-correlation to measure how closely groundwater levels matched ocean levels at the Sumner tide gauge over 25 to 31 December 2019.

What we found

Across 265 sites, the mean depth to groundwater was 1.39 m. The median depth was 1 m or less below about 62% of the interpolated area, 1 to 2 m below 35%, and more than 2 m below only 3%. At 91% of sites, the standard deviation of depth (a measure of the size of water-level fluctuations) was 0.3 m or less.

From the coast, along the estuary and inland to Ensors Road bridge on the Heathcote River and Fitzgerald Avenue bridge on the Avon River, the median water table lies below mean high water springs (MHWS). MHWS is the average high tide during spring tides, the tides with the largest range. This part of the eastern suburbs is likely to experience shoaling (a rise of the water table towards the surface) and flooding as the sea rises.

Neighbouring sites tended to have similar fluctuation sizes. Small fluctuations clustered north of the central city, in Fendalton, Merivale, St Albans and Edgeware, and near Travis Wetland. Large fluctuations clustered in Avonside, Richmond and Hoon Hay. Of the sites with the smallest fluctuations, 80% had a median depth within 1.5 m of the ground.

From 2016 to 2020, the water table was probably rising at 40% of the 245 automatic sites and probably falling at 46%, with no clear trend at the rest. Long-term manual wells showed 32% rising and 50% falling before the CES. The recent falls are consistent with a gradual decrease in rainfall. However, the rises did not follow rainfall. Their cause is unexplained, but one possible cause is the sealing of sewer networks (Becker et al., 2022).

Clustering sorted the 210 records into groups of 15, 43, 106 and 46. The group of 15 had flashy records, with frequent, sharp peaks, from sites concentrated in a small area north-west of the central city, all within 200 m of surface water. The other three groups had no clear relation to depth, fluctuation size or distance from surface water. Clustering successfully classified responses to rainfall and stream flow, but dewatering (pumping out groundwater) was harder to pick out because its signal overlaps with tides and rainfall. The similar shapes of most records hint at the potential to move some loggers to new locations, from sites whose trends and statistics resemble others and that lie outside the shallowest areas.

The closer a site was to a tidal river, the more closely its groundwater level followed the ocean tide. The ocean seemed to influence shallow groundwater mainly through the estuary and tidal rivers. These preliminary results come from one week of records.

Limitations

Monitoring has several large gaps along the coast and near the estuary. We also assumed a depth of 0 m at the coast. As a result, the interpolation placed the water table at 0.7 and 0.9 m near the Heathcote River mouth and in South New Brighton, against measured medians of 1.5 and 2.4 m. Over less than 4 years of records, our methods did not detect interactions between groundwater and underground infrastructure. Numerical modelling may represent these interactions better (Attard et al., 2016a, b). We recommend continued monitoring with continuous sensors, with priority near the tidal rivers, the estuary and the coast, so that changes in groundwater under a rising sea can be detected.

Glossary

Groundwater is water held in the pores of soil and sediment below the ground surface.

The water table is the top of the groundwater, where the sediment becomes saturated.

Depth to groundwater is the distance from the ground surface down to the water table, here in metres below ground level.

The Canterbury Earthquake Sequence (CES) is the series of earthquakes that struck Christchurch in 2010 and 2011.

A hectare (ha) is 10,000 m², and 100 ha make 1 km².

Interpolation estimates values between measurement points to make a continuous map.

Standard deviation measures how far a record varies about its average.

Moran's I is a statistic that measures whether values at nearby sites are more similar than values at distant sites.

The seasonal Kendall test detects a trend in a time series while allowing for regular seasonal variation.

K-means clustering is a machine-learning method that sorts records into a chosen number of groups by similarity.

Cross-correlation measures how closely two time series match, allowing one to lag behind the other.

Mean high water springs (MHWS) is the average height of high tide during spring tides, 1.368 m in the New Zealand Vertical Datum 2016.

Spring tides are the tides with the largest range between high and low water.

Shoaling groundwater is groundwater whose level is rising towards the ground surface.

A flashy record rises and falls quickly, with frequent, sharp peaks.

Dewatering is the removal of groundwater by pumping or drainage.