Modelling water quality in New Zealand rivers from catchment-scale physical, hydrological and land-cover descriptors using random forest models

This report is one of two prepared for the Ministry for the Environment by NIWA, which update national indicator data for river water quality in New Zealand and help provide the evidence base for policy and decision makers.

This report models water quality for all river reaches in New Zealand using water quality data and a range of catchment features such as climate, geology, topography and land cover. The models incorporate river water quality data from up to 601 sites between 2003 and 2007, including regional council and National River Water Quality Network data.

The second report presents information of the state of river water quality and recent trends for both regional council sites and the national network sites from 1998 to 2007.

Key findings

The state of water quality in New Zealand’s rivers is highly variable.

Rivers in urban and rural areas generally have poorer water quality compared to rivers in native forest.

Areas where river water quality is significantly deteriorated include lowland areas of Northland, Auckland, Waikato, the east coast of the North Island, Taranaki, Manawatu-Wanganui, Canterbury and Southland.

Based on the modelled water quality, 57 per cent of lowland rivers (rivers lower than 150 metres above sea level) exceed the New Zealand guideline value for total nitrogen and 54 per cent of lowland rivers exceed the total phosphorus guideline value.

Over 70 per cent of the variation in some aspects of water quality, such as nutrient and bacterial levels, can be explained by catchment features (eg, land-cover, climate and geology). This highlights the large influence these features have on water quality and therefore the significant contribution of non-point sources (such as run-off from agriculture) to poor water quality.

The variation in water quality explained by each of the models ranged from 40 per cent for the poorest fitting model (suspended solids) to 78 per cent for the best fitting model (total nitrogen). The models explained greater than 60 per cent of the variation for eight of the 11 parameters and greater than 70 per cent for four parameters.

For the strongest models (eg, conductivity, total nitrogen and total phosphorus) water quality can be identified down to about a 1 km scale. For example, isolated pockets of farmland along the west coast of the South Island and forest remnants in intensively farmed areas of Bay of Plenty, Waikato and Northland are evident.