Evidence for the effects of land use on freshwater ecosystems in New Zealand

ABSTRACT

To meet the challenges of preventing and reversing adverse effects of land use on ecosystems, management actions need to be founded on strong evidence. We used the pressure-state-impact (PSI) framework to assess evidence of land-use effects on New Zealand freshwater ecosystems. The evidence consisted of published quantitative and categorical associations linking land-use pressures to state changes and ecological impacts in rivers, lakes and aquifers. There was substantial evidence of land-use effects, particularly where land use/land cover (LULC) classes were used as pressure variables. Proportions of catchment area in urban and pastoral LULC were consistently, positively correlated with contaminant levels in water bodies and negatively correlated with ecological-health indicators. Other consistent PSI associations included positive correlations between cattle stocking rates and river contaminant levels, increased fine sediment and decreased ecological-health scores in rivers following forest harvest, and increased river contaminant levels at sites with stock access. Despite these consistent associations, the evidence base has four general shortcomings that should be addressed: (1) inadequate integration of data and models that link land use and contaminant loss to state changes and impacts in freshwater ecosystems; (2) weak inferences based on LULC; (3) reliance on categorical PSI associations; (4) gaps in reported PSI associations.

KEYWORDS:

• Contaminants
• evidence
• freshwater ecosystems
• land use
• land cover
• land management practice
• pressure-state-impact

Introduction

Land areas used for agricultural, forestry, and urban (commercial, residential, industrial, transport) purposes are major sources of exogenous sediment and nutrients in freshwater environments in New Zealand (Basher 2013; Dymond et al. 2013, 2016). While soils and nutrients are valuable resources on land, they are stressors in aquatic ecosystems when input rates exceed natural assimilative capacities and cause water-quality degradation. In turn, these changes in environmental state cause proliferations of phytoplankton and benthic algae, loss of sensitive species, and other ecological impacts (Biggs 2000; Schallenberg and Sorrell 2009; Reid et al. 2011). In addition to ecological impacts, degraded environmental state has negative impacts on cultural and socio-economic values such as customary harvests, contact recreation and food safety (Donnison et al. 2009; Harmsworth et al. 2011; McBride et al. 2012).

Preventing degradation of aquatic ecosystems by land-derived stressors and reversing degradation where it has already occurred are among the greatest challenges in environmental management. Meeting these challenges requires management actions and policies such as setting limits on contaminant losses. The most effective management actions and polices are based on predictive relationships that link land use, stressor levels in receiving environments, and the impacts of those stressors on ecological and socio-economic values (Snelder et al. 2004; Clapcott et al. 2011; Schallenberg et al. 2017). These relationships are needed to identify land-use intensity levels beyond which stressor levels are excessive and ecological values are degraded, and to forecast the impacts of future land-use change (Larned and Schallenberg 2019). There is widespread acknowledgement of the need for evidence-based environmental management and policies, where the evidence consists of or includes stressor-response relationships (Hill and Arnold 2012; Norris et al. 2012; Nichols et al. 2017).

Relationships that link land uses to stressor levels in receiving environments, and stressor levels to ecological impacts are presumed to form ordered sequences of causally connected relationships or ‘causal chains’ (Walmsley 2002; Odada et al. 2004). However, most reports of land-use effects on aquatic ecosystems are limited to bivariate relationships and categorical associations between land uses and stressors or between stressors and impacts, without consideration of causal chains. In rare cases, path analyses and simulation models have provided evidence of causal chains by coupling land use-stressor relationships and stressor-impact relationships in multiple-step sequences (e.g. Greenwood et al. 2012; Villeneuve et al. 2018). While these sequential approaches are improvements on single bivariate relationships, they invariably omit steps such as stressor attenuation between source areas and receiving environments. Despite the missing steps, the land use–stressor–impact sequence is useful for conveying information about land-use effects and for risk assessments.

The land use–stressor–impact sequence corresponds closely to the ‘pressure-state-impact’ (PSI) framework, which is widely used to guide policy development, impact assessments and management responses (Walmsley 2002; Oesterwind et al. 2016). The PSI framework is also used for environmental reporting in New Zealand, as required by the Environmental Reporting Act 2015 (MFE 2019; PCE 2019). We used the PSI framework in this study to assess published evidence about land-use effects.

Weaknesses and gaps in the evidence of land-use effects on aquatic ecosystems are impediments to environmental policy development and implementation in New Zealand (Duncan 2014, 2017; Rouse and Norton 2017; PCE 2019). In this paper, we assessed the current evidence of land-use effects in New Zealand rivers, lakes and aquifers based on an assessment of >100 published papers and technical reports. We identified major weaknesses and gaps and set out recommendations for strengthening the evidence base.

Land use, land cover and land management practices as pressures

Although land-use pressures are frequently investigated in New Zealand, the term ‘land use’ is not always defined explicitly. Instead, it is used to refer to a wide range of activities and spatial characteristics and is often used interchangeably with ‘land cover’ and ‘land management practice’. Where they are distinguished, land use (LU) describes the purpose for which a land parcel is used (e.g. urban residential, dairy farming), land cover (LC) describes observable features on the land surface (e.g. exotic grassland, plantation forest), and land management practice (LMP) refers to activities and inputs and outputs used to achieve a given land use (e.g. fertiliser application, clear-fell harvesting) (ACLUMP 2010; Verburg et al. 2011).

LU and LC are closely related variables; both concern the types and spatial distribution of vegetation and structures on land, and analyses of both are based on classification systems that assign areal cover or spatially dominant-use classes to land parcels, map polygons or catchments (Jansen and Gregorio 2002; Rutledge et al. 2009; Verburg et al. 2009). The most recent national LC database for New Zealand is the Land Cover Database (LCDB), based on satellite imagery from the summers of 1996/97, 2001/02, 2008/09, and 2012/13 (https://lris.scinfo.org.nz). The most recent national LU databases for New Zealand are the Land Use and Carbon Analysis System (LUCAS) and Land Use New Zealand (LUNZ) (Manderson et al. 2018). Both databases incorporate data from the LCDB and Agribase, a national spatial farms database. The classifications used for LCDB, LUCAS and LUNZ are broadly overlapping, and some classes (e.g. planted forest) are common to all three. Many studies of land-use effects in New Zealand have used combinations of LU and LC classes as pressure variables. In this review, we pooled studies based on LU and/or LC classes into a single group, LULC.

In contrast to LULC, there are no standard procedures for classifying LMPs in New Zealand. In the absence of standard procedures, we grouped LMPs into three broad classes: agricultural, forestry and urban.

Mitigation systems and strategies to prevent or reduce contaminant loss (e.g. stock exclusion, deferred effluent irrigation) may be considered LMPs, as they affect contaminant input to, and ecological impacts in, aquatic receiving environments. Mitigation systems were excluded from this review, as their use in New Zealand has been reviewed recently (e.g. Van Roon 2011; McDowell, Monaghan et al. 2018; McDowell, Schallenberg et al. 2018). However, some mitigation studies were used to infer land-use effects in the absence of mitigation. For example, information about the effects of livestock access to rivers was inferred from published comparisons of river conditions with and without stock access.

Methods

We used the PSI framework to organise and assess evidence. As used here, ‘pressures’ in the PSI framework are levels of land-use activities or the areal extent of land use or land cover, ‘states’ are levels of stressors in aquatic environments, and ‘impacts’ are responses in ecological, cultural and socio-economic values to changes in state levels. In this study, state variables were dominated by physical–chemical variables such as sediment and nutrient concentrations and impact variables were dominated by ecological and human health-related variables such as macroinvertebrate abundance, faecal bacteria concentration, and metrics used as indicators of ecological health (e.g. lake trophic state).

Google Scholar and Web of Science searches were used to locate published papers and publicly available technical reports dated after 1975. We screened the searches for publications that contained quantitative or categorical PSI associations in New Zealand. These publications included observational studies and process-based and statistical models that were set up and run for New Zealand locations. Quantitative associations were based on bivariate relationships between continuous variables, most of which were characterised using correlation or regression analyses. Categorical relationships were based on differences between the levels of state and impact variables in different LULC or LMP classes and between different ‘treatments’ (e.g. upstream and downstream of a land-use activity).

We focused on three classes of freshwater receiving environments (rivers, lakes and aquifers) and five classes of land-derived contaminants (nitrogen (N), phosphorus (P), sediment, metals, faecal bacteria). Associations between metals and land use were limited to urban centres and adjacent rural areas due to data shortages for other areas. Other land-derived contaminants such as pesticide residues were excluded due to data shortages. We excluded hydrological alterations as a land-use pressure because water abstraction for agricultural and urban land use has been poorly quantified in New Zealand (Booker 2018).

We considered wastewater management practices that involve land-application (e.g. land-disposal of municipal sewage effluent) to be LMPs and included reports concerning these practices. We excluded reports of point-source effluent discharges to aquatic receiving environments (e.g. direct discharge of effluent and stormwater to rivers and lakes), as the land involved is limited to that used for wastewater treatment plants and other infrastructure.

Quantitative assessments of the consistency of results in the compiled reports (e.g. meta-analyses) were not undertaken due to high variability in study designs and insufficient numbers of reports for some combinations of receiving environments and pressure, state and impact variables. In contrast, numerous reports had correlations between proportions of agricultural and urban land cover and river state and impact variables. High consistency in correlation direction across these reports made meta-analyses unnecessary.

Results

Rivers

We first summarise PSI associations between LULC classes and river contaminants and ecological impacts, then PSI associations between LMPs and river contaminants and impacts. The evidence we compiled concerning rivers came from both multi-site studies and single-site case studies. Case studies have limited use for making broad inferences about land-use effects; we only considered case studies when no multi-site or large-scale reports were available for a given PSI association.

LULC classes

Associations between LULC classes and state and impact variables are scale-dependent and the studies of New Zealand rivers reflect this dependence. Analyses of LULC-river associations have been based on spatial LULC data for the catchment areas upstream of sampling sites and for riparian zones of varied width and length (Table 1). Association strength at catchment and riparian scales have been compared in several studies (e.g. Buck et al. 2004; Burrell et al. 2014). LULC variables have been used both as continuous predictors in correlation analyses, and as categorical predictors in comparisons between LULC classes. In the latter cases, sites were assigned to LULC classes based on the spatially dominant class in the catchment or on rules that reflect the disproportionate effects of different LULC classes (e.g. Snelder and Biggs 2002).

Table 1. New Zealand studies of associations between land use/land cover (LULC) and state and impact variables in rivers. Abbreviations: TN – total nitrogen, NO₃-N – nitrate nitrogen, NH₄-N – ammoniacal nitrogen, TP – total phosphorus, DRP – dissolved reactive phosphorus, SS – suspended sediment, EPT – Ephemeroptera, Plecoptera and Trichoptera, MCI – macroinvertebrate community index, QMCI – quantitative MCI, IBI – index of biological integrity, UCI – urban community index, QUCI – quantitative UCI.

LULC variables	Spatial extent	Summary	Reference
LULC in upstream catchments
Native forest, planted forest, urban, pastoral, tussock cover (%)	National (101 sites)	Mean DRP concentrations were positively correlated with % planted-forest cover, mean TN concentrations were negatively correlated with % native-forest cover.	Close and Davies-Colley 1990
Pastoral cover (%)	National (51 sites)	Median invertebrate taxon richness, Shannon diversity, predator biomass and EPT abundance were higher at sites with <30% pastoral cover than at sites with >30% pastoral land cover.	Quinn and Hickey 1990
Natural, planted forest, urban, crop land, pastoral cover (%)	National (77 sites)	Median TN and NO3-N concentrations were positively correlated with % pastoral, planted-forest, cropland and urban cover; median TP and DRP concentrations were positively correlated with % pastoral, planted-forest and urban cover.	Ballantine and Davies-Colley 2014
LULC classes based on dominant cover: horticulture, forestry, urban, industrial, dairy, non-irrigated grazing	National (723 sites)	Median DRP concentrations at dairy and industrial sites were higher than all other land-use classes.	McDowell et al. 2015
Urban, high-intensity agriculture (high-producing exotic grassland, short-rotation crops, orchards, vineyards) (%)	National (365–882 sites)	Median TN, NO3-N, NH4-N, TP, DRP and E. coli concentrations were positively correlated and visual clarity and MCI negatively correlated with % urban and intensive agricultural cover.	Larned et al. 2016, 2019
High-producing exotic grassland, native and exotic deciduous forest, planted forest, scrub-grassland cover (%)	National (77 sites)	Median TN, NO3-N, TP and DRP were positively correlated with % planted-forest and high-producing exotic-grassland cover, and negatively correlated with % scrub-grassland cover. Median TN, NO3-N, TP concentrations were negatively correlated with % native-forest cover.	Julian et al. 2017
Intensive agriculture (high-producing exotic grassland, short-rotation crops, orchards, vineyards) cover (%)	National (592 sites)	Catchment yields of TN, NO3-N, TP and DRP were positively correlated with % intensive-agricultural cover.	Snelder et al. 2017
Pastoral, urban, scrub, native and planted forest cover (%)	National (1130 sites)	Densities of 14 out of 21 native fish species were correlated with one or more cover classes.	Jowett and Richardson 2003
LULC classes (<33, 34–66, >67% cover) in native forest, urban, planted forest, pastoral	National (5497 sites)	Mean fish IBI scores were lower in the higher urban, planted forest and pastoral classes and higher in the higher native forest classes.	Joy and Death 2004
Native forest cover (%)	National (9351–11843 sites)	Probability of occurrence of lamprey, giant kokopu and common smelt were positively correlated with % native-forest cover.	Crow et al. 2014
Urban, impervious surfaces cover (%)	National urban centres (32–72 sites)	Median dissolved copper and zinc concentrations were positively correlated with % urban cover. Median zinc and NH4-N concentrations were positively correlated with % impervious surface cover.	Gadd 2016, 2019
LULC classes based on dominant cover: forest (natural and planted), dairy, pastoral non-dairy	Regional (29 sites)	Median NO3-N concentrations in Southland streams were highest at dairy sites and lowest at forest sites; median DRP concentrations were highest at forest sites and lowest at non-dairy pastoral sites.	Hamill and McBride 2003
LULC classes: reference (>95% native vegetation cover), planted forest (>30%), rural (>13% agriculture, <30% forestry, <5% urban), urban (>10% urban, <30% planted forest)	Regional (66 sites)	Mean macroinvertebrate taxon richness in hard-bottom streams decreased in the following order of LULC classes: reference, planted forest, rural, urban. Mean macroinvertebrate taxon richness in hard-bottom streams decreased in order: planted forest, reference and rural (tied), urban. Mean MCI scores in both hard-bottom and soft-bottom streams decreased in order: reference, planted forest, rural, urban.	Moore and Neale 2008
LULC classes: natural (100% cover), pastoral (10–49%, 50–90%, >90% cover), urban.Continuous native forest and scrub cover (%)	Regional (91 sites)	Mean invertebrate taxon richness, EPT taxon richness, %EPT individuals, and MCI scores in Waikato streams decreased in the following order of LULC classes: natural, pastoral, urban. The same metrics increased with increasing proportions of native forest and scrub cover.	Collier and Hamer 2010; Collier and Olsen 2013
LULC classes: dairy, sheep	Regional (24 sites)	Mean TP and orthophosphate concentrations in sediments were higher in first-order streams draining south Otago catchments used for dairy farming than in catchments used for sheep farming.	McDowell 2009
Pastoral (pastoral + bare ground) cover (%)	Regional (21 sites)	Mean DIN, TP, DRP and SS concentrations, fine sediment cover, periphyton biomass and macroinvertebrate density and biomass in third- and fourth-order streams in Otago were positively correlated with pastoral cover. EPT density and MCI scores had unimodal relationships with pastoral cover.	Niyogi et al. 2007
LULC classes based on dominant cover: native forest, planted forest, pastoral	Catchment (9–11 sites)	Mean TN, NO3-N and NH4-N concentrations and yields and turbidity in streams at the Whatawhata Research Centre were higher and DRP concentrations and visual clarity lower at pasture and planted-forest sites than at native-forest sites. Invertebrate density was higher and EPT density and QMCI lower at pasture sites than at planted- and native-forest sites. Median E. coli concentrations were higher at pasture sites than at planted-forest and native-forest-sites. Native fish densities and biomass were higher at pastoral sites than at native or planted-forest sites, and mean eel productivity was 7X greater at pasture sites than at forested sites.	Quinn et al. 1997; Quinn and Stroud 2002; Hicks and McCaughan 1997; Donnison et al. 2004; Dodd et al. 2008
LULC classes based on dominant cover: native forest, pastoral	Catchment (6 sites)	Areal densities of crayfish in Whatawhata streams were higher at native-forest-sites than pastoral sites; peak biomass and juvenile growth rates were higher at pasture sites.	Parkyn et al. 2002
LULC classes based on dominant cover: natural (native forest, tussock, scrub), planted forest, pastoral, horticulture	Catchment (10–18 sites)	Median TN, NO3-N, E. coli and Campylobacter concentrations were higher at pasture and horticulture sites in the Motueka catchment than at native forest and planted forest sites. MCI and invertebrate shredder abundance were higher at native forest sites than at pasture or planted forest sites.	Young et al. 2005; Shearer and Young 2011
Natural (native forest, tussock, scrub) cover (%)	Catchment (25 sites)	Cultural stream health measure values in the Motueka catchment were positively correlated with % native-forest cover.	Harmsworth et al. 2011
LULC classes based on dominant cover: native forest, grazed tussock, planted forest, pastoral	Catchment (60 sites)	Mean TP concentrations were higher at pastoral and native-forest sites in the Taieri catchment than at planted-forest sites. Mean invertebrate densities were higher at pastoral sites than native-forest sites. Relative abundance of shredders and filterers were greater at native- and planted-forest sites than pastoral sites.	Townsend et al. 1997
High-producing exotic grassland cover (%)	Catchment (43 sites)	Probability of trout presence in the Manuherikia River and tributaries increased as high-producing exotic grassland cover increased from 0–20% cover, then decreased to absence at 40% cover.	Lange et al. 2014
LULC classes based on 100% cover: pastoral, native forest, planted forest	Catchment (3 sites)	Mean TN, TP, DRP concentration in streams in the Purukohukohu catchment were highest in the pastoral subcatchment, and mean NO3-N concentrations were highest in the native-forest subcatchment. TN and TP annual yields were highest in the pastoral subcatchment.	Cooper and Thomsen 1988
LULC classes based on 80-100% cover: pastoral, native forest, planted forest	Catchment (7 sites)	Mean areal density and abundance of native fish species in eastern North Island catchments were higher at pastoral sites than forested sites, mainly due to higher short-finned eel abundance at pastoral sites. Banded kokopu densities were higher at forested sites than pastoral sites.	Rowe et al. 1999
Dairy farming, forest (native and planted) cover (%)	Catchment (9 sites)	Densities of brown trout, native fish, and the dominant benthic invertebrate Deleatidium and five invertebrate-based stream health indicators were negatively correlated with % dairy farm cover in the Pomohaka catchment.	Ramezani et al. 2016
Pastoral, tussock, native forest cover (%)	Catchment (8 sites)	Mean TN and NO3-N concentrations were positively correlated with % pastoral cover. Mean DRP concentrations were positively correlated with % tussock cover in the Lake Wanaka catchment.	Weaver et al. 2017
Urban, impervious surfaces cover (%)	Urban centre (64 sites)	Macroinvertebrate taxon richness, MCI and UCI scores in Auckland City streams were negatively correlated with % impervious cover	Allibone et al. 2001
Urban, impervious surfaces cover (%)	Urban centre (34 sites)	Sediment zinc, copper and lead concentrations in the Avon River, Christchurch were positively correlated with % impervious cover.	Gadd and Sykes 2014
LULC classes based on dominant cover: urban, planted forest, pastoral	Urban centre (19 sites)	Median NO3-N, E. coli, dissolved copper and zinc, and sediment zinc and lead concentrations were highest at urban sites, lowest at planted forest sites and intermediated at pastoral sites. Median macroinvertebrate taxon richness, EPT richness, MCI and %EPT were highest at planted forest sites, lowest at urban sites and intermediated at pastoral sites.	Stansfield 2016
LULC in upstream catchments and riparian corridors
Agriculture (high-productivity grassland, short-rotation crops), forest (native, planted, deciduous), scrub (gorse, manuka, low-productivity grassland) cover (%).	Regional (21 sites)	Stream metabolism in Canterbury streams was more strongly associated with agricultural land cover at the catchment scale, more strongly associated with forest cover at the reach-riparian zone scale, and intermediate at the network-riparian zone scale.	Burrell et al. 2014
Pastoral, tussock, scrub-tussock, planted forest cover (%)	Regional (24 sites)	TN and TP concentrations in east Otago streams were positively correlated with catchment-scale pastoral cover and negatively correlated with catchment-scale tussock and scrub cover. Correlations between land cover and TN and TP were generally stronger at the catchment scale compared to the riparian scale.	Buck et al. 2004
Natural (native forest, scrub) cover (%)	Regional (138 sites)	Sixteen invertebrate-based diversity and ecosystem health metrics were positively correlated with catchment-scale natural land cover in Waikato streams. For all metrics, catchment-scale correlations were higher than segment-scale correlations.	Death and Collier 2010
LULC classes: upland forest (>200 m asl), lowland forest (<200 m asl), riparian forest present or absent. Total forest cover (%)	Catchment (172 sites)	Koaro occurrence in Banks Peninsula streams was positively correlated with catchment forest cover, but only at sites with forested riparian zones. The effect of catchment forest cover was greatest when most of the forest was at low elevations.	Eikaas et al. 2005
Impervious surface area (%)	Urban centre (25 sites)	Invertebrate taxon richness, EPT richness, UCI and QUCI in Hamilton City streams decreased with increasing % impervious cover at all spatial scales; impervious cover at upstream corridor scales was a stronger predictor of EPT richness and QUCI than at catchment scales.	Collier and Clements 2011
LULC in riparian corridors
LULC classes: continuous native riparian forest, native riparian forest fragment, pastoral riparian vegetation	Catchment (12 sites)	Reaches in Banks Peninsula streams with continuous native riparian zones had lower light levels, lower water temperatures and higher invertebrate taxon richness, EPT taxon richness and Margalef’s biodiversity index than reaches with pastoral and native-fragment riparian zones.	Harding et al. 2006

LULC in upstream catchments

Correlations between proportions of catchment LULC classes and contaminant levels in rivers have been reported periodically for almost 30 years (Table 1). The spatial extent of these correlations ranged from single catchments to multiple catchments distributed across New Zealand. The reported correlations were highly consistent: concentrations and loads of total N (TN), total P (TP), nitrate-N (NO₃-N) and dissolved reactive P (DRP) were positively correlated with pastoral, urban and planted-forest land cover, and negatively correlated with natural (i.e. native forest, scrub, tussock) land cover. Differences in correlation directions for planted forest cover and native forest cover may be related to soil disturbance during forestry rotations, or to legacy nutrients from previous agricultural land use and the high TP concentrations in the central North Island volcanic geology where the majority of planted forests are located (Julian et al. 2017).

In national studies that included Escherichia coli (E. coli) and other faecal bacteria, bacterial concentrations were positively correlated with pastoral and urban land cover (Table 1). In most of the national studies that included ecological impact variables (e.g. macroinvertebrate community metrics), the impact variables were negatively correlated with urban and pastoral land cover, and positively correlated with natural land cover (e.g. Quinn and Hickey 1990; Larned et al. 2016, 2019).

Several national assessments of associations between LULC and fish metrics were based on data from the New Zealand Freshwater Fish Database (Table 1). Jowett and Richardson (2003) reported a mixture of positive and negative correlations between native fish densities and proportional land cover (e.g. shortfin eel and common bully densities were negatively correlated with native forest cover and positively correlated with pastoral land cover; koaro and banded kokopu densities were positively correlated with native forest cover and negatively correlated with pastoral land cover). Crow et al. (2014) reported that the probabilities of occurrence of three out of 31 fish species were positively correlated with native forest cover. Joy and Death (2004) reported that values of a fish-based index of biotic integrity differed systematically among some land-cover classes; index values at native-forest dominated sites were 15%–20% higher than at urban and pasture-dominated sites.

Urban land cover and impervious cover (i.e. urban cover consisting of concrete, asphalt and roofs) have been associated with contaminants and ecological metrics in streams in New Zealand’s major urban centres (Gadd 2016, 2019; Table 1). These analyses were carried out separately from other national analyses because data characterising impervious surfaces and some contaminants (e.g. copper, zinc) are limited to urban centres. The analyses indicated that dissolved zinc and copper concentrations were positively correlated with urban land cover, and dissolved zinc and ammoniacal N (NH4-N) concentrations were positively correlated with impervious cover. No other statistically significant correlations were detected, due in part to high between-site variation within urban centres.

LULC-river associations have been assessed in several regional investigations. Most of these investigations used macroinvertebrate-community metrics (hereafter ‘macroinvertebrate metrics’) to assess land use impacts (Table 1). For brevity, we summarise associations between LULC and macroinvertebrate metrics in Waikato Region streams, where Collier and colleagues have surveyed streams across a wide range of catchment conditions. In these streams, multiple macroinvertebrate metrics (e.g. percent of invertebrates from the insect orders Ephemeroptera, Plecoptera and Trichoptera (%EPT), macroinvertebrate community index (MCI)) were positively correlated with natural land cover and negatively correlated with pastoral land cover (Collier 2008; Death and Collier 2010). Thresholds were evident in some plots of macroinvertebrate metrics versus natural land cover, indicating rapid declines in % EPT and MCI scores as natural land cover decreased from 100% to 80% and from 60% and 0% (Collier and Hamer 2010). These non-linear relationships suggest that the ecological health of streams is resistant to minor losses in native land cover, but degrades rapidly with decreasing natural land-cover below a land-cover threshold.

The remaining references in Table 1 reported associations between LULC and river state and ecological impact variables in single- and multiple-catchment studies. These associations have been a focal issue at the Whatawhata Research Centre, which encompasses several Waipa River subcatchments (see references in Table 1). In the Whatawhata studies, stream NO₃-N, NH₄-N and TN concentrations were higher and DRP concentrations and visual clarity were lower in catchments dominated by pasture and planted forest compared with native forest-dominated catchments. Native fish densities and biomass and median E. coli concentrations were higher and macroinvertebrate metrics were lower in the pasture-dominated catchments compared with the planted-forest and native -forest dominated catchments.

The Motueka River catchment has also been used to identify associations between LULC variables and water quality (Young et al. 2005), ecological conditions (Shearer and Young 2011) and cultural health (Harmsworth et al. 2011). In the water quality study, NO₃-N, TN, E. coli and Campylobacter concentrations were higher at pasture- and horticulture-dominated sites than at native forest and planted forest sites. In the ecology study, MCI scores and proportions of macroinvertebrate detritivores were higher at native forest sites than at pasture or planted forest sites.

The study by Harmsworth et al. (2011) is noteworthy as it was the first to link land use to Māori cultural values in rivers. Results of the study included a positive correlation between cultural stream health measure (CSHM) scores and native forest land cover. CSHM is a metric for evaluating river health based on cultural indicators such as clarity, suitability for swimming and cultural harvests (Tipa and Teirney 2006).

Several monitoring studies in individual urban centres reported associations between catchment LULC and contaminant levels and ecological impacts. Gadd and Sykes (2014) reported that stream sediment zinc, copper and lead concentrations at 35 sites in the Avon River, Christchurch were positively correlated with impervious cover in the upstream catchment. Allibone et al. (2001) reported that three macroinvertebrate metrics in Auckland streams were negatively correlated with impervious cover in the upstream catchment. Stansfield (2016) reported that, across 19 stream sites in the Henderson and Huruhuru Stream catchments in Auckland, NO₃-N, E. coli, and metals concentrations were highest and macroinvertebrate metric scores lowest at urban sites. Concentrations of the same contaminants were lowest and macroinvertebrate metric scores highest at planted forest sites; all state and impact variables were intermediated at pastoral sites.

LULC in riparian zones

Catchments with contrasting patterns of riparian-zone and upland LULC (e.g. grassland-dominated catchments with forested riparian zones) have been used in nested studies to compare the strengths of associations between LULC and river conditions based on catchment LULC versus riparian LULC (Table 1). In two nested studies, catchment LULC explained more variation in stream contaminant concentrations and invertebrate metrics than riparian LULC (Buck et al. 2004; Death and Collier 2010). Burrell et al. (2014) reported that instream metabolism was positively correlated with catchment-scale agricultural cover and negatively correlated with riparian-scale forest cover; in this study, high primary production in streams with low riparian cover and high agricultural catchment cover was symptomatic of nutrient-driven eutrophication. Eikaas et al. (2005) reported that the distribution of the migratory galaxid Galaxias brevipinnis was positively correlated with catchment forest cover, but only at sites with intact riparian forest.

Collier and Clements (2011) compared the strengths of correlations between four macroinvertebrate metrics in Hamilton City streams and impervious land cover at multiple spatial scales: 30-, 50- and 100-m wide corridors along stream segments and networks, and in entire catchments. In this study, all metric scores decreased with increasing impervious cover at all spatial scales, and impervious cover in 50- and 100-m wide corridors was a stronger predictor of two metrics than catchment-scale impervious cover.

In lieu of a nested design, Harding et al. (2006) compared macroinvertebrate communities and water quality in three riparian land-cover classes (pastoral, native forest fragments, continuous native forest). Sites with continuous native riparian zones had lower water temperatures and higher macroinvertebrate metric scores than sites with pastoral and native fragment riparian zones.

Land management practices

Most of the New Zealand reports with associations between LMP variables and river state and impact variables used one of three categorical designs: comparisons of river conditions before and after LMP occurrence (e.g. before and after fertiliser application); comparisons of sites upstream and downstream of LMPs (e.g. upstream and downstream of irrigation return flows); and comparisons of catchments with different LMPs in effect (e.g. forest harvest with and without riparian buffers).

Agricultural LMPs

The agricultural LMPs that have been associated with state variables in New Zealand rivers fall into four general classes: stocking rates, stock access to rivers, fertiliser use and water irrigation (Table 2). We located no reports of associations between river conditions and some widespread agricultural LMPs that are likely to affect river ecosystems (e.g. soil tillage, winter forage, land application of dairy effluent). All of the reports that we located concerned associations between agricultural LMPs and river state variables; none included ecological impacts in rivers. Most of the evidence in Table 2 concerns improved and unimproved pastoral agriculture; there was little evidence concerning horticulture and other non-pastoral agricultural sectors.

Table 2. New Zealand studies of associations between agricultural land management practices and state and impact variables in rivers. Abbreviations: NRWQN – National River Water Quality Network, SS – suspended sediment. See Table 1 for additional abbreviations.

Land management practice	Summary	Reference
Stocking rate	Stocking rate (stock units ha^−1) was positively correlated with river TN, NH4-N, TP concentrations and turbidity in two east Otago catchments.	Buck et al. 2004
Stocking rate	Cattle stocking rate (animals ha^−1) was positively correlated with river TN, NO3-N, TP, DRP concentrations and turbidity across 77 NRWQN sites.	Julian et al. 2017
Dairy cow access	A herd of 246 dairy cows crossing the Sherry River, Tasman caused E coli, SS and TN concentrations to increase 400%, 54% and 10%, respectively, and water clarity to decrease 10%.	Davies-Colley et al. 2004
Deer access	Instream NO3-N, NH4-N, TP, DRP, SS and E. coli loads decreased by 78–98% after fencing farmed deer out of a small Otago stream.	McDowell 2007, 2008
Beef and dairy cow access	E. coli input to river reaches in the Ruamahanga catchment through direct stock access was estimated to be 15 × 10^15 CFU yr-1, based on a GIS model.	Dymond et al. 2016
Sheep and beef access	Following stock exclusion at four sites in the Whatawhata Research Centre, Waikato, median water clarity increased at all four sites, and median NO3-N and DRP concentrations increased at three and two sites, respectively.	Hughes and Quinn 2014
Sheep and beef access	Southland stream reaches with grazed riparian zones had less shade, higher maximum water temperatures, larger seasonal temperature variation, and higher periphyton biomass than streams with retired riparian zones, and most grazed reaches had lower macroinvertebrate abundances, taxon richness and QMCI scores.	Quinn et al. 1992
Fertiliser application	Shifting P fertiliser from superphosphate to reactive phosphate for three years resulted in lower DRP and TP loads in the outlet streams of two sheep-grazed Hawke’s Bay catchments.	McDowell et al. 2010
Water irrigation	A comparison of sites upstream and downstream of irrigation return flows on four Otago rivers indicated that downstream DRP concentrations were higher than upstream reaches at all sites, downstream E. coli concentrations were highest at two sites, and downstream NO3-N concentrations were highest at two sites.	Wilcock et al. 2011
Water irrigation	A comparison of sites upstream and downstream of irrigation return flows in tributaries of the Taieri River, Otago indicated that downstream TN and TP concentration were highest at two out of two sites, and downstream DRP, E. coli and SS concentrations were highest at one site.	McDowell and Kitto 2013
Dairy cow access, effluent discharge and application	Median and 95th percentile E. coli concentrations in stream reaches downstream of a dairy farm in the Toenepi catchment, Waikato were estimated from a budget model composed of the following terms: direct input from cow access and stream crossing, direct input from ducks, dairy effluent input from either a two-pond treatment systems, an advanced pond system, or high- or low-rate land application, stream flow and instream attenuation. The estimated median E. coli concentration was very close to the observed median, but the modelled range was greater than observed due to averaging the spatial variability in inputs.	Muirhead et al. 2011

Stocking rate

Correlations between stocking rates and river contaminant levels have been reported in a study of two Otago catchments and a study of the catchments above the 77 monitoring sites in the National River Water Quality Network (NRWQN) (Table 2). Stocking rates in the Otago catchments were positively correlated with river TN, NH₄-N and TP concentrations and turbidity (Buck et al. 2004). In the NRWQN study, stocking rate was positively correlated with river TN, NO₃-N, TP, DRP concentrations and turbidity (Julian et al. 2017).

Stock access

N, P and faecal bacteria inputs to rivers from stock defecation have been estimated using faecal production rates and bacterial, N and P concentrations in faecal material (Collins et al. 2007; Bagshaw et al. 2008; Dymond et al. 2016). These inputs have been estimated at reach- (Muirhead et al. 2011) and catchment-scales (Dymond et al. 2016) (Table 2). The effects of stock crossing on river contaminant concentrations and loads were measured directly by Davies-Colley et al. (2004). In that study, a dairy herd crossing a river caused E. coli, SS and TN concentrations to increase by 400%, 54% and 10%, respectively, and water clarity to decrease by 10% (Table 2).

In the remaining stock-access studies, the effects of stock access were inferred from comparisons of river state and impact variables with and without stock access, or before and after stock exclusion. In a comparison of Southland streams with and without stock access, most sites with stock access had less shade, higher seasonal temperature variation, higher periphyton biomass, and lower macroinvertebrate metric scores (Quinn et al. 1992). In a second study at Whatawhata, water clarity increased in three streams after stock exclusion, but DRP and NO₃-N concentrations also increased (Hughes and Quinn 2014). The increases in DRP and NO3-N concentrations were attributed to reduced instream N and P uptake due to increased riparian shade, and to reduced N and P retention due to afforestation of seepage wetlands.

The effects of deer access to headwater streams have been investigated at two Otago deer farms. NO₃-N, NH₄-N, TP, SS and E. coli concentrations in these streams were significantly higher during deer-wallowing periods compared with deer-free periods, under all flow conditions (McDowell 2007). In a subsequent study at one of the farms, mean annual NO₃-N, NH₄-N, TP, DRP, SS and E. coli loads measured during a period when deer accessed a stream were 78%–98% higher than when deer were excluded (McDowell 2008).

Fertiliser use

We located a single report linking agricultural fertiliser use to river state (Table 2). In this study, McDowell et al. (2010) compared DRP and TP loads in the outlet streams of catchments that had been top-dressed with two forms of P fertiliser in alternate years: superphosphate and reactive phosphate rock. Over the four-year study, the use of high-solubility superphosphate was associated with higher instream DRP and TP loads than low-solubility rock phosphate.

Water irrigation

Runoff and drainage following irrigation in excess of soil-water capacity are well-studied in New Zealand (e.g. Carey et al. 2004). However, the direct effects of irrigation return flows on river state and ecological impacts are rarely measured. Return flows from spray irrigation to rivers is usually via groundwater and diffused, which makes the effects of irrigation on river conditions difficult to detect (Baalousha 2009). In contrast, flood and border-dyke irrigation with drainage across the land-surface and spray irrigation leading to saturation-excess runoff lend themselves to comparisons of river conditions upstream and downstream of return-flow sites. Wilcock et al. (2011) and McDowell and Kitto (2013) compared contaminant concentrations at paired sites upstream and downstream of return-flows on five Otago streams. In most of these streams, N, P and E. coli concentrations were higher at the downstream site (Table 2).

Forestry LMPs

Typical planted forest rotations comprise six general LMPs: land preparation, tree planting, thinning and pruning, fertilisation, weed and pest control, and harvesting, which includes road building, felling, yarding and transport (Maclaren 1993). With the exception of weed and pest control, effects of each LMP on New Zealand rivers have been reported in at least one publication. The single report on effects of land preparation concerned afforested low-gradient wetlands; we found no reports linking impacts to the effect of land preparation on hillslopes, where most plantation forestry occurs in New Zealand. The lack of information about effects of hillslope land preparation is a concern, as this LMP class includes several activities that accelerate sediment and nutrient transport to surface water (Aust and Blinn 2004).

Land preparation

In the single study of effects of land preparation, Collier et al. (1989) compared stream chemistry and macroinvertebrate communities in Westland streams, where pakahi wetland had been cleared and drained, with control streams (Table 3). Streams in the cleared and drained areas were more acidic and had higher aluminium concentrations and lower invertebrate taxon richness than control streams. Results from a follow-up study eight years later indicated that macroinvertebrate communities in the cleared and drained sites were similar to those in control streams (Harding et al. 2000).

Table 3. New Zealand studies of associations between forestry land management practices and state and impact variables in New Zealand rivers. See Tables 1 and 2 for abbreviations.

Land management practice	Summary	Reference
Land preparation	Streams in northern Westland pakahi wetland areas that had been ripped, drained and humped for afforestation were more acidic, had lower buffering capacity, higher aluminium concentrations, and lower invertebrate taxon richness than streams draining unprepared sites.	Collier et al. 1989
Tree planting	Median stream NO3-N, TN, DRP and TP concentrations were significantly higher for three years post-planting than pre-planting concentrations in one Whatawhata catchment. A positive trend in stream DRP concentration of approximately 1% yr^−1 was detected for the 3-year post-planting period in the same catchment.	Hughes and Quinn 2019
Fertiliser application	Peak N and P concentrations in adjacent streams were observed within 1 d of aerial application of urea and superphosphate to five forest blocks across New Zealand, and returned to baseline levels within two months. An estimated 0.3% of the fertiliser N reached the stream, of which 50% was from direct aerial deposition.	Leonard 1977; Neary and Leonard 1978
Tree thinning	Median stream NO3-N concentrations increased by 13-17% yr^−1 and TN concentrations increased by 9-15% yr^−1 for three years after stand thinning in three catchments with intact riparian margins. Median TP and DRP concentrations increased by 47 and 225%, respectively, after thinning in one catchment	Hughes and Quinn 2019
Forest harvest	Mean NO3-N, TP and SS concentrations in four Motueka River tributaries in the Golden Downs Forest were higher, and seasonal water temperature fluctuations were larger at recently clear-felled sites that at an unharvested site. One harvested site had lower macroinvertebrate taxon richness and lower long-finned eel and dwarf galaxiid abundance than the unharvested site.	Graynoth 1979
Forest harvest	Five Tairua Forest, Coromandel Peninsula streams were monitored before and after clear-felling. Riparian buffers were retained at two streams and no differences in water temperature, clarity, substrate size, or macroinvertebrate taxon richness, % EPT or IBI scores were detected at these sites before and after harvesting. In streams where forest was clear-felled to the margins, water clarity and substrate size declined, annual maximum water temperature increased, and macroinvertebrate taxon richness, % EPT or IBI scores decreased,	Harding et al. 2000
Forest harvest	Four tributaries of the Rangitaiki River were monitored to assess effects of forest harvest and post-harvest slash management. Post-harvest DRP concentrations initially increased to 2–5 times greater than in the preharvest period, then returned to pre-harvest levels within four months. Streambed sand and fine gravel cover decreased from 90% in the pre-harvest period to 20% in the post-harvest period. Benthic organic matter content at sites with logging slash recovered to pre-harvest levels within three years. Changes in macroinvertebrate community structure, density and taxon richness were greatest at sites with the greatest logging slash and channel disturbance; macroinvertebrate communities at all sites were similar to unharvested levels after three years.	Collier and Bowman 2003
Forest harvest	Three sites in the Pakuratahi Stream catchment, Hawke’s Bay were monitored before and 1.5–2.5 years after approximately 85% of the catchment was clear-felled to the stream margins. Macroinvertebrate communities were dominated by mayflies in the pre-harvest period, and by caddisflies and gastropods in the post-harvest period. Macroinvertebrate taxon richness and MCI and QMCI scores decreased in the post-harvest period.	Death et al. 2003
Forest harvest	Stream reaches in four Whangapoua Forest catchments were monitored to compare pre-harvest and post-harvest (<4 years after harvest) conditions. Periphyton abundance, light levels, summer water temperatures, and diurnal temperature fluctuations were lowest at pre-harvest sites and highest at harvested sites with no riparian buffers. Macroinvertebrate taxon richness, QMCI, IBI, %EPT density, EPT taxa and Shannon’s diversity were lower at harvested sites with no riparian buffers compared with sites with continuous riparian buffers or unharvested forest. Sites with discontinuous riparian buffers had intermediate macroinvertebrate metric scores. Mean redfin bully abundance was significantly higher at harvested sites with riparian buffers than at unharvested sites, and mean banded kokopu abundance was significantly higher at harvested sites with riparian buffers than at harvested sites with no buffers. Long-finned and short-finned eel abundance did not vary significantly between treatments. Total fish abundance was highest at harvested sites with riparian buffers and lowest at unharvested sites.	Rowe et al. 2002; Boothroyd et al. 2004; Quinn et al. 2004
Forest harvest	Stream reaches in the Mahurangi Forest, Northland were monitored to compare conditions at sites clear-felled to the stream margin with an unharvested site. Average light levels and mean annual and monthly maximum water temperatures were higher and average dissolved oxygen saturation was lower at the harvested sites. The harvested sites had higher macroinvertebrate densities and lower % EPT density than the unharvested site.	Baillie et al. 2005
Forest harvest	In a study of progressive harvesting in the Pokairoa and Taneatua Forests, Bay of Plenty, % of the catchment area harvested was negatively correlated to %EPT density and positively correlated with elmid beetle density and % silt cover on stream beds.	Collier and Smith 2005
Forest harvest	Mean densities of long-finned eel increased by 75% and redfin bullies by 560% after clear-felling to the margin of Ngakaroa Stream, Bay of Plenty. There were no detectable changes in densities of giant kokupu, brown and rainbow trout or common smelt.	Kusabs and Mitchell 2005
Forest harvest	In a monitoring study of water quality at the outlet of the Pakuratahi catchment, Hawke’s Bay, no statistically significant changes in TN, NH4-N, NO3-N, TP, DRP, SS or faecal coliform concentrations were detected in the 2-10-year period after 85% of the catchment area was clear-felled.	Fahey and Stansfield 2006
Forest harvest	Mean NO3-N and DRP concentrations in three out of four Otago streams were significantly higher two year after clear-felling compared with the preharvest period. Invertebrate collector-gatherer biomass increased at all sites one year after harvest, and shredder biomass increased two years after harvest. Two invertebrate species (Aphrophila neozealandica (tipulid fly), Stenoperla prasine (stonefly)) that were present at all sites prior to harvest were absent post-harvest. Potamopyrgus antipodarum (gastropod) was absent prior to harvest but abundant in the post-harvest period.	Thompson et al. 2009
Forest harvest	Water temperature were monitored for up to 10 years after forest harvest in Whangapoua Forest stream reaches. Temperature elevation ranged from 2 to 3.8 C for summer daily means and 4–7.3 C for summer daily maxima. Summer daily mean and maximum temperatures declined during riparian vegetation regrowth by 0.18 and 0.47 C yr^−1, respectively, for the largest stream and 1.4 and 1.9 C yr^−1 in the smallest stream.	Quinn and Wright-Stow 2008
Forest harvest	An analysis of a 17-year macroinvertebrate dataset from 15 Coromandel Peninsula streams indicated that QMCI and IBI scores were reduced shortly after clear-felling commenced. The effects were most severe in small catchments that were clear-felled to the stream margin. Macroinvertebrate community recovery was faster in small catchments (≤8 years post-harvest) than in large catchments (>17 years post-harvest), and faster where riparian buffers were retained.	Reid et al. 2010
Forest harvest	Mean annual SS yield from the Herring subcatchment, Motueka River increased from 30 tonnes km^−2 yr^−1 in the pre-harvest period to 152 km^−2 yr^−1 during the harvest period.	Basher et al. 2011
Forest harvest	% disturbed plantation forest cover (a proxy for harvested forest area) in the catchments upstream of the National Water Quality Monitoring Network sites was negatively correlated with visual water clarity.	Julian et al. 2017
Forest harvest	Post-harvest median NO3-N concentrations were significantly higher than pre-harvest concentrations in one Whatawhata catchment. A negative trend in stream DRP concentration (−16% yr^−1) was detected for the 3-year period after harvest in the same catchment.	Hughes and Quinn 2019

Planting

In the single study of effects of forestry planting on river conditions, Hughes and Quinn (2019) monitored water quality in two Whatawhata subcatchments before and after first-rotation planting. Median NO3-N, TN, DRP and TP concentrations in the streams draining one of the catchments were significantly higher 3-years after planting than in the pre-planting period (Table 3). As in their previous Whatawhata study (Hughes and Quinn 2014), the authors attributed the increased N and P concentrations to reduced instream N and P uptake due to increased riparian shade, and reduced N and P retention in wetlands after afforestation.

Fertiliser application

Two reports on the effects of planted forest fertilisation on New Zealand stream water quality were based on forest blocks fertilised with urea and/or superphosphate (Leonard 1977; Neary and Leonard 1978). In-stream N and P concentrations increased by as much as two orders of magnitude in the first 1–3 days after fertiliser application, but returned to baseline levels within two months (Table 3). As is the case for agricultural land, fertiliser leaching losses from planted forests have been measured, but not linked to state changes or impacts in aquatic ecosystems (Davis et al. 2012).

Thinning

The Whatawhata study by Hughes and Quinn (2019) was the only report located with associations between tree thinning and stream conditions (Table 3). In that study, stream NO3-N and TN concentrations increased for three years after thinning in three planted catchments, and DRP and TP concentrations increased after thinning in one catchment These changes were attributed to reduced soil N and P uptake after tree density was reduced.

Harvesting

Clear-fell harvesting is considered the forestry LMP with the greatest potential for adverse effects on rivers; it is also the most common harvesting method in New Zealand planted forests (Maclaren 1993). Clear-felling can cause soil destabilisation, accelerate sediment input to streams, reduce soil nutrient uptake, add logging slash to channels, and alter stream light and thermal conditions. In addition to harvested areas, forestry roads and landings are sources of eroded sediment (Fransen et al. 2001; Marden et al. 2006). The effects of alternative methods such as selective logging may be less severe than clear-felling, but are limited to native forestry in New Zealand (Hawes and Memon 1998).

In Table 3, forest harvesting by clear-felling to stream margins is distinguished from clear-felling with riparian buffers retained. In several studies, clear-felling to stream margins led to increased solar insolation and water temperatures, and decreased daily average and minimum dissolved oxygen concentrations (Collier and Bowman 2003; Baillie et al. 2005; Quinn and Wright-Stow 2008). Increased solar radiation and post-harvest nutrient pulses also led to increased periphyton abundance after riparian clear-felling (Boothroyd et al. 2004; Thompson et al. 2009; Reid et al. 2010). In a comparative study of forest harvest effects in Whangapoua Forest, Coromandel Peninsula, stream reaches where trees were felled to the margin had higher light levels, higher bank erosion and periphyton biomass, fewer banded kokopu (Galaxius fasciatus) and lower macroinvertebrate-metric values than reaches with riparian buffers (Rowe et al. 2002; Boothroyd et al. 2004; Quinn et al. 2004).

In four of the five studies that compared instream N and P measurements before and after harvesting, and in harvested and unharvested catchments, N and P concentrations increased during post-harvest periods ranging from four-months to two-years (Graynoth 1979, Collier and Bowman 2003; Thompson et al. 2009; Hughes and Quinn 2019). In contrast, Fahey and Stansfield (2006) reported that difference in stream N, P, SS and faecal bacteria concentrations before and 2–10 years after 85% of a forested catchment was clear-felled were statistically undetectable.

Comparisons of sediment yields in harvested and unharvested catchments and in pre- and post-harvest periods indicate that hillslope sediment losses increase after harvest, then decline to pre-harvest levels two to six years after replanting (Fahey and Marden 2006; Basher et al. 2011; Quinn and Phillips 2016). Elevated sediment yield is generally accompanied by increased instream SS concentrations and fine deposited sediment cover (Graynoth 1979; Collier and Bowman 2003; Basher et al. 2011; Table 3). The time-courses of sediment yield and instream sediment levels following harvesting vary with local storm intensity and timing, and connectivity between sediment source areas and stream channels (Kamarinas et al. 2016).

Multiple assessments of ecological impacts of forest harvesting have been based on comparisons of macroinvertebrate communities before and after harvest and between harvested and unharvested areas (Table 3). Collectively, these studies indicate that macroinvertebrate communities at sites where forests were clear-felled to stream margins have high proportions of stress-tolerant taxa (e.g. midges, snails, segmented worms), increased total invertebrate abundance, and reduced ecological-health scores. In a subset of these studies, sites that were clear-felled to stream margins were compared to clear-felled sites with intact riparian buffers and to unharvested sites (Harding et al. 2000; Quinn et al. 2004; Reid et al. 2010). Macroinvertebrate community structure and ecological-health scores at harvested sites with intact buffers were more similar to those at unharvested sites than to those at sites that were clear-felled to the margins.

The study by Collier and Smith (2005) is notable because it is one of the few studies of forest harvest effects with continuous relationships between harvesting and its effects. In that study, % catchment area harvested was negatively correlated to %EPT density and positively correlated with elmid beetle density and % silt cover on stream beds (Table 3).

Two studies of the effects of forest harvesting on freshwater fish were located (Table 3). The first reported that long-finned eel and dwarf inanga abundances were lower in a clear-felled forest stream than in an unharvested control stream; this study also reported a brown-trout kill and attributed it to hypoxia following clear-felling (Graynoth 1979). In the second study, redfin bully abundance was higher at harvested sites with riparian buffers than at unharvested sites, and banded kokopu abundance was higher at harvested sites with riparian buffers than at harvested sites without buffers. Long-finned and short-finned eel abundance did not vary significantly between harvesting methods (Rowe et al. 2002).

Urban LMPs

We located two studies that reported associations between urban LMPs and state and impact variables in rivers. The first study concerned land-application of municipal sewage effluent from the city of Rotorua in planted forest as an alternative to direct discharge to Lake Rotorua (Tomer et al. 2000). The forested catchment used for effluent application drains to Waipa Stream, a tributary of Lake Rotorua. Data from six years (1991–1997) of monitoring effluent N and P concentrations, effluent application rates, and Waipa Stream flows and N and P loads were used to estimate the loss of N and P to the stream. On average, 17% of the N and 2% of the P applied to land were exported; these exports increased the N and P loads of Waipa Stream by approximately 64% and 30%, respectively.

The second study concerned impacts of road use on aquatic macroinvertebrate communities (Shaver and Suren 2011). Macroinvertebrates were monitored upstream and downstream of six high-use state highway crossings in the Wellington and Auckland Regions. Contaminant runoff was not measured directly, but upstream-downstream differences were used to infer contaminant effects. Macroinvertebrate community composition and ecological health metrics were highly similar at most of the paired upstream-downstream sites, indicating that road runoff had generally weak effects on macroinvertebrates at these sites.

We found no published evidence of the effects of other urban LMPs that can affect river conditions, including earthworks, landfill operations, and the use of zinc and copper-yielding roofs and guttering. A large proportion of the recent urban water-quality research in New Zealand concerns contaminant transport in urban infrastructure (e.g. sewage systems and stormwater drains). However, we excluded point-source discharges from sewage and stormwater outfalls from this review.

Lakes

Investigations of the effects of land-use on New Zealand lakes have focused almost entirely on N and P inputs and subsequent changes in lake state and impact variables. This focus reflects the risk of eutrophication posed by historical increases in catchment nutrient loading to lakes and widespread reports of phytoplankton proliferations and elevated trophic-level index (TLI) values (e.g. Drake et al. 2011; Smith et al. 2016; Schallenberg et al. 2017). In a recent analysis of monitored lakes across New Zealand, 60% of lakes were classed as eutrophic, based on median TLI > 4 (Larned, Snelder et al. 2018). Despite reports of elevated E. coli and SS in some New Zealand lakes (Hicks et al. 2013; Dada and Hamilton 2016), we found no published associations between land use and faecal bacteria or sediment in lakes. All reports that included associations between land use and lake state or impact variables were based on LULC variables; we found no studies that used LMPs to represent land use.

Eight studies reported continuously varying associations between LULC and lake state or impact variables (Table 4). In these reports, N and P concentrations in lake waters were positively correlated with catchment-scale pastoral, urban and impervious surface cover. In addition, TP concentrations were positively correlated with planted-forest cover in one study of 101 lakes (Abell et al. 2010).

Table 4. New Zealand studies of associations between land use/land cover (LULC) and state and impact variables in lakes. Abbreviations: TLI – trophic level index; Chl a – phytoplankton Chlorophyll-a concentration. See Tables 1 and 2 for more abbreviations.

LULC variables	Spatial extent	Summary	Reference
LULC-lake condition correlations
High producing grassland, arable, low-producing grassland, urban, planted forest cover (%)	National (101 lakes)	Mean TN concentrations were positively correlated with % high-producing grassland and urban cover; mean TP concentrations were positively correlated with % high-producing grassland and planted-forest cover.	Abell et al. 2010
Pastoral, native (native forest, scrub), alpine cover (%) Land-cover classes: pastoral, native, alpine, planted forest	National (112 lakes)	Median TLI scores were positively correlated with % pastoral cover and negatively correlated with % native and alpine cover.Median TN and TP concentrations were highest in lakes in the pastoral class and lowest in lakes in the native and alpine classes.	Verburg et al. 2010
Non-indigenous (non-native forest, pastoral), cover (%), impervious surface (buildings, roads, rail) cover (%)	National (45 lakes)	Mean TN, TP, phytoplankton Chl a concentrations in shallow, coastal lakes were positively correlated with % non-indigenous land cover; invertebrate taxon richness was negatively correlated with impervious cover.	Drake et al. 2011
High producing grassland, impervious surface (buildings, roads, rail) cover (%)	National (25 lakes)	Mean TN, TP, NH4-N, DRP and TLI in deep, monomictic lakes were positively correlated with % pastoral and impervious cover; invertebrate and macrophyte taxon richness negatively correlated with % pastoral and impervious cover.	Özkundakci et al. 2014
Pastoral, forest (native and planted) cover (%)	National (37 lakes)	The probability of shallow lakes undergoing shifts from macrophyte dominance to phytoplankton dominance was positively correlated with % pastoral cover and negatively correlated with % forest cover.	Schallenberg and Sorrell 2009
Native (native forest, scrub, tussock) cover (%)	National (69 lakes)	Ecological integrity (an ecological health indicator) in shallow lakes, deep lakes and brackish lagoons was positively correlated and TLI negatively correlated with % native cover.	Schallenberg 2019
Native forest, planted forest, pasture, scrub, tussock, urban cover(%)	Regional (45 lakes)	Mean TN, TP, and DRP concentrations in lakes and ponds in the Otago Region were positively correlated with % pastoral land cover.	Galbraith and Burns 2007
Pastoral cover (%)	Regional (10 lakes)	Mean potential denitrification rates in lake sediments and mean TN, DRP, and phytoplankton Chl a concentrations in Bay of Plenty lakes were positively correlated with % pastoral land cover; mean bottom-water dissolved oxygen concentrations were negatively correlated with % pastoral land cover.	Bruesewitz et al. 2011
Native forest, planted forest, pastoral, scrub, urban cover(%)	Regional (11 lakes)	The relative cell abundance of planktonic cyanobacteria in Bay of Plenty lakes was negatively correlated with % native forest cover and positively correlated with % pastoral cover; the relative cell abundance of planktonic chlorophyta was positively correlated with % native forest and urban cover and negatively correlated with % pastoral cover.	Paul et al. 2012
Lake N and P budgets
Land-cover classes: pastoral, native forest, tussock and bare, planted forest	Single lake	Proportions of the annual TN load to Lake Taupo from different LULC classes were: 29% from pastoral land (23% of the catchment area), 33% from native forest and tussock (55% of the catchment area), 14% from planted forest (22% of the catchment area). The remaining 24% of the annual TN load was from sewage discharge and direct rainfall	Vant and Huser 2000
Pastoral, native (native and planted forest, scrub, tussock) cover (%)	Single lake	Pastoral land makes up 23% of the Lake Brunner catchment, but the annual TN and TP loads originating on pastoral land are six times greater than the loads originating on the remaining, natural land.	Rutherford et al. 2008
Pastoral, native forest and scrub, planted forest, horticulture, urban, lifestyle block cover (%)	Single lake	Proportions of the annual TN and TP loads to Lake Rotorua from different LULC classes were: 75% of TN and 46% of TP from pastoral land (45% catchment cover); 25% of TN and 54% of TP from forested and urban land (31% catchment cover).	PCE 2006
Pastoral (livestock), native forest, bare ground, planted forest, urban cover (%)	Single lake	Proportions of the annual TN and TP loads to Lake Tarawera from different LULC classes were: 29% of TN and 34% of TP from pastoral land (13% catchment cover); 9% of TN and 11% of TP from urban land (10% catchment cover); 9% of TN and 11% of TP from planted forest (10% catchment cover); 35% of TN and 30% of TP from native forest and bare ground (43% catchment cover);	Hamilton et al. 2006

The ecological impact variables in New Zealand lake studies were phytoplankton concentration, TLI, ecological integrity (EI, a normative ecological-health indicator; Schallenberg et al. 2011), invertebrate and macrophyte taxon richness, and the probability of lake ‘regime shifts’ from clear-water, macrophyte-dominated conditions to turbid, phytoplankton-dominated conditions. In these studies, TLI scores were positively correlated with pastoral land cover and negatively correlated with native land cover, EI scores were positively correlated with native land cover, and the probability of lake regime shifts was positively correlated with pastoral land cover. Invertebrate and macrophyte taxon richness were negatively correlated with pastoral and impervious cover in lake catchments (Table 4).

The remaining reports in Table 4 used nutrient budget methods to estimate N and P loads to lakes and partitioned those loads into LULC classes. In each case, the proportion of the total load originating from agricultural land exceeded the proportion of catchment area used for agriculture, even when point-source inputs were included.

Aquifers

Land-derived contaminants reach aquifers by multiple pathways, including deposition on the land surface and subsequent leaching, by-pass flow and river recharge (Close 1989; White et al. 2012; Cameron et al. 2013). In several contaminant-leaching studies, losses from plant root zones were equated with inputs to aquifers, but the contaminants were not directly measured in groundwater (e.g. Houlbrooke et al. 2004). Actual contaminant input to groundwater and changes in groundwater contaminant levels cannot be determined from root-zone losses alone due to attenuation, dispersion, and other processes (Stenger et al. 2008, 2012). Compared with rivers and lakes, we located few studies of New Zealand aquifers with associations between LULC or LMPs and groundwater state and/or impacts.

LULC classes

We located three reports with associations between LULC classes and groundwater contaminant levels or ecological variables (Table 5). In a study of DRP in National Groundwater Monitoring Programme (NGMP) wells distributed across New Zealand, McDowell et al. (2015) reported that DRP concentrations were highest in anoxic groundwater below dairy-farms. In a study of groundwater NO₃-N in NGMP wells in the Waikato Region, McLay et al. (2001) reported that NO₃-N concentrations were significantly higher in market-gardening areas compared with sheep and beef areas, and intermediate in areas dominated by other LULC classes. Boulton et al. (1997) assessed associations between LULC classes and invertebrate communities in hyporheic zones (shallow groundwater beneath streambeds) at the Whatawhata Centre. Hyporheic invertebrate abundance and taxon richness were highest at sites in native forest-dominated catchments and lower at sites in planted-forest and pastoral catchments.

Table 5. New Zealand studies of associations between land use/land cover (LULC) and state and impact variables in New Zealand aquifers. NGMP: National Groundwater Monitoring Programme. See Table 1 for more abbreviations.

LULC variables	Spatial extent	Summary	Reference
LULC classes based on dominant cover: horticultural, forestry, urban, industrial, dairy, non-irrigated grazing	National (540 sites)	Median DRP concentrations at dairy sites were higher than at sites in other land-use classes.	McDowell et al. 2015
LULC classes based on dominant landcover in 78.5 ha buffer areas: dairy, cattle grazing, sheep and beef grazing, market gardens, orchards	Regional (88 sites)	Mean groundwater NO3-N concentrations at NGMP sites in the Waikato Region were higher at market garden sites than at sheep and beef sites. NO3-N concentrations were positively correlated with % market garden cover.	McLay et al. 2001
LULC classes based on dominant catchment cover: native forest, planted forest, pastoral	Catchment (5 sites)	Mean hyporheic water temperature was higher in streams at pastoral sites at the Whatawhata Research Centre than at native- and planted forest sites. Mean invertebrate abundance and taxon richness were lowest at the pastoral sites.	Boulton et al. 1997

In several analyses of groundwater state at NGMP sites, associations between LULC variables and groundwater state variables were tested but were not statistically significant (Daughney et al. 2012; Sirisena et al. 2013; Moreau and Daughney 2015). However, groundwater at approximately 40% of the NGMP sites appears to affected by unspecified human activities, as indicated by elevated NO₃-N and sulfate concentrations (Daughney and Randall 2009; Moreau and Daughney 2015).

LMPs

The LMPs that have been associated with state changes and ecological impacts in New Zealand aquifers fall into six classes: land-application of municipal sewage effluent, septic tank operation, irrigation, land-application of agricultural effluent, fertiliser application and feedlot operation (Table 6). The state variables in these associations were limited to NO₃-N and faecal bacteria concentrations and loads. Two reports included ecological impacts in aquifers, both based on groundwater invertebrate abundance and composition.

Table 6. New Zealand studies that report associations between land management practices and state and impact variables in aquifers. WWTP: wastewater treatment plant. See Tables 1 and 2 for other abbreviations.

LMP variable	Summary	Reference
Land application of sewage effluent	Groundwater NO3-N, faecal coliform, feacal streptococci and Clostridium perfringens spore concentrations were monitored in bores upgradient and downgradient of dyked paddocks used for primary effluent disposal at Canterbury WWTPs. Groundwater NO3-N concentrations immediately downgradient of two WWTP disposal areas were approximately 25% higher than upgradient concentration, and remained elevated for at least 2 km. Downgradient faecal bacterial concentrations were only elevated after heavy rains, and when effluent was applied close to the monitoring wells.Tracer bacteria, bacterial endospores, and enteroviruses were added to sewage effluent prior to land application, to estimate transport velocities and attenuation rates in vadose and groundwater zones. The microbial tracers were attenuated by 90–99%, primarily in the vadose zone. Despite the high attenuation, viable pathogens were transported >400 m downgradient.	Martin and Noonan 1977; Quin 1978; Sinton et al. 2000, 2005
Land application of sewage effluent	Groundwater invertebrates were monitored within and upgradient of the sewage effluent plume at land application sites in Templeton and Leeston, Canterbury. At both sites, groundwater invertebrate composition was similar within and outside of the sewage plumes. However, invertebrate densities within the Leeston effluent plume were three times higher that upgradient densities. This pattern was attributed to the food subsidy for invertebrates consisting of sewage-derived particulate organic matter and bacteria.	Sinton 1984; Hartland et al. 2011
Septic tank discharge in leach fields	Groundwater NO3-N and faecal coliform concentrations were monitored in 25 water supply wells in a rural area in Canterbury used as a leach field for household septic systems. NO3-N concentrations increased along the groundwater flow path through the leach field. A transport distance >3 km was required for regional groundwater to dilute NO3-N to background levels. Reduction in faecal coliform concentrations occurred over a shorter distance due to rapid attenuation.	Sinton 1982; Pang et al. 2006
Border strip irrigation	E. coli and Campylobacter were monitored in six wells in border-strip irrigated dairy farms in Canterbury. E. coli and Campylobacter were detected in 75% and 12% of groundwater samples, respectively. E. coli concentrations and Campylobacter detections were highest following ‘high-risk’ events when irrigation, rainfall and stock presence coincided.	Close et al. 2008
Spray irrigation	Faecal bacteria were monitored in groundwater under a centre-pivot irrigated dairy paddock in Lincoln and in lysimeter drainage under a South Canterbury dairy farm irrigated by travelling irrigator. Elevated bacterial concentrations at the centre-pivot irrigated farm only occurred after high rainfall events, with minimal effect of irrigation. At the farm with a travelling irrigator, the mean flux of faecal coliforms from ground surface was 33,000 cfu m^−2 following high irrigation application, over 10 times the mean flux following normal irrigation application and rain events.	Close et al. 2010
Land application of dairy effluent	The effects of land application of dairy effluent on groundwater NO3-N concentrations were monitored for 12 years at a Waikato dairy farm. NO3-N concentrations increased rapidly for 4–5 years after land application commenced, and continued to increase gradually for the next eight years. Mean groundwater NO3-N concentrations were not statistically correlated with annual N application rates. Instead, peaks in groundwater NO3-N followed periods of high loading (e.g. effluent applications preceding rainfall recharge).	Selvarajah et al. 1994
Fertiliser application	Groundwater NO3-N and NH4-N were monitored at a Hamilton dairy farm. Paddocks were assigned to three urea fertiliser treatments: 0, 200 or 400 kg N ha^−1 yr^−1. Groundwater NO3-N concentrations remained low for the first year of urea applications, then increased under the treated paddocks, with the highest concentrations under the highest urea application. Groundwater NH4-N did not vary between urea application levels, probably due to rapid nitrification in the volcanic soils.	Ledgard et al. 1996
Feedlot operations	Groundwater chemistry was monitored in wells upgradient and downgradient of a sheep feedlot in Maraekakaho, Hawke’s Bay while it was in operation (1994–1995) and for three years after closure (2001). During feedlot operation, groundwater NO3-N concentrations in downgradient wells were >5 times higher than in upgradient wells. Downgradient NO3-N concentrations remained elevated for at least three years after closure, and risk-assessment modelling indicated that another 3–5 years would be required for groundwater NO3-N to return to pre-feedlot levels.	Rosen 1996; Rosen et al. 2004

The most frequently studied associations between LMPs and groundwater NO₃-N and faecal bacteria concern land-application of sewage effluent from wastewater treatment plants (WWTPs) and the use of land for residential septic-tank leachfields (Table 6). The primary aim of these studies was to assess risks of potable groundwater contamination. The WWTP studies were carried out at land-application sites in Canterbury where effluent is pumped from oxidation ponds to dyked paddocks, percolates to the water table, and is transported downgradient in groundwater flow. The effects of effluent input were assessed by comparing groundwater NO₃-N and faecal bacteria concentrations in wells upgradient and downgradient of the disposal areas (Martin and Noonan 1977; Quin 1978). At both sites, mean downgradient NO₃-N concentrations were elevated year-round. Downgradient faecal bacteria concentrations were elevated after heavy rains, and during dry periods when effluent was applied close to the downgradient wells. Tracer experiments indicated that water-supply wells needed to be separated from the land-application sites by up to 4 km to ensure adequate attenuation of faecal pathogens (Sinton et al. 2000, 2005). Ecological impacts of land-application of sewage effluent disposal were assessed using groundwater invertebrates (Sinton 1984; Hartland et al. 2011). Invertebrate diversity was reduced in the groundwater effluent plumes, but abundance was increased (Table 6), presumably due to food subsidies in the form of effluent organic matter.

In the septic-tank studies, data from drinking-water wells and a transport model were used to assess groundwater NO₃-N and faecal bacteria contamination from the use of household septic tanks (Sinton 1982; Pang et al. 2006). The model indicated that NO₃-N dilution by regional groundwater to background levels required a transport distance of nearly 3 km. Faecal bacteria contamination was more localised, due to rapid attenuation.

The remaining LMPs in Table 6 concern agricultural practices. The effects of three water irrigation systems on faecal bacteria in groundwater have been measured at dairy farms. Under border-dyke irrigation, E.coli detections peaked following ‘high risk’ events when irrigation water occurred within a few days of stock grazing near monitoring wells (Close et al. 2008). A second study assessed effects of travelling and centre-pivot spray irrigation (Close et al. 2010). Under a travelling water irrigator, the flux of faecal coliforms to groundwater ranged from 4600 to 33,000 cfu m⁻² after irrigation events. In contrast, E. coli and Campylobacter were rarely detected in groundwater after centre-pivot irrigation. The results of these three surveys suggest that centre-pivot irrigation with water minimises bacterial fluxes to groundwater, but the surveys also differed in well depths, irrigation rates, stocking rates, soil types and indicator bacteria.

Changes in groundwater NO₃-N concentrations in response to fertiliser application, effluent application and feedlot operations have each been assessed in single cases studies (Table 6). In the fertiliser study, groundwater NO₃-N concentrations increased with increasing urea application rates, after a lag time of 18–24 months (Ledgard et al. 1996). In the effluent study, groundwater NO₃-N concentrations increased for 12 years after the commencement of dairy factory waste application on a Waikato farm (applied in each irrigation season from 1982 to 1994), but there was no clear relationship between temporal variation in application rates and groundwater NO₃-N (Selvarajah et al. 1994). Instead, groundwater NO₃-N fluctuated in response to rainfall recharge, as high recharge transported soil N to the water table. In the feedlot study, groundwater NO₃-N was monitored upgradient and downgradient of a sheep feedlot while it was in operation and three years after it closed (Rosen 1996; Rosen et al. 2004). NO₃-N in downgradient groundwater was very high (15–40 mg L⁻¹) during the operating period and declined after closure, but remained elevated compared to upgradient concentrations. The effects of the feedlot were predicted to persist for a further 3–5 years after closure, as vadose-zone N was depleted.

Discussion

PSI associations are essential evidentiary tools for land and water management (Millennium Ecosystem Assessment 2005; Larned and Schallenberg 2019). Examples of recent applications of PSI associations in New Zealand include the nitrogen allocation policy for the Lake Taupo catchment (Jenkins 2016) and the estimation of reference conditions for state and impact variables in New Zealand rivers and lakes (McDowell et al. 2013; Clapcott et al. 2017; Schallenberg 2019). Estimated reference conditions have been used for setting freshwater objectives under New Zealand’s resource management legislation and in regional land and water plans.

Ideally, strong evidence (i.e. accurate, mechanistic PSI associations) would be available for all aquatic ecosystems affected by land use. In reality, the observational data and/or modelling results needed to develop such relationships are limited to a small number of locations, and management plans and actions are usually based on weaker evidence. Weak evidence includes correlative PSI associations where causation is not clearly established and association-strength is low, and qualitative results from categorical comparisons (Suter et al. 2002; Cottingham et al. 2005; Rehme et al. 2011). The preponderance of weak evidence has led to guidance for its use in environmental management; the essential guidance is to compile PSI associations from multiple sources, then base management decisions on the most consistent associations (Downes et al. 2002; Norris et al. 2012; Schallenberg et al. 2017). Here, consistency refers to the degree to which PSI associations from different sites are consistent in direction (e.g. MCI scores decline after forest clear-felling at multiple study sites). Consistent PSI associations can be distilled from the preceding summaries.

Consistent associations

In multiple studies, proportions of upstream catchment area with agricultural and urban land-cover were positively correlated with contaminant concentrations and loads in rivers and lakes, and negatively correlated with trophic-, macroinvertebrate- and fish-based ecosystem health metrics (Tables 1 and 4). In contrast to the high consistency, association strength tends to be low for PSI associations based on proportional land-cover (i.e. correlation coefficients tend to be <0.5). Two general reasons for low association strength are: (1) simple correlations do not account for multivariate interactions; and (2) land cover is an imprecise proxy for the LMPs that generate and transfer contaminants to aquatic receiving environments. Single land-cover classes often encompass a wide range of LMPs (Bai et al. 2008; Verburg et al. 2009). The utility of land cover as a proxy for LMPs is considered further below.

Several agricultural and forestry LMPs were consistently associated with increased contaminant levels in rivers and aquifers. Among the agricultural LMPs, cattle and deer access to rivers was associated with increased N, E. coli and suspended sediment concentrations, irrigation return flows were associated with increased downstream N and P concentrations, and stocking rates were positively correlated with river N and P concentrations and turbidity.

Faecal bacteria input to land from cattle and N input to land from livestock and fertiliser have been associated with elevated bacterial and N concentrations in downgradient groundwater in multiple studies, but contaminant input and groundwater responses are decoupled by transient storage in unsaturated zones; effects in groundwater may only be apparent after recharge events mobilise and transport contaminants to the water table (Close et al. 2008, 2010).

Clear-fell harvesting in plantation forests is one of the most frequently studied LMPs in New Zealand river science. Collectively, these studies indicate that clear-felling to stream margins leads to increased deposited fine sediment, SS concentrations and loads, water temperature and light levels at stream surfaces, and decreased macroinvertebrate-based ecosystem-health metrics. The adverse effects indicated by macroinvertebrate metrics persisted for 1–8 years after harvesting, depending on catchment size and the presence of riparian buffers. In cases where riparian buffers were retained, adverse effects were consistently reduced (Boothroyd et al. 2004; Quinn et al. 2004; Quinn and Wright-Stow 2008; Reid et al. 2010).

Despite the consistent associations listed above, shortcomings remain in the evidence base for New Zealand land-use effects. In the remainder of this section, we discuss four major shortcomings and approaches to overcome them: (1) inadequate integration of data and models to link land use and contaminant loss to impacts and state changes in aquatic environments; (2) weak inferences based on LULC classes; (3) reliance on categorical PSI associations; and (4) gaps in PSI associations.

Inadequate integration of data and models to link land use and contaminant loss to impacts and state changes in aquatic environments

There is a large body of research in New Zealand focused on quantifying contaminant losses from land via surface runoff, leaching and soil erosion (e.g. Aislabie et al. 2001; Quine et al. 2003; Monaghan et al. 2007, 2009; Hoogendoorn et al. 2011). It is frequently stated in these reports that the contaminant losses affect conditions in freshwater receiving environments, but data from the receiving environments are rarely included, which precludes PSI associations. This approach, where losses are measured but inputs and responses in receiving environments are assumed, is particularly prominent in studies of agricultural LMPs like fertiliser use, livestock grazing and tillage. There is also a large body of research in New Zealand focused on responses to contaminants in aquatic ecosystems (e.g. Abell et al. 2010; Reid et al. 2011; Ramezani et al. 2014). Many of the latter papers refer to, but do not incorporate, land use and contaminant loss data. These uncoupled studies reflect the challenges of research that accounts for the entire causal chain from land use to ecological impacts in aquatic ecosystems (Brook and Blomqvist 2016). Despite the challenges, fully integrated empirical studies and models are needed to ensure that land-based policies and management actions achieve their intended effects in receiving environments. Recent progress has been made to address this challenge in New Zealand, using coupled models to link land-use, contaminant loss and transport to aquatic receiving environments, and effects on water quality, periphyton, phytoplankton and macroalgae (Trolle et al. 2014; Elliott et al. 2016; Kamarinas et al. 2016; McBride and Hamilton 2017; Me et al. 2018).

Weak inferences based on LULC classes

Although the PSI framework is useful for organising information, and its focus on bivariate relationships reflects the approach used in most New Zealand studies, it has limitations. One limitation is the lack of distinction between direct PSI associations (i.e. causal relationships between pressure and state or impact variables) and indirect PSI associations (i.e. pressure variables are correlated with state and impact variables, but are not the proximate causes of change). Many of the PSI associations in this review are indirect; this is most evident in associations that use LULC classes as pressure variables. For example, agricultural and urban land cover has been correlated with water quality and ecological responses in numerous studies, but land cover per se is not the proximate cause of those responses. Instead, land cover is a proxy for a wide range of activities that affect contaminant losses (Verburg et al. 2009). As proxy variables, LULC classes permit only weak inferences about changes in aquatic systems, and weak inferences may be too unreliable for policy makers and managers (Allan 2004; Rehme et al. 2011). In contrast, direct PSI associations generally come from site-specific, controlled studies of causal variables and their effects (e.g. Davies-Colley et al. 2004; Matthaei et al. 2006). Despite their increased reliability, these site-specific PSI associations are difficult to generalise for policy-making.

Between the extremes of large-scale, weak inferences based on LULC classes and small-scale, strong inferences from controlled studies are correlative PSI associations that use LMPs as pressure variables. As LMPs represent activities that can generate and mobilise contaminants on land, they are more directly linked to responses in receiving environments than LULC classes, and are better able to provide reliable, scalable PSI associations (Erb et al. 2017). Shifting from the current reliance on LULC classes to LMPs could improve the assessment and prediction of land-use effects in freshwater ecosystems in New Zealand. We recognise that many of these ecosystems are affected by multiple land-derived stressors, which are in turn associated with multiple LMPs (Matthaei and Piggott 2019). In addition to shifting to LMPs as pressure variables, a shift in data analysis and reporting approaches is needed, from the current focus on single stressors in PSI associations to a more realistic focus on multiple stressors (Larned and Schallenberg 2019).

There are some impediments to compiling and using LMP data in New Zealand, including limited data access for privacy protection and a lack of standard procedures for characterisation and classification (Statistics New Zealand 2017). However, New Zealand LMP inventories have been developed and standard procedures are under investigation (MacLeod and Moller 2006; Manderson et al. 2018). These recent advances have focused on agricultural LMPs; comparable advancements are also needed for urban and forestry LMPs.

Reliance on categorical PSI associations

Most of the PSI associations summarised in Tables 1–6 were based on categorical comparisons between LULC and LMP classes (e.g. natural versus pastoral land-cover, harvested versus unharvested forest catchments). A smaller set of PSI associations were based on correlations using LULC or LMP gradients (e.g. percent impervious surface area, percent of forest area harvested). Continuous correlations based on gradient more informative than categorical associations as they characterise progressive changes in state and impact over the observed pressure gradients, and they can be used for interpolation and for identifying pressure thresholds (Cottingham et al. 2005; Kreyling et al. 2018). In contrast, the information from categorical comparisons is limited to inter-class differences. When continuous data are available to characterise LULC and LMPs, then categorical analyses are unnecessary simplifications. Given that most categorical and quantitative PSI associations permit only weak inferences about land-use effects, researchers should develop the most informative, albeit correlative, PSI associations possible. To do so, we recommend shifting from categorical to continuous associations.

Gaps in PSI associations

In addition to the methodological shortfalls discussed above, there are combinations of land use pressures and environmental responses for which there are few or no reliable PSI associations. In the remainder of this discussion, we identify some high-priority shortages in PSI associations, based in part on a recent gap analysis (Larned, Booker et al. 2018). Four general shortages were identified in that analysis: (1) PSI associations based on LMPs were lacking in studies of New Zealand lakes; (2) PSI associations linking LMPs to state and impact variables in rivers and urban streams are limited to a narrow range of LMPs, particularly at regional and national scales; (3) PSI associations of all forms are very scarce for aquifers; (4) for many widespread LMPs, PSI associations are lacking for all receiving environments.

Widespread LMPs for which no published PSI associations were located include soil tillage, winter grazing on forage crops, fertiliser use on horticultural and cropping land, earthworks, landfill operations, and the use of pesticides, other organic toxicants and veterinary pharmaceuticals. Widespread LMPs for which only one or two small-scale case studies have been reported include fertiliser use on pastoral land and planted forests, land application of livestock effluent, forestry land preparation, and road use. These rare case studies are not amenable to assessments of consistent associations as noted above.

Of all freshwater receiving environments in New Zealand, aquifers are the least studied in terms of PSI associations. While it is certain that aquifers are receiving environments for land-derived contaminants, few PSI associations have been reported, and large-scale data analyses have not yielded detectable correlations between LULC variables and groundwater state variables (Daughney et al. 2012; Moreau and Daughney 2015). There are several reasons for the insidious effects of land use on groundwater: capture zones are unknown for most monitoring sites; contaminant losses from land and groundwater responses are decoupled by long lag times and distances; and the effects of aquifer geochemistry, recharge and other factors mask land-use effects. The most expedient ways to improve our ability to link land-pressures to groundwater state and impacts may be delineating capture zones in national and regional groundwater monitoring networks, and advancing groundwater contaminant models that can integrate the resulting land use and groundwater data (e.g. Bidwell and Good 2007).

Concluding remarks

In this review, we assessed the evidence for land-use effects on freshwater ecosystems set out in over 100 New Zealand studies. Despite the multitude of studies, the evidence base is not comprehensive and is dominated by weak inferences, which are typical of observational studies (Downes et al. 2002; Norris et al. 2012). The most consistently observed PSI associations involved proxy land-use variables (e.g. proportional land cover) that were associated with state and impact variables by correlation or categorical comparisons. The limitations of these approaches are discussed in the preceding section. The consequences of weak inferences are: (1) uncertainty about the causes of observed degradation; (2) uncertainty about the degree to which land-use effects are reversible; and (3) uncertainty about the effects of future land-use changes. These uncertainties limit the utility of the existing evidence to guide environmental policy-making and management actions (Suter et al. 2002; Adams 2003).

Uncertainty about causes of degradation poses risks that management actions will target pressures that have little effect on the impacts of concern. For example, negative correlations between agricultural land-cover and ecological health metrics often have multiple proximate causes (e.g. nutrient enrichment, sedimentation, flow alteration) that vary in their strength of association with the response metric (Riseng et al. 2011; Schmidt et al. 2018). Management actions aimed at reducing nutrient losses and loads are unlikely to be effective at sites where flow alteration is the dominant stressor. In other cases, associations between land-use pressures and state or impact variables are spurious (i.e. both the pressure and response variables are causally related to a third variable, but not to each other). Management actions based on spurious associations are also unlikely to be effective (Garsd 1984; Suter et al. 2002; Adams 2003).

Uncertainty about the degree to which land-use effects are reversible poses risks that reducing the causal land-use pressure or pressures will not elicit the intended improvement in environmental conditions. For example, many lakes that have been degraded by excessive nutrient loading failed to recover when nutrient loads were reduced (Jeppesen et al. 2005; Hilt et al. 2018). Resistance to recovery when the causal stressor is reduced is indicative of stabilising feedback in degraded ecosystems, such as competitive exclusion of lake macrophytes by phytoplankton (Larned and Schallenberg 2019). In these cases, management interventions in addition to reduced land-use pressures are required.

Preventing adverse impacts of land use before they occur is one of the fundamental aims of freshwater policies in New Zealand (Duncan 2014; Rouse and Norton 2017). To achieve that aim, the effects of land-use change on receiving environments must be predicted in advance (e.g. Trolle et al. 2014). However, the combination of uncertainty about causation and reversibility impedes the prediction of future land-use effects.

The New Zealand government recently set out a strategic plan to ensure that the evidence base needed to inform environmental policy is fit for purpose (MFE 2017). Our assessment indicates that this objective, as it applies to land-use effects on freshwater ecosystems, has not yet been achieved. However, strengthening the evidence base as recommended here can reduce uncertainty and increase confidence about the outcomes of policies and management.

Acknowledgements

We thank Clive Howard-Williams, Ross Monaghan, John Quinn and Ton Snelder for interminable discussions and helpful manuscript reviews.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by Our Land and Water National Science Challenge, Ministry of Business, Innovation and Employment [grant number C10X1507].