A national-scale trophic state analysis to prioritize lakes for restoration in Aotearoa New Zealand

ABSTRACT

Pressure on lakes in Aotearoa New Zealand is increasing because of elevated catchment nutrient loads, establishment of non-native species, and climate change. Current government legislation requires that pressures are managed to avoid eutrophication and degradation of lake health. This approach requires information on the state of lakes at regional and national scales, which is challenging because <5% are currently monitored. In this study, we (1) modelled lake trophic status at a national scale using a highly representative dataset and lake characteristics, land use, and environmental parameters as predictor variables; and (2) identified lakes that should be prioritized for protection to prevent further degradation. Six statistical models were evaluated, with extreme boosting producing the highest predictive power and lowest error. This model indicated that for the 3738 lakes in the dataset, 44% were eutrophic or higher trophic state, 22% mesotrophic, and 34% oligotrophic or lower trophic state. These data provide a benchmark to guide management and support the call for more resourcing to restore lakes in Aotearoa New Zealand. To identify lakes to prioritise for protection, we focused on the ∼800 mesotrophic lakes. We used (1) the portion of lake catchment not within conservation estate, and (2) road access as proxies for the likelihood of land-use intensification and the introduction of non-native species, respectively. We identified >170 lakes with limited catchment protection and easy human access. Immediate attention should be given to protecting these waterbodies to prevent the need for costly and resource-intensive remediation in the future.

KEYWORDS:

• Aotearoa New Zealand
• environmental DNA
• extreme boosted modelling
• lakes
• national-scale predictions
• trophic level index

Introduction

Aotearoa New Zealand has an estimated 3820 lakes >1 ha in area (Leathwick et al. 2010). These lakes are nationally significant fresh waterbodies that provide critical ecosystem services and hold high cultural importance (Te Wai Māori 2008, Schallenberg et al. 2013). Like most lakes worldwide, lakes in Aotearoa New Zealand are impacted by multiple stressors, including water level fluctuations, nutrient and sediment runoff, introductions of non-native species, and climate change (Schallenberg and Sorrell 2009, Drake et al. 2011). Concerns about the health of lakes in Aotearoa New Zealand are long-standing, with several reports documenting eutrophication beginning in the late 1960s and early 1970s (Fish 1969, McColl 1972, Burns and Mitchell 1974). Additionally, a shift to intensive agricultural production land use in the past 2 decades has increased irrigation and fertilizer use at a rate greater than any other nation in the Organisation for Economic Co-operation and Development (OECD 2021). In recent decades, increasing attention has been given to the plight of freshwater in Aotearoa New Zealand, including lakes, with remedial actions now being taken by government, industry groups, Māori (the indigenous people of Aotearoa New Zealand), community groups, and environmental managers in central and local government.

The National Policy Statement for Freshwater Management (NPS-FM; MFE 2020a) was first introduced by the New Zealand government in 2011 and has been updated several times, most recently in 2020. The NPS-FM is intended to arrest degradation of freshwater quality by providing a mechanism to control diffuse pollution. It requires the environmental management authorities (regional and unitary councils) in Aotearoa New Zealand to “maintain or improve” water quality across “freshwater management units.” The NPS-FM is supported by the National Objectives Framework (NOF; MFE 2015), which sets values, attributes, and objectives for water quality. In the NOF, bands are designated to waterbodies from A, relating to ecological conditions close to reference state, to D, indicating unacceptable conditions and requiring actions to achieve a higher band (examples for total phosphorus [TP] and chlorophyll a [Chl-a] in Supplemental Table S1). The bands are defined by thresholds of total nitrogen (TN), TP, Chl-a in surface waters, dissolved oxygen in bottom waters, and LakeSPI, which is a submerged macrophyte index (Clayton and Edwards 2006). In 2017, the New Zealand Government also announced the Clean Water package, which aims to make 90% of lakes and rivers swimmable by 2040 (MFE 2017). For lakes, maps for “swimmability” are produced based on modelled levels of toxic cyanobacteria (Snelder et al. 2016, MFE 2018). The NPS-FM requires that lake restoration plans be developed for lakes that fall below the designated standard in the NOF and that lakes in good condition be protected from degradation.

Implementation of the NPS-FM through the Clean Water package requires national and regional lake managers to have information to support the designation of bands to fresh waterbodies under their jurisdiction. Current regional or national scale assessments have been hampered by small and biased datasets. Only ∼120 of the 3820 (>1 ha) lakes in Aotearoa New Zealand have at least 3 years of seasonal or monthly trophic state data (i.e., TN, TP, Chl-a), and LakeSPI (data exist for ∼300 lakes; MFE 2020b), equating to <4% and 8% of all lakes in Aotearoa New Zealand, respectively. The monitoring data are also heavily biased towards larger, low-altitude lakes, and only a few of the monitored lakes have catchments dominated by alpine or native vegetation.

Statistical modelling has previously been used to predict water quality in lakes at a national scale. Snelder et al. (2022) used 5 years of monitoring data from 120 lakes to predict water quality variables, showing that lakes in low-elevation catchments were generally in a eutrophic or worse trophic condition. Another study estimated nutrient concentrations in ∼1000 lakes using nutrient mass loading models to show that just over 50% of lakes were in a eutrophic or higher trophic state (Abell et al. 2019). The latter study was extended to predict Trophic Level Index (TLI; Burns et al. 1999) in the 1000 lakes and show that trophic state had increased ∼1 categorical unit (e.g., oligotrophic to mesotrophic, mesotrophic to eutrophic) compared with a modelled reference condition. These studies each acknowledged limitations in their datasets and the underrepresentation of certain lake types in their analyses.

Pioneering limnologists such as E. Naumann (e.g., Naumann 1922, 1924) had a strong interest in regional limnology (Jones 2022). However, despite concerted efforts by many (e.g., Magnuson and Kratz 2000) and an increasing array of technologies, such as remote sensing, in-depth regional assessments are lacking for many parts of Earth. The “Our lakes health: past, present, future” programme, also known as Lakes380 (www.lakes380.com), undertook the largest national scale survey to measure the current and historical health of lakes in Aotearoa New Zealand to date. Water, surface sediment, and sediment core samples were collected from ∼10% of natural lakes; samples were carefully selected to represent typologies and environmental gradients. Lakes could only be sampled once because of the logistics of sampling such a large number and the remoteness of many.

A new method, known as the Sediment Bacterial Trophic Index (SBTI), was developed and used as a proxy for TLI, which requires monthly monitoring of nutrients and Chl-a over 3–5 years to provide a representative value for each lake (Snelder et al. 2022). The SBTI infers lake trophic status based on different indicator bacteria present in lake surface sediment (Pearman et al. 2022). Unlike water samples, in which physicochemical attributes and lake ecological communities are often highly variable because of climate conditions and seasonality, surface sediment provides a temporally integrated representation of the lake conditions and organisms as well as inputs from the catchment (Schallenberg and Kalff 1993). The SBTI uses 16S ribosomal RNA (16S rRNA) metabarcoding to characterize the bacterial community, from which key indicator bacteria are identified and the index calculated. The SBTI was compared to TLI during its development, but without Secchi depth. For 100 monitored lakes, the SBTI and TLI were strongly correlated (r² = 0.842, p < 0.001; Pearman et al. 2022). The requirement for only a single sampling time point makes the SBTI a rapid and cost-effective tool for inferring trophic state in large numbers of lakes.

In the present study, we aimed to (1) predict lake trophic status at a national scale using the relationship between the SBTI obtained from the highly representative Lakes380 dataset and nationally available data on lake and catchment characteristics, and (2) identify lakes that should be prioritized for protection. For the second aim, we focused on mesotrophic lakes. We considered the amount of a lake’s catchment not in protected land as a proxy for the likelihood of potential land-use intensification that will impact external loads. About a quarter of lakes in Aotearoa New Zealand have >30% of their catchment in publicly owned conservation estate (land that has conservation purposes and is protected for its natural and intrinsic resources) administered by Te Papa Atawhai – Department of Conservation (DoC). Although this conservation land has varying degrees of protection, any land-use changes are subject to evaluation. We also assessed road access as a proxy for recreational activity, which is known to increase the risk of incursions of non-native species (Kilroy and Unwin 2011, Darwall et al. 2018, Panlasigui et al. 2018).

Methods

Lakes with predicted trophic level index

Surface sediment samples (top 5 mm) were collected from 265 lakes in Aotearoa New Zealand across a variety of land-use and environmental gradients, as described in Pearman et al. (2022) (Supplemental Table S2). DNA was extracted from these samples and the 16S rRNA gene amplified and sequenced, as detailed in Pearman et al. (2020). Sequences were processed in DADA2 (Callahan et al. 2016), bacterial indicator Amplicon Sequence Variants (ASVs) were identified, and an SBTI was calculated as detailed in Pearman et al. (2022). This index uses the proportion of bacterial trophic state indicator ASVs from a single mid-lake sediment sample to infer the trophic status of a lake. The indicator ASVs were determined based on the investigation of the bacterial community in 96 monitored lakes with a TLI assignment across Aotearoa New Zealand. The index was validated using 5-fold repeated cross-validation, with stringent filtering of ASVs to remove rare and underrepresented taxa. The index was developed and calibrated based on TLI. The SBTI was calculated by the following equation as developed, validated, and described in Pearman et al. (2022):

SBTI = 0.017 × %microtrophic + 0.026 × %oligotrophic + 0.031 × %mesotrophic + 0.040 × %eutrophic + 0.055 × %supertrophic + 0.067 × %hypertrophic, (1)

where, for example, %microtrophic is the percentage of indicator taxa in the sample that are microtrophic indicators. SBTI values were categorized into 6 trophic states based on the classifications of Burns et al. (1999) (Supplemental Table S3): microtrophic (SBTI 1–2), oligotrophic (SBTI 2–3), mesotrophic (SBTI 3–4), eutrophic (SBTI 4–5), supertrophic (SBTI 5–6), and hypertrophic (>6).

Environmental predictor data

Lake and catchment characteristics data were obtained from Freshwater Ecosystems of New Zealand (FENZ) geodatabase (Leathwick et al. 2010), which contains descriptors of climatic, geological, topographical, and hydrological conditions for 3782 lakes in Aotearoa New Zealand classified as >1 ha in area. Land-use characteristics of the catchment of each lake were retrieved from the Land Cover Database v5 (Landcare Research New Zealand, Ltd.; https://lris.scinfo.org.nz/layer/104400-lcdb-v50-land-cover-database-version-50-mainland-new-zealand/) and categorized as (1) native vegetation, (2) native forest, (3) urban, (4) non-native vegetation, (5) water, (6) forestry, (7) high production grassland, (8) low producing grassland, and (9) other (category composition in Supplemental Table S4).

Data exploration, including outlier detection, was conducted following Zuur et al. (2010). Correlations among land and catchment data (from FENZ) and land-use variables were examined using R (R Core Team 2022) packages lares (Lares 2022) and ggcorrplot (Kassambara 2019). For correlation coefficients >0.9, only the most ecologically relevant variable was retained, leaving 21 environmental predictors (Table 1) after the removal of 3 variables (percentage of catchment in water or urban land use and estimated catchment percentage cover of peat soil) with near 0 variation.

Table 1. Freshwater Ecosystems of New Zealand (FENZ) and land-use characteristics from the Land Cover Database v5 selected for use in modelling.

FENZ characteristics	Unit	Mean	Median	Q25	Q75
Max depth	m	19.2	18.1	9.0	26.0
Lake area	ha	99.8	2.5	1.4	6.4
Lake elevation	m	462.3	280.2	52.9	787.2
Catchment slope	degrees	11.9	7.5	2.4	21.7
Catchment area	ha	51 608 113	694 800	261 038	2 681 325
December solar radiation	MJ/m^2/d	2209.3	2265.4	2129.3	2333.4
December air temperature	°C	14.3	14.9	11.2	17.3
Mean wind	m/s	4.3	4.1	3.2	5.2
Catchment flow	m^3/s	791 936	760 970	137 185	4 760 519
Catchment phosphorus	Categorical 1–5	2.2	2.0	1.0	3.0
Catchment calcium	Categorical 1–4	1.4	1.1	1.0	2.0
Catchment particle size	Categorical 1–5	3.2	3.5	2.0	4.9
Direct distance to the coast	km	27.3	17.0	4.5	42.2
Residence time	years	27.8	1.2	0.3	6.6
Land-use characteristics
High productivity exotic grass	%	30.4	0.0	0.0	73.7
Low productivity grass	%	5.8	0.0	0.0	0.0
Forestry	%	5.9	0.0	0.0	0.8
Native	%	25.5	2.0	0.0	48.3
Native forest	%	15.8	0.0	0.0	15.5
Non-native	%	3.1	0.0	0.0	0.0
Other catchment type	%	11.5	0.4	0.0	8.8

The mean, median, and 25th (Q25) and 75th (Q75) quantiles for the data included in the modelling.

Environmental predictors were centred and scaled using the R package caret (Kuhn 2008) to allow comparisons. Because data were missing for the selected environmental variables, 3738 total lakes were retained for use in the modelling analysis. Histograms showed that the sampled lakes provided a good representation of the land-use and catchment characteristics across the country. The deviation among all land-use categories was <10% (Supplemental Fig. S1). Shallow (<10 m) and small (1–10 ha) lakes and those with small catchment areas and low catchment phosphorus and calcium levels were slightly underrepresented (Supplemental Fig. S2). Similarly, medium-sized lakes (10–100 ha) with high summer solar radiation, mid-range catchment areas, and moderate catchment calcium levels were also slightly underrepresented (Supplemental Fig. S2).

Each lake was placed into 1 of 9 different geomorphic classes (coastal/shoreline, dam, glacial, landslide, riverine, swamp/wetland, tectonic, volcanic, and wind/aeolian) as defined in FENZ (Leathwick et al. 2010).

Modelling

Six different machine learning methodologies (Random Forest, Boosted Regression Tree [BRT], extreme boosting, linear regression, support vector machine, and neural networks) were investigated to identify the best-performing model to predict SBTI based on the 21 selected environmental predictors using the R package caret. Cross-validation took a stratified 5-fold approach based on trophic state categories of the SBTI using the trainControl function (method = cv) within caret. Model performance was assessed using the mean absolute error (MAE), root mean square error (RMSE), and R-squared (r²) values. Extreme boosting and BRT had the highest performance; however, we used extreme boosting for inference and predictions because of its slightly lower MAE and RMSE values (Supplemental Table S5, Supplemental Fig. S3). Machine learning algorithms can be individually trained on data or aggregated with other models to form an ensemble. Ensembles involve integrating multiple basic models to generate a stronger collective model. Extreme boosting represents a technique used to establish such ensembles. Initially, a base model is trained using the data, and a second model is then constructed to enhance predictions for observations that challenged the first model. The amalgamation of these 2 models aims to outperform each standalone model. This boosting iteration is replicated multiple times, with each successive model aiming to rectify deficiencies in the combined boosted ensemble encompassing all prior models. Advantages of using these machine learning approaches also include the ability to automatically handle interactions between predictor variables, to rank the predictors according to their relative contributions, and to describe their potentially complex nonlinear response curves, which are key to detecting appropriate thresholds on the effect of important predictors.

To minimise over-fitting, the tolerance function within caret was used to select the least complex model within 2% of the best model. The final options used for the extreme boosting model were nrounds (number of passes of the data) = 50, max_depth (depth of tree) = 1, eta (learning rate) = 0.3, gamma (minimum loss reduction to make split) = 0, colsample_bytree (fraction of columns to be random samples for each tree) = 0.6, min_child_weight (minimum sum of weights of all observations required in a child) = 1, and subsample (fraction of observations to be random samples for each tree) = 1.

Sediment Bacterial Trophic Index predictions

SBTI values were predicted for all lakes using the selected extreme boosting model. First, 265 lakes with surface samples were selected, and the predicted SBTI was regressed against the observed SBTI value. Using the caret package, performance metrics (MAE, RMSE, and r² values) were calculated using the function postResample, and the importance of predictor variables on the model’s predictive performance was investigated using the function varImp. Partial plots for the environmental variables were calculated in package pdp (Greenwell 2017) and plotted with ggplot2. Absolute residuals were calculated from the linear equation within the R environment.

Vulnerable lakes – conservation estate and road access

To help prioritize lakes for protection, we used (a) the portion of the catchment not in conservation estate as a proxy for the likelihood that land-use intensification would impact external loads to the lake, and (b) road access as a proxy for recreational activity, a major pathway for the introduction of non-native species to lakes (Trombulak and Frissell 2000). These variables were not included in the extreme boosting model because the intention was to conduct a separate, easy-to-apply analysis to guide the prioritization of lake protection efforts. This component of the analysis primarily focused on lakes predicted to be mesotrophic.

Spatial data for protected areas and selected lake catchment boundaries were sourced from DoC Public Conservation Areas dataset (https://koordinates.com/layer/754-doc-public-conservation-areas/data/) and the FENZ geodatabase, respectively. FENZ lake areas were removed from both datasets, and areas of lake catchments were recalculated. To streamline processing, this dataset was cropped and filtered to exclude conservation areas that did not overlap with the lake catchments of interest. Conservation area polygons within selected lake catchments were extracted, and areas were summed for the polygons, along with the proportion of each lake catchment covered by these polygons. This analysis used package sf (Pebesma 2018) within the R environment for statistical computing (2022).

We obtained data on road access from Kilroy et al. (2021), defined as the presence of a road within 200 m of the lake, a distance deemed relevant to locations where recreational equipment could be transported to the lake shore (Kilroy et al. 2021).

Results

An SBTI was derived from surface sediment for all 265 sampled lakes. Of these, the SBTI estimates were 5% microtrophic, 25% oligotrophic, 25% mesotrophic, 20% eutrophic, 22% supertrophic, and 5% hypertrophic (Supplemental Table S3).

Overall, mean model predictive performance was high and comparable across the 6 candidate algorithms, except for linear regression, which had relatively low r² and high MAE and RAE values (Supplemental Table S5, Supplemental Fig. S4). Cross-validation results for the extreme boosting algorithm had a mean MAE of 0.51 (min = 0.43, max = 0.57), RMSE of 0.67 (min = 0.52, max = 0.77), and r² of 0.71 (min = 0.63, max = 0.85; Supplemental Table S5). The final extreme boosting model fitted with all the data had an MAE of 0.28, ranging from 0.25 in microtrophic lakes to 0.31 in eutrophic lakes, an RMSE of 0.37, and an r² of 0.91. A strong and significant relationship existed between observed and predicted SBTI (r² = 0.91, p < 0.001, slope = 1.03; Fig. 1). Overall, the SBTI category was correctly predicted for 73% of all sampled lakes, with only 1 lake (<0.5%) >1 trophic level away from the observed classification. The highest proportion of misclassifications occurred for lakes with extreme trophic states, a generally underrepresented group, with only 42% and 58% correctly classified for microtrophic and hypertrophic lakes, respectively.

Figure 1. Regression analysis showing the observed Sediment Bacterial Trophic Index (SBTI, y-axis) against the predicted SBTI (x-axis) for 265 lakes. The blue line depicts the regression line (slope = 1.03) with the grey shading the 95% confidence limits, while the dotted line represents the 1:1 line.

The percentage of high-productivity grassland (HPG) in the catchment of the lake was ranked first in predictive importance, followed by summer temperature, lake summer solar radiation, and lake elevation (Supplemental Fig. S4). Partial plots of the top 3 characteristics of the importance plot indicated that increases in these variables resulted in higher SBTI values (Fig. 2) and that the conversion of a lake’s catchment from lacking HPG to complete (now defined above) coverage will result in an average increase of ∼1 unit in SBTI. Similarly, a temperature shift from 10 to 17.5 °C will lead to ∼1 unit elevation in SBTI.

Figure 2. Partial plots from the extreme boosting model showing the effect on the Surface Bacterial Trophic Index (SBTI) of predictor variables that had importance >0.05. The predictors are ordered by their relative contribution (Supplemental Fig. S4). HPG = high productivity grassland, rad. = radiation. The blue line represents loess smoothing of the data with the grey shading representing the 95% confidence limits.

Of the 3738 lakes in our dataset, 44% were predicted to be eutrophic or higher trophic state, 22% mesotrophic, 32% oligotrophic, and 2% microtrophic (Fig. 3). Of lakes on North Island, 83% were predicted to be eutrophic, supertrophic, or hypertrophic, compared with 23% on South Island (Fig. 3).

Figure 3. The predicted trophic states based on the Sediment Bacterial Trophic Index for 3738 lakes in Aotearoa New Zealand shown (a) geographically, and (b) as the proportion of lakes in each trophic level.

Regional comparisons showed that Taranaki had the highest percentage of lakes in eutrophic condition or higher (91%), followed by Hawkes Bay (90%) and Gisborne (89%; Fig. 4, Supplemental Table S6), all regions on the North Island. Southland had the highest proportion of lakes in an oligotrophic or microtrophic condition (81%), which was a markedly higher proportion than for the next region (Nelson & Tasman 42.7%; Fig. 4; Supplemental Table S6), all regions on the South Island.

Figure 4. The predicted percentage of lakes in 3 trophic classes across 15 regions in Aotearoa New Zealand. Lakes in oligotrophic and microtrophic state are combined, as well as those in eutrophic state or higher, to aid visualization (full data provided in Supplemental Table S4).

Based on geomorphic class, the highest proportion of oligotrophic or lower trophic state lakes were of glacial origin (80%; Table 2). The highest proportion of eutrophic or higher trophic state was wind/aeolian and dams (77% and 74%, respectively; Table 2).

Table 2. Percentage of lakes in Aotearoa New Zealand predicted for each trophic status based on the modelled Sediment Bacterial Trophic Index (SBTI; Pearman et al. 2022), grouped by geomorphic lake type.

Geomorphic class	Microtrophic/oligotrophic	Mesotrophic	Eutrophic, supertrophic, hypertrophic
Coastal/shoreline	11.1	16.2	72.6
Dam	4.6	21.6	73.8
Glacial	79.8	17.0	3.3
Landslide	25.6	46.2	28.2
Riverine	5.2	24.9	70.0
Swamp/wetland	14.3	36.0	49.7
Tectonic	0	28.6	71.4
Volcanic	10.6	25.0	64.4
Wind/aeolian	2.3	20.3	77.4

Lakes with a low proportion (<33%) of their catchment in conservation land had elevated trophic states, with 73% of the lakes predicted to be eutrophic or higher (Fig. 5; Supplemental Table S7). By contrast, a high proportion of lakes with >66% of their catchment area in conservation land were either oligotrophic (71%) or microtrophic (4%). Of the 818 lakes predicted to be mesotrophic, 22% have less than a third of their catchment in conservation land and were deemed vulnerable to degradation due to land-use intensification in their catchments.

Figure 5. Percentage of lakes predicted in each trophic state dependent on the percentage of their catchment in conservation estate: Low ≤ 33%, Medium = 33–66%, and High ≥ 66%.

Only 27% of all lakes in Aotearoa New Zealand have road access. Of the 818 lakes predicted to be mesotrophic, 30% have road access. The region with the highest number of mesotrophic lakes with road access was Otago (50 lakes), followed by Canterbury (46) and the West Coast (40; Table 3), all regions of the South Island, but these lakes accounted for <50% of the mesotrophic lakes in the region. In Auckland, 8 of 10 mesotrophic lakes had road access, while Waikato (58%) and Bay of Plenty (58%) also had road access to 50% of the mesotrophic lakes, all regions on the North Island. [Model outputs for all lakes are given in Supplemental Table S8; recent monitoring data are accessible at https://lakes380.upshift.co.nz/].

Table 3. Number of lakes modelled as mesotrophic in each region, and percentage with road access.

Region	No. of mesotrophic lakes	No. with road access	% with road access
Auckland	10	8	80.0
Canterbury	169	46	27.2
Bay of Plenty	19	11	57.9
Waikato	38	22	57.9
Gisborne	6	3	50.0
Hawke’s Bay	22	4	18.2
Manawatū-Wanganui	18	10	55.6
Marlborough	13	3	23.1
Northland	67	22	32.8
Otago	121	50	41.3
Southland	117	18	15.4
Taranaki	7	0	0.0
Nelson & Tasman	29	1	3.4
Wellington	17	4	23.5
West Coast	165	40	24.2

Discussion

A call for greater management action to protect and restore lakes in Aotearoa New Zealand

Our analysis provides regional and national scale predictions of trophic status for lakes in Aotearoa New Zealand. Of greatest concern was the high number of lakes nationally (∼44%), particularly on the North Island (∼83%), predicted to be eutrophic or higher. This finding generally aligns with previous studies that predicted TLI at a national scale (Snelder et al. 2022). Abell et al. (2019) modelled reference (i.e., prehuman/natural) conditions for >1000 larger lakes in Aotearoa New Zealand and predicted that ∼10% of lakes (depending on the nutrient being considered) were eutrophic, only 1–3% were supertrophic, and none were hypertrophic. A comparison between the data of Abell et al. (2019) and the predictions in the present study highlights the transformation of lake water quality that has occurred largely related to anthropogenic activities, albeit we note that a small portion of lakes in Aotearoa New Zealand were naturally eutrophic. A national-scale paleolimnologically focused study is currently underway (www.lakes380.com) that aims to provide information on reference conditions and the timing and rate of change and drivers of decline in lakes across Aotearoa New Zealand.

In the present study, most lakes predicted to be in a degraded state are warmer, at low elevations, and in highly modified catchments where agriculture is the dominant land-use, consistent with multiple studies globally showing that human activity has transformed landscapes, leading to an increase in nutrient in streams and lakes (Smith 2003, Dodds et al. 2009, Carpenter et al. 2011, Schindler 2012, Jenny et al. 2020). Numerous lake-specific or regional studies have also highlighted associations between high-intensity land use and poor lake water quality within Aotearoa New Zealand (Verburg et al. 2010, Abell et al. 2011, Paul et al. 2012).

The NPS-FM (MFE 2020a) requires that regional councils develop plans to improve lakes in their jurisdiction to meet or exceed the national bottom-line standards (i.e., above a D band). These standards are defined by concentration thresholds of TN, TP, and Chl-a (TP ≥ 50, TN ≥ 750, and Chl-a > 12 mg/m³) for the C–D band threshold rather than TLI. However, these concentrations approximately align with the values commonly observed in lakes classified as supertrophic or hypertrophic when using the TLI (TP = 50–100 mg/m³, TN = 500–1500 mg/m³, and Chl-a 15–30 mg/m³, Waikato Regional Council 2022). Our modelling suggests that ∼660 lakes fail the NPS-FM “bottom line” nationally and require management and restoration plans that, according to the NPS-FM, should be implemented by 2026.

Lake restoration is a complex and costly undertaking that requires in-depth knowledge of the drivers of change and an understanding of the desired restored state (Spears et al. 2022), and it could take many decades or centuries, depending on the degree of degradation (Søndergaard et al. 2007, Hamilton et al. 2016). Numerous options for lake restoration now exist, including biomanipulation (Jeppesen et al. 2012); engineering solutions such as hypolimnetic withdrawal, artificial mixers, or aerators (Grochowska and Gawrońska 2004); chemical flocculants or binding agents (Reitzel et al. 2005); and nutrient reduction strategies including artificial wetlands and biomanipulation (Mehner et al. 2002, Bozic et al. 2013). Methods for lake restoration exhibit varying degrees of success. For instance, published studies using biomanipulation have demonstrated a success rate of ∼60% (Mehner et al. 2002). Studies documenting the effects of hypolimnetic withdrawal mostly indicate a decrease in phosphorus and chlorophyll concentrations as well as a reduction in hypolimnion anoxia, although the approach is not commonly applied (Nürnberg 2020). Although hypolimnetic aeration has been shown to lead to substantial reductions in sediment phosphorus release, with some cases showing up to a 90% decrease (Walker et al. 1989), other studies have indicated it often falls short in increasing oxygen levels in the bottom waters, preventing the desired effects from being achieved (Niemistö et al. 2020, Ruuhijärvi et al. 2020). Without actions taken to control external nutrient inputs, mitigation measures such as the use of flocculants and sediment capping typically yield only short-term improvements (van Oosterhout et al. 2022). In Aotearoa New Zealand, few examples of successful improvements in lake water quality exist, largely involving ongoing costs and major human interventions such as continued alum dosing of inflows into Lake Rotorua (Smith et al. 2016), engineered inflow diversion in Lake Rotoiti, and artificial destratification in Mangatangi Reservoir (Hamilton et al. 2016). Despite considerable investment, current lake restoration in Aotearoa New Zealand is largely failing to meet its objectives. For example, the Te Arawa-Rotorua lakes restoration programme has cost ∼NZD$140 million over 25 years and has not achieved all its desired outcomes (MFE 2021). Recent government funding for restoration demonstrates the scale of economic investment that will be required in the decades ahead; for example, NZD$11.2 million was allocated to restore Lake Horowhenua in 2021 (The Beehive 2021), and groups involved in the restoration of Te Waihora received NZD$4.2 million in 2020 (MFE 2023).

The magnitude of the task ahead is daunting for many regions. The Manawatū-Whanganui region has a population of only ∼240 000 and limited resources for environmental improvement. Our data indicate that an estimated 40% of lakes in this region will fail the NPS-FM “bottom-line.” Developing restoration plans for these lakes individually will be challenging and time consuming. Successful restoration at regional and national scales will require a combination of methods, including reducing external and internal loads, biomanipulation, hydrological modification, and new innovative methods, with a selection of approaches specifically tailored for each lake (Hamilton et al. 2016). Plans will also require appropriate sociopolitical frameworks, including leadership by water management agencies and iwi (Māori tribes), engaged and motivated local groups, and changes in regulation.

Our model identified summer air temperature as one of the predictors of trophic state, with higher temperature related to poor TLIs. Summer temperature will likely increase in many regions with climate change and impact lake trophic state (Trolle et al. 2011). The effects of climate change on lakes in Aotearoa New Zealand will likely be profound and include increased temperatures, changes in stratification patterns, increased invasions by non-native species, and extreme climate events (e.g., storms and droughts), which influence external nutrient and sediment inputs (Hamilton et al. 2012). The potential implications of climate change urgently need to be incorporated into the future management and restoration of lakes in Aotearoa New Zealand.

Protecting the most vulnerable lakes

Our model predicts ∼800 mesotrophic lakes are located in Aotearoa New Zealand. Action taken now to prevent the degradation of these lakes will reduce the need for expensive and time-consuming restoration in the future. To assist in prioritizing which lakes should be targeted for protection, we focused on (1) the portion of lake catchment not protected as a proxy for the likelihood of land-use intensification, which would impact external loads entering a lake; and (2) road access, shown to increase incursions of non-native species.

Intensification of land use in Aotearoa New Zealand over the last 20 years, in particular dairy farming and urbanization, has resulted in increases in catchment nitrogen and phosphorus loads entering waterways (Parliamentary Commissioner for the Environment 2013, Snelder et al. 2018). The Department of Conservation administers public conservation land in Aotearoa New Zealand. Although the degree of protection varies depending on category (e.g., National Park, Conservation Area, Reserve and Marginal strip) or status the land holds under various legislation, we predict heightened levels of scrutiny if land-use intensification were to occur. We acknowledge other land management types that promote conservation, including land in private ownership, regional parks, and QEII national trusts (https://qeiinationaltrust.org.nz/); however, their evaluation was beyond the scope of this work. Of the 818 lakes predicted to be mesotrophic, 463 had <33% of their catchment in conservation estate. We recommend further evaluation of land-use in the catchment of these lakes and, where feasible, action to ensure protection.

The introduction of non-native species can have substantial impacts on the ecology of the lake (Craig 1992, Mills et al. 1994, Rowe 2007, Kilroy et al. 2021). Non-native macrophyte introductions, for example, have been associated with regime shifts that have ultimately led to a resilient turbid state in shallow lakes (Schallenberg and Sorrell 2009). While we acknowledge that nonindigenous species may have already established in many lakes, biosecurity actions to prevent the spread of these species will help prevent further ecological degradation. Pearman et al. (2022) showed that lakes with public access on the South Island of Aotearoa New Zealand, especially road access, had distinct lake communities. While humans are not the only mechanism for dispersal of non-native species (e.g., mammals, birds; Kilroy et al. 2021), they play a substantial role. Of the lakes modelled to be mesotrophic, 242 had road access. We urge lake managers to continue with and/or undertake new biosecurity actions to limit the introduction and spread of new species.

A total of 176 mesotrophic lakes had road access, and <33% of their catchment was also in conservation estate; these lakes should be prioritized for actions to ensure no further degradation. Acting now to explore mechanisms to reduce external nutrient loads and prevent incursions of non-native species will likely prevent future resource-intensive restoration actions.

Study limitations

Our sample set of 265 lakes was highly representative of lakes in Aotearoa New Zealand across a range of lake attributes, land-use, and other catchment parameters. Of note was an underrepresentation of small lakes (<1–10 ha), many of which are constructed farm dams not considered during lake selection because the Lakes380 project aimed to reconstruct changes in historic lake health over a 1000-year period, which includes pre-human settlement in Aotearoa New Zealand. This evaluation is clearly not possible for constructed lakes, which were excluded during the sample selection process, but they remain in the national dataset of predicted trophic status.

The approach used in this study allows us to model broad-scale patterns in TLI using catchment characteristics and lake-scale descriptors as predictors. However, the processes influencing lake water quality are complex and include variables not accounted for in our approach, such as groundwater inputs and internal cycling; therefore, some unexplained variation in our model is expected. The MAE, RMSE, and r² values for the selected model are considered good (Moriasi et al. 2015), indicating that the predicted TLIs reflect broad-scale differences between lakes. While the modelling approach used in this study is useful for providing a national and regional scale overview, caution should be taken when interpreting the data at an individual lake level. The data generated by this model are currently being used by lake managers across Aotearoa New Zealand to prioritize lakes for protection or restoration. These lakes are then targeted for sampling to confirm the TLI model predictions and to obtain further information on lake health and drivers of decline.

Also note that our model was informed by SBTI rather than water quality data. While this enabled us to better represent lakes across different environmental and land-use gradients, slight differences exist between the TLI as determined using water quality data and as estimated using the SBTI as analysed and discussed in Pearman et al. (2022). Pearman et al. (2022) showed the SBTI was strongly correlated with TLI from 96 lakes with multiple years of data; however, the sediment samples for the SBTI were only taken from a single time point. Further research involving analysing sediment samples from multiple seasons is needed to confirm seasonal stability in the SBTI.

Conclusions

This study estimates the current trophic state of all lakes in Aotearoa New Zealand. The estimates offer valuable information for strategic purposes, such as identifying high-priority areas for restorative and conservation interventions in catchments and within lakes. Combined with previous studies using similar approaches, our results highlight the bleak current state of lakes in Aotearoa New Zealand, with ∼44% of lakes nationally (∼1700 lakes) estimated to be eutrophic or worse. Recent evidence clearly shows that climate change will bring new challenges and exacerbate existing threats. Urgent action is needed to develop effective restoration approaches at the lake, regional, and national scales. We advocate for immediate attention to the 818 lakes in mesotrophic condition and, in particular, the 176 lakes with limited catchment protection (<33% in conservation land) and easy human access (road accessible).

Acknowledgements

We thank all members of the Lakes380 team for field assistance (www.lakes380.com/about-the-project/the-team). We thank Sophie Young and Eloise Beattie (Cawthron) for laboratory assistance. We thank Robyn Jones from “Friends of Mangarakau” for helping to sample Lake Mangarakau. We acknowledge the support of New Zealand regional authorities that provided data and permission to use it and assisted with access to the sampling sites: Northland Regional Council, Auckland Council, Waikato Regional Council, Bay of Plenty Regional Council, Hawkes Bay Regional Council, Taranaki District Council, Horizon Regional Council, Greater Wellington Regional Council, Marlborough District Council, Tasman District Council, West Coast Regional Council, Environment Canterbury, Otago Regional Council and Environment Southland. The authors thank iwi and landowners nationwide for their assistance with sampling, accessing sites and guidance throughout this work. The Department of Conservation is acknowledged for assistance with permitting.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was funded by the New Zealand Ministry of Business, Innovation and Employment research programme – Our lakes’ health; past, present, future (Lakes380; C05X1707) and additional support from Cawthron Internal Investment Fund (2015–2021) and the Strategic Science Investment Funding to GNS Science in the framework of the Global Change Through Time research programme.