Diverse contributions of nature to climate change adaptation in an upland landscape

ABSTRACT Nature provides numerous functions and services that can contribute to societal climate change adaptation. These ‘adaptation services’ can be sustained, latent, or novel, depend on persistence or transformation of ecosystems, and require varying co-production by people. Adaptation services also include climate mitigation. We present an approach to explore an extensive set of land use and climate scenarios that outline possible futures for a landscape, and quantify the contributions of adaptation services. We quantified adaptation success across six criteria relevant to the region, characterised the contributions of different types of adaptation services, and mapped spatial variation in contributions across the landscape. We built an integrated model of the Mackenzie District (an upland landscape of Aotearoa New Zealand), and analysed 1200 hypothetical scenarios for the period 2060–2070. We found many adaptation options, with 46% of scenarios meeting the criteria for successful adaptation. Four sustained, two latent and five novel services co-produced through financial input made diverse contributions to economic profit, profit resilience, climate change mitigation, climate risk adaptation, landscape cultural heritage and biodiversity. Successful adaptation scenarios were multifunctional, relying on alternative combinations of services allowed by spatial heterogeneity. By accounting for the numerous relationships between people and natural components within complex landscape systems, our advanced simulation approach can inform participatory pathway development by quantifying the potential for nature to contribute to future climate change adaptation.


Introduction
Global environmental change challenges societies to maintain and improve human and planetary health (Sachs et al. 20222022).There is increasing policy and practical interest in nature-based adaptation to future environmental change (Cohen-Shacham et al. 2016;Seddon et al. 2020;Pörtner et al. 2021), but we need to build systematic evidence for the role of ecosystems in supporting pathways towards successful adaptation (Cohen-Shacham et al. 2016;Seddon et al. 2020).Analysing this potential to support adaptation to future climate change must consider the numerous pathways through which adaptation could be achieved, and the complex interactions between climatic change, ecological responses and human coproduction they represent (Colloff et al. 2020;Lavorel et al. 2020).
Adaptation services are ecosystem processes or services that increase the ability of people to adapt to environmental change (Colloff et al. 2015;Lavorel et al. 2015Lavorel et al. , 2020)).Achieving successful adaptation will require balancing a range of environment, social, and economic objectives, therefore adaptation services include not only the role of ecosystems in directly contributing resilience to climate change-induced pressures, but also contributions to climate mitigation, and fulfilling the myriad needs and desires of society -for example in providing ecosystem services, supporting cultural practices, and protecting biodiversity (Lavorel et al. 2015(Lavorel et al. , 2020;;Bruley et al. 2021).Quantifying outcomes of future scenarios of nature-based adaptation is essential to guide the development of adaptation pathways but has not yet been done in previous studies addressing.To quantify outcomes, adaptation services can be described according to three key elements (Lavorel et al. 2020).First, we characterise the biophysical components that give rise to future services by defining them as either persistent or transformed since the pre-adaptation reference point in time (Lavorel et al. 2020).Second, we classify adaptation services as either sustaining current ecosystem services, developing latent services that currently exist but are not recognised, or delivering novel services that have not been produced before in the region (Colloff et al. 2015;Lavorel et al. 2015Lavorel et al. , 2020)).Finally, recognising that all services are the outcome of socio-ecologicaltechnological systems that require human coproduction, we quantify for the first time the relative contributions from humans to their supply (Lavorel et al. 2020).
It is essential to quantify potential adaptation scenarios and their key characteristics, to support evidencebased decision and anticipatory policy and action (Donatti et al. 2020;Seddon et al. 2020;Pettorelli et al. 2021;Pörtner et al. 2021).Landscapes are complex systems with varied environmental conditions, diverse human actors who have contrasting perspectives, and varying capacity and demand for different adaptation services (Fischer et al. 2015;Colloff et al. 2017).For a given landscape and under a specified future climate and socio-economic scenario, there are therefore multiple possible pathways to successful adaptation.Each pathway relies on a specific combination of adaptation services, with bundles of services that can be produced simultaneously, as well as some trade-offs between what can be achieved (Vallet et al. 2018;Lavorel et al. 2019).To address the complexity of landscape adaptation, we can explore possible alternatives by simulating possible scenarios and assessing their multidimensional performance (Herzig et al. 2018;Vallet et al. 2018).Scenario generation has been used to analyse potential futures within the field of ecosystem services (Elkin et al. 2013;Harrison et al. 2019;Knoke et al. 2020;Dominati et al. 2021) so may be extended to inform adaptation by running scenarios that investigate the combined impacts of different land use and climate change scenarios.Whether a landscape has successfully adapted to change is a subjective judgement that can only fully be made with hindsight by the residents of a place (Doria et al. 2009;Westoby et al. 2020), and developing indicators and thresholds for adaptation success is complicated by the fact that people's opinions of adaptation success will vary across stakeholder groups and over time (Palomo et al. 2019;Bardgett et al. 2021).Nonetheless, we can set thresholds based on our current understanding of the system and its values to indicate successful adaptation (Adger et al. 2005;Bardgett et al. 2021).A broad range of criteria that consider multiple aspects of ecological and human well-being to assess adaptation projects have been developed, for example indicators of changes in crop production, income, and soil erosion (Donatti et al. 2020).
Much of the research into future adaptation pathways has taken a qualitative, participatory approach, yet is generally not supported by quantitative modelling of the likely adaptation service outcomes (Cradock-Henry et al. 2020, 2021;Werners et al. 2021;Bergeret and Lavorel 2022).Participatory research into adaptation pathways will benefit from quantitative modelling tools to explore the range of possible options for future transformation (Verkerk et al. 2018;Frantzeskaki et al. 2019;Pedde et al. 2019), and characterise the contributions of different adaptation services in these alternative future scenarios (Lavorel et al. 2019).Analyses of adaptation services in a future-facing, holistic sense have not yet been fully put into practice -although some retrospective studies have examined historical cases of adaptation (Colloff et al. 2020;Bruley et al. 2021), and there are examples of quantitative modelling to investigate facets of adaptation potential (Colloff et al. 2015;Lavorel et al. 2019).
This study showcases an approach to explore future scenarios and quantify the contributions and characteristics of different adaptation services towards achieving successful adaptation.We illustrate the approach for the Mackenzie District in Aotearoa New Zealand, where strong climate change exposure combines with tensions around potential futures for the region, and where knowledge on potential nature-based adaptation could contribute to ongoing political processes (Upper Waitaki Shared Vision Forum 2020).We generated and evaluated a wide variety of potential future climate and land use scenarios to: (1) quantify their adaptation success; (2) characterise the contributions of different types of adaptation services to successful adaptation; (3) identify bundles of adaptation services that were commonly produced simultaneously; and (4) quantify spatial variation in the impacts of different adaptation services, by mapping the frequency of spatial patterns across the successful adaptation scenarios.

Study area
Mackenzie District is an administrative region of New Zealand's South Island, with a population of just over 5000 people and area of 7339 km 2 .Mackenzie District incorporates three distinct topographic regions, with two flatter regions separated and surrounded by mountain ranges (Figure 1).The dominant land covers are tussock grasslands, gravel, and productive grasslands (Figure 1; Supplementary Table S1).The dryland basin in the west of the district is known as Te Manahuna or the Mackenzie Country, and has a landscape heritage value associated with its tussock grasslands, wide open vistas, and open skies (Thompson-Carr 2012;Upper Waitaki Shared Vision Forum 2020).Dryland pastoral agriculture has been historically important to its economy, and continues alongside hydropower production and recreation (Abbott et al. 2019;Upper Waitaki Shared Vision Forum 2020).Irrigation of the dryland basin for dairy farming has increased over recent decades, leading to considerable conflict between proponents of agricultural intensification and those with nature conservation and cultural heritage concerns (Brower et al. 2018;Upper Waitaki Shared Vision Forum 2020).The eastern region of gently rolling hills is less arid and supports sheep and beef, dairy, and crop production.The mountain region incorporates part of the Southern Alps mountains, including Aoraki/Mount Cook which is the highest point in the range at 3754 m.It also includes the Albury and Two Thumb ranges, which separate the dryland basin from the eastern region (Figure 1).As a whole, the Mackenzie District is expected to face changing pressures caused by climate change, including elevated temperatures and decreased rainfall (Tait et al. 2016; Supplementary Table S2).Climate change will present a range of challenges to the district, and a corresponding variety of new and continued land uses that could be applied to aid adaptation (Abbott et al. 2019).

Land use and climate scenarios
To spatially represent the Mackenzie District we used a land cover typology based on the New Zealand Land Cover database (Dymond et al. 2017).Additional datasets used to parameterise some models included travel accessibility, the locations of tracks and visitor infrastructure, and topography (Supplementary Table S5).
Climate projections were provided by the New Zealand downscaled version of the NorESM1-M global model projections for the period 2060-2070, under the following representative concentration pathways (RCPs); RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5 (Tait et al. 2016; Ministry for the Environment 2018).We selected the NorESM1-M model arbitrarily from the other global models available, due to computational constraints restricting multi-model runs.NorESM-1 is a moderatelow climate sensitivity model.The climate scenario data are available on a daily timestep for the 10-year period of interest, at a spatial resolution of approximately 0.5 by 0.5 decimal degrees.For all climatic parameters included in the models, we quantified the parameters annually and used the mean of these annual values over the 10 year period.All spatial datasets were resampled to a resolution of 100 m by 100 m for analyses.
To generate and assess future scenarios we integrated a series of models that represent key socioecological processes, including projected climate conditions, land use change, the dynamic response of semi-natural vegetation to changing climate and land management, and the performance of the landscapes in terms of adaptation services (Figure 2).
We generated 1200 scenarios of hypothetical futures by factorially combining three attributes: the climate scenario (four alternatives), the overall likelihood of human land use changes occurring at each land parcel (three alternatives), and the relative frequency of different land use changes (100 alternatives).The four climate scenarios were the representative concentration pathways RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5 (Tait et al. 2016; Ministry for the Environment 2018).The overall likelihood of land use change from the current land use within each parcel represents the overall degree of human land use change in the system, and was modelled as one of three probabilities of conversion (0.1, 0.3, or 0.6) representing relatively conservative, intermediate, and dynamic human land use changes respectively (Supplementary Methods 1).We identified 13 potential alternative land uses in the system (Table 1), and generated an indicator of their relative probability of uptake by drawing a number randomly from a uniform distribution between zero and one.We recalculated the relative probability of uptake for each land use 100 times, to generate 100 alternative iterations of this attribute in which different land use types are most likely to be adopted.

Land use change within parcels
The 13 potential land use conversion options (Table 1) include those that are currently common in the study region, such as sheep production, dairying, and forestry.Other land uses highlight potential new opportunities identified nationally, such as chestnut or truffle.We also included exotic plant species removal and agricultural abandonment options (Table 1).Land use change was simulated

Land cover consequences of land use change
Areas under urban, crop, and production tree plantations were assigned uniform cover with an associated land cover type (Table 1).Under grazed and conservation land uses, land cover change was simulated dynamically depending on dispersal, establishment, and competition between vegetation types (Table 1).
We modelled expected shifts in the spatial distribution of broad vegetation types using the FATE-HD model (Boulangeat et al. 2014

Adaptation services assessment
We applied a suite of models to each scenario to estimate 21 biophysical and socio-ecological processes that contribute to 13 indicators of adaptation services (Table 2).Indicators were selected to include a combination of services currently important in the region that stakeholders might wish to sustain in the face of climate change, as well as some latent services that may become increasingly realised, or novel services that may develop in the future from transformed ecosystems (Table 2; Supplementary Table S5).Adaptation services were modelled using the mean annual values for the period 2060-2070.

Sustained adaptation services
Sustained adaptation services are those that are at present economically, environmentally, or socially important in the Mackenzie District, and which may in the future be supported by persistent or transformed ecosystems (Lavorel et al. 2020).Exotic tree species forestry, sheep production, and dairy farming are key productive industries in Mackenzie District (Fairweather and Swaffield 1995;Brower et al. 2018;Upper Waitaki Shared Vision Forum 2020).In addition, carbon sequestration from forested areas larger than 1 ha that were established since 1989 are currently eligible for including in New Zealand's carbon emissions trading scheme (Leining et al. 2020; Supplementary Table S6).The landscape cultural heritage value of the central Mackenzie Basin sub-region was quantified in relation to established cultural norms that identify with the rugged, tussock grassland nature of the place (Thompson-Carr 2012;Abbott et al. 2019).We generated a grid of viewing locations every 500 m within the Mackenzie Basin and modelled the observable viewshed from each location (Schirpke et al. 2013).We then quantified the average proportion of the viewsheds that were observing tussock grassland or depleted tussock grassland as an indicator of this parameter.
The Mackenzie District holds significant value for outdoor recreation (Thompson-Carr 2012).The relative value for outdoor recreation was mapped using a recreational opportunity spectrum approach (Byczek et al. 2018;Lavorel et al. 2022).The developed index combined protected area status, proximity to visitor infrastructure such as walking tracks and huts, proximity to key tourist hotspots as identified and weighted by their relative number of ratings on TripAdvisor, and relative attractiveness of the landscape viewed from each location scored using a subjective relative attractiveness index (Lavorel et al. 2022; values given in Supplementary Table S6).We generated a grid of viewing locations every 500 m within the Mackenzie Basin and used a 3-dimensional representation of topography and vegetation to model the relative attractiveness score of the observable viewshed from each location (Schirpke et al. 2013).We used a recreational opportunity spectrum approach because this method provides an indicator that is spatially explicit and influenced directly by changes in land cover (Byczek et al. 2018;Lavorel et al. 2022).
Soil erosion has been a significant issue in New Zealand over the past decades, and the importance of soil conservation has therefore been increasingly recognised (Mather 1982;Beattie 2003) The total annual loss of soil due to surficial erosion processes was modelled using the revised universal soil loss equation, parameterised for New Zealand soil and vegetation types and accounting for annual precipitation (Dymond 2010).

Latent adaptation services
Latent services are those that may be currently provided by ecosystems of the Mackenzie District, but their value is not yet appreciated or realised.Such services with direct economic benefits could include mānuka honey production, as mānuka vegetation is present within the region and honey production is substantial elsewhere in the country (Table 1).In addition, carbon sequestration from vegetation not currently captured by New Zealand's carbon emission trading scheme could yield economic returns given future policy changes (Burrows et al. 2018; Supplementary Table S6).
Tree and scrub shelter is important in providing refugia to livestock under extreme weather conditions, including high wind, precipitation, temperature, and solar radiation (Gregory 1995;Hawke and Dodd 2003), conditions which are projected to increase in frequency and magnitude with climate change in future (Thornton et al. 2021).The role of vegetation in providing shelter to livestock was quantified as the proportion of the grazed area that was within or adjacent to an area of scrub or woodland.Trees and scrub within pastoral systems may help mitigate some of the increased thermal discomfort projected to impact livestock as climate change progresses (Zeppetello et al. 2022).
Stormwater-driven flood events are not a common occurrence in the Mackenzie District at present, but the risk of such flood events may increase due to future climate change (Ministry for the Environment 2010).Ecosystems can reduce flood risks by retaining rainfall and thus attenuating high flows.Rainfall retention by ecosystems was modelled using a curve number approach parameterised from a lookup table developed by synthesising stormwater runoff modelling guidelines published for the Auckland, Waikato, and Wellington regions of New Zealand (Auckland Regional Council 1999; Lockyer 2019; Waikato Regional Council 2020).The hydrologic soil group class across the study area was extracted from a global mapping dataset class dataset (Ross et al. 2018).Runoff retention was modelled under a high rainfall event, equivalent to the 99 th percentile daily rainfall conditions calculated over the 10 years of the climate scenario.The proportion of the total rainfall retained across the study region was quantified as the indicator of this parameter.
New Zealand has historically experienced relatively infrequent wildfires, but climate change is projected to increase wildfire risk, particularly in the Mackenzie dryland basin (Melia et al. 2022).The relative fire risk across the study region was quantified as the mean value of a modelled spatially-explicit fire risk index that was designed to integrate a weather index of fire risk, and a standing fuel stock index of fire risk.The weather index of fire risk was the build-up index (BUI) developed by the Canadian Forest Service ( Van Wagner 1987), quantified as the maximum 95 th percentile BUI experienced annually over the 10 projected years of the climate scenarios.The standing fuel stock was mapped across the vegetation cover using lookup tables of reference values collated in a national report (Pearce and Anderson 2008).In some cases the available fuel is influenced by the BUI, so this was cross-referenced when calculating the standing fuel stock (Pearce and Anderson 2008).For grassland and pasture ecosystems, the approximate height of the different vegetation covers was also taken into account when calculating fuel stocks (Pearce and Anderson 2008).The BUI and standing fuel stock indicators were scaled between 0 and 1, in the case of BUI by dividing by 100, and in the case of fuel stock by dividing by the maximum possible fuel stock for any vegetation cover in the scenario.The product of these two scaled indicators was taken as our indicator of fire risk, because it weights equally the relative risk of fire according to weather conditions, and the fuel stock present.The average fire risk indicator score across the study area was taken as our indicator of this parameter.
Pollinators are currently present within Mackenzie District, but play only a limited role in providing pollination services due to the low area of pollination-requiring crops.If future land uses include horticultural production, the benefit provided by pollinators may become more important (Thomas et al. 2020).The potential carrying capacity of beehives was modelled based on climatic and land cover parameters using an existing model, and the pollination of crops by bees was indicated as the proportion of the total production crop area that was serviced by at least one modelled beehive (Ausseil et al. 2018).
The stock of carbon held within soils and vegetation within Mackenzie District may be increasingly appreciated as a contributor to climate change mitigation in future.The total soil and biomass carbon stock was modelled using a nationally-developed lookup table with parameters for most vegetation types (Carswell et al. 2008; Supplementary Table S6), supplemented by an additional study for urban vegetation (Schwendenmann and Mitchell 2014).The lookup table values for each vegetation type were fixed across climate scenarios (Supplementary Table S6).

Novel adaptation services
Novel adaptation services are those that are predominately absent from Mackenzie District at present, but may become economically, environmentally, or socially important in a climate change future.Novel production systems that have previously been highlighted as potential future options and scoped for their potential across New Zealand include chestnut, truffle, onion, pea, and potato (Table 1, Thomas et al. 2020).Suitability for these production systems is modelled depending on soil and climatic conditions specific to the requirements of each product (Thomas et al. 2020).In addition, we consider the importance of habitat heterogeneity in supporting indigenous biodiversity to be a novel indicator in the Mackenzie District.A broad interpretation of biodiversity and its underpinning in different indigenous ecosystems is recognised nationally (Department of Conservation 2020), yet the discourse in Mackenzie District has predominately focused on conservation of tussock grasslands -thus under-representing indigenous forest, scrub, and wetland types that were historically present (McGlone and Moar 1998;Abbott et al. 2019).The potential of the region to support a broader range of indigenous biodiversity was quantified as the Simpson's diversity index of indigenous land cover types within 1 km by 1 km grid cells (Morelli et al. 2013;Yoshioka et al. 2017).This index is a measure of available habitat heterogeneity and has been shown to be a proxy for taxonomic diversity (Morelli et al. 2013;Yoshioka et al. 2017), including in New Zealand (Ewers et al. 2005).

Adaptation success
Based on prevailing societal values in New Zealand, we developed six criteria for adaptation success, representing economic production, climate change mitigation, resilience to climate change-induced risk, human and environmental well-being (Table 3), because successful climate change adaptation must meet the needs of diverse stakeholders (Donatti et al. 2020;Lavorel et al. 2020;Ausseil et al. 2022).We identified six key criteria based on prevailing values rather than requiring successful adaptation to meet thresholds for all indicators included in the study, because enforcing overly strict criteria can constrain the variability in successful options and lead to monofunctional outcomes.The criteria include a measure of economic and livelihood improvement, quantified as a required increase in the total profitability from primary industry and carbon trading due to the importance of agriculture and forestry to New Zealand's economy, and welldeveloped carbon emissions trading system (Cradock-Henry et al. 2019;Leining et al. 2020).We required successful adaptation to profit at least 110% of the baseline economic profit, to ensure that the

Adaptation criteria Threshold value
Economic profit Total profits must be more than 110% of the current value.

Profit resilience
Total profits must be diversified, with less than 50% of the total net profit coming from the dominant product.

Climate change mitigation
The amount of carbon stored across the landscape must increase by at least 25% over the baseline.

Climate risk adaptation
Any change in relative fire risk must be less than 175% of the baseline fire risk.

Landscape cultural heritage
The average tussock view within the Mackenzie Basin must not be less than 66% of the current value.

Biodiversity
The biodiversity habitat indicator must be within the top 66% of all the future landscape scenarios.
successful scenarios bring improved profitability from primary industry (Table 2).We also included a criterion for the reliance of profitability on one sector to be no greater than 50% of the total profit (Table 2), as diversified economic production may be more resilient to unexpected future shocks and pressures (Hopkins et al. 2015;West et al. 2021).Climate change mitigation has been a focus of international commitments and national policy (Hopkins et al. 2015), so we included one criterion to increase the carbon stocks across the district by at least 25% (Table 2).Furthermore, we developed one criterion regarding wildfire risk due to increasing national policies highlighting the importance of climate change adaptation (Ministry for the Environment 2022).The Mackenzie District is an arid region and is one of the hotspots of projected increases in wildfire risk in New Zealand (Melia et al. 2022).As some increase in wildfire risk was projected across almost all scenarios, we set the threshold to limit this increase to no more than 175% of the baseline (Table 2).The human well-being criterion was set in relation to the landscape cultural heritage indicator, which we set a threshold of change of no less than 66% of the baseline value (Table 2).Tussock grassland vistas are associated with the national perception of the Mackenzie Country, and maintaining some of this character is a consideration of future planning (Abbott et al. 2019; Upper Waitaki Shared Vision Forum 2020).The environmental well-being criterion was set in relation to the biodiversity indicator, due to the importance of indigenous biodiversity and continued losses of indigenous ecosystem cover (Cieraad et al. 2015;Norton et al. 2018).Due to widespread land cover changes across the district, we set the biodiversity criteria threshold in comparison to the distribution of scores across the scenarios (Table 2).To assess the sensitivity of the number of successful adaptation scenarios to the selected criteria thresholds, we quantified the number of scenarios meeting each criterion across a continuous range of criteria thresholds determined by the range of possible values (Supplementary Figure 24).

Typology of adaptation services
We characterised the reliance of each scenario on different types of adaptation services according to three criteria: the contributions of persistent or transformed ecosystems, the provision of sustained, latent, or novel services, and the human inputs required for co-production (Lavorel et al. 2020).The relative impact of persistent and transformed ecosystems on the outcomes of the scenario was quantified as the proportion of the land area that was persistent (identical to the baseline land cover), transformed by predominately spontaneous vegetative processes (different to the baseline and derived from the outputs of the FATE-HD modelling), or transformed by direct human conversion (different to the baseline and derived from direct human conversion land use changes).
The relative importance of sustained, latent, and novel services in each scenario was quantified by defining each indicator as one of these categories depending on whether the service is currently acknowledged as important in the study region, currently provided in the study region but not widely considered a key aspect of the system, or largely absent from the study region at present, respectively (Section 2.5, above).For comparison of all 13 adaptation service indicators across scenarios (Figure 2; Table 2), the ranks of each scenario and each indicator were calculated and scaled between zero and one, with one indicating the highest score across all scenarios.Finally, the mean scaled rank was taken for each of the sustained, latent, and novel categories.The novel, latent, and sustained economic production indicators were taken as the sum of the multiple forms of production within each category (Supplementary Methods 3).
The human inputs required for co-production of each scenario were indicated as the total anthropogenic capital inputs required for the scenario, including the costs of maintaining land uses and conversion costs (Supplementary Table S4).We estimated the human capital inputs required to convert and maintain the scenario over the 40 years between 2020 and 2070 (Supplementary Table S4) including the sum of the conversion and irrigation costs, plus 40 years of annual maintenance costs (Howes et al. 2014;Thomas et al. 2020).

Ordination analysis of scenario outcomes
To analyse the set of future possibilities explored through the simulations, we performed a canonical correspondence ordination of the 13 adaptation service outcomes.Ordination methods have previously been used to describe the main axes of variation and extract bundles of co-occurrent services in ecosystem service studies and simulation experiments (Mouchet et al. 2014;Vallet et al. 2018).In addition, canonical correspondence analysis can also quantify the degree to which environmental gradients explain observed variation in services (Mouchet et al. 2014).We analysed the influence of five composite land cover indicators that summarise land cover variation in the simulated scenarios, quantifying the proportional coverage of tussock grasslands ('tussock'), other grassland types ('grassland'), agricultural crops ('crop'), exotic trees and scrub ('exotic woody'), and indigenous trees and scrub ('indigenous woody').These composite landcovers were combinations of the more detailed vegetation types (Supplementary Table S1).All variables were scaled and centred prior to ordination.To quantify bundles of adaptation services that were commonly provided simultaneously in scenarios, we used k-means clustering on the first two axes extracted from the canonical correspondence analysis to extract four clusters.The number of clusters was selected based on visual selection (Supplementary Figure 21).

Spatial variation in adaptation services
To map spatial variation in the contributions of different types of adaptation services across the Mackenzie District, we summarised the maps from the successful adaptation scenarios by taking their mean indicator values.For the persistence and transformation of land cover indicators, the spatial indicator was thus the proportion of the successful scenarios in which the land cover was either persistent, transformed by predominately vegetative processes, or transformed by direct human conversion.For the human capital input, the spatial indicator was the average human capital input in New Zealand Dollars per hectare.For the latent, sustained, and novel service contributions, the spatial indicator was the mean of the latent, sustained, and novel indicator scores.The latent, sustained, and novel indicator scores were the mean of the ranked and rescaled indicator scores for the indicators contributing to these three types of adaptation, as described above.These scores are therefore scaled between zero and one, with a higher score indicating a greater relative contribution from latent, sustained, or novel services.

Results
Of the 1200 future scenarios, 541 (46%) met the criteria for successful adaptation (Figure 3(a); Supplementary Table S7).The number of successful scenarios was greatest under RCP 6.0, followed by RCP4.5, RCP 2.6, and RCP 8.5 (Supplementary Table S7).The number of successful scenarios was greatest under the intermediate (0.3) probability, followed by the higher (0.6) probability of land use change (Supplementary Table S7).Most of the scenarios that did not meet the adaptation criteria (90%) fell short on only one criterion, with most of the remaining scenarios failing to meet two criteria (9.5%).The threshold which the largest number of scenarios did not meet was the biodiversity criterion (33% did not meet the threshold), followed by carbon stock (22% did not meet the threshold).The criteria threshold sensitivity analysis indicated nonlinear sensitivities for most criteria (Supplementary Figure 24).Most scenarios improved performance in comparison to the baseline for livestock shelter, beehive density, pollination, soil erosion reduction, carbon stock, runoff retention, landscape heritage, latent, and novel economic production indicators (Supplementary Table S8).Conversely, most scenarios declined in performance compared to the baseline for fire risk, biodiversity, relative recreation score, and sustained economic production (Supplementary Table S8).Successful adaptation scenarios had significantly greater coverage of tussock grassland and indigenous woody vegetation, and significantly lower coverage of crops, exotic woody vegetation, and grassland, than unsuccessful scenarios (Figure 3(a), Supplementary Figure 22).The area of grassland and cropland was lower in most scenarios than in the baseline, while the area of tussock grassland, exotic woody vegetation, and indigenous woody vegetation was higher in most scenarios as compared to the baseline (Supplementary Figure 22).
Ordination of the scenario performance indicators and land cover highlighted their correlations (Figure 3(a,b)).The first canonical correspondence axis (CC1) accounted for 36% of the variance explained, and described a gradient strongly driven by the relative coverage of exotic woody vegetation (largely exotic forest but also including gorse and broom shrubland, and deciduous exotic woodland) compared to other vegetation types (Figure 3(a)).The second canonical correspondence axis (CC2) was not strongly associated with variation in land cover but showed variation in adaptation service indicators.The land cover gradient influenced the performance indicators (Figure 3(b)), among which we identified four main bundles (Supplementary Figure 21).The first bundle of adaptation services included the biodiversity, recreation, and tussock visibility indicators, and was supported in scenarios with lower exotic woody cover and greater cover of grassland, tussock grasslands, crops, and indigenous woody vegetation (Figure 3(a)).The second bundle of adaptation services included the livestock shelter, fire risk reduction, carbon stock, and hive density indicators, and was supported by greater coverage of exotic woody vegetation (Figure 3(a,  b)).The third bundle of adaptation services included the pollination, runoff retention, and profitability from latent, novel, and sustained service indicators, and was supported in scenarios with intermediate exotic woody vegetation and other land covers (Figure 3(a,b)).The final bundle of adaptation services included only the soil erosion protection indicator and was supported in scenarios with high exotic woody vegetation cover (Figure 3  (a,b)).Due to the strong associations between exotic woody vegetation cover and the adaptation services, we include a plot comparing each indicator across a gradient of exotic woody vegetation cover (Supplementary Figure 23).
The scenarios were analysed according to three characteristics of adaptation services (Figure 3(c-e)).Scenarios with more positive CC1 scores had larger areas transformed by direct human conversiontypically the planting of exotic forestry plantations (Figure 3(c)).Scenarios with more negative CC1 scores had greater coverage of persistent ecosystems, commonly tussock and indigenous scrubland (Figure 3(c)).Scenarios with intermediate CC1 scores often had higher coverage of ecosystems that had transformed largely by spontaneous vegetation dynamics; these could be either indigenous or exotic, depending on the location and level of management influence through exotic species removal or extensive grazing (Figure 3(c)).Other scenarios with intermediate CC1 scores had a balance of transformed and persistent vegetation cover (Figure 3(c)).Consequently, scenarios with more positive CC1 scores were overall associated with higher human capital inputs, and scenarios with more negative CC1 scores typically had lower inputs of human capital.However, some scenarios requiring high human capital inputs were found across the gradient (Figure 3(d)).
The relative contribution of novel, latent, and sustained adaptation services was variable across the canonical correspondence axes (Figure 3(e)).In general, scenarios with higher performance for latent indicators were associated with more positive CC1 scores (e.g.exotic forest plantations) and more negative CC2 scores (Figure 3(e); Supplementary Figure 25).Scenarios with higher performance for sustained indicators were associated with more positive CC2 scores.Scenarios with higher performance for novel indicators were associated with more negative CC1 and CC2 scores (Figure 3(e); Supplementary Figure 25).
The ordination space of scenarios was influenced by the input parameters, namely the probability of land use change and climate projection used to generate the scenarios (Figure 3(f); larger version shown in Supplementary Figure 26).Scenarios that had a greater probability of change were associated with greater human capital inputs, direct human conversion of land cover, and more extensive exotic woody vegetation cover (Figure 3).Spatial mapping of average adaptation service scores across the successful adaptation scenarios highlighted some frequently occurring spatial patterns (Figure 4).The flatter parts of the dryland basin and eastern region commonly supported land cover persistence and transformation by direct human conversion (Figure 4(a,b)), with some areas of current low producing grassland transformed by vegetative change into tussock grassland or exotic woody cover (Figure 4(c); Supplementary Figure 22).In the mountain zone, spontaneous vegetation cover transformation was most common in current areas of bare ground, grassland and tussock, particularly at higher elevation where climatic habitat suitability is projected to change, and where the opportunities for human-dominated direct land use conversion are low (Figure 4  with greater contributions in parts of the dryland basin and eastern region which are most suitable for novel agricultural crops, and in mountainous regions with heterogeneous environmental conditions, where different native ecosystems were more likely to be found adjacent to each other (Figure 4(g)).The general spatial patterns observed across all successful adaptation scenarios were similar when analysing the successful scenarios for each of the four representative concentration pathways separately (Supplementary Figures 27-30).

Adaptation in Mackenzie District
Our adaptation simulation approach explored a broad range of land use and climate scenarios for the Mackenzie District.Variation in the potential outcomes was driven largely by land use change and vegetation dynamics, in combination with variation across climate scenarios.Projected changes in climatic conditions directly influenced some adaptation services; for example the fire risk indicator increased in almost all scenarios due to reduced precipitation and increases in temperature (Supplementary Table S2).However, there were also indirect climatic effects on the suitability for some crops (Supplementary Figures S13:20), and vegetation types (Supplementary Figures S1:8), which impacted the frequency with which these land covers were selected or established through vegetation succession respectively.
The exotic woody vegetation cover was a major driver of the adaptation service outcome, explaining much of the variation in realised adaptation services (Figure 3; Supplementary Figure S23).Exotic woody vegetation cover was greater in all modelled scenarios than in the baseline, and established over large areas of the landscape in some scenarios because it can arise through both intentional human planting for exotic tree forestry (West et al. 2020), and unintentional spontaneous establishment under grazed and ungrazed land use management (Mason et al. 2017(Mason et al. , 2021)).Furthermore, projected climate change resulted in more suitable conditions for gorse scrub and exotic forest in the mountain region, although habitat suitability was reduced, particularly for exotic forest, in the eastern and dryland basin regions (Supplementary Figure S3; Supplementary Figure S5).
Exotic woody vegetation had a large influence on the adaptation services provided, with positive impacts on our indicators of carbon stock and soil erosion prevention, and negative impacts on our indicators of recreation, landscape heritage, and biodiversity value (Figure 3(b)).Exotic woody vegetation thus presents potential opportunities for adaptation in the Mackenzie District but also brings trade-offs.Care must be taken when establishing exotic trees for carbon sequestration or production forestry purposes (Dymond et al. 2012)practices which are currently encouraged by the structure of New Zealand's carbon emissions trading scheme (Leining et al. 2020).Furthermore, exotic woody species such as conifers (Mason et al. 2021) and grazing-resistant shrubs (Lee et al. 1986) can rapidly invade native and semi-natural vegetation types.Our simulation projected widespread invasion by exotic trees, and localised invasion by gorse scrubland, whether grazing is maintained or not (Supplementary Figures 9-12).To manage the spread of exotic woody vegetation and protect adaptation services such as recreation, landscape heritage, and biodiversity value, active destruction or removal of plants must be carried out (Mason et al. 2021).This example demonstrates that human inputs and management are critical in supporting a range of adaptation services (Lavorel et al. 2020).
Despite the overall tendency for the Mackenzie District to develop exotic woody vegetation, other land covers also contributed strongly to influencing the adaptation service outcomes.Scenarios with larger areas of grazed land and crop cover commonly performed well for pollination, fire risk, and economic performance, but required greater human capital inputs (Figure 3; Supplementary Figure 25).Exotic species removal protected existing depleted and tall tussock grasslands from invasion and supported range increases (Supplementary Figure 22); their range increase was however limited over the timescale of the simulation by the slow maturation and expansion rates of these species.Exotic plant species removal also resulted in increased areas of native scrubland -particularly mānuka and matagouri scrub 22).If it is desired to further increase the area of tussock grasslands, it may be necessary to take a more active role in addition to the land abandonment, extensive grazing, and exotic plant species removal management options simulated in this study; for example by managing other introduced grazing mammals such as rabbits (Meurk et al. 2002) or seeding and maintaining tussock species (Norton et al. 2006).Scenarios with more extensive coverage of native vegetation types commonly scored highly for the biodiversity, landscape cultural heritage, and recreation indicators (Figure 3).Some indicators recorded future declines compared to the current baseline in some scenarios (Supplementary Table S8).Almost all future scenarios performed less well than the baseline for fire risk reduction, due to the change in modelled weather conditions (Melia et al. 2022).While some increase in weather wildfire risk is likely inevitable, stakeholders in the most fire-prone regions may manage for plants with lower potential fuel loads or flammability, including where possible indigenous forest (Pearce and Anderson 2008;Mason et al. 2016;Wyse et al. 2016), to mitigate future increases in risk.In addition to the differences in vegetation type fuel load compared by our wildfire risk indicator, there are further opportunities for land management to mitigate or reduce wildfire risk; for example through vegetation thinning or prescribed burning to manage connectivity (Halofsky and Peterson 2016;Francos and Úbeda 2021).

Diverse outcomes are facilitated by heterogeneity
Adaptation services provide diverse opportunities for the Mackenzie District to adapt to future environmental change.Each successful adaptation scenario depends to a greater or lesser extent on different adaptation services, and the suite of adaptation services provided may rely more heavily on sustained, persistent, or novel services (Colloff et al. 2020;Lavorel et al. 2020).While we found clear bundling of individual adaptation services (Figure 3), there were only relatively weak associations between the scores for the relative contributions of sustained, latent, and novel services (Supplementary Figure 25).This indicates that there are not necessarily trade-offs between these different types of adaptation, and that a combination of sustained, latent, and novel services can support adaptation success.One mechanism through which the landscapes can provide multiple types of adaptation service concurrently is spatial heterogeneity, as different types of service can be provided in different parts of the landscape (van der Plas et al. 2019).The Mackenzie District is topographically and environmentally heterogeneous, with some areas more suitable for some land uses, and less suitable (or completely unsuitable) for others (Brower et al. 2018;Abbott et al. 2019).Spatial diversity in native vegetation types may arise naturally if indigenous vegetation is encouraged in areas with varied environmental conditions; historically, tussock grasslands and wetlands were present in flatter areas, with indigenous scrub and forest on valley slopes (McGlone and Moar 1998).Human interventions can also create spatial variation in environmental conditions, for example by varying grazing regimes, shade, and management of exotic plants and animals to increase diversity among plant communities (Norton et al. 2006;Norton and Young 2016).The spatial heterogeneity in capacity and suitability to provide adaptation services, resulted in the observed spatially consistent patterns in different types of adaptation services (Figure 4).

Simulating nature-based adaptation requires increasingly complex models
It is becoming increasingly important to understand how adaptation services may contribute to solving the challenge of climate change.This is inherently a forward-looking, exploratory field of research, so requires simulation approaches to scope and quantify the performance of hypothetical future scenarios (Herrero et al. 2014;Kwakkel et al. 2015).To gain insights into likely pathways to adaptation, it is necessary to characterise the different types of adaptation services and quantify their contributions; for example using the typology developed by Lavorel et al. (2020).Our study systematically applied this typology to analyse potential future scenarios, revealing the diverse and complementary opportunities for adaptation through different mechanisms.Future applications of the typology to different landscapes are needed to establish whether patterns observed for Mackenzie District are generalisable.Key adaptation services and their bundling are likely to be contextdependent, with expected climate change impacts on precipitation and temperature, the local history of land use and ecosystem service extraction determining how services are classified.For example, we classified runoff retention as a latent adaptation service in this case due to the relatively low historical and current risk of stormwater-driven flooding, whereas runoff retention may be classified as a sustained service in regions that already recognise these benefits (Barral et al. 2015).Similarly, here the human capital input was strongly correlated with the degree of human-driven direct land cover transformation and latent ecosystem services, which is a result of the relatively higher costs of direct conversion to crop, dairy, forestry, or production tree systems, and the relative contributions of latent production goods and carbon sequestration (Table 1, Supplementary Table S3).These patterns may therefore not hold in regions for which adaptation services have different human input requirements, impacts on land cover, and contributions towards adaptation.

From exploratory simulation to co-developing pathways
Our study explored a broad range of opportunities for climate change adaptation in the Mackenzie District, providing a diverse range of options that could lead to successful adaptation.To reach application, this diverse range of options must be further refined, requiring in-depth engagement with the community and governing entities, creative inputs from landscape design, and further iterative simulation to provide feedback on the impacts of proposed decision on multiple adaptation services and functions (Bhave et al. 2018).In particular, participatory methods are needed to co-develop pathways towards adaptation (Cradock-Henry et al. 2020;Bruley et al. 2021;Werners et al. 2021).Participation will be required from critical partners in governance, including institutions of the New Zealand Government, and representatives of indigenous mana whenua, principally the Māori iwi Ngāi Tahu (Thompson-Carr 2012; Magallanes 2021).In addition, engagement from primary (agriculture, aquaculture, forestry, irrigation), tourism and hydropower industries, and representatives from environmental and recreation interest groups (fishing, hunting, outdoor sports), and from residents will be critical (Upper Waitaki Shared Vision Forum 2020).This participatory work must first identify the performance criteria that are prioritised by the community, as well as the potential land uses that could feasibly be implemented in the district, are acceptable to at least some of the stakeholders, and could impact adaptation services.Scoping of potential land uses could include novel, sustained, and latent crops and rotation systems (Gardner et al. 2021), or novel spatial configurations of existing production systems, such as agroforestry (Paul et al. 2017).Next, proposed pathways must be co-developed with participants, accommodating the demands of different groups (Cradock-Henry et al. 2020;Werners et al. 2021).These proposed pathways may then be spatially represented and their adaptation service performance modelled, to provide feedback for future iteration and improvement by the co-development group (Nay et al. 2014;Bhave et al. 2018).

Limitations
The classification of adaptation services into the Lavorel et al. (2020) typology requires subjectivity, depending on individual interpretation of the 'latent', 'sustained', or 'novel' nature of each service.We assessed this characteristic of the services following our understanding of the context and values in the Mackenzie District and New Zealand, but different interpretation of the history and future potential of each adaptation service in the region may alter the classification, as would the use of alternative value scenarios (Pereira et al. 2020).The definition of 'persistence' or 'transformation' of an ecosystem can be more objectively quantified, although depends on the breadth of the vegetation types that are defined and interpretation of ecological change or stasis -which can be complex when considering natural dynamic ecosystems (White and Walker 1997).Similarly, co-production by people is complex and multifaceted, involving a mixture of social, human, financial and technological capital (Palomo 2016), while in this study we chose a relatively narrow economic indicator.The relative contributions of different types of adaptation services to adaptation success may be altered using a different definition of these characteristics of the services.
Our study was limited in considering only 13 potential land uses for the study system, while there are a broader range of potential land uses that may be relevant.For economic production there are many other potential sources of income that would further broaden the range of opportunities available; including deer and camelid production (Alonso 2009;Vink and Kittelberger 2013), other crops currently grown in other parts of the world (Merlos and Hijmans 2022), and diversification out of the primary sector to tourism or recreation income streams (Fountain et al. 2020).Each of these potential land uses will have different economic, environmental, and societal costs, benefits, and risks.Our study used only one set of climate scenarios, while projections using other climate models may give slightly different outcomes (Tait et al. 2016).
Landscapes are complex, so it is important that models of landscape change can represent key attributes of complex systems (Bennett 2017); including emergent phenomena, path dependency, and reciprocity between model sub-systems (O'Sullivan 2004).Our model representation of the Mackenzie District exhibits some of these characteristics to some extent, due to the interplay between the climate projections, vegetation dynamics model, and probabilistic model of land use change.For example, the emergence of exotic woody vegetation as a dominant land cover in many scenarios is due to the interplay between the vegetation model and assignment of land uses.While our results show only one time point, the establishment of exotic woody vegetation is in reality a path dependent phenomenon -making it difficult to for native vegetation cover to naturally re-establish, and constraining the options for alternative land uses (Peltzer 2018).However, our current model does not include many reciprocal impacts of one model sub-system on another, meaning that feedback relationships (such as changes in land use or management that respond to changing environmental conditions) are generally not represented.Future development of complex systems models to simulate landscape adaptation may incorporate such feedback mechanisms, for example by using agent-based modelling to better represent human decision making and its evolution in response to changing conditions over time (Pan et al. 2019;Villamor et al. 2022).Future refinement could also better represent subtle and potentially non-linear impacts, such as the effect that invasion of exotic woody vegetation on pasture may have on pasture production and livestock performance (Hawke 1991).

Conclusion
Upland landscapes will benefit from a broad range of adaptation services as they respond to future climate change, providing a challenging array of opportunities for future planning.Our model representation of the Mackenzie District allowed us to systematically explore a range of opportunities, highlighting possible options and the tendency of the system to develop towards a particular trajectory.Applying a typology of adaptation services allowed us to compare the relative contributions of different types of nature-based adaptation to climate and land use pressures and in different parts of the landscape.The simulation approach demonstrated here is broadly applicable, and may be implemented in other places to inform societal adaptation to environmental change.

Figure 1 .
Figure 1.(a) Land cover in the Mackenzie District in 2018, and three topographically distinct regions.(b) Photograph showing the flat character and tussock grasslands of the Mackenzie country dryland basin, with the Southern Alp range in background.(c) Photograph taken in the eastern region showing rolling hills with crop and livestock agriculture.Photographs by Philip Capper and Francis Vallance, used under CC by 2.0 license.Full details of photographs can be found in supplementary information.

Figure 2 .
Figure 2. Modelling workflow for land use, land cover, and adaptation service indicator simulation.

Figure 3 .
Figure 3. Canonical correspondence analysis of 1200 hypothetical future land use and climate change scenarios for Mackenzie District.(a) Scenarios plotted by their scores for the first two canonical correspondence axes, according to adaptation criteria success.(b) Loading scores for the 13 variables included in the ordination; Shelter = shelter for livestock, Fire = relative reduction in fire risk, Hives = number of beehives supported, Pollination = crops receiving pollination, Soil = relative reduction in soil erosion, Carbon = carbon stocks, Retention = runoff retention, Recreation = recreation opportunity spectrum score, Biodiversity = habitat diversity index, Tussock = tussock view index, LEcon = profit from latent primary production and carbon sequestration, SEcon = profit from sustained primary production and sequestration, NEcon = profit from novel primary production.Text colour indicates cluster grouping (c) the relative coverage of land cover types resulting from persistence (P), transformation by predominately vegetative processes (V) and transformation by direct human conversion (H).(d) the human capital input over approximately 40 years required in each of the scenarios.(e) the relative ranking of each scenario in providing novel (N), latent (L), and sustained (S) contributions to people, visualised across a ternary colour space with each ranking scaled between the minimum and maximum ranks.(f) Parameters used to generate each scenario, including background climate scenario and approximate proportion of land cover change.Larger version available as Supplementary Figure 26.
(c)).Direct human conversion in the mountain region (Figure 4(b)) was mainly driven by cases of plantation forestry.Higher human capital inputs were more common in the flatter parts of the dryland basin and eastern regions that are more suited to agricultural production, mirroring the areas where direct human conversion of land cover was more likely (Figure 4(b)).Latent adaptation services made on average greater contributions in the eastern region and parts of the dryland basin (Figure 4(e)).Sustained adaptation services made on average greater contributions in the dryland basin, eastern region, and the lower elevation parts of the mountain region -such as the Albury and Two Thumb ranges (Figure 4(e)).Novel adaptation services made spatially scattered contributions,

Figure 4 .
Figure 4. Spatial patterns in adaptation services averaged over 541 successful adaptation scenarios.(a) the proportion of the successful adaptation scenarios in which vegetation cover was persistent from the baseline.(b) the proportion of the successful adaptation scenarios in which vegetation cover was transformed by direct human conversion.(c) the proportion of the successful adaptation scenarios in which vegetation cover was transformed by predominately vegetative processes.(d) the average human capital input across the successful adaptation scenarios.(e) the average scaled score for latent adaptation services across the successful adaptation scenarios.(f) the average scaled score for sustained adaptation services across the successful adaptation scenarios.(g) the average scaled score for novel adaptation services across the successful adaptation scenarios.Black lines indicate the three sub-regions described in Figure 1.

Table 1 .
Land use options for Mackenzie District, factors influencing suitability for these land covers, and methods of conversion between land use and land cover map.
Mason et al. 2017)s.extensiveexotic plant species control (modelled as removal of exotic forest and gorse vegetation cover types;Mason et al. 2017).Each run of the model assumed that the management interventions were established uniformly across the study area.Full details on the implementation of FATE-HD is included in Supplementary Methods 2, maps of the projected vegetation cover under each combination can be viewed in Supplementary Figures 9-12.

Table 2 .
Modelled indicators, their contributions to adaptation services, and incorporation in calculating six criteria for adaptation success.

Table 3 .
Adaptation criteria and threshold values for successful adaptation.