To supply growing global demand, agricultural systems must produce 60% more food by 2050 relative to 2005–2007 levels (Food Agriculture Organisation of the United Nations, 2015). Much of this production shortfall must be met by smallholder farmers (Herrero et al., 2017), while adapting to the largely negative impacts of climate change and variability, using resources more efficiently, and reducing greenhouse gas emissions (Keating et al., 2014; Lipper et al., 2014; Steffen et al., 2015). Yet livelihoods based on smallholder agriculture and fisheries in developing countries and regions are amongst the most vulnerable to climate and other drivers of change, due to low adaptive capacity and persistent poverty (Carter and Barrett, 2006; Ensor, 2011; Porter et al., 2014). Recent evidence indicates that the necessary transformations in food production are not occurring amongst these smallholders (Thornton et al., 2018), and this may be reflected in rising levels of undernourishment in many regions (Food Agriculture Organisation of the United Nations, 2018).

For decades it has been assumed that to achieve rapid and widespread changes in production amongst smallholders, science-driven technologies must be “scaled-out,” defined as delivering more benefits to more people over wider geographical areas, more quickly, equitably and lastingly (Franzel et al., 2001). This paradigm has been referred to as the linear “pipeline” model of technology innovation and diffusion (Biggs, 2007). However, there are growing concerns that blanket recommendations lead to low adoption amongst smallholders, and hence there are no “silver bullets” that can generate transformations in climate adaptation and resource efficiency (Giller et al., 2009; Biermann et al., 2012; Descheemaeker et al., 2016).

Instead, alternative approaches to achieving necessary transformation have been suggested. It has been argued that food production is a component of complex, adaptive social-ecological systems (Walker et al., 2010; Reed et al., 2013; Foran et al., 2014), and therefore multiple technological and social interventions are required simultaneously at different scales to generate systemic change (Giller et al., 2009; Scherr et al., 2012; Food Agriculture Organisation of the United Nations, 2013; Steenwerth et al., 2014; Westermann et al., 2018). In parallel, there has been a shift from dependence on the pipeline model to an “actor innovation” model, which deliberately aims to change social and power dynamics, and institutions and governance that are shaped by these relationships (Biggs, 2007). Consequently approaches based on collaborative learning, which integrate stakeholders' knowledge to stimulate social innovation and action (Kristjanson et al., 2014; Ensor and Harvey, 2015), and hence induce practice change in policy and institutional mechanisms (Gillespie et al., 2015; Dinesh et al., 2018) are gaining in popularity, and sometimes referred to as “scaling-up.” Riddell and Moore (2015) go further and advocate that for durable change to occur in complex systems, “scaling-deep” is also required, which transforms values and cultural norms by tackling the root causes of problems. Nonetheless, even when taking a participatory, systems-based perspective that engenders scaling-up and scaling-deep, the question remains to what extent the solutions formulated by such nuanced approaches can be scaled-out to generate widespread impact.

Islands provide an extreme challenge for transforming livelihoods and food production. By nature they are surrounded by sea and often isolated, and therefore frequently remote from markets and political centres; transport and other logistical costs are high; and natural resources, infrastructure and livelihoods are particularly exposed to both climatic and other natural hazards (Pelling and Uitto, 2001; Tompkins and Adger, 2002; Bunce et al., 2009). Consequently, many islands are highly vulnerable to damage and loss from climate change and related sea level rise, and their sustainable development is severely compromised [Climate Development Knowledge Network, 2014; UN-OHRLLS (Office of the High Representative for the Least Developed Countries, Landlocked Developing Countries and Small Island Developing States), 2015]. This is especially the case for the tropical island archipelagos of the Asia-Pacific nations (Hoegh-Guldberg et al., 2009; Asian Development Bank, 2011; Foale et al., 2013; Butler et al., 2014a,b; Bell et al., 2016), where rising levels of undernourishment are of growing concern (Food Agriculture Organisation of the United Nations, 2018).

Consequently, the necessity for transformation amongst rural communities is paramount in islands of the Asia-Pacific region. Yet the pipeline model of technical innovation and diffusion remains implicit in pan-regional food security programs, such as the Pacific Adaptation to Climate Change Program (2018), and the Coral Triangle Initiative on Coral Reefs, Fisheries and Food Security (2018), while systems-based approaches to planning and implementing adaptation and transformation of livelihoods and food systems are not common practice (Butler et al., 2015).

In this paper we assess the feasibility of scaling-out adaptation strategies formulated through a systems-based approach which exemplified scaling-up and scaling-deep. The study focussed on two island provinces in the Asia-Pacific region: Nusa Tenggara Barat in Indonesia, and West New Britain in Papua New Guinea. We compare the provinces' social-ecological characteristics, and the resulting adaptation strategies developed by a participatory, multi-stakeholder planning process carried out in 2011–2014. Our results highlight the limitations to scaling-out in island social-ecological systems due to the heterogeneity of socio-cultural contexts, combined with steep climate gradients and diverse resource use. Instead, we argue for the need to catalyse transformation by scaling-up iterative social learning and capacity building into existing community development decision-making, and reconfiguration of donor funding mechanisms to support this process.

In this study we applied a participatory, systems-based approach combined with a sub-district resource use typology to analyse rural livelihoods and food production, current and future climate and other pressures, in order to design “no regrets” strategies which would potentially generate systemic change in the case study sub-districts. The process was replicated identically in all case study sub-districts. Fifty-six strategies were formulated for 31 villages in NTB and WNB. We tested two assumptions about the feasibility of scaling these out across the islands.

First, we assumed that our adaptation planning process would yield similar adaptation strategies for the villages concerned. Clearly this assumption was not tenable. A diverse range of strategies were formulated, and the highest priority strategies included social interventions, natural resource management, training and capacity building, diversification of income and construction of infrastructure. Only 16% of these related directly to food production (although some training was aimed at farmers), and only one directly targeted climate impacts. The similarity between the villages was very low, although there appeared to be greater similarity in WNB than in NTB. This may have been a function of the smaller number of villages assessed in WNB (10) than in NTB (21), and the fact that both sub-district case studies in WNB were coastal, while two of the five in NTB were in-land. There were some clusters of higher similarities amongst villages in both WNB and NTB, but these were the exception. The only strategies which were relatively common, and therefore could potentially be scaled-out to some degree were sustainable fisheries management, and improved freshwater supplies and sanitation, road and irrigation infrastructure. However, fisheries-related interventions would only be relevant to coastal villages where livelihoods involved fishing, fish-processing or aquaculture.

The second assumption was that in the expectation that province-wide scaling-out was not possible, the sub-district resource use typologies in NTB and WNB would provide a basis for scaling out amongst villages within them. This was underpinned by a further implicit assumption that villages in each type would have similar agricultural or other natural resource use regimes, and pressures from climate change and population growth on the resources would be locally-specific. This assumption was partially tenable. The similarities amongst the sub-district types were double those of all villages combined, but still low, and only Bali-Witu sub-district in WNB had an SSI approaching 0.5. It may be relevant that this sub-district was a group of off-shore islands, and consequently livelihoods, climate change and population growth impacts may have been more uniform in this discrete location. Importantly, there was little similarity between the two pairs of sub-districts in the same types in NTB, suggesting that scaling-out across sub-districts within the same type may not always be appropriate. This finding was reinforced by the low IndVal results, which indicated that while most strategies were unique to one type, they were usually selected by only one village each, and were therefore not necessarily suitable for all villages in a type.

These results suggest that, perhaps with the exception of smaller off-shore islands, scaling-out strategies produced by a systems-based planning process is not feasible, even amongst villages within sub-district types. The primary explanation is that by analysing livelihoods holistically, and therefore considering multiple drivers of change, including both climate change and human population pressures, plus the diversity of socio-cultural issues, highly variable responses are generated at the village level. The socio-cultural heterogeneity amongst rural communities in NTB and WNB was marked, with an overall Islamic culture in NTB overlaying three ethnic groups plus other immigrant entities, and in WNB a Christian culture overlaying animism and 25 language groups, resulting in likely variations in values assigned to ecosystem services, livelihood priorities and decision-making processes. In addition, population densities varied across the islands, from 86 people per km² in Sumbawa to 671 people per km² in Lombok, and up to 188 people per km² in WNB's Bali-Witu islands. For small islands in PNG, such as Bali-Witu, it has been estimated that at current levels of development, there is a threshold population density of 120 people per km², above which food and water security declines (Butler et al., 2014a). Likewise, population growth rates varied considerably between the provinces, with 1.2% per annum in NTB, and the higher rate of 2.5% per annum in WNB may lead to a doubling of the population by 2040. Furthermore, this variability in local contexts was exacerbated by the range of participants attending the workshops, who in NTB had differing “knowledge cultures” which influenced their perceptions and priorities (Bohensky et al., 2016).

Overlaying the socio-cultural diversity are the steep climate gradients typical of the islands, illustrated by the down-scaled climate simulations, plus the range of natural resources available for utilisation. During the workshops the climate projections were integrated with population projections to identify the ecosystem goods and services most likely to be impacted at the sub-district and village scale (Skewes et al., 2016). Consequently, even within sub-district types, participants could identify markedly different responses to mitigate impacts from one village to the next. These results therefore emphasise the utility of downscaled climate simulations for informing adaptation planning across island archipelagos in the Asia-Pacific region. However, there may be some scales and topographies for which climate gradients are less tangible, for example small off-shore islands, and particularly those that are low-lying (e.g., coral atolls) and therefore do not generate orographic effects (McGregor et al., 2016), such as Bali-Witu in WNB.

To ameliorate the diversity of strategies and the resulting impediment to scaling-out, we pooled the strategies using Biagini et al.'s (2014) classes. The result was higher levels of similarity across all villages, and within sub-district types, but wide variation still remained, further undermining the tenability our assumptions. Furthermore, the generalisation of the strategies would obscure their intentional specificity to local contexts. It is notable from this analysis, however, that strategies predominately fell into the practice and behaviour class, as did many priority strategies, and there was only one technological strategy. Also, none of the strategies were classified as policy, information, warning or observing system, or financing. This reveals participants' focus on social and institutional issues within their systems, and the need to address these rather than introduce technological innovations, either for agriculture or other livelihood activities. However, whether the practice and behaviour interventions will generate the necessary systemic change is questionable: in NTB's Janapria case study the majority of strategies addressed the symptoms of vulnerability, rather than their root causes (Butler et al., 2016a). For change to occur at the scale and degree required to transform livelihoods and food systems, identifying and tackling the underlying institutional and political drivers of vulnerability is paramount (Lemos et al., 2007; Pelling, 2011; Rodima-Taylor et al., 2012).

To address the effects of social-ecological heterogeneity, we developed resource use typologies in NTB and WNB. In sub-Saharan Africa, failure of the pipeline model of technological innovation and diffusion has been attributed to the diversity of agro-ecological, cultural, economic and institutional factors, plus the geographically-specific manifestations of climate change (Descheemaeker et al., 2016). Tools to overcome these constraints have included the definition of local social-ecological “niches” (Ojiem et al., 2006; Franke et al., 2014), and participatory farming systems analysis (Descheemaeker et al., 2016). Our typologies aimed to address this challenge by assessing ecosystem goods and services utilised by communities at the sub-district scale; however, unlike Ojiem et al.'s (2006) niches we could not account for the intricate socio-cultural dimensions of livelihoods overlaying the resource base due to a lack of suitable information (Rochester et al., 2016). Another shortcoming was that due to the diversity of the resource use patterns in NTB the typology captured only 42% of the variation in that province's data, compared to 85% in WNB, which potentially undermined the typology's predictive power and utility for scaling-out.

It is arguable that few technological interventions were identified, particularly for food production, because participants were not aware of available options and innovations. Although the workshops included sources of technical expertise (e.g., local government agriculture and fisheries agencies, civil society organisations), and their knowledge (Butler et al., 2015; Bohensky et al., 2016), it is possible that they were not aware of contemporary techniques and solutions, either. However, prior workshops were held at the provincial level amongst higher-scale stakeholders (e.g., national government, universities) which assessed the sub-district types using the same livelihood adaptation pathways process. With less awareness of local sub-district and village contexts, their strategies tended to emphasise scientific and technically-based strategies (Butler et al., 2016a), some of which were subsequently trialled within case study sub-districts in partnership with local farmers (Butler et al., 2016b; Liu et al., 2016). The strategies from the provincial scale and sub-district case studies were also integrated in a third stage of workshops (Butler et al., 2015), resulting in the inclusion of some technological interventions, but the emphasis on social and behavioural interventions identified by the sub-district and village level participants remained strong (Wise et al., 2016).

It is acknowledged that adaptation and transformation of smallholder's livelihoods and food production requires an integrated approach based on collaborative learning which can integrate stakeholders' knowledge to stimulate social innovation and action (Kristjanson et al., 2014; Ensor and Harvey, 2015), and hence induce practice change in policy and institutional mechanisms (Gillespie et al., 2015; Dinesh et al., 2018), or scaling-up. Riddell and Moore (2015) propose that scaling-deep is also required, whereby values, cultural practices and the quality of relationships are transformed. Thus, a combination of tactics are required that can simultaneously achieve scaling-out, scaling-up and scaling-deep, including analysing root causes of problems, and building networks and partnerships (Riddell and Moore, 2015). In the case of livelihood and food systems in island archipelagos of the Asia-Pacific region, we suggest that NTB and WNB are representative of the heterogeneity caused by diverse socio-cultural, climate and natural resource profiles that are characteristic of these locations (Hoegh-Guldberg et al., 2009; Butler et al., 2014a; Bell et al., 2016). Consequently, even when adopting the necessary systems approach to address localised livelihood challenges and to identify systemic solutions, the feasibility of scaling-out resulting strategies is limited, perhaps with the exception of smaller off-shore islands. In these situations, we suggest that scaling-up and scaling-deep will be of even greater importance than in other contexts.

To this end, it is the mainstreaming of systems-based social learning processes, such as livelihood adaptation pathways, into existing development decision-making cycles that is most likely to lead to the requisite transformation to climate compatible and enhanced food production. This has two implications for government and donor support for initiatives aiming to establish scaling-up and scaling-deep. First, prerequisite capacity building will have to be delivered to all district and sub-district stakeholders within a province, and on a recurring basis for elected officials and other temporary community representatives. Second, our evaluation of the NTB process indicated that while capacity for transformation can be primed amongst communities and stakeholders, realising necessary governance reconfiguration requires consistent support for change agents, particularly in conditions of institutional flux caused by decentralisation, and hence long-term investments (Butler et al., 2016c).

Some important further research is also necessary to inform the design of scaling-out and scaling-deep initiatives. First, the length of time required to instigate institutional change within local planning processes is not clear, and may vary between contexts. In NTB we estimated that at least 5 years of ongoing engagement is necessary (Butler et al., 2016c), while in countries with lower capacity such as PNG this may require 10 years or more (Stone-Jovicich et al., 2015). In addition, knowledge integration and power dynamics in participatory decision-making are key determinants of institutional change, and require greater understanding and engagement by researchers (Tschakert et al., 2016; Butler et al., 2020). Tools and processes which can trigger double- and triple-loop learning, and thus the identification of transformational strategies are also imperative, but the social psychology necessary to design such tools and accurately attribute their impact is nascent (Butler et al., 2017b). Overall, if these research questions are to be answered, evaluation approaches must be employed which can track and record change from a complex systems perspective (Maru et al., 2018).

All datasets generated for this study are included in the article/supplementary files.

This study was carried out in accordance with the recommendations of the Australian National Statement on Ethical Conduct in Human Research, CSIRO Human Research Ethics Committee with verbal consent from all subjects. All subjects gave informed consent in accordance with the Australian National Statement on Ethical Conduct in Human Research (2007). The protocol was approved by the CSIRO Human Research Ethics Committee.

JB conceived, designed and led the study and wrote the paper. WR and TS carried out the statistical analyses. RW, EB, NP, WS, YY, TH, PH, IJ, YS, BM-L, DV, and KD co-designed, prepared, facilitated, and collated data from the workshops. JK and DK prepared downscaled climate projection data.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.