Carbon benefits through agroforestry transitions on unmanaged fallow agricultural land in Hawaiʻi

Abstract

There are growing efforts to incorporate agroforestry into ecosystem service incentive programs. Indigenous and other place-based multi-strata agroforestry systems are important conservation and agricultural strategies, yet their ecosystem services, including carbon sequestration benefits, have received little research attention. To fill this gap, we draw on interviews with agroforestry practitioners and ecosystem service modeling in Hawaiʻi to: (1) create future scenarios of where fallow unmanaged agricultural and non-native dominated conservation lands could be transitioned to multi-strata agroforestry under current and future climates; and (2) quantify the potential above-ground carbon and soil carbon benefits and tradeoffs of transitions across these scenarios. We found that about half of unmanaged fallow agricultural lands, representing >1,500 km² , was suitable for agroforestry transitions under current rainfall and over a third, representing >1,200  km², remained suitable under a dry climate change scenario, RCP 8.5 mid-century. Mean above-ground carbon in modeled agroforestry systems was estimated to be 92–125 Mg C ha^-1 (337–458 Mg CO₂ ha^-1) with ~75%  of the potential restoration area projected to significantly increase above-ground carbon storage. Considering both above-ground and soil carbon, overall carbon benefits are expected across over a third of the potential restoration area with just 5% of the area with expected overall losses. These results provide evidence for potential carbon hotspots for agroforestry transitions, as well as to the need for further study of soil carbon changes with multi-strata agroforestry transitions across varying climates and soil types. With potential carbon sequestration similar to or greater than that of native forest restoration, restoration through agroforestry represents an important pathway to achieving carbon benefits through multi-benefit forest-agricultural systems on large areas of unmanaged agricultural lands, offering a pathway to support inclusive and effective natural climate solutions.

Introduction

There are growing efforts around the world to incorporate agroforestry systems into ecosystem service incentives, such as Payments for Ecosystem Services (PES) focused on carbon, water, biodiversity, and other ecosystem benefits¹. Agroforestry systems incorporate trees and crops or other tended and harvested products, and vary widely from several trees in a field to biodiverse, multi-strata systems of trees, shrubs, and understory plants². Agroforestry, as a land-use option, continues to gain traction in agricultural-focused ecosystem service incentives, such as the United States Department of Agriculture’s Environmental Quality Incentive Program (EQIP), in part because of its contribution to agricultural and biodiversity complexity of working landscapes³. Within forest-based PES programs, agroforestry systems are also increasingly seen as alternatives to removing land from production, which can have adverse social and ecological consequences, including loss of local management and livelihood systems⁴. PES programs in Latin America, for example, have added agroforestry systems as eligible for compensation in an effort to expand access to and benefits of incentive programs for lower-income, smaller landholders and rural communities, while providing numerous ecological benefits^4,5.

While the inclusion of agroforestry into ecosystem incentive programs is more recent, the practices that support agroforestry systems are intergenerational—dating back in some cases millennia. Widespread, Indigenous agroforestry systems have long produced food, medicine, and fiber, increased or maintained tree cover, supported local biodiversity, and contributed to local livelihoods^6–9. These approaches are based on generations of local knowledge, practices, and interests, and can offer important strategies for addressing concerns over justice and effectiveness raised in evaluations of ecosystem incentive programs and restoration efforts more broadly^4,10. Improved equity and justice outcomes occur, in part, because Indigenous and place-based approaches inherently allow for local autonomy in design and placement, have community buy-in, and build on generational ecological and socio-cultural knowledge and values^11,12.

The importance of Indigenous lands, knowledge, and practices has been elevated in the context of biodiversity conservation^13,14, and Indigenous rights are recognized as critical components of effective and equitable climate policy from local to international scales¹⁵. However, the potential of restoration through Indigenous and other place-based multi-strata agroforestry systems to contribute to climate change mitigation through carbon sequestration (and reciprocally the potential ways that carbon ecosystem service incentive programs inclusive of place-based agroforestry could benefit local communities) remains under-studied¹⁶. This is despite acknowledgement of the importance of agroforestry, in its broadest definition, as a natural climate solution^17,18. Barriers to greater perpetuation and expansion of Indigenous and other place-based agroforestry practices include structural factors such as land tenure, labor and financial capital availability, and need for appropriate recognition of Indigenous knowledge and cultural responsibilities¹⁹. A paucity of data on carbon sequestration potential of these systems also presents a barrier to entering into carbon ecosystem incentive programs, which are growing around the world and have the potential to support the expansion of agroforestry²⁰.

The State of Hawaiʻi has committed to carbon neutrality by 2045 (State of Hawaiʻi Act 15), and will be unable to reach this goal without terrestrial carbon sequestration actions in addition to reductions in fossil fuel use²¹. Hawaiian forests have undergone vast environmental degradation, face some of the highest resource management costs in the world, and there is a need to develop reforestation strategies that provide carbon sequestration and other ecosystem services along with economic opportunities²². Across Hawaiʻi, there are many examples of restoration through Indigenous management practices, including agroforestry^12,19, and their expansion offers important, yet under-studied, potential for ecosystem restoration and land-based carbon sequestration¹⁶. As such, Hawaiʻi provides an important case study to evaluate the potential for carbon sequestration of these place-based practices to inform emerging carbon ecosystem incentive programs focused on reforestation. In the current post-plantation, tourism-dependent economy, the decline of large colonial sugar and pineapple plantations alongside the high costs of land and labor facing farmers²³ has left over 40% of agricultural lands as unmanaged fallow, often with degraded soils²⁴. The challenging economics of agriculture in Hawaiʻi continue to perpetuate extremely high reliance on food imports²⁶. Largely dominated by nonnative grasses and other non-native vegetation, unmanaged fallow agricultural lands also create conditions of extreme fire risk, particularly in dry areas²⁷ and are the leading cause of devastating fire events in the islands^27–29.

At the same time, Indigenous and other local community groups and practitioners are beginning to transition these unmanaged lands to place-based multi-strata agroforestry systems to increase local food production while supporting a suite of societal values¹⁹. Multi-strata agroforestry systems, which are forests tended for a suite of reciprocal benefits for people and place, historically played a critical role in food production in Hawaiʻi^9,30. These place-based systems were developed to meet local community needs while operating within environmental constraints and often prioritized the protection of existing native forest patches³¹. Today there is great potential for^9,30 and interest in¹⁹ multi-strata agroforestry across the state’s unmanaged fallow agricultural lands. Multi-strata agroforestry systems are also abundant across Pacific Islands and play critical roles in supporting communities and biodiversity^7,8,32,33, yet they remain understudied. Given strong contemporary interest in land care through multi-strata agroforestry systems and, where possible, supporting climate mitigation through these systems, the potential of multi-strata agroforestry to sequester carbon is an important knowledge gap.

In this study, we describe the potential of agroforestry systems to sequester carbon across Hawaiʻi’s fallow, unmanaged agricultural lands, non-native dominated conservation lands, and undeveloped urban lands. In addition to the vast areas of unmanaged fallow agricultural lands, an estimated 40% of areas zoned conservation across the state are dominated by non-native vegetation³⁴, offering important potential lands for restoration through agroforestry³⁵. We employ mixed methods including local and Indigenous practitioner interviews, literature review, and spatial modeling to: 1) build future scenarios of the potential for multi-strata agroforestry systems; and 2) estimate where above-ground and soil carbon benefits and tradeoffs are most likely to accrue in order to explore the potential for multi-strata agroforestry to contribute to carbon sequestration in Hawaiʻi under current and future climates. As governmental, NGO, and private entities are rapidly mobilizing to initiate carbon credit projects in Hawaiʻi³⁶, this is a critical and timely opportunity to facilitate the inclusion of these important land management practices that provide multiple ecological and social benefits and the potential for more durable carbon sequestration efforts.

Results

Agroforestry species mixes

We based our design of multi-strata agroforestry land use options on existing systems that are being tended across Hawaiʻi (Fig. 1). These systems are site-specific and diverse relative to other agriculture in Hawaiʻi (> 10 species per site), including both native and non-native, culturally important species¹⁹. In interviews with agroforestry practitioners across Hawaiʻi, Hastings et al.¹⁹ found that while each agroforestry system is unique, shared values motivate many people to practice multi-strata agroforestry, including restoring relationships to ʻāina (land), ancestors, and/or culture and strengthening local communities¹⁹. Specifically, some of the most mentioned themes of motivations raised by practitioners interviewed included to “reverse damage of plantation agriculture and ranching”, because of “kuleana (responsibility) to ʻāina”, to “feed our community”, and for “community’s health and wellness”. Practitioners interviewed in Hastings et al.¹⁹ also expressed agroforestry specific motivations including to build on and perpetuate Indigenous and local knowledge and practices because “the template [for agroforestry] was created by our ancestors” and to “bring the forest back.”

Fig. 1.

Example multi-strata agroforestry systems in restoration today across Hawaiʻi include, Top left: dry system at Kahaluʻu, Hawaiʻi Island [~ 6 years old], Top right: mesic system at Puʻulani, Heʻeia, Oʻahu [~ 4 years old]; bottom left: wet system at Honoliʻi, Hawai’i Island [> 10 years old] stewarded by Dr. Orlo Steele; bottom right: wet system at Waipā, Kauaʻi [1 year-old system].

Specifically, we designed three multi-strata agroforestry species mixes suitable for dry (550–1,500 mm/year), mesic (1,500–3,000 mm/year), and wet (> 3000 mm/year) conditions based on common species tended by current practitioners in these rainfall zones and an understanding of ʻŌiwi (Native Hawaiian) agroforestry system types (Table 1). Species selected are only representative of the structure and diversity of current place-based multi-strata agroforestry in Hawaiʻi today; in practice, each system would likely contain a different mixture of the 137 + plants mentioned in interviews from Hastings et al.¹⁹, representing 15 dry, 11 mesic, and four wet multi-strata agroforestry sites (Table S1.1).

Table 1.

Ōiwi (Native Hawaiian) agroforestry types and representative species mixes of dry, mesic, and wet multi-strata agroforestry systems in Hawaiʻi based on interviews from Hastings et al.¹⁹.

	ʻŌiwi agroforestry system types			Dominant species in each land use option
	Name	Range	Description	Overstory (> 8 m)	Midstory (8—2.5 m)	Understory (< 2.5 m)
Dry multi-strata (550—1,500 mm / yr)	–	Kaupō, Maui	depression-planting within cinder/ash layers	ʻulu (Artocarpus altilis)^ koa (Acacia koa) +  niu (Cocos nucifera)^ avocado (Persea americana) wiliwili (Erythrina sandwicensis) +	cacao (Theobroma cacao) alaheʻe (Psydrax odorata)* ʻiliahi (Santalum ellipticum) +  tangerine (Citrus reticulata) ʻaʻaliʻi (Dodonea viscosa)* wauke (Broussonetia papyrifera)^	ipu (Lagenaria siceraria)^ lilikoi (Passiflora edulis) ʻuhaloa (Waltheria indica)* nīoi (Capsicum frutescens) lemongrass (Cymbopogon citratus)
Mesic multi-strata (1,500—3,000 mm / yr)	pāhala kauluʻulu	Puna, Hawaiʻi mesic midlands	hala, kalo agriculture practiced where tees/branches cleared (Handy & Handy 1972) mixed, open canopy dominated by ʻulu	ʻulu (Artocarpus altilis)^ koa (Acacia koa) +  niu (Cocos nucifera)^ kukui (Aleurites moluccanus)^ mango (Mangifera indica)	maiʻa (Musa spp.)^ coffee (Coffea arabica) hala (Pandanus tectorius)* papaya (Carica papaya) māmaki (Pipturus albidus) +  kō (Saccharum officinarum)^	kalo (Colocasia esculenta)^ ʻōlena (Curcuma domestica)^ ʻuala (Ipomoea batatas)^ ʻawapuhi (Zingiber zerumbet)^ palapalai (Microlepia strigosa)* maile (Alyxia stellata) *
Wet multi-strata (> 3,000 mm / yr)	ʻāpaʻa - pākukui	rainforest belt old-growth forest patchwork Hāmākua, Hawaiʻi	native canopy is maintained, subcanopy altered highly tended forests to augment ecosystem services at landscape scale novel forest of kukui, swidden agriculture practiced where trees felled (Lincoln 2020)	ʻōhiʻa (Metrosideros polymorpha) +  kukui (Aleurites moluccanus)^ kamani (Calophyllum inophyllum)^ koa (Acacia koa) +  niu (Cocos nucifera)^	milo (Thespesia populnea)^ moringa (Moringa oleifera) ʻōhiʻa ʻai (Syzygium malaccense)^ kou (Cordia subcordata)^ hāpuʻu (Cibotium spp.) māmaki (Pipturus albidus) +	ʻawa (Piper methysticum)^ ʻuluhe (Dicranopteris linearis)* kī (Cordyline fruticosa)^ kupukupu (Nephrolepsis spp.)* lau pele (Abelmoschus manihot) lauae (Microsorum spectrum) +

* native; + endemic; ^ Polynesian introduction.

Potential restoration area

We then developed spatial scenarios for the dry, mesic, and wet multi-strata agroforestry land use options and modeled their potential extent across Hawaiʻi. We used the current climate as a low-range potential future climate, and RCP 8.5 mid-century as a mid to upper range climate projection. We found a total of ~ 1,654 km² and 1,342 km² of potential area for agroforestry transitions across the state under current³⁷ and projected RCP 8.5 mid-century rainfall,³⁸ respectively (Fig. 2; Figure SI.1). We found that 91–93% of potential sites occurred on unmanaged, zoned agricultural lands (1,227–1,536 km²), 6–8% on non-native dominated conservation lands (102–104 km²), and the rest on undeveloped urban and rural zoned lands (< 1%). This potential restoration area represents about half and over a third of currently unmanaged fallow agricultural lands in Hawaiʻi, for the current and future climate respectively. Under the current climate, 53%, 26%, and 21% of the restored area corresponds to dry, mesic, and wet multi-strata agroforestry, respectively. With projected reductions in rainfall, there is a shift towards drier systems (i.e., wet to mesic (~ 30 km² ); and mesic to dry (~ 74 km²)) as well as a loss of ~ 208 km² of potential restoration areas that drop below 550 mm/year rainfall.

Fig. 2.

Potential areas for multi-strata agroforestry transitions under projected RCP 8.5 mid-century rainfall. See Figure SI.1 for current rainfall projections.

Change in above ground carbon

We estimated per area mean above-ground carbon (AGC) of restored agroforestry systems to be 92–125 Mg C ha^-1 (338–459 Mg CO₂ ha^-1; Table 2; Dataset SI.1) based on calibrations from our representative species mixes (Dataset SI.2).

Table 2.

Mean AGC by agroforestry climate type.

System	Mean Mg C ha-1 (Mg CO2/ha-1)	Maximum (Mg C ha-1 (Mg CO2/ha-1)	Minimum (Mg C ha-1/ Mg CO2/ha-1)
Dry	97.5 (357.5)	131.3 (481.4)	63.4 (232.5)
Mesic	92.3 (338.4)	113.0 (414.3)	70.4 (258.1)
Wet	125.3 (459.4)	153.4 (562.5)	90.3 (331.1)

Note: In comparison, mean Mg C ha^-1 of native dry forest = 6.6; invasive dry forest = 15.7; native wet-mesic = 72.4, invasive wet-mesic = 90.9 (Selmants et al. ²¹).

Overall, transitioning land to agroforestry resulted in a significant increase in AGC across 72% and 78% of the restoration area under the current rainfall and RCP 8.5 mid-century rainfall respectively; in contrast, only 7% and 8% of areas show a significant decrease in AGC (Fig. 3; Figure SI.2). The only areas without a significant increase in AGC through agroforestry transitions are areas currently classified as non-native forest. The greatest gains in AGC are in transitions from sparsely vegetated areas, followed by grassland and shrubland areas to agroforestry, whereas transitions from non-native forest to agroforestry vary (Table 3).

Fig. 3.

Projected changes in AGC with multi-strata agroforestry transitions under RCP 8.5 mid-century rainfall. See Figure SI.2 for current climate projections.

Table 3.

Mean change in AGC (Mg C ha^-1) with a transition to multi-strata agroforestry by initial land cover type.

Initial land cover	Mean change in AGC current climate (Mg C ha-1; SD)	Area current climate (ha)	Mean change in AGC Under RCP 8.5 (Mg C ha-1; SD)	Area future climate (ha)
Forest	17.9 (52.1)	619.5	12.7 (49.7)	579.8
Shrubland	94.1 (6.7)	277.1	94.2 (7.5)	156.2
Grassland	97.9 (11.4)	358.1	98.9 (12.2)	275.0
Sparsely veg	105.4 (13.7)	399.8	107.1 (14.2)	331.5

SD = standard deviation

Overall, transitioning the entire potential restoration area to agroforestry, is projected to sequester 11.6 million Mg (42.4 million Mg CO₂) and 8.6 million Mg C (31.5 million Mg CO₂) under current and RCP 8.5 mid-century rainfall projections respectively (with an average per area sequestration of 63.9–69.9 Mg C ha^-1 (234.3-256.3 Mg CO₂ ha^-1)). However, if only including areas with an expected increase (either trend increase or significant increase; 1,425 km² or 1,111 km²) sequestration is projected to reach 12.2 million Mg C (44.7 million Mg CO₂ ha^-1) and 9.5 million Mg C (34.9 million Mg CO₂ ha^-1), under current and future rainfall, respectively (with an average per area sequestration of 85.7 Mg C ha^-1 (314.2 Mg CO₂ ha^-1) under both climates). Assuming a 20-year growth period of agroforestry systems, the areas where AGC increase have an average per area sequestration rate of 15.7 Mg CO₂ ha^-1 yr^-1 . Over the entire restoration area, this would translate to a total per year carbon sequestration of 2.2 million Mg CO₂ yr^-1 under the current climate and 1.7 million Mg CO₂ yr^-1 under the future climate.

Changes in soil carbon

We based estimated changes in soil C primarily on data from global meta-analysis of soil C impacts of land use transitions to agroforestry^39–41 (Table SI.2). These estimates show increases in soil C with transitions from cropland to agroforestry, but uncertain impacts with transitions from pasture or grassland, and little to no data on transitions from shrubland. Across all types of agroforestry systems, transitions from forest to agroforest generally reduce soil C, but for the subgroup of multi-strata agroforestry systems, Chaterjee et al.⁴⁰ found forest to agroforestry transitions can increase soil C (see Table SI.2 for a summary of meta-analysis conclusions).

We complemented meta-analysis findings with a review of studies that focused on either multi-strata agroforestry transitions in the tropics or from Hawaiʻi-based land use change studies. Dataset SI.3 provides the matrix used to compile available studies within each land-use transition (monoculture cropland, shrubland, grassladn, and forest to agroforestry), climate zone (wet, mesic, dry), and soil type (Table SI.3) ⁴² (Dataset SI.3; Table SI.3). The availability and directional change of these studies was used to assign low, medium, and high confidence levels (see Methods).

About a third (i.e., 35% and 37% under the current and RCP 8.5 mid-century rainfall, respectively) of the modeled agroforestry restoration areas are projected to increase in soil C (Table SI.4). These are areas formerly in intensive plantation agriculture—primarily sugarcane and pineapple. Areas classified as "increase high confidence," correspond to wet climates with poorly crystalline and non-crystalline mineral soils (Table SI.3; PNCM, mainly Andisols), which is also where the majority of land-use change studies have been done in Hawaiʻi (Fig. 4; Figure SI.3; Dataset SI.3).

Fig. 4.

Projected changes in soil C with multi-strata agroforestry transitions under RCP 8.5 mid-century climate. See Figure SI.3 for current climate projections.

Our results show about a quarter (23% and 28% under the current and future climate, respectively) of the modeled area would have “no change” in soil C, corresponding to areas where forest is transitioned to agroforestry. These were typically sites dominated by low activity clay soils in mesic systems and PNCM dry systems that have relevant local studies supporting no change in soil C and thus classified as "no change medium confidence" (Fig. 4; Fig SI.3; Table SI.4; Dataset SI.3), whereas the rest are classified as "no change low confidence."

Approximately 14% and 13% of the modeled area, under the current and future climate respectively, have “uncertain” soil C outcomes because of conflicting evidence, corresponding to pasture and grassland and reflecting the diversity in outcomes reported in global and local literatures. Finally, about a quarter (28% and 23% under the current and future climate respectively of modeled area) are classified as “unknown” soil C outcomes because of a lack of data, with these areas generally corresponding to shrubland and sparsely vegetated areas (Fig. 4; Figure S1.4; Table SI.4; Dataset SI.3 ).

Synergies & Tradeoffs of AGC & Soil C

We then combined AGC and soil C analyses to classify areas by the combined carbon outcome (Fig. 5; Figure SI.4; Table SI.5). We project agroforestry transitions to significantly increase both soil C and AGC in ~ 28% of restoration areas. In another ~ 10% of the area, AGC increases significantly, but there is no change expected in soil C. Another 6% of areas are not expected to result in significant shifts in AGC, but are projected to increase in soil C. Together these three groups represent over a third of the restoration area (~ 43%) and are given a “green light,” in that carbon benefits are highly certain to accrue with a transition to agroforestry.

Fig. 5.

Projected synergies and tradeoffs in AGC and soil C with multi-strata agroforestry transitions under RCP 8.5 mid-century climate. See Figure SI.4 for current climate projections.

Conversely, only 4% of the areas are clear no-go zones where AGC is expected to significantly decrease and no change in soil C is expected (Fig. 5; Figure S1.4; Table SI.5). Eleven and 14% of the modeled area under current and future climate, respectively, are projected to not see changes in soil C or AGC with agroforestry transitions.

Another 41% and 35% of modeled areas, under current rainfall and future rainfall, respectively, can be considered “yellow” zones where more information is needed to understand the potential impacts of a transition to agroforestry on soil carbon. For these areas, directional change in soil C is either uncertain (conflicting evidence) or unknown (insufficient evidence). This includes areas that are projected to significantly increase in AGC, with unclear impacts on soil C. Another 3% of areas show a decrease in AGC, but a potential increase in soil C, representing non-native forest with intensive cultivation histories (Fig. 5; Figure SI.4; Table SI.5).

Discussion

Place-based agroforestry systems have important potential as equitable and effective land-use practices in both agriculture and forest-focused carbon incentive programs. However, a paucity of data on carbon sequestration of place-based multi-strata systems can be an obstacle to their inclusion in these programs^1,43. While agroforestry generally, and multi-strata agroforestry, in particular, is recognized as a promising natural climate solution in terms of carbon sequestration potential¹⁸, estimates are globally variable and largely lacking for Hawaiʻi and other Pacific islands^17,18,39. We based our scenarios and carbon estimates on actual agroforestry systems being tended today by communities and other groups who support restoration through agroforestry because of a suite of interrelated social, cultural, ecological, and economic motivations¹⁹. Extending a previous effort to model the historical extent of colluvial agroforestry in Hawai’i⁹, we find that restoration through Indigenous and place-based multi-strata agroforestry is suitable across large areas of fallow agricultural lands and non-native dominated conservation lands. This is consistent with a recent modeling effort of a diversity of historical Indigenous agroforestry systems across Hawaiʻi³⁰. While we find that the potential restoration area decreases with a drying climate, there is still a substantial area of land suitable for agroforestry transitions across these vast currently unmanaged lands.

An important modeling result is ~75% of the modeled restoration area show the potential for significant increases in above-ground carbon. The greatest gains, unsurprisingly, are in areas sparsely vegetated or dominated by invasive grasslands and shrublands. Conversely, the lowest above-ground carbon gains, including losses, are in non-native forest to agroforest transitions. We also find areas, primarily in current dry invasive forest areas, where transitions to agroforestry show increases in above-ground carbon even in areas that are currently forested.

The overall mean change in AGC with transitions to agroforestry found in this study is similar to mean estimates of multi-strata agroforestry carbon gains from available studies from Africa, Asia, and Latin America (79 Mg C ha^-1)^18,39; our projected gains in AGC when restoring grasslands, shrublands, or sparsely vegetated areas are higher, whereas our estimates are lower when transitioning already forested land. Our mean estimates of potential AGC in agroforests are also higher than mean estimates of carbon stored in native dry and mesic-wet forest types in Hawaiʻi⁴⁴, suggesting that agroforests offer an important reforestation strategy that can sequester relatively large amounts of carbon while also providing a suite of other benefits in areas that were traditionally and historically used for food production. A focus on unmanaged fallow agricultural land also avoids land competition with other agricultural land uses, and agroforestry, as a food producing system itself, is in line with the activities of the agricultural land use district.

Another clear finding is that the greatest benefits for soil C will be in areas that were formerly intensively cultivated under plantation agriculture, representing over a third of the restoration area. This is in line with broader literature on the influence of both agroforestry and other reforestation projects on soil C^39–41,45, which suggest the greatest benefits happen where projects occur on cultivated or highly altered lands⁴⁶. In Hawaiʻi, the greatest benefits for overall carbon will be seen with restoration of invasive grasslands or sparsely vegetated non-native systems that were formerly in plantation agriculture given that high above-ground C and soil C gains are expected. These fallow grassland areas are also some of the areas now presenting the highest fire risk in Hawaiʻi as a result of land use and climate drivers⁴⁷. Given that replacing grasslands with closed-canopy woody vegetation is a key strategy to reducing fire risk⁴⁸, agroforestry presents a viable fire mitigation strategy while also sequestering carbon and providing broad social value to these lands. The economic and material benefits from agroforestry are providing incentives for communities elsewhere to integrate the practice as a fire risk strategy in economic and ecological contexts that have proven similarly difficult to manage⁴⁹.

Only 4% of the total area are considered "no go zones" where AGC is expected to decrease with uncertain or unknown benefits for soil C and another 11–14% with no expected change in soil C or AGC. These are predominantly in areas with non-native wet and mesic forests as well as unmanaged plantation forests (primarily Eucalyptus spp.) with higher or similar levels of above-ground carbon than agroforests, but that were never used for intensive cropland agriculture. Whereas it makes little sense to prioritize restoring these areas through agroforestry for carbon sequestration benefits alone, transitioning non-native forests and unmanaged plantations to agroforestry is still important for other objectives such as biodiversity, food production, and cultural benefits³⁵. It is also important to note that some multi-strata agroforestry systems have been found to have similar to higher soil C than paired forests⁴⁰. Thus, further research may shed light on the potential for soil C sequestration in this context¹⁶.

In other areas, including grasslands and shrublands which were never used for intensive agriculture, but often were used for pasture, there is wide uncertainty over the likely influence of transitions to agroforestry on soil C. In general, the global literature is mixed on the influence of reforestation and transitions from pasture or grassland to agroforestry on soil C^39–41,45, and there is little evidence for transitions from shrubland to agroforestry. However, given that above-ground C increases in these transitions, if soil C increases or stays the same, these areas would also be viable for carbon sequestration and likely expand the extent of positive carbon benefits. Overall, our study highlights gaps in understanding of the likely impact of transitions to multi-strata agroforestry on soil C across diverse land-use transitions, climate zones, and soil types (Dataset SI.3), but also demonstrates a method to incorporate qualitative directional change in soil C based on existing evidence. Empirical research on agroforestry transitions in Hawaiʻi and soil C outcomes will help to further understanding of these shifts and help to validate results of this approach.

Currently, nature-based carbon initiatives in Hawai’i are focused on restoring higher elevation pastures to koa (Acacia koa) as some growth and yield data and established silviculture practices exist for this native species⁵⁰. Koa is also a fast growing, native keystone canopy species with high cultural and economic value. While these initiatives offer important opportunities in Hawaiʻi, there are vast lowland areas that have the potential for nature-based interventions and many different culturally relevant species, in addition to koa, that may provide similar C benefits. Agroforestry transitions in the lowland areas surrounding communities are a particular priority in the current context and urgent need to reduce fire risk on fallow agricultural lands²⁹. In this context, restoration through agroforestry offers a land use strategy that produces food and other products on agriculturally zoned land, much of which is otherwise left as unmanaged fallow, posing enormous fire threat to homes and adjacent ecosystems. As Hawaiʻi confronts this problem after disastrous fires in 2023, combining carbon sequestration with these other benefits may offer pathways to finance restoration of broad social value to these lands while also reducing fire risk. Moreover, prioritizing these unmanaged agricultural areas avoids tradeoffs in both C storage as well competing interests that often arise in reforestation and afforestation projects, such as impacts on native ecosystems and food production⁵¹.

While our study advances understanding of the potential carbon sequestration benefits of Indigenous and other place-based multi-strata agroforestry systems, there are important uncertainties that require further research to refine estimates. In addition to the soil C uncertainties described above, above-ground C estimates are based on potential systems and use generalized allometric equations rather than specific allometric equations. Thus, developing species specific allometric equations to better estimate potential benefits over time will be critical for future work. As land use transitions to agroforestry systems continue to expand in Hawaiʻi and beyond, efforts to track changes in soil C and above-ground C over time will be imperative, and opportunities to do so will also increase. This will help to address the uncertainty in our soil C analyses which are not based on empirical data on agroforestry transitions in Hawaiʻi due to the paucity of available data. At the same time, work with practitioners to define priority areas for potential agroforestry transitions will inform where hotspots for carbon sequestration align with community goals and other factors influencing land use change. While a registered carbon project through the voluntary carbon market in Hawaiʻi may be unlikely in the near future due to scale and data challenges, there are emerging ways to support carbon sequestration in these systems such as government funding mechanisms for readying underserved landowners in emerging ecosystem service markets.

Conclusion

We demonstrate the carbon sequestration potential of transitions to place-based multi-strata agroforestry on unmanaged fallow agricultural and non-native dominant conservation lands across the Hawaiian Islands. While carbon data on complex multi-strata systems is limited, it is clear they can have important soil and above-ground C benefits, while also being a multi-benefit land use that produces food, connects people to place, and provides increasingly important fire mitigation benefits. The approach we developed to assess the potential carbon sequestration benefits of place-based agroforestry at scale can be applied in other places where interest in these systems is high, but where site-specific carbon data may be limited.

Investing in land management and restoration transitions up front is difficult for any restoration effort (including forest carbon projects). In the context of agroforestry systems, which require time to generate income from agricultural production, carbon incentives may help to complement revenue from harvests and other value-added products. Over the long-term, the direct economic and cultural value derived from these systems may help to ensure their durability.

An important additional question will be how to support place-based systems and local stewards at a scale amenable to carbon projects, which will likely entail addressing costs of land and tenure¹⁹, building support networks across practitioners (e.g., malaoiwi.org), and exploring policies and economic programs that will facilitate these initiatives. Aggregated projects with multiple landowners is challenging, but there are successful examples (https://www.forestfoundation.org/what-we-do/increase-carbon-storage/family-forest-carbon-program/). While carbon will never be a primary motivation for these transitions, understanding the ways that carbon ecosystem incentive programs can support this type of multi-benefit restoration leading to more effective and equitable outcomes is a critical part of just climate policy moving forward in Hawaiʻi and beyond. As such, carbon incentives can be conceptualized as one tool in a broader restoration financing toolkit.

Methods

Study area

We examined the potential for multi-strata agroforestry transitions across the main Hawaiian Islands. Statewide, land is zoned either as conservation (~ 49%), agriculture (~ 46%), urban (~ 5%), or rural (< 1%)⁵². Over 40% of agricultural lands are un-managed^24,25 and 40% of conservation lands are dominated by non-native vegetation³⁴, offering important potential lands for agroforest restoration across vast areas of the islands.

Potential agroforestry transitions scenarios

We first developed three land use options, or representative species mixes, of multi-strata agroforestry systems suitable for dry (550–1500 mm/yr), mesic (1,500–3,000 mm/yr), and wet (> 3,000 mm/yr) rainfall zones⁵³. We primarily based the agroforestry species mixes on interviews with multi-strata agroforestry practitioners from 30 sites across Hawaiʻi¹⁹. We analyzed semi-structured interview transcripts and extracted plant species mentions from each interview and noted any indication of the relative abundance of the species at the site. We identified the average annual rainfall of each site based on their location on the Hawaiʻi Rainfall Atlas³⁷ and categorized species as pertaining to dry, mesic, and wet systems according to the rainfall zones in Price & Jacobi (2012) (Table 1).

In the three land use options, we included species that were the most frequently mentioned across sites (Table S1.1) within each rainfall zone, and also took into consideration broader knowledge of the sites, from in-person visits since the interviews, and our team’s collective knowledge of multi-strata agroforestry in Hawaiʻi. We selected species adapted to the particular rainfall zone, assuming that irrigation would only be used in the establishment phase and potentially during severe drought.

Although species composition of agroforestry systems is dynamic and successional, we developed the mixes based on the composition at maturity (> 20 years since establishment). We estimated the total number of trees and shrubs per hectare in the mid- and overstory using the ‘four-layer complex’ pattern–the pattern most closely approximating multi-strata systems–in AgroforestryX, an online design tool developed for Pacific Island agroforestry systems that produces counts of trees in each layer for a 30 × 30 m plot⁵⁴. Based on AgroforestryX, each 30 × 30 m plot included 189 overstory (mix of five species) and 200 midstory (mix of four species) individuals (Dataset SI.1). We combined the counts for the ‘emergent’ and ‘high’ layers given by AgroforestryX into one ‘overstory’ layer as pruning is the main factor distinguishing these layers⁵⁴, and this level of detail was not feasible to include in our model. For the midstory, we used the counts given for the ‘medium’ layer in AgroforestryX. We included the same number of species and number of trees across each rainfall zone. We assumed an understory layer of non-woody species whose compositions vary by climatic zone.

Spatial extent

We then projected where on the landscape each agroforestry type could be feasible under the current climate³⁷ and under a future climate scenario (Representative Concentration Pathway 8.5 mid-century)³⁸ (Fig. 1). RCP 8.5 mid-century is based on statistical downscaling of Coupled Model Intercomparison Project phase 5 (CMIP5) for Hawaiʻi, projecting an overall dryer climate, but greater contrasts between the wet and dry regions³⁸. While initially considered extreme, RCP 8.5 is already expected to be overshot⁵⁵. Accordingly, we use the current climate as a low-range potential future climate, and RCP 8.5 mid-century as a mid to upper range climate projection. We omitted Niʻihau from the analysis due to the lack of available data for soil and future climate.

Most transitions to multi-strata agroforestry today occur when practitioners gain new access to primarily agricultural zoned land, in large part because of the history of Indigenous land dispossession and accumulation of land during the plantation era⁵⁶. All 30 sites practicing multi-strata agroforestry Hastings et al.¹⁹ interviewed had a history of plantation agriculture or ranching and were fallow prior to undergoing restoration by practitioners. Prior to restoration, approximately half of the sites (n = 17), were dominated by non-native grasses, while the rest were restored from non-native secondary forest.

Accordingly, we assumed that multi-strata agroforestry transitions could occur on environmentally feasible land that was either: (1) zoned agriculture, but not used currently for agriculture (i.e., fallow or unmanaged); (2) zoned conservation, but considered low priority given a dominance of non-native species; (3) zoned urban and rural, but undeveloped.

To determine unmanaged or fallow agricultural lands, we used the 2020 State of Hawaiʻi Agricultural Baseline²⁴, which identified areas in active agricultural production, and state land use zoning maps which delimits agricultural zoned land⁵². Those lands zoned agriculture, but were not in production in 2020 were considered unmanaged agricultural lands. Within these areas, we excluded any areas classified as developed or as native vegetation in the Hawaiʻi Carbon Assessment land cover map⁵⁷ and young lava flows⁵⁸. Following the Kurashima et al.⁹ spatial model of colluvial agriculture (agroforestry) systems, we constrained the scenarios from sea level to 855 m in line with crop growth restrictions and excluded areas with slopes over 30 degrees. We assumed multi-strata agroforestry would not occur below 550 mm rainfall per year, since long-term irrigation is often cost-prohibitive for agroforestry practitioners¹⁹.

In addition to fallow agricultural lands, we also considered undeveloped urban and rural zoned areas as well as non-native dominated conservation zoned lands, which overlap with projections of suitable area for historical Indigenous colluvial agriculture (agroforestry)⁹. This aligns with several examples of restoration through agroforestry on conservation lands dominated by invasive species⁵⁹, and on urban or rural, but not developed lands². Scenarios were created using ArcGIS Desktop version 10.8.2.

Change in above-ground carbon (AGC)

We estimated the AGC (Mg C/ha) of trees in the overstory and midstory for the three types of multi-strata agroforestry. We did not include understory species or shrubs in the AGC calculations given that the majority of AGC in Hawaiian forests are found in tree biomass⁶⁰ and the limited data available to include shrub and herbaceous biomass. Given the paucity of species-specific allometric equations for mature trees in our agroforestry land use options, we used a general allometric equation for tropical forest trees to estimate the above-ground carbon for a mature (~ 20 years) tree⁶¹:

where D = diameter at breast height (DBH in cm), p = wood density (g/cm3), H = height (m) However, we used species-specific equations for two species that have tree growth forms, but are not woody: niu (Cocos nucifera;⁶² and mai’a (Musa spp.;⁶³). All data and sources for height, DBH, and wood density values are in the supplementary material (Dataset SI.2).

We calculated a range of AGC estimates for each multi-strata agroforestry type by varying the abundance of overstory and midstory species. We calculated AGC for (1) an even distribution of individuals across overstory and midstory species (i.e., average estimate), (2) a skewed distribution of individuals in which the three species with the highest carbon per tree in the overstory each made up 30% of the total trees in that layer and the two highest carbon trees in the midstory each making up 45% of the total individuals in that layer (i.e., maximum estimate), and (3) a skewed distribution of individuals such that the three species with the lowest carbon per tree in the overstory made up 30% and the midstory the top two 45% (i.e., minimum estimate) (Dataset SI.1). Due to the goals of the interviewed practitioners (high diversity, biocultural, etc.), we did not look at dominance of a single overstory species. Varying the distribution of species abundance in this way allowed us to estimate a range of values for AGC reflective of the diversity of planting designs followed by agroforestry practitioners.

Next, we assigned the AGC estimates for each multi-strata agroforestry type to the future spatial scenarios to create a map of the potential AGC storage under potential agroforestry transitions. This was compared to a baseline AGC map, which combined an aboveground carbon density layer for forested lands in Hawaiʻi⁶⁴, and mean biomass estimates for non-forest lands in Hawaiʻi⁴⁴ (grassland = 2.5 Mg ha^-1; shrubland 4.4 Mg ha^-1). In order to incorporate a measure of uncertainty, only where the low range estimates exceeded the baseline AGC was the increase deemed significant. Likewise, a decrease was considered significant where the high range estimate was lower than the baseline AGC. Analyses were done in R software (R Core Team 2022)⁶⁵.

Change in soil C

Given the complexity of shifts in soil C with land-use change and a lack of data on changes in soil C with agroforestry transitions in Hawaiʻi^16,66,67, we developed an approach to estimate the likelihood of directional change in soil C under varying combinations of initial land cover, soil type, and rainfall. As a first layer, we drew on a global meta-analysis of changes in soil C with agroforestry transitions^39–41. These studies broadly find that soil C generally increases when transitioning intensive agriculture to agroforestry, decreases when transitioning from natural forest, and is mixed or no significant change with pasture or grassland transitions to agroforestry. Chaterjee et al.⁴⁰ explicitly included data on transitions from agriculture and forests to multi-strata agroforestry systems in the lowland humid tropics and subtropics (15°N to 25°N and 15°S to 25°S), finding that multi-strata systems have more soil C than paired agricultural systems, and similar soil C compared to natural forests (see Table SI.2). De Stefano & Jacobson⁴¹ include agrisilviculture, which includes multi-strata systems, but also includes other systems including wind breaks and plantation crops.

There are several limitations of the meta-analyses. First, none of the meta-analyses includes comparisons of agroforestry to non-native forest or to shrublands. The analyses are also not disaggregated by soil type, which along with land use history and climate, likely influences the impacts that these transitions have on soil C^66,67. Accordingly, we created a matrix and conducted a literature review on multi-strata agroforestry transitions classified by climate, soil type, agricultural land use history, and current land cover (Dataset S1.3). While there are no data on agroforestry transitions in Hawai’i, we also include land-use change studies in Hawaiʻi which compare cropland (sugar) or pasture to paired native forest as multi-strata forests are similar in structure to native forests. We describe each component of the matrix below.

Climate

We used the rainfall zones described above to classify existing studies into: dry (550–1500 mm/yr), mesic (1,500–3,000 mm/yr), and wet (> 3,000 mm/yr) rainfall zones⁵³.

Soil type

Soil type groupings (Table SI.6) were delineated by spatial data developed for the Hawai′i Soil Atlas Order Series, “Fertility Class” layer. Based on Hawai’i Soil Atlas classifications⁴² for mineral fertility class, soils were grouped as: high activity clays (HAC, including Mollisols, Vertisols, and Aridisols), low activity clays (LAC, including Oxisols and Ultisols), poorly and non-crystalline minerals (PNCM, including Andisols), and organic soils which include all Histosols (HIST). Soils that did not fall into these categories include Entisols and Spodosols, which are classified as ‘Other’. Inceptisols were grouped based on mineralogy (Table SI.3).

Agricultural land use history and land cover

Given the importance of the presence or absence of cultivation history in soil conditions, we classified any area with a history of intensive cultivation (primarily sugar and pineapple). To do so, we used maps of historical sugar and pineapple lands, including the 1978–1980 Agricultural Land Use Maps (ALUM)⁶⁸ and the 2020 agricultural baseline²⁴. If there was 'no production history’, we used current land cover as the basis of comparison. Accordingly, we considered transitions from former intensively cultivated agricultural lands, non-native forests, non-native shrublands, non-native grasslands, and from sparsely vegetated land.

Literature review

We then considered whether each of the cells in the above-matrix (climate, soil type, agricultural history and land cover) has existing studies on multi-strata agroforestry transitions or from Hawaiʻi-based land cover studies.

To do so, we specifically considered:

1. Land-use change studies from Hawaiʻi comparing cropland or pasture to native forest or other restoration (including land uses that restore perennial vegetation without disruption of the belowground system) from the Hawaiʻi Soil Carbon Database⁶⁹.

2. Global studies of transitions to multi-strata agroforestry in tropical regions with similar climate and soil types. We extracted multi-strata agroforestry transition studies from, a meta-analysis of changes in soil C with agroforestry transitions globally^39–41. Studies completed post-2018 and thus not included in the global meta-analyses were identified using Web of Science and Google Scholar using topic search (TS) term: TS = (soil carbon* + tropics OR agroforestry AND SOC* + tropics OR agroforestry). Abstracts were screened and articles included if they were located within tropical climates and involved a transition from intensive agriculture, non-native forest, shrubland, or grassland, or from sparsely vegetated land to multi-strata agroforestry.

Based on the literature available, we classified each point in the matrix in terms of the likelihood of directional change in soil C (Table 4). We classified studies as increase, decrease, no change, mixed evidence, or insufficient evidence.

Table 4.

Rules to delineate projected soil C shift with transitions to agroforestry.

	Criteria
Increase (HC)	Meta-analyses conclude significant increase AND more than two Hawaiʻi-based studies and/or tropical multi-strata studies find significant increase in soil C
Increase (MC)	Meta-analyses conclude significant increase AND 1–2 Hawaiʻi-based studies OR tropical multi-strata studies find significant increase in soil C
Increase (LC)	Meta-analyses conclude significant increase OR one or more Hawaiʻi-based OR tropical multi-strata studies find significant increase in soil C
No change (HC)	meta-analyses conclude no significant change AND more than two Hawaiʻi-based studies and/or tropical multi-strata studies find no significant difference in soil C
No change (MC)	Meta-analyses conclude no significant change AND 1–2 Hawaiʻi-based studies and/or tropical multi-strata studies find no significant change in soil C;
No change (LC)	One or more Hawaiʻi-based OR tropical multi-strata studies find no significant change in soil C
Decrease (HC)	Meta-analyses conclude significant decrease AND more than two Hawaiʻi-based studies and/or tropical multi-strata studies find significant decrease in soil C
Decrease (MC)	Meta-analyses conclude significant decrease AND 1–2 Hawaiʻi-based studies OR tropical multi-strata studies find significant decrease in soil C
Decrease (LC)	Meta-analyses conclude significant decrease OR one or more Hawaiʻi-based OR tropical multi-strata studies find significant decrease in soil C
Uncertain (conflicting evidence)	Meta-analyses OR existing Hawaiʻi-based and tropical multi-strata studies are conflicting
Unknown (Insufficient evidence)	No available data in Hawaiʻi, tropical, & global studies AND no relevant meta-analysis comparison

HC high confidence, MC medium confidence, LC low confidence.

Tradeoffs in AGC and soil C

To evaluate the synergies and tradeoffs in soil C we characterized each pixel in the restoration scenarios as pertaining to an AGC category (AGC increase, AGC decrease, or AGC no change) and to a soil C category (soil C increase, soil C no change, or soil C uncertain/unknown). Pixels were only categorized as AGC increase or decrease if classified as significantly increasing or decreasing AGC (see Sect. 4.3); otherwise they were categorized as AGC no change. No areas were projected to decrease soil C with restoration, so there were none classified as soil C decrease). Analyses were done in R 4.3.3 using the ‘raster’ (v. 3.6–26) and ‘sp’ (v. 2.1–3) packages..

Author contributions

Conceptualization: LB, CT, TT, NK, ZH, KW; Funding acquisition: LB, TT, CT, KW, NK; Methodology: LB, GM, CT, ZH, SEC, CG; Analysis: LB, GM, CT, ZH, CT, ND; Writing (first draft): LB, GM, ZH; review: all.

Declarations

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-87891-y.