Introduction

The Andean headwaters of the Amazon are emblematic for the challenge of meeting demands from the interconnected water-energy-food sectors while enabling ecosystem conservation1. Headwater basins in the Andes-Amazon are home to millennia old, yet dynamic, smallholder agro-food systems2,3,4, contribute most to the Amazon River’s discharge, are critical to producing hydropower for the basin’s riparian countries1,5, and are home to unique montane forests and grassland ecosystems6. In recent decades, these areas have experienced rampant deforestation and degradation4,7,8, with severe consequences for local biodiversity, people, and sectors that depend on ecosystem services and regional and global hydro-climatic processes9,10.

While policies in the countries of the Andes-Amazon recognize the need to preserve natural ecosystems11, demand across water-energy-food sectors is growing12,13,14, all in the face of imminent climate change15,16. Exploring the environmental challenges and opportunities to reconcile the dual imperatives of nature conservation and restoration along with human development is urgent in these geographies. These investigations resonate with global needs to find equitable and robust approaches to meet fundamental societal demands without exceeding planetary boundaries17.

Conflicts between sectors—which reflect tensions between people and nature’s diverse needs and potential opportunities to navigate those trade-offs through proactive and robust management approaches—remain understudied for the region. For the Amazonian lowlands, studies have highlighted the vulnerability of sectors, notably hydropower, to land degradation and deforestation10,18,19. However, these studies adopted a top-down perspective20 based on limited landuse and climate scenarios and did not explore adaptation strategies across sectors—i.e., how different landuse/management or hydropower operation strategies could be actively managed to reduce trade-offs, but also how adaption in one sector might impact others. Regional studies elsewhere also found strong trade-offs between land management and interests of downstream water users that are difficult to resolve, particularly under more extreme climate change scenarios21, however, representation of sectors such as hydropower is often highly simplified. In general, identifying sustainable development opportunities that internalize sectoral interdependencies and climate and socio-economic uncertainties remains a research frontier22,23,24,25.

We herein explore opportunities to reduce sectoral conflicts at the water-energy-food-ecosystems (WEFE) nexus for a typical basin of the Andes-Amazon. The upper Huallaga Basin is part of the Marañón River in Peru, which is itself a major tributary of the Amazon (Fig. 1A), representing multiple pressures and the need to reconcile conflicting objectives for human development and conservation. The Huallaga Basin is characterized by a patchwork of native Andean ecosystems and family farming systems within “campesino” communities (Fig. 1B)26. Agricultural productivity in the Huallaga Basin and the entire region is low, and levels of economic poverty27 and undernutrition are high, which makes improving agricultural livelihoods a central issue for improving food security and advancing human development28,29.

Fig. 1: Modeling robust adaptation strategies for the water-energy-food-ecosystem (WEFE) nexus for basins in the Andes-Amazon.
Fig. 1: Modeling robust adaptation strategies for the water-energy-food-ecosystem (WEFE) nexus for basins in the Andes-Amazon.
Full size image

A Topography and SWAT model set up for case study, the Huallaga River Basin upstream of the Chaglla Dam (thick dashed outline), and 3 stations used for model calibration. B Landuse map of the upper Huallaga River. C Modeling the WEFE nexus under climate change is of regional importance. The region has significant existing (blue square) and future (orange circle) hydropower potential, accounting for nearly 42 GW33,34,69, and offers opportunities to increase food production through expanding irrigation (green)29.

The upper Huallaga River also provides water to Chaglla Dam, Peru’s third largest hydroelectric plant (450 MW installed capacity). Despite impounding a major reservoir, the plant has limited capacity to buffer inflows because it is operated like a run-of-river plant30. Energy generation at weekly to monthly scales thus relies mainly on the hydrological functioning of the upper Huallaga Basin. Specifically, dry season low flows (from June to November) determine the plant’s capacity to deliver firm power, an indicator for the reliability of power generation31, which is critical for the plant’s operational and financial baseline32.

Based on the case-study of the upper Huallaga Basin we highlight how watershed processes create biophysical connections between sectors under climate change, and we demonstrate that understanding such connections are critical to managing WEFE systems. Our approach reveals opportunities to integrate high-resolution modeling of watershed processes into decision-making approaches under uncertainty, making those strategies more rooted in the biophysical processes that underpin the WEFE nexus. Using these advances, we address three main questions. Firstly, which sectors (water, energy, food) are most vulnerable to climate change? Secondly, how does climate adaptation, notably in the food sector create sectoral interdependencies and trade-offs? Thirdly, are there opportunities to find robust strategies for synergistic management of critical sectors in the basin?

Our study deploys an integrated framework consisting of a high-resolution catchment model and a detailed model for hydropower design and operation. Firstly, biophysical modeling is conducted to understand how climate change, land practices, and hydropower operation affect food, hydropower production, and ecosystem conservation (Fig. 2A). Then, This biophysical modeling framework is embedded in a bottom-up analysis of uncertainty due to climate change and future socio-economic development(Fig. 2B)23. Note that we herein use the terms “bottom up” versus “top down” to distinguish between approaches focusing on scenarios (“top down”) and approaches that center around vulnerability and adaptive capacity (“bottom up”)18,20,23. Using this framework, we analyzed thousands of strategies for agricultural development, ecosystem conservation, and dam operations under projected climate change. We then explore the possibility of achieving an equitable distribution of climate impacts and growth opportunities across sectors. Our results highlight the strong dependence of hydropower outcomes on the management of upstream watersheds and notably on how upstream potato agriculture, the dominant crop in the region, is managed. This dependence creates major trade-offs between increasing upstream irrigation, which is critical for food systems’ climate adaptation, and unlocking potential land restoration and downstream hydropower, which are critical for Peru’s climate mitigation efforts.

Fig. 2: Performance metrics for the energy, ecosystems, and food dimensions are influenced by both climate uncertainty and sectoral decision.
Fig. 2: Performance metrics for the energy, ecosystems, and food dimensions are influenced by both climate uncertainty and sectoral decision.
Full size image

Using an integrated biophysical and energy infrastructure modeling approach, we characterized the metrics for a wide range of combinations of climate uncertainty and sectoral decisions (A). We then used a three-step decision analysis process to identify the sectoral decision options that allow for coherent and robust outcomes under climate uncertainty. The process includes (i) identifying the subset of decisions that are incompatible with the performance levels of a given set of targets (B1, B2, the targets are shown as green, yellow, and pink line) for at least one climate condition, (ii) assessing the robustness of the non-incompatible decisions to meet the targets across all climate uncertainties (B2, B3), (iii) by applying this framework to multiple frames of reference (different sectoral targets), we can identify the subset of decisions that are ultimately compatible with the targets across all climate uncertainties and provide options that inform a sectoral—bottom-up—decision making process.

Our findings are of importance for the entire Andes-Amazon region, where current (6700 MW) and future (up to 39,000 MW) hydropower production33,34,35 intersects with opportunities to intensify agriculture2,29 and with governments’ ambitions to restore and conserve ecosystems26,36. The findings also highlight overlooked challenges for trans-sectoral management of other working landscapes and the role of sustainable agricultural irrigation practices, particularly where watersheds contribute to future expansion of irrigation and hydropower, across the world.

Results

Sectoral impacts of climate and landuse change

We find limited climate change impacts on hydropower outcomes under two common Shared Socio-economic Pathways (SSPs), SSP126 and 585, representing the lowest and highest anthropogenic radiative forcing, respectively, and thus enveloping future climate change impacts. Chaglla Damonly started operation in 2018, but had it been operational in the past (1985–2014), the median firm power would have been 47.7 MW (Fig. 3A, centerline of boxplots represents the median). Changes under the evaluated climate scenarios are relatively small, and only by the end of the century will there likely be a clear decline in firm power (Fig. 3A). Median firm power would increase slightly to 48.0 MW (SSP126) and to 50.2 MW (SSP585) in the near future (2015–2044) and slightly increase to 49.5 MW (SSP126) or decrease to 45.4 MW (SSP585) for the mid-century period (2045–2074). Only for the end-of-century period would there be a decrease in median firm power values across both SSPs (44.1 MW for SSP126 and 42.5 MW for SSP585) below current levels (SSP126, 2075–2099).

Fig. 3: Climate adaptation in the food sector could have major impacts on downstream hydropower in the upper Huallaga River.
Fig. 3: Climate adaptation in the food sector could have major impacts on downstream hydropower in the upper Huallaga River.
Full size image

Hydropower outcomes would remain relatively stable for the analyzed climate future and if current landuse/management systems are maintained (A) while crop yields could see a major decrease (B). Common measures for climate adaptation in the food sector (expansion, intensification) could hedge against worst climate outcomes (D) but risk incurring major impacts on hydropower (C). The data include all landuse/management practices over 18 climate scenarios (3 time horizons X 2 SSP X 3 general circulation models (GCMs)) in boxplot (C, D). Centerlines in boxplots indicate the median, box extents are defined by the 25th and 75th percentile, whiskers extend to the largest value no further than 1.5 × IQR (interquartile range) from the box outline, results outside of that range are shown as individual points.

In contrast to the hydropower sector, rainfed crop production in the Huallaga Basin is highly vulnerable to climate change. Modeled median yields of potatoes, the dominant crop in the region26 and what we used to represent food sector performance, were 19.2 t/ha/yr for the 1985–2015 period (Fig. 3B) in the upper Huallaga Basin. In the near future, for SSP126 and SSP585 scenarios, total median crop yield could slightly increase by 0.36–0.4 t/ha/yr, but then decrease to 17.2–16.4 t/ha/yr for the mid-century period, and decrease to 13.6–10.4 t/ha/yr for the end-of-century period. In terms of total crop production, the 1985–2014 median was 26.8 × 105 t/yr (see SI Fig. 7A), which could increase to 26.9 × 105 t/yr in the near future (SSP585), and then decrease to 24.8 × 105–24.0 × 105 t/yr by mid-century, and to 19.6 × 105–14.8 × 105 t/yr for both SSP scenarios by the end of the century. Compared to firm hydropower, where different SSPs result in different trends, mid-century and end-of-century futures see a consistent negative trend in food production under the current landuse/management system for both SSPs.

Food system adaptation could create major conflicts with the hydropower sector. Farmers could respond to the climate pressures either through (1) intensification (e.g., irrigation of the currently rainfed croplands), (2) expansion of croplands by encroaching remaining natural ecosystems, or (3) a combination of both. At the same time, pressure on stakeholders to restore degraded forest lands will increase26, resulting in a potential decrease in cropped areas. Across the 16 landuse/management options, median total crop production for the near future could increase from 19.6 × 105 t/yr (SSP126) to 20.4 × 105 t/yr (SSP585), 8.4 × 105 t/yr (SSP585)–10.4 × 105 t/yr (SSP126) for the mid-century future. By end of century, median crop production increased by 2.4 × 105 t/yr (SSP126) and decreased by 19.6 × 105 t/yr (SSP585) (see SI Fig. 7B).

Intensification and expansion of agriculture enables increasing yields in the near future and allows farmers to adapt to a changing climate by mitigating losses towards the middle and end of the century. However, such strategies would lead to major losses for downstream hydropower generation. Firstly, median firm power would decrease in all future and SSP scenarios, e.g., by up to 32.8% (end-of-century value for SSP126) (Fig. 3C). Secondly, we find that the interannual variability (measured as standard deviation of firm power) would be 316.0–436.0% higher than the historic baseline. Thirdly, Chaglla Dam would see years with zero firm power generation (the minimum points in Fig. 3C) because dry season discharge falls below the threshold for environmental flow release so that no generation is possible, a condition that has not been observed in the past. These three mechanisms would have major impacts on the economic baseline of Chaglla Dam.

Conflicting outcomes for hydropower, food, and ecosystems

Understanding opportunities to balance outcomes between hydropower generation, food production, and ecosystems requires a granular analysis of different management strategies. These opportunities become evident when comparing the sectoral outcomes across different landuse/management scenarios (Fig. 4A, B). For instance, for the mid-century future (2045–2074, SI Fig. 8 for all three time periods) all scenarios without irrigation (yellow area in SI Fig. 8) result in firm power outcomes like today, but with increases in crop yields per hectare. Those increases can occur for different reasons, either because of the restoration of lower productivity lands with secondary vegetation (S3 scenarios) or because of the expansion of agriculture (S1, S2 scenarios). Implementing efficient irrigation allows slight increases in crop yields over current, rainfed conditions while sacrificing some hydropower outcomes. Full implementation of irrigation results in maximum yields (Fig. 4C), but also in the greatest impacts on hydropower generation. Scenarios with any conversion of secondary vegetation (through reforestation of conversion to cropland) lead to an increase in relative crop yields because we assume that crop yields are lower on land classified as “secondary vegetation” (more information in Methods “Hydrologic model set-up”).

Fig. 4: Tade-off analysis in food and hydropower sectors under different climate and landuse/management adaptations.
Fig. 4: Tade-off analysis in food and hydropower sectors under different climate and landuse/management adaptations.
Full size image

Sustainable intensification will be critical for adaptation and to balance food and energy sector outcomes in the mid-term (2045–2074). Food systems could adapt to climate change through a range of changes in cropped areas (A), which can be combined with different management strategies, resulting in 16 landuse/land management scenarios (B). Each of these scenarios will result in different trade-offs between food, energy, and ecosystems (C: representing food system outcomes as yield, D: representing food system outcomes as total crop production). The values shown for each landuse/management scenario are the median over 18 climate scenarios.

Crop yields and firm power are affected by irrigation, but total crop production is mostly driven by changes in cropped areas. Under maximum agricultural expansion (Fig. 4D, landuse S1), total crop production could greatly increase even under climate change (+255.2 to +284.5%), but such an expansion implies conversion of 53.4% of the basin’s native ecosystems (3089.51 km2) to cropland. Limited agricultural expansion with a full conversion of secondary vegetation only (Fig. 4A, landuse S2 scenario) would result in a +16.2 to +33.6% change in total crops. Restoration of secondary vegetation (Fig. 4A, landuse S3 scenario) would lead to a 12.6-26.7% decrease in median total crop production.

Efficient irrigation enables balancing trade-offs between sectors. For instance, sustainable irrigation with restoration (S3-e) would increase natural vegetation (+8.0%) at the costs of some food production (−21.7%) and energy (−18.9% in firm power). The same scenario without irrigation (S3-a, S3-b) would lead to a greater reduction in food production (−26.3 to −26.7% in median total production). And the same scenario with traditional irrigation (S3-c, S3-d) would lead to major losses for hydropower (−77.7 to −78.0 % of median firm power). For all those aspects, meeting sectoral demands will become more challenging in the far future (SI Fig. 8C), when irrigation expansion would continue to greatly impact hydropower but with reduced benefits for crop yields. While this analysis of median outcomes highlights general interdependencies and trade-offs between sectors, better (or worse) strategies, and outcomes might be hidden by aggregating across years, SSPs, and climate models, and by assuming that some sectors (hydropower) continue business-as-usual operations into the future.

Robust and equitable solutions to food-energy-ecosystem trade-offs

By exploring 10,080 possible adaptation pathways, we find that robust trans-sectoral strategies, i.e., strategies that maintain or increase all sectors’ current performance consistently until the end of the century over a wide range of climate uncertainties, require coordinated action between sectors. While our analysis finds the possibility of highly unequitable outcomes and suggests ways to mitigate them, certain trade-offs are unavoidable according to our modeling results.

We approach the identification of robust strategies by exploring a sequence of steps shown in Fig. 5. First, we set specific targets that satisfy a given criterion (sectoral or system level, shown in Fig. 5 with pink line). The criterion for agriculture, hydropower, ecosystems conservation (Fig. 5B–D) to maintain the current status-quo outcomes for food, hydropower, and ecosystems. The criterion for the integrated strategy scenario (Fig. 5C) focusing on the maximum benefits in all sectors is set to current status-quo outcomes for food, ecosystem, and total energy and to a slight decrease in firm power (less than 25%).

Fig. 5: Decision analysis framework for water-energy-food-ecosystem trade-offs in the upper Huallaga Basin.
Fig. 5: Decision analysis framework for water-energy-food-ecosystem trade-offs in the upper Huallaga Basin.
Full size image

The approach is based on three steps, (1) identifying objectives and filtering desirable outcomes for each objective (2) finding management actions that are compatible with these outcomes and (3) analyzing the full set of outcomes for compatible management actions. A full set of outcomes for 10800 futures. B–E searching for management actions compatible with avoiding or limiting future losses in (B) Agriculture, (C) Hydropower, (D) conservation,and across sectors (E). These searches are represented by pink lines that indicate filters set for specific outcomes.

Considering the trade-off in food and hydropower sectors, the criterion for multiple sectors is the change in food, ecosystem, and total energy above zero, and the change in firm energy above −25%; next, we select the strategies (the combination of decisions) compatible with the stated targets; finally, we conduct the bottom-up analysis to identify the susceptibility of missing the targets of the compatible strategies across planning timeframes and climate uncertainties.

Figure 5 illustrates this process. Starting from the domain of all 10,080 possible adaptation pathways (Fig. 5A), we identify strategies that exceed defined targets for individual sectors—agriculture (5B), hydropower (5C), or ecosystem conservation (5D) and lastly across all three sectors (5E) (different colors in Fig. 5 highlight near future, mid-century, and end-of-century outcomes under different SSPs). These strategies also include two additional management levers to represent engineering and economic adaptation at Chaglla Dam. These levers are firstly allowing wider variability in the storage water level of the reservoir impounded by Chaglla Dam and second, allowing greater hydropower sales to the grid.

Focusing policies solely on food, energy, or ecosystems would lead to major impacts on the other sectors. For instance, expanding croplands, irrigating, and  implementing measures to minimize soil erosion, would lead to a substantial and consistent increase in food production across the range of climate uncertainties (excluding end-of-century SSP585 climate conditions) (Fig. 5B). Yet, all those strategies would also cause the loss of firm power (up to −90%) and total generation (up to −30 %) and encroachment on  native ecosystem areas (up to −53.4%).

Focusing policies on the hydropower sector (Fig. 5C) would require limiting expansion of agricultural lands, keeping agriculture rainfed, and implementing terracing practices on remaining lands to avoid sediment delivery. These agriculture requirements would be in conjunction with expanded allocation of generation to the grid and increased use of the reservoir for more frequent storage level regulation. This strategy would maintain firm power and could increase total energy production. However, achieving those outcomes would come at the costs of reduced crop yields under many future conditions.

Focusing only on restoration of terrestrial ecosystems (Fig. 5D) would entail restoring native vegetation on lands recently converted to agriculture (i.e., “secondary vegetation” in today’s landcover map, Fig. 1), without constraining allocations of remaining land, water, or energy. The results are in a wide range of food and energy outcomes, with a high likelihood of a substantial reduction in hydropower outcomes (95.2% in firm power and 9.4% mean generation).

Figure 5E shows the trans-sectoral policy pathways that aim to mitigate and distribute climate change impacts across sectors in an equitable manner. This includes intensification of existing croplands through efficient irrigation practices and constraints to agricultural expansion. In the hydropower sector, measures also include allowing for greater hydropower energy sales during the wet season. Integrated multisectoral strategies (Fig. 5E) result in some losses to firm power (up to −38%) in the hydropower sector but hedges those losses to smaller magnitudes than scenarios, like no climate adaptation (Fig. 5A) and only prioritizing ecosystems conservation (Fig. 5D) future. At the same time, it robustly increases food sector outcomes until the middle of the century for the SSP585 climate action scenarios and to the end of the century for the SSP126 climate action scenarios.

Actions at a single infrastructure scale have limited potential to mitigate impacts on catchment processes. Decision levers associated with on-site management such as the use of active reservoir storage (by increasing the permissible fluctuation in water levels to balance a greater variability in inflows) cannot compensate for the effect of catchment management. Demand allocation to the Chaglla Dam has the potential to increase energy performance at the project scale (Fig. 5C). Specifically, mean annual generation in Chaglla could be increased if there was higher energy demand from the grid, particularly in the wet season, so that water could be used for generation rather than being spilled. Such opportunities for selling more electricity during the wet season could offset some revenue losses from dry season generation and firm power, although the loss of grid services in the form of firm power delivery from Chaglla dam would need to be compensated for using other mechanisms.

Discussion

The environmental challenges facing the Andean headwater basins of the Amazon have been under-explored in the existing literature. These basins are not only key for providing food and income for thousands of rural and indigenous families, but could also play a key role in the regional transition to renewable energy (Fig. 1C)33,37. In this study, we focused on identifying and quantifying challenges and opportunities for climate adaptation, particularly the key trans-sectoral policy dimensions for water provisioning, energy generation, food production, and ecosystem conservation in the Andes.

Accelerating climate change impacts call for the comprehensive analysis of interdependencies across this WEFE nexus. From a single sector perspective, we found family farming at much greater climate risk than other sectors, notably hydropower. Irrigation offers an effective sectoral adaptation strategy for farmers in the region. However, without regulation and efficient irrigation practices, a widespread and unplanned adaptation of irrigation could hamper downstream hydropower generation, a major source of renewable energy. Similarly, targets for ecosystem restoration are obvious in a basin like the Huallaga, where extensive secondary vegetation is testament to recent deforestation. Crop losses from taking secondary crop lands out of production could be offset through intensification of remaining cropland, but this would imply similar losses for the hydropower sector.

Trade-offs between hydropower and agriculture are often framed with respect to the potential conflicts of upstream dam operations with downstream irrigation38. However, impacts are mutual, with a major dependence of dams on the management of upstream catchments. This is particularly evident for hydropower facilities with little regulation capacity which are vulnerable to small changes in the timing and magnitude of low flows. For the lowland Amazon, previous studies highlighted the risk that deforestation creates for hydropower outcomes9,10. However, these findings did not shed light on complex upstream-downstream trade-offs nor identify opportunities to actively influence hydropower outcomes through proactive catchment management.

Compared to the Amazonian lowlands, trade-offs are much harder to navigate in the more densely populated landscapes of the Andes-Amazon, since large-scale conservation might benefit hydropower and ecosystems but would also negatively impact many rural and indigenous families’ livelihoods and food security. Our results highlight the need to study linkages between catchment management, e.g., expansion of irrigation29,39, or restoration40,41,42, and downstream water users. Deploying high resolution models for natural and engineered systems is critical in this context, since trade-offs might be created by small mismatches in water supply and demand43,44.

Mainstreaming robust trans-sectoral assessments is critical for informing policy processes and aquatic and terrestrial conservation. Firstly, there is increasing momentum for strategic planning of hydropower in the region1,5,45,46, aiming to select portfolios of dam projects that meet future hydropower demands with the least impact on downstream rivers. Our results provide evidence that such approaches must incorporate possible impacts from both climate and upstream landuse change to judge future hydropower outcomes in a robust manner. Notably, projects that have reservoir storage may help accommodate adaptation in other sectors, and storage provision, rather than installed capacity or generation, could become a critical objective in strategic infrastructure planning47,48. However, reservoir projects can cause additional impacts, e.g., on sediment connectivity and freshwater biodiversity, compared to run-of-river schemes33,49, which needs to be accounted for in strategic planning. Secondly, Peru and other Andean countries have passed legislation to enable “payments for ecosystem service” schemes50,51. While there is often little coordination between sectors, the case of Chaglla highlights the need for science-informed dialogs between upstream and downstream actors and robust approaches to navigate overlooked trade-offs52 and enable the fair and sustainable use of basin resources.

This study also faces several limitations. Firstly, our scenario analysis assumes that landuse/management configurations remain constant throughout a specific climatological period. As a result, the metrics obtained reflect the steady state of the system after implementation but fail to capture the transient effects of transitioning from the current state, such as the water balance changes associated with afforestation. Secondly, our modeling of the food sector represents only potato yields. While Potatoes are the dominant crop in the region, this simplification overlooks the diversity of Andean agriculture systems. We also did not include adaptation strategies that involve incorporating crop switching (e.g., cereals, which might have lower water demands). Thirdly, the energy system model used in our study maintains the current seasonal patterns of electricity demand but future changes in electricity supply (e.g., from increasing photovoltaic installation) and demand (e.g., from electrified transportation or hydrogen generation) might reduce the seasonality in demand for hydropower. Lastly, our framework poses opportunities to integrate stakeholder needs. A more “bottom up” integration of local actors could help to refine current and future landuse scenarios (regarding cropping patterns, cropping practices), adaptation strategies (e.g., regarding crop switching or alternative livelihoods), and acceptable performance thresholds in a more participatory manner. It can leverage the herein presented numerical approaches to find equitable and robust adaptation strategies for Andean watersheds.

Our findings, as well as the study’s limitations, also highlight the need for further research to incorporate WEFE nexus dimensions into comprehensive studies at larger scales. Previous work has demonstrated trade-offs between WEFE objectives and opportunities to resolve them through optimizing landuse and management, e.g., through optimizing irrigation practices53 and cropping patterns54. Yet, our results highlight the need to represent sectoral objectives (such as hydropower) with high fidelity. This is because trade-offs might manifest through specific sectoral objectives such as firm power generation, which cannot be represented without a high temporal resolution and a detailed understanding of demands for sectoral services (e.g., food products or hydropower).

Overall, our results indicate that win-win situations across all sectors are elusive in the Andean Amazon. Nonetheless, robust planning with a trans-sectoral approach to watershed processes, climate, and engineered systems will be critical for the effective and equitable climate mitigation and efforts to restore terrestrial ecosystems in the Andes-Amazon and beyond.

Materials and methods

Our analysis uses a watershed model to quantify hydrologic processes and crop yield along with a model for hydropower operation. The coupled framework is informed by the latest IPCC-CMIP generation (CMIP6)55 climate scenarios. We then embed results in a multi-objective, bottom-up framework to identify management solutions that result in acceptable outcomes across sectors and across many landuse and climate futures.

Hydrologic model set-up

We model runoff and crop yields using the Soil Water Assessment Tool (SWAT)56. We set up SWAT using a daily time-step for simulation periods from 1985 to 2100 and report results aggregated over four periods: historic reference (1985–2014), near future (2019–2044), mid-century (2045–2074), and end-of-century (2075–2100) climatological periods. Using standard SWAT methods, we segmented the upper Huallaga Basin into 1371 hydrologic response units (HRU), delineated based on slope, soil, and landuse. Input data for modeling and validation of the model are shown in SI Table 1 and SI Table 2.

SWAT model calibration and validation

Daily discharge data are available from 2016–2020 at the Chaglla Dam site and from 2001 to 2015 at the Tingo María gauging station, 48 km downstream of Chaglla Dam, and from 2016–2020 at Higueras station, 110 km upstream of the dam (Fig. 1). Given the limited data availability at the dam site, we apply a space-for-time substitution approach for calibration, combining data from multiple stations to overcome the data scarcity in timeseries at Chaglla57 (more details in performance evaluation in SI Table 3). For this study, we used the SUFI 2 algorithm to automatically calibrate parameter values (see SI Methods 3.3) and perform a sensitivity and uncertainty analysis (SI Fig. 2). We first calibrate the model for Chaglla Dam (R2 = 0.57, Nash–Sutcliffe Efficiency (NSE) = 0.50) and Higueras (R2 = 0.60, NSE = 0.42) in the 2016–2018 period and for Tingo Maria in the 2001–2019 period (R2 = 0.70, NSE = 0.53). We also validate this model in all three sites, with Chaglla Dam (R2 = 0.70, NSE = 0.53) and Higueras (R2 = 0.60, NSE = 0.46) in the 2019–2020 period, and Tingo Maria in 2010–2015(R2 = 0.64, NSE = 0.58). The model achieves good performance (SI Fig. 1, for calibrated SWAT parameters in SI Table 4) with respect to three criteria, coefficient of determination (R2), NSE, and percent bias (PBIAS) in SI table 3.

Modeling crop yields

SWAT calculates crop yield from above-ground dry biomass on the day of harvesting and a harvest index58. Farmers in the Huallaga Basin grow a diverse range of crops for subsistence. The most widely grown and economically significant are rainfed potatoes50. We thus focus crop modeling on rainfed potatoes, reproducing the local cropping schedule with planting in May and October and harvesting in November and April50. For crops grown on land classified as “secondary vegetation” (see Figs. 1B and 4A) we multiplied modeled crop yields by a factor of 0.5, because not all secondary vegetation lands are used for cultivation. However, the available landuse data did not allow us to distinguish between patches of natural vegetation and crops within lands classified as secondary vegetation.

There are no production figures for potatoes specifically for the Huallaga Basin. Average annual yields for smallholder potato crops in the Peruvian-Andes have been reported to be in the range of 14–15 tons per hectare59. SWAT results in crop yields are measured as whole plant dry matter while observed potato yields are measured as fresh weight of the tubers. We therefore adjust modeled crop yields from SWAT by first multiplying them by the proportion of tuber mass to total biomass percentage (80%)60,61. We then convert this modeled yield of fresh potatoes to dry tuber biomass by using a conversion factor of 0.262,63. We find that SWAT predictions are in good agreement with general observations (observed: 14–15 t/ha/yr59, modeled: 20 t/ha/yr).

Climate and landuse scenarios

We generate landuse/management scenarios from Peru’s ecosystems map64 using biophysical suitability for agriculture and natural vegetation from their observed current distribution. We use our calibrated SWAT model to quantify agricultural outcomes and inflows to the Chaglla reservoir for those landuse scenarios under many climate futures. Future climate conditions are derived for the period 2019–2100 from the latest CMIP6. To represent an envelope of climate action trajectories, we use SSP (Shared Socioeconomic Pathways) scenarios including both a low (SSP126) and a high (SSP585) climate change pathway in our analysis55.

Climate change scenarios

We use a two-stage procedure for selecting general circulation models (GCMs) and downscaling methods. Firstly, we benchmark 11 GCMs (SI Table 5) from CMIP6 by how well they represented observed historic precipitation records with respect to six indicators (SI Table 6 & SI Methods 3.1). Historic precipitation and air temperature data are obtained from Peru’s national gridded meteorological dataset, PISCO65, for the years 1980–2020. We calculate an unweighted compound rank for each model across the six indicators and select the three best performing models (MIROC6, MPI-ESM1-HR, and MPI-ESM1-LR) (SI Fig. 3). We then downscale climate data for these three models using the PISCO data as local reference. Based on these outputs, we select the K-nearest neighbor resampling approach from a number of well-documented downscaling approaches (SI Fig. 4 and SI Methods 3.2)66. This results in a total of 18 future climatologies of daily precipitation and temperature from 2019 to 2100 (one downscaling method for three GCMs times two SSPs, and times three prospective periods: 2019–2044, 2045–2074 and 2075–2100) and four historic back casts (one downscaling method for three GCMs, and the historic reference climatology in years 1985–2014).

Landuse and management scenarios

For this study, we developed a total of 15 landuse/management futures, representing different ways water and land resources could be managed in the future. We consider four types of landuse change (S0: no change/current, S1: full expansion of agriculture, S2: limited expansion of agriculture, and S3: partial restoration) (see Fig. 4A for more details). For S1, we determine slope and altitude limits of current agriculture and assume that all land below those thresholds is converted to cropland. For S2, we assume that current croplands remain in production and areas of secondary vegetation, i.e., degraded forest with partial agricultural use, are fully converted to agriculture. S3 maintains current croplands but restores secondary vegetation to natural vegetation. We then combine each land-cover scenario with five agricultural practices (a: rainfed, b: rainfed and terracing, c: irrigated, d: irrigated and terraced, e: efficiently irrigated and terraced). Irrigation is modeled under the assumption of an irrigation efficiency of 0.4, and irrigation is set to reach zero water stress for crops if water is available. For efficient irrigation, we increase irrigation efficiency to 0.8, or approximately the average between drip and sprinkler irrigation systems in Latin America67 and allowed a water deficit of 30%. For all irrigation practices we set the environmental flow (eflow) threshold to 10% of the natural flow or 4 m3/s (whichever is higher, where 4 m3/s is the eflow mandate for Chaglla dam). Note that future scenarios focus only on potatoes as the dominant crop and do not consider crop switching, i.e., that farmers could move to other, more water efficient crops in response to climate change. Future landuse scenarios also maintain the current cropping calendar (as described above).

Overall, we model a total of 16 scenarios, including the baseline and 15 combinations of S1, S2, and S3 with landuse practices a–e. In this nomenclature, for instance, scenario S1-c indicates both an expansion of agriculture and intensification through irrigation. Note that we refer to the baseline as S0-a.

Ecosystem impact

We represent the impact of different landuse/management and climate scenarios on ecosystems and the services they provide using two indicators: (1) the extent of remaining native vegetation (including montane forest, Andean grasslands, and shrublands) and (2) sediment retention. For the former, we consider all land-use classes other than croplands and secondary vegetation (Fig. 1). Ecosystem services can be measures in many other domains (e.g., carbon storage and biodiversity); however, for this study, we select sediment retention to represent ecosystem services other than hydrologic regulation (which is represented by the SWAT model). The reason for this is that sediment retention links to both food security (via retention of soils) and energy (via avoided storage loss and maintenance for downstream hydropower). For details on erosion and sediment retention modeling, (please see SI Methods 3.5 and SI Table 7).

Hydropower services modeling

We model hydropower using a dynamic daily timestep simulation of the water balance and reservoir allocations68. The operations model considers the effects of (i) inflows (reflecting upstream hydrological conditions combining prospective changes due to climate and landuse/management conditions), (ii) reservoir management (reflecting on-site operational decisions, constraints, and targets), and (iii) demand-side considerations (reflecting the effect of grid-scale energy and power commitments and market competition). The model functioning and relevant parameters are described in more detail in Sl Methods 3.4. The model generates information on both mean hydroelectric generation and firm power availability. Compared to the observation data, the model shows good performance (R2 = 0.91, PBIAS = 1.7 %, SI Fig. 6) For more details on the hydropower operation model refer to SI Methods 3.4, SI Table 8, and SI Fig. 5.