Publications

Index insurance, while heralded as a potential solution to alleviate poverty and food insecurity among agricultural households, has its own set of challenges, notably basis risk. Basis risk is the discrepancy between the insurance payout and losses incurred, posing a significant deterrent to the adoption of index insurance. This study investigates the effect of basis risk training on the uptake of rainfall index insurance among rural farmers in Senegal, employing an educational simulation game as a pedagogical tool. Our results suggest that a lack of transparency and a misunderstanding of basis risk contribute to low participation and a lack of trust in index insurance.

Evaluating sustained community engagement in co-management is critical for designing durable governance with conservation potential, yet such persistence remains rarely assessed. We analyze 750 co-management initiatives established under Chile's Territorial Use Rights for Fishing policy (1998–2021), examining persistence across conditions theorized to shape collective action in social-ecological systems. Survival analysis shows that initiatives had a 75% probability of persisting beyond 15 years. Abandonment risk declined nonlinearly with initial exploitable abundance and increased with monitoring distances. Exploring combinations of variables through cluster analysis reveals that, while initiatives starting with high abundance showed the highest persistence, some initiatives with lower initial abundance also endured at comparable rates when combined with shorter monitoring distances, proximity to large markets, higher poverty, and stronger upwelling. These findings suggest diverse pathways to sustained community engagement in co-management, generate hypotheses for future research, and show how tracking persistence can inform strategies for durable and equitable conservation.

The location of a wind or solar project is one of the most consequential decisions a renewable energy developer can make when planning a new project, as it shapes numerous aspects of project economics and local impacts on ecosystems and host communities. This study examines key factors influencing the siting of utility-scale solar and wind projects in the contiguous United States using multiple statistical and machine learning models and a robust sampling approach. We find that while proximity to transmission lines or substations and existing projects is critical for both technologies, solar locations are more flexible and are shaped by a greater diversity of factors including higher population density, greater road accessibility, and lower ecological impact. In contrast, wind locations are primarily driven by wind resource quality and agricultural land use. There are notable regional nuances in these national trends, with some variables like population density having greater effects in the Midwest. While individual models emphasize different predictors and regional patterns, the ensemble reveals consistent tendencies that increase confidence in interpretation. Lastly, the probabilities of wind and solar being sited in disadvantaged (DAC) and non-DAC areas vary across regions, suggesting that the Midwest and Northeast in particular may see disproportionately less development pressure from solar and wind projects in disadvantaged community census tracts.

Agricultural field boundaries define the fundamental spatial units for measuring agricultural production, management practices, and farmer behavior. However, mapping agricultural field boundaries in smallholder systems remains one of the most challenging problems in remote sensing due to their small parcel sizes, irregular shapes, and lack of labeled training data. Deep learning has improved boundary delineation in large, uniform farming systems, yet these approaches depend on extensive local annotation and fail to generalize where data are scarce. Foundation models offer a potential solution by transferring knowledge from large-scale pretraining, but their real-world performance in fine-grained agricultural settings remains unclear. This study evaluates the Segment Anything Model (SAM) to benchmark its intrinsic zero-shot capacity for delineating field boundaries in a highly fragmented landscape (<0.1 ha). Zero-shot evaluation is critical because it defines the practical lower bound for application in regions where any task-specific annotation or fine-tuning is infeasible. Using high-resolution SkySat imagery and over 8000 manually digitized reference fields in Bihar, India, we systematically quantify how inference strategies such as model checkpoints, input tile sizes, multi-temporal observations, and edge enhancement influence segmentation performance. Without any fine-tuning, SAM identifies 57% of reference fields Rtwowith a mean Intersection over Union (IoU) value of 0.73, and performance further improves when predictions from multiple model checkpoints, image sizes, and temporal observations are combined. While smallholder fields (<0.1 ha) are often dismissed as noise in large-scale analyses, this study demonstrates that foundation models can recover meaningful boundaries even in such fragmented landscapes, advancing the goal of globally inclusive agricultural monitoring. The results establish a quantitative baseline for zero-shot generalization and provide specific insights into the optimal inference strategies of foundation models for Earth observation.

In its April Proposed Amendments to the cap-and-invest (C&I) program, CARB introduces the Manufacturing Decarbonization Inventive (MDI) program, which creates a new tranche of allowances in addition to the official statewide cap. These MDI allowances may be granted to industrial compliance entities if they undergo specific clean investments and can be used to meet emissions obligations under C&I. MDI has state revenue implications. By creating new allowances above the cap that are freely available for industrial compliance entities, the MDI lowers demand for CARB-auctioned allowances. With recently auctioned allowances selling at the price floor, this suggests future unsold auctioned allowances. Auction revenue funds the Greenhouse Gas Reduction Fund (GGRF) and California Climate Credit (CCC).

We examined GGRF and CCC implications of the MDI under different MDI usage scenarios. If fully utilized over the next four years, the MDI program could lower C&I auction revenue by $4 billion. For comparison, total auction revenue in FY24-25 was $5.8 billion, with $3.4 billion going to GGRF and $2.4 billion to CCC. If this baseline revenue and spending share were unchanged during '27-'30, a fully utilized MDI would lower GGRF funds by $2.3 billion and CCC by $1.7 billion, a 17% reduction for each. Actual fiscal impact will depend on MDI adoption, which depends on macroeconomic conditions and program implementation details. One can look to recent California industrial emissions changes to understand potential future MDI usage. For example, if the '22-'23 drop in industrial emissions were to continue in '27-'30 and granted MDI allowances, that would be consistent with a 25% MDI usage scenario, with a $1 billion drop in auction revenue during '27-'30. This is likely an underestimate as clean energy adoption should accelerate under the MDI. Finally, the MDI is likely to push allowance prices, as reflected in the secondary market, below the price floor. This is because MDI allowances can be sold in the secondary market, outside the price floor maintained in CARB auctions.

Growing demand for food coupled with climate commitments to reduce emissions will result in more land development for agriculture and renewable energy. Simultaneously, conserving land for biodiversity and nature’s contributions to people (NCP) is imperative for achieving international climate, sustainable development, and biodiversity goals. Meeting these interconnected objectives requires efficient land allocation across sectors. Here, we present a flexible, multiple-objective framework for strategically allocating land to mitigate threats to biodiversity and NCP under climate change while supporting development. Application of this framework at a global scale through country-level targets shows that if future development is planned without consideration of nature, demands for land could impact nearly 1 million km2 of high-priority conservation areas. Multi-sector planning can mitigate potential conflict, reducing carbon loss and species exposure. Our findings underscore the need to conserve critical areas for nature, reduce land demand for food and energy, and intentionally coordinate land use across sectors.

Farm size is a key characteristic of agricultural systems, closely linked to farming and land management practices. However, data on farm size are often difficult to obtain, especially in smallholder-dominated regions, and where available, they are rarely spatially explicit. This study examines the use of field size as a proxy for farm size, aiming to identify field size thresholds that can distinguish farms of different sizes. Using household-level data from Zambia’s Crop Forecast Survey, we applied both continuous and categorical methods to quantify the field-to-farm size relationship. Non-parametric Theil–Sen regression revealed a generally linear relationship between log-transformed field and farm sizes, particularly at the survey enumeration area scale, where farm size is proportional to field size (slope ≈ 1) and field size alone predicts farm size with a median bias of 0.017 ha. A Gaussian Naive Bayes classifier was also developed to assign fields to three categories: very small (A: < 2 ha), small (B: 2–5 ha), and medium (C: 5–20 ha), and a coarser two-class scheme (S: < 5 ha; M: 5–20 ha) that has been applied in previous studies. The Gaussian model identified field size thresholds at 0.58–0.6 ha (A–B) and 1.18–1.69 ha (B–C), depending on the prior farm size distribution. For the S–M scheme, thresholds ranged from 0.96–1.4 ha. Using these thresholds, national classification achieved F1 scores of 0.78, 0.59, and 0.77 for A, B, and C, respectively, with B more prone to misclassification. The two-class model achieved 0.87 and 0.88 for S and M, indicating more robust classification performance under a coarser scheme. Kernel density analysis shows that field-size distributions within farm categories and the separability of these distributions, vary across regions, with the underlying factors responsible for these differences remaining an open question for future research. These results demonstrate the potential of field size as a spatial proxy for farm size to support remote sensing based farming system classification in data-scarce regions.

In applications across agriculture, ecology, and human development, machine learning with satellite imagery (SatML) is limited by the sparsity of labeled training data. While satellite data cover the globe, labeled training datasets for SatML are often small, spatially clustered, and collected for other purposes (e.g., administrative surveys or field measurements). Despite the pervasiveness of this issue in practice, past SatML research has largely focused on new model architectures and training algorithms to handle scarce training data, rather than modeling data conditions directly. This leaves scientists and policymakers who wish to use SatML for large-scale monitoring uncertain about whether and how to collect additional data to maximize performance. Here, we present the first problem formulation for the optimization of spatial training data in the presence of heterogeneous data collection costs and realistic budget constraints, as well as novel methods for addressing this problem. In experiments simulating different problem settings across three continents and four tasks, our strategies reveal substantial gains from sample optimization. Further experiments delineate settings for which optimized sampling is particularly effective. The problem formulation and methods we introduce are designed to generalize across application domains for SatML; we put special emphasis on a specific problem setting where our coauthors can immediately use our findings to augment clustered agricultural surveys for SatML monitoring in Togo.

Carbon finance initiatives such as Reducing Deforestation and Forest Degradation (REDD+), designed to mitigate climate change, offer an opportunity to also protect biodiversity. However, managing forests to store and sequester carbon does not necessarily conserve biodiversity. We evaluated the biodiversity co-benefits of the Gola-REDD+ initiative in the tropical forests of Sierra Leone, using bioacoustics and DNA metabarcoding under a quasi-experimental study design. We used soundscape saturation (SS) as a measure of vocalizing diversity, and e-DNA arthropod community as a complementary measure of biodiversity to examine whether a Gola-REDD+ financed protected area (Treatment-PA) conserved biodiversity more than (1) a multiuse community land (Control-CL) and (2) a PA without REDD+ finance (Control-PA). We found that REDD+ financing is associated with additional biodiversity co-benefits in the Treatment-PA compared to both control areas. Our study makes three key contributions. First, we provide concrete evidence on a carbon finance (REDD+) project's effectiveness in conserving faunal diversity while sequestering carbon. Second, we present a gold-standard causal inference study design for evaluating biodiversity co-benefits of conservation strategies. Third, we highlight the role of conservation technologies like bioacoustics and DNA metabarcoding in informing conservation policy.

Weather conditions can drastically alter the state of crops and rangelands and, in turn, impact the incomes and food security of individuals worldwide. Satellite-based remote sensing offers an effective way to monitor vegetation and climate variables on regional and global scales. The annual peak normalized difference vegetation index (NDVI), derived from satellite observations, is closely associated with crop development, rangeland biomass, and vegetation growth. Although various machine learning methods have been developed to forecast NDVI over short time ranges, such as one-month-ahead predictions, long-term forecasting approaches, such as one-year-ahead predictions of vegetation conditions, are not yet available. To fill this gap, we develop a two-phase machine learning model to forecast the one-year-ahead peak NDVI over high-resolution grids, using the Four Corners region of the southwestern United States as a testbed. In phase one, we identify informative climate attributes, including precipitation and maximum vapor pressure deficit, and develop the generalized parallel Gaussian process that captures the relationship between climate attributes and NDVI. In phase two, we forecast these climate attributes using historical data at least one year before the NDVI prediction month, which then serve as inputs to forecast the peak NDVI at each spatial grid. We develop open-source data and tools that outperform alternative methods for both gross NDVI and grid-based NDVI one-year forecasts, providing information that can help farmers and ranchers make actionable plans a year in advance.

Understanding food demand patterns is critical for developing countries in Sub-Saharan Africa (SSA), which are facing increasing urbanization, intensifying climate change, and rapid transitions in food production systems. This study models food demand in Zambia and Tanzania using data from the Living Conditions Monitoring Survey (LCMS) (2015 and 2022) and the Living Standards Measurement Survey (LSMS) (2015 and 2021). This study applies the Quadratic Almost Ideal Demand System (QUAIDS) model to analyze demand behavior and calculate expenditure and price elasticities of both food and nutrition. Urbanization is linked to significant shifts in dietary preferences and greater sensitivity to the prices of non-staple, nutrient-dense foods, reflecting the diversifying food demand in urbanized regions. Households allocate more resources to calories with increases in income, but the rate of increase slows as basic caloric needs are met. Conversely, expenditure nutrition elasticity for Vitamin B12 is particularly high, as rising incomes enable households to spend more on animal-based foods such as meat, fish, eggs, and dairy. Once basic caloric needs are satisfied, additional income is more likely to be directed toward diversifying the diet with higher-value, nutrient-dense foods rather than increasing carbohydrate consumption. The positive expenditure elasticities for both food consumption and nutrient intake in both countries emphasize that increasing household income remains a crucial strategy for improving food security.

Increasing shares of wind and solar generation, together with rising electricity demand, introduce growing uncertainty into power system operations. Accurate day-ahead forecasts of electricity demand and renewable generation are essential for system operators to coordinate electricity markets and maintain reliability at low cost. Here, we show that forecasting based on joint probability distributions of demand and renewable supply can substantially improve system-level forecasting performance using publicly available weather data. We develop multiple day-ahead forecasting models that combine machine learning methods to identify relevant weather variables with probabilistic approaches to quantify forecast uncertainty, and we evaluate these models using proper scoring rules. Applied to the three zones of the California Independent System Operator, the best-performing model improves forecast skill by 25% relative to current benchmarks. We further show that forecasts based on joint probability distributions enable a more effective allocation of operating reserves than conventional deterministic approaches, highlighting the potential of probabilistic machine learning to enhance market efficiency and grid stability in increasingly decarbonized power systems.

Large-scale deployment of wind and solar power is essential for climate change mitigation but requires substantial land and offshore area due to their relatively low installed capacity densities. However, wide variation in installed capacity density estimates creates major uncertainty in projecting future spatial requirements. Here, we apply multiple spatial methods to satellite imagery to estimate installed capacity densities for 60 onshore wind, 24 offshore wind, and 54 utility-scale solar PV plants (including 36 fixed-tilt and 18 single-axis tracking) across six countries: Australia, China, Germany, India, United Kingdom, and the United States. For wind power plants, mean estimates of installed capacity density using the 8D buffer method (eight times the rotor diameter), the convex hull method with a 5D buffer, and the median Voronoi polygon (bounded by an 8D buffer) method are similar (4.0, 4.4, and 4.9 MW km⁻² for onshore wind and 4.6, 5.2, and 5.4 MW km⁻² for offshore wind, respectively) but the 5D buffer method provides a relatively larger estimate because of unaccounted areas within power plant boundaries. Single-axis tracking solar PV power plants have an approximately 33% lower mean installed capacity density than fixed-tilt power plants, with estimates of 36 and 29 MW km⁻² (without and with a 50 m buffer, respectively), compared to 54 and 43 MW km⁻² for fixed-tilt systems. Estimates vary more within countries and across methods than between countries, underscoring the importance of methods and assumptions in estimating spatial requirements for large-scale wind and solar deployment.

Rigorous evidence detailing the myriad interconnections between humans and ecosystems will be critical to slow the loss of biodiversity. Effective conservation interventions will depend upon a detailed understanding of the benefits that biodiversity provides to people, the ways that human activities drive biodiversity decline, and the potential for conservation policies to stem this decline. Although hundreds of papers explore these relationships, a careful review of this literature shows that the overwhelming majority of studies fall short of documenting causal relationships with biodiversity per se. However, a combination of data and methodological advances has led to rapid growth of quasi-experimental analyses that are advancing our understanding of human interactions with biodiversity in three domains. First, economists have provided valuable insights into the causes of biodiversity loss, which complement traditional ecological experiments by better reflecting real-world conditions. Second, quasi-experimental studies have begun to identify which policy interventions, in what contexts, have slowed biodiversity loss. Finally, recent quasi-experimental studies have shown that the loss of species can impose extremely large costs on humanity but that these costs vary widely depending upon the species and opportunities for human adaptation.

Climate change and anthropogenic activities threaten biodiversity and ecosystem services. Climate-smart conservation plans address these challenges by ensuring protection of some climate-resilient areas. However, integrating climate change in the design of conservation plans is often deemed too expensive, as it may require larger networks or protecting more costly sites from a conservation perspective. Using mangroves as a case study, we evaluate the efficiency of protecting mangroves in climate-smart versus climate-naïve reserve networks. We find that climate-smart conservation plans could provide sizable benefits (13.3%) for relatively moderate increases in protected area (+7.3%). Moreover, transboundary plans, involving cooperation among countries, require less area and protect more climate-resilient mangroves than nation-by-nation plans. Implementing these strategies would improve the current protected area network for mangroves, which currently has poor climate resilience. Our methodology could potentially be tested on other ecosystems, assuming sufficient information exists regarding their distribution, biodiversity, and resilience to climate change.

The 2025 U.S. Clean Competition Act (CCA) is designed to address the twin challenges of accelerating industrial decarbonization and maintaining U.S. industrial competitiveness. It does so by pairing a U.S. domestic carbon performance fee applied to dirtier than average U.S. firms with a carbon import tariff in carbon intensive, trade-exposed (CITE) sectors. The CCA also contains “climate club” provisions that waive carbon tariffs for trade partners implementing comparable domestic climate policies. This policy brief analyzes the CCA’s economic and climate impacts in the U.S. and around the world. It uses a general equilibrium global trade model designed for analyzing climate and trade policies, calibrated to disaggregated sectoral data. We analyze the initial year features of the CCA as applied to the aluminum, iron and steel, cement, chemicals, glass, nitrogen-based fertilizers, paper and pulp sectors. Our modeling results suggest that the CCA can jointly achieve U.S. industrial decarbonization and competitiveness goals. Moreover, the climate club provisions in the CCA could serve as a foundation for large-scale global GHG reductions. For both unilateral and multilateral CCA results, the domestic performance fee is critical: without the domestic fee, CCA’s economic and climate benefits are significantly dampened.

Forests are central to climate, biodiversity, and development goals, but effective monitoring and evaluation of their contributions depends on reliable data. Satellite-derived global forest cover and change datasets (GFDs) are widely used to address this need. However, differences in resolution, forest definition, and methodology challenge their use in research and policy, yet how GFD differences affect key forest-related estimates remains poorly understood. Here, we quantify global area-based spatial congruence among 10 GFDs and test their influence on three national policy-relevant estimates: carbon accounting, forest-poverty mapping, and biodiversity habitat. We find only 26% spatial congruence among GFDs at native resolution. This low congruence translates to an order-of-magnitude difference in national case study indicator estimates. We demonstrate that GFD selection fundamentally shapes monitoring and evaluation outcomes, particularly in biomes with fragmented or sparse tree cover. We provide a decision-support framework to guide GFD selection according to different forest-related science and policy applications.

Heat extremes pose substantial health risks during pregnancy and early childhood. High humidity exacerbates heat strain, but its long-term effects on health remain poorly understood. We compare the effect of prenatal exposure to extreme humid heat versus heat alone on child growth in South Asia, where high rates of child stunting meet rapidly accelerating hot-humid extremes. After adjusting for sociodemographic, seasonal, and spatial confounders, we use within-community variation in children’s ages to isolate the impact of prenatal exposures. We find that hot-humid exposures are much more detrimental to health than hot temperatures alone, with the potential to increase stunting in South Asia by over 3 million children by 2050. These findings underscore the importance of accounting for humidity when estimating and localizing climate change impacts.

The world relies on analyses by the United Nations-facilitated Integrated Food Security Phase Classification (IPC) to identify where populations are food insecure and to quantify the severity of these crises. IPC sub-national analyses are designed to be comparable over space and time in the more than 30 countries in which the IPC operates. Humanitarian agencies appear to regard these findings as authoritative and comparable, and as of 2024, used IPC analyses to guide more than six billion dollars of annual aid allocations. We study the consistency and comparability of IPC food insecurity analyses across time and space. Drawing on 1,849 IPC subnational analyses covering 742 million people from fifteen countries between 2019 and 2023, we show that IPC analyses face significant challenges related to data availability and food security measurement, resulting from underlying food security indicators that are often discordant. We find that the vast majority of IPC subnational analyses are consistent with IPC technical guidance, but that this guidance permits a wide range of classifications for a given set of food security indicators. We also find evidence that IPC subnational analyses vary in the way they use food security data, often weighing food security indicators differently across locations. While variation in how analyses use food security indicators can plausibly reflect varying contextual factors across countries, we find evidence that analyses weight indicators differently across time for the same location. Finally, we show that analyses do not treat closely correlated food security indicators as substitutes, suggesting inconsistency in the treatment of food security indicators across analyses. We discuss implications of these findings for policy and for the interpretation and use of IPC analyses by researchers and policymakers.

The Integrated Food Security Phase Classification (IPC) system is the official global method for classifying food insecurity. As of 2023, international agencies and governments use IPC analyses to allocate more than US$6 billion of humanitarian assistance annually. Here we evaluate data from approximately 1 billion people in more than 10,000 IPC subnational analyses conducted between 2017 and 2023. We find that IPC estimates understate the extent and severity of crises. Our primary estimates indicate that IPC subnational analyses underestimate the number of acutely hungry people in the world, missing approximately one in five. We find evidence of under-classification around the IPC threshold that determines whether an area is classified as ‘stressed’ or ‘in crisis’—a threshold meant to trigger deployment of humanitarian resources. Contrary to widely held assumptions, our findings suggest that IPC analyses are conservative; the prevalence and severity of acute hunger is probably considerably higher than global estimates indicate.