Research

Abstract: Monitoring systems for disaster prevention are costly, and measuring benefits is difficult when monitoring effort is endogenous. We provide the first causal estimate of one such system's impact using three decades of desert locust monitoring data. We document conflict-induced interruptions to monitoring in remote breeding areas, reconstruct how infestations spread to populated areas, and show that exposure to locust swarms around birth decreases child height-for-age, increasing stunting risk by over 7 percentage points. Eliminating the locust monitoring system would induce annual losses of US$25 billion, implying a benefit-cost ratio between 160:1 and 680:1 from child nutrition benefits alone.

Abstract: As biodiversity protection and restoration goals grow, so must impact assessments that can guide conservation efforts on the ground. This is especially true in agricultural areas, where competing pressures on use of the land leave little room for manoeuvre for conservation actions. As much as they are needed, biodiversity impact assessments are limited by the difficulty of tracking wildlife and associated ecosystem services at scale. This Special Collection features a series of original articles, which contribute broadly to the topic of monitoring biodiversity dynamics in agroecosystems. They are rooted in Southern Ontario, a densely populated and intensively farmed region with a rich ecological history, and programs like the Alternative Land Use Services (ALUS) that support on-farm conservation practices, including habitat restoration on agriculturally marginal field areas. Contributions fall under two broad themes: (1) impact of habitat management and agricultural practices on biodiversity, with a particular focus on regionally-important taxa and ecosystems (arthropods in grasslands, aquatic food webs in riparian ecosystems); and (2) challenges and opportunities for monitoring arthropods across agricultural landscapes. The collection illustrates the value of partnerships between regional organizations and researchers for producing locally-applicable insights with relevance to the broader literature.

Abstract: Global action to slow biodiversity loss is critically needed but comes at substantial cost. In this article, we assess the effectiveness of local ballot measures for land-based conservation projects in the United States as a way to increase bird abundance, a key conservation indicator. Using a citizen science dataset of bird observations, we employ an abundance model to estimate relative abundance conditional on observer effort and exploit a sharp discontinuity in land-based conservation funding at the vote threshold to estimate plausibly causal effects. We find that an approved ballot measure has modest but significant impacts that accrue over ten years.

Genetically modified (GM) crops have been adopted by some of the world’s leading agricultural nations, but the full extent of their environmental impact remains largely unknown. Although concerns regarding the direct environmental effects of GM crops have declined, GM crops have led to indirect changes in agricultural practices, including pesticide use, agricultural expansion, and cropping patterns, with profound environmental implications. Recent studies paint a nuanced picture of these environmental impacts, with mixed effects of GM crop adoption on biodiversity, deforestation, and human health that vary with the GM trait and geographic scale. New GM or gene-edited crops with different traits would likely have different environmental and human health impacts.

Abstract: We propose Hedgementation: a new benchmark to evaluate machine learning models for hedgerow mapping from remote sensing data at country scale and 10m2 spatial resolution. We combine and harmonize multiple remote sensing data products and ground truth labels sourced from a hedgerow inventory in France. We measure the ability of three baseline models to generalize across spatial distance, and across climatic zones, a more explicitly challenging task. Our benchmark tests both supervised and self-supervised learning approaches for remote sensing, applied to tracking fine-scale features of high agricultural importance.

Abstract: Detecting temporal shifts in plant flowering times is of increasing importance in a context of climate change, with applications in plant ecology, but also health, agriculture, and ecosystem management. However, scaling up plant-level monitoring is cost prohibitive, and flowering transitions are complex and difficult to model. We develop two sets of approaches to detect the onset of flowering at large-scale and high-resolution. Using fine grain temperature data with domain knowledge based features and traditional machine learning models provides the best performance. Using satellite data, with deep learning to deal with high dimensionality and transfer learning to overcome ground truth label sparsity, is a challenging but promising approach, as it reaches good performance with more systematically available data.

Monitoring shifts in flowering phenology using satellite imagery and deep learning. (with Mathias Lécuyer)

Shifts in the timing of plant flowering are a key signal of an ecosystem’s response to environmental change. Consequences are far-reaching: the timing of plant flowering affects the synchronization between plants and pollinators; it influences the exposure of crops to weather extremes during flowering, the most sensitive stage of development, with implications for later yield. A consistent and large-scale measure tracking the timing of flowering across years would be of great use for assessing risks of plant-pollinator desynchronization or risks of subpar crop yields. Current efforts, ranging from local to regional scales, do not quite achieve this goal: ground observations, while expanding, remain limited in their spatial and temporal coverage; process-based or statistical models are not flexible enough to capture local acclimation or adaptation to environmental factors; and remotely sensed estimates of plant development stages are often disconnected from the ground. We propose a method that combines spectral reflectance data from satellite imagery, with a deep learning model, to create a large-scale, high-resolution proxy for the timing of flowering. We illustrate this approach with two use cases. First, over cropland in Illinois and Iowa, we construct a proxy measure of the evolution of crop flowering—silking for corn, flowering for soybean—across the growing season. Second, over the Eastern US, we construct a proxy measure of the onset of spring flowering, which reflects the phenology of early-flowering, temperature-sensitive species.