About

Maggi Kelly is a Professor and Cooperative Extension Specialist in the Department of Environmental Science, Policy and Management at UC Berkeley. She holds a PhD in Geography from the University of Colorado, an MA in Geography from the University of North Carolina, Chapel Hill, and a BA in Geography from the University of California, Berkeley. Her research focuses on studying the drivers, patterns, and consequences of environmental change across California’s complex, socially diverse, and dynamic landscapes. Kelly's technological expertise in GIScience includes remote sensing analysis, object-based image analysis, geospatial modeling, lidar analysis, participatory webGIS, and field-based monitoring. She often employs a suite of these tools to address environmental problems and engage with stakeholders. Her work encompasses environmental change, vegetation dynamics, fire regimes, coastal marsh resilience, and the use of drones in research and extension activities.

Research topics

• Environmental science
• Remote sensing
• Geography
• Computer science
• Ecology

Selected publications

• Mapping Orchard Trees from Uav Imagery Through One Growing Season: A Comparison Between Obia-Based and Three CNN-Based Object Detection Methods

  Preprints.org · 2025-06-13 · 4 citations

  preprintOpen access1st authorCorresponding

  Extracting the irregular and complex shapes of individual tree crowns from high-resolution imagery can play a crucial role in many applications, including precision agriculture. We evaluated three CNN models - MASK R-CNN, YOLOv3, and SAM - and compared their tree crown results with OBIA-based reference datasets from UAV imagery for seven dates across one growing season. We found that YOLOv3 performed poorly across all dates; both MASK R-CNN and SAM performed well in May, June, September, and November (Precision, Recall and F1 scores over 0.79). All models struggled in the early season imagery (e.g., March). MASK R-CNN outperformed other models in August (when there was smoke haze) and December (showing end of season red leaf senescence). SAM was the fastest model, and as it required no training, it could cover more area in less time; MASK R-CNN was very accurate and customizable. In this paper, we aimed to contribute insight into which CNN model offers the best balance of accuracy and ease of implementation for orchard management tasks. We also evaluated their applicability within one software ecosystem, ESRI ArcGIS Pro, and showed how such an approach offers users a streamlined, efficient way to detect objects in high resolution UAV imagery.

  PublisherOA PDFDOI

• Mapping Orchard Trees from UAV Imagery Through One Growing Season: A Comparison Between OBIA-Based and Three CNN-Based Object Detection Methods

  Drones · 2025-08-22 · 1 citations

  articleOpen access1st authorCorresponding

  Extracting the shapes of individual tree crowns from high-resolution imagery can play a crucial role in many applications, including precision agriculture. We evaluated three CNN models—MASK R-CNN, YOLOv3, and SAM—and compared their tree crown results with OBIA-based reference datasets from UAV imagery for seven dates across one growing season. We found that YOLOv3 performed poorly across all dates; both MASK R-CNN and SAM performed well in May, June, September, and November (precision, recall, and F1 scores over 0.79). All models struggled in the early season imagery (e.g., March). MASK R-CNN outperformed other models in August (when there was smoke haze) and December (showing end-of-season variation in leaf color). SAM was the fastest model, and, as it required no training, it could cover more area in less time; MASK R-CNN was very accurate and customizable. In this paper, we aimed to contribute insight into which CNN model offers the best balance of accuracy and ease of implementation for orchard management tasks. We also evaluated its applicability within one software ecosystem, ESRI ArcGIS Pro, and showed how such an approach offers users a streamlined efficient way to detect objects in high-resolution UAV imagery.

  PublisherOA PDFDOI

• Object-Based Image Analysis

  2024-10-29 · 3 citations

  book-chapter

  This chapter explains OBIA (Object Based Image Analysis) methods as used in the geospatial domain and elsewhere. We will start from the quest to partitioning geospatial data into meaningful image-objects, and the needs and possibilities to assessing their characteristics through spatial, spectral and temporal scale. At its most fundamental level, OBIA requires image segmentation, attribution, classification and the ability to query and link individual objects (aka segments) in space and time. We will elucidate the evolution of this approach, its relatively short history and its older origins. Instead of a comprehensive state-of-the-art analysis we refer to the key literature and try to summarize the core concepts for the reader in an understandable way, with a particular emphasis on a common nomenclature, definitions, and reporting procedures. Ultimately, we will ask where this development will lead to in terms of applications, research questions and needs in education, training and professional workforce development, and we conclude with the main advances and recommendations for future work.

  PublisherDOI

• Using Sentinel-2 Imagery to Measure Spatiotemporal Changes and Recovery across Three Adjacent Grasslands with Different Fire Histories

  Remote Sensing · 2024-06-19 · 1 citations

  articleOpen accessSenior authorCorresponding

  As a result of the advocacy of Indigenous communities and increasing evidence of the ecological importance of fire, California has invested in the restoration of intentional burning (the practice of deliberately lighting low-severity fires) in an effort to reduce the occurrence and severity of wildfires. Recognizing the growing need to monitor the impacts of these smaller, low-severity fires, we leveraged Sentinel-2 imagery to reveal important inter- and intra-annual variation in grasslands before and after fires. Specifically, we explored three methodological approaches: (1) the complete time series of the normalized burn ratio (NBR), (2) annual summary metrics (mean, fifth percentile, and amplitude of NBR), and (3) maps depicting spatial patterns in these annual NBR metrics before and after fire. We also used a classification of pre-fire vegetation to stratify these analyses by three dominant vegetation cover types (grasses, shrubs, and trees). We applied these methods to a unique study area in which three adjacent grasslands had diverging fire histories and showed how grassland recovery from a low-severity intentional burn and a high-severity wildfire differed both from each other and from a reference site with no recent fire. On the low-severity intentional burn site, our results showed that the annual NBR metrics recovered to pre-fire values within one year, and that regular intentional burning on the site was promoting greater annual growth of both grass and shrub species, even in the third growing season following a burn. In the case of the high-severity wildfire, our metrics indicated that this grassland had not returned to its pre-fire phenological signals in at least three years after the fire, indicating that it may be undergoing a longer recovery or an ecological shift. These proposed methods address a growing need to study the effects of small, intentional burns in low-biomass ecosystems such as grasslands, which are an essential part of mitigating wildfires.

  PublisherOA PDFDOI

• High-resolution mapping of grassland canopy cover in China through the integration of extensive drone imagery and satellite data

  ISPRS Journal of Photogrammetry and Remote Sensing · 2024-09-12 · 14 citations

  article
  PublisherDOI

• Social–ecological predictors of spotted hyena navigation through a shared landscape

  Ecology and Evolution · 2024-04-01 · 6 citations

  articleOpen accessSenior author

  Abstract Human–wildlife interactions are increasing in severity due to climate change and proliferating urbanization. Regions where human infrastructure and activity are rapidly densifying or newly appearing constitute novel environments in which wildlife must learn to coexist with people, thereby serving as ideal case studies with which to infer future human–wildlife interactions in shared landscapes. As a widely reviled and behaviorally plastic apex predator, the spotted hyena ( Crocuta crocuta ) is a model species for understanding how large carnivores navigate these human‐caused ‘landscapes of fear’ in a changing world. Using high‐resolution GPS collar data, we applied resource selection functions and step selection functions to assess spotted hyena landscape navigation and fine‐scale movement decisions in relation to social–ecological features in a rapidly developing region comprising two protected areas: Lake Nakuru National Park and Soysambu Conservancy, Kenya. We then used camera trap imagery and Barrier Behavior Analysis (BaBA) to further examine hyena interactions with barriers. Our results show that environmental factors, linear infrastructure, human–carnivore conflict hotspots, and human tolerance were all important predictors for landscape‐scale resource selection by hyenas, while human experience elements were less important for fine‐scale hyena movement decisions. Hyena selection for these characteristics also changed seasonally and across land management types. Camera traps documented an exceptionally high number of individual spotted hyenas (234) approaching the national park fence at 16 sites during the study period, and BaBA results suggested that hyenas perceive protected area boundaries' semi‐permeable electric fences as risky but may cross them out of necessity. Our findings highlight that the ability of carnivores to flexibly respond within human‐caused landscapes of fear may be expressed differently depending on context, scale, and climatic factors. These results also point to the need to incorporate societal factors into multiscale analyses of wildlife movement to effectively plan for human–wildlife coexistence.

  PublisherOA PDFDOI

• Centering Amah Mutsun voices in the analysis of a culturally important, fire‐managed coastal grassland

  Ecological Applications · 2024-07-14 · 5 citations

  articleOpen accessSenior author

  Indigenous communities throughout California, USA, are increasingly advocating for and practicing cultural fire stewardship, leading to a host of social, cultural, and ecological benefits. Simultaneously, state agencies are recognizing the importance of controlled burning and cultural fire as a means of reducing the risk of severe wildfire while benefiting fire-adapted ecosystems. However, much of the current research on the impacts of controlled burning ignores the cultural importance of these ecosystems, and risks further marginalizing Indigenous knowledge systems. Our work adds a critical Indigenous perspective to the study of controlled burning in California's unique coastal grasslands, one of the most biodiverse and endangered ecosystems in the country. In this study, we partnered with the Amah Mutsun Tribal Band to investigate how the abundance and occurrence of shrubs, cultural plants, and invasive plants differed among three adjacent coastal grasslands with varying fire histories. These three sites are emblematic of the state's diverging approaches to grassland management: fire suppression, fire suppression followed by wildfire, and an exceedingly rare example of a grassland that has been repeatedly burned approximately every 2 years for more than 30 years. We found that Danthonia californica was significantly more abundant on the burned sites, whereas all included shrub species (Baccharis pilularis, Frangula californica, and Rubus ursinus) were significantly more abundant on the site with no recorded fire, results that have important implications for future cultural revitalization efforts and the loss of coastal grasslands to shrub encroachment. In addition to conducting a culturally relevant vegetation survey, we used Sentinel-2 satellite imagery to compare the relative severities of the two most recent fire events within the study area. Critically, we used interviews with Amah Mutsun tribal members to contextualize the results of our vegetation survey and remote sensing analysis, and to investigate how cultural burning contrasts from typical Western fire management approaches in this region. Our study is a novel example of how interviews, field data, and satellite imagery can be combined to gain a deeper ecological and cultural understanding of fire in California's endangered coastal grasslands.

  PublisherOA PDFDOI

• Tree mortality during long-term droughts is lower in structurally complex forest stands

  Nature Communications · 2023-11-17 · 96 citations

  articleOpen access

  Increasing drought frequency and severity in a warming climate threaten forest ecosystems with widespread tree deaths. Canopy structure is important in regulating tree mortality during drought, but how it functions remains controversial. Here, we show that the interplay between tree size and forest structure explains drought-induced tree mortality during the 2012-2016 California drought. Through an analysis of over one million trees, we find that tree mortality rate follows a "negative-positive-negative" piecewise relationship with tree height, and maintains a consistent negative relationship with neighborhood canopy structure (a measure of tree competition). Trees overshadowed by tall neighboring trees experienced lower mortality, likely due to reduced exposure to solar radiation load and lower water demand from evapotranspiration. Our findings demonstrate the significance of neighborhood canopy structure in influencing tree mortality and suggest that re-establishing heterogeneity in canopy structure could improve drought resiliency. Our study also indicates the potential of advances in remote-sensing technologies for silvicultural design, supporting the transition to multi-benefit forest management.

  PublisherOA PDFDOI

• Structurally complex forests are more resilient to extreme droughts

  2023-02-26

  preprintOpen accessCorresponding

  The increasingly frequent and severe droughts caused by global warming is threating the forest ecosystem health with pervasive tree mortality. Canopy Structure is one of the important factors that regulating drought-induced tree mortality. However, how tree structural influences the spatial and temporal patterns of tree mortality during droughts remains controversial. Through an analysis of nearly 1.5 million trees during the 2012-2016 drought in California, USA, we found tree mortality first decreased with height for small trees, then increased with tree height in the middle sized trees, and decreased again with tree height for matured big trees. We also found relative tree canopy size compared to neighboring trees demonstrates a consistent negative relationship with tree mortality across species. This new finding may be explained by the fact that trees in a structurally complex forest with tall neighboring trees may have higher crown shadow ratio and less water loss to evapotranspiration during the drought. Therefore, the relatively smaller trees in a structurally complex forest have higher survival rate even during an extreme drought. Our findings suggest that a new forest management strategy that re-establishes heterogeneity in tree species and forest structure could improve forest resiliency to severe and extended droughts.

  PublisherDOI

• Spotted hyena navigation of social-ecological landscapes on a coexistence frontier

  2023-10-23

  preprintSenior author

  “Coexistence frontiers”, or regions where human infrastructure and activity are increasing rapidly or newly appearing, constitute novel environments where wildlife must learn to navigate and coexist with people. It is widely recognized that behaviorally flexible species are more likely to persist in these human-dominated landscapes. Nevertheless, we do not fully understand how these animals navigate landscapes shaped by infrastructure, human activity, and human tolerance. As a widely reviled and behaviorally plastic apex predator, the spotted hyena (Crocuta crocuta) is a model species for understanding how wide-ranging large carnivores navigate social-ecological landscapes in an urbanizing world. Using high-resolution (minimum 5-min fix rates) GPS collar data and supplemental camera trap imagery, we applied resource selection and step selection functions to assess spotted hyena landscape navigation and fine-scale movement decisions in relation to social-ecological features in Lake Nakuru National Park and Soysambu Conservancy, Kenya. Second, we used camera traps and barrier behavior analysis (BaBA) to further examine hyena interactions with barriers. Our results show that environmental covariates—including NDVI, terrain, and proximity to water—were the best predictors of landscape-scale resource selection by hyenas, while human infrastructure and the likelihood of conflict with humans or livestock predicted fine-scale hyena movement decisions. We also found that hyena selection for these characteristics changed seasonally and across land management types. Camera traps documented an exceptionally high number of individual spotted hyenas (234) approaching the national park fence at 16 sites during the study period, and BaBA results suggested that hyenas perceive protected area boundaries’ electric fences as risky but may cross them out of necessity. Our results highlight that wildlife adaptability in coexistence frontiers may be expressed differently depending on context and scale. These results also point to the need to incorporate societal factors into multiscale analyses of carnivore movement to effectively plan for human-carnivore coexistence.

  PublisherDOI

Recent grants

• SEES Fellows: Sustainability and Safety in the Pacific West's National Parks

  NSF · $475k · 2013–2017

Frequent coauthors

• Qinghua Guo

  China National Petroleum Corporation (China)

  68 shared

• Stephen Allen

  Edward Francis Small Teaching Hospital

  64 shared

• Steve Narolski

  University of Georgia

  64 shared

• Frederick W. Cubbage

  North Carolina State University

  64 shared

• James B. McCarter

  64 shared

• Mike Clutter

  64 shared

• Richard Bin Mei

  University of California, Berkeley

  64 shared

• Yanjun Su

  Chinese Academy of Sciences

  63 shared

Awards & honors

• Excellence in Education Award for geospatial literacy traini…
• Athletic Hall of Fame - University of California - 2007