ACAnabelle Williamson Cardoso

Anabelle Williamson Cardoso

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Anabelle Williamson Cardoso has a total of 143 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Remote Sensing for Biodiversity: Insights from the BioSCape Campaign

    by Adam Wilson, Erin Hestir, Jasper Slingsby, Anabelle Cardoso

    BioSCape (Biodiversity Survey of the Cape) is NASA’s first biodiversity-focused airborne and field campaign, designed to advance understanding of ecosystem structure, function, and composition, and how these are changing over time and space in South Africa’s Greater Cape Floristic Region. This international collaboration involving ~160 US and South African scientists across 19 sub-projects produced an unprecedented, open-access dataset spanning 45,000 km², integrating imaging spectroscopy (AVIRIS-NG, PRISM, HyTES), LiDAR (LVIS, ELMAP-V), and extensive ground data from ~2,600 plots. Data are publicly archived at ORNL DAAC and accessible via a cloud computing environment, fostering open science and reproducibility. In this presentation, we will summarize the results to date across the BioSCape project, highlighting key scientific outcomes from both terrestrial and aquatic landscapes. We will also discuss some of the broader implications of the project, including laying the groundwork for achieving ethical and equitable benefit-sharing via compliance with the Nagoya Protocol. A core principle of BioSCape is ”use-inspired science,” which involves co-developing research questions with local partners to directly address pressing conservation and resource management challenges. This ensures that our scientific findings translate into actionable data products for decision-makers in South Africa and globally. For example, BioSCape’s data and products are already informing conservation and resource management efforts, including kelp range maps for ecosystem classification and woody invasive alien plant maps that feed into national ecosystem condition assessments. The project’s contributions have already been recognized in the 2024 State of Environment Outlook Report for the Western Cape Province.

  • Open Access

    DeepForestVision : Automated wildlife identification for camera traps of African tropical forests

    by Hugo Magaldi, Raphaël Cornette, John Justice Tibesigwa, Raymond Katumba, Harold Rugonge, Bala Amarasekaran, Naomi Anderson, Noémie Cappelle, Anabelle W. Cardoso, Daniel Cornélis, et al.

    Tropical forests are rich in biodiversity but face the rapid loss of their wildlife due to increasing anthropogenic pressure, underscoring the urgent need for effective monitoring. Remote‐sensing tools such as camera traps offer faster, less invasive alternatives to human observations. These technologies can provide complementary insights into elusive species, especially in habitats that are difficult to access through direct observation. However, analysing the large volumes of data they produce is labour‐intensive, often leaving datasets underutilised due to limited human resources. Deep learning algorithms can automate aspects of data analysis, but their value to research and conservation efforts depends on their ability to reliably identify target species and be easily deployed in field conditions. To improve wildlife monitoring in African forests using camera trap data, we develop DeepForestVision, the first deep learning algorithm tailored to these challenging habitats that can be used in the field to process both photographs and videos. DeepForestVision was trained on an unprecedented dataset of 2,775,671 photographs and 221,982 videos gathered from camera traps from more than 63 research sites across 11 African countries. It identifies 33 non‐human vertebrate taxa, including 31 mammal taxa, from the most common to the most threatened ones observed on ground‐level camera traps, as well as humans, vehicles and blank photographs or videos. Classification tests demonstrate that DeepForestVision achieves an accuracy of 87.7% on the video test set with 23 taxa. It outperforms the three existing species identification algorithms applicable to these environments: Zamba by 13.1%, Mbaza by 45.0% and SpeciesNet by 37.7%. We provide the model weights for researchers and developers and offer DeepForestVision through a free offline interface. The interface is designed to function in the field in a low‐resource setting and requires no programming expertise. Solution : DeepForestVision is a reliable field tool for monitoring species observed on camera trap photos and videos in African tropical forests. Used by research, conservation, private or political actors, it can guide conservation strategies inside and outside of protected areas, and thus contribute to reducing the loss of biodiversity in African tropical forests.


  • Open Access

    Author Correction: BioSCape combines local knowledge and remote-sensing technology for inclusive biodiversity science

    by Anabelle W. Cardoso, The BioSCape Consortium, Erin L. Hestir, Jasper A. Slingsby, Cherié J. Forbes, Glenn R. Moncrieff, Woody Turner, Andrew L. Skowno, Jacob Nesslage, Philip G. Brodrick, et al.

  • Open Access

    BioSCape combines local knowledge and remote-sensing technology for inclusive biodiversity science

    by Anabelle W. Cardoso, The BioSCape Consortium, Erin L. Hestir, Jasper A. Slingsby, Cherié J. Forbes, Glenn R. Moncrieff, Woody Turner, Andrew L. Skowno, Jacob Nesslage, Philip G. Brodrick, et al.


  • Open Access

    How forests survive alongside flammable open ecosystems: conservation implications for Africa

    by Anabelle Williamson Cardoso, Heath Beckett, William John Bond

    Mosaics of closed, fire-sensitive forests and open flammable ecosystems are common across Africa and other parts of the world. The open ecosystems have long been interpreted as products of deforestation, but diverse lines of evidence point to their origins millions of years before humans. There is widespread concern over the survival of fire-sensitive forest in a flammable matrix, leading to diverse measures, including fire suppression, to protect forests. But if both systems are ancient, how did fire-sensitive forests survive the regular fires in the flammable open ecosystems? Here we discuss recent research on this topic, with a focus on Africa, including the stability of these mosaics through time and the factors accounting for this. These include local topography, variation in grass flammability, the presence of fire-tolerant forest margin tree species, and fauna that create firebreaks along the forest edge. We go on to discuss the conditions under which mosaics are less stable, for example during extreme fires, and consider the uncertain future of mosaicked landscapes under climate change. Finally, we suggest a set of guidelines for consideration by conservation managers concerned about fire damage to forest patches.


  • Open Access

    Real‐time alerts from AI‐enabled camera traps using the Iridium satellite network: A case‐study in Gabon, Central Africa

    by Robin C. Whytock, Thijs Suijten, Tim van Deursen, Jędrzej Świeżewski, Hervé Mermiaghe, Nazaire Madamba, Narcisse Mouckoumou, Joeri A. Zwerts, Aurélie Flore Koumba Pambo, Laila Bahaa‐El‐Din, et al.

    Efforts to preserve, protect and restore ecosystems are hindered by long delays between data collection and analysis. Threats to ecosystems can go undetected for years or decades as a result. Real‐time data can help solve this issue but significant technical barriers exist. For example, automated camera traps are widely used for ecosystem monitoring but it is challenging to transmit images for real‐time analysis where there is no reliable cellular or WiFi connectivity. We modified an off‐the‐shelf camera trap (Bushnell™) and customised existing open‐source hardware to create a ‘smart’ camera trap system. Images captured by the camera trap are instantly labelled by an artificial intelligence model and an ‘alert’ containing the image label and other metadata is then delivered to the end‐user within minutes over the Iridium satellite network. We present results from testing in the Netherlands, Europe, and from a pilot test in a closed‐canopy forest in Gabon, Central Africa. All reference materials required to build the system are provided in open‐source repositories. Results show the system can operate for a minimum of 3 months without intervention when capturing a median of 17.23 images per day. The median time‐difference between image capture and receiving an alert was 7.35 min, though some outliers showed delays of 5‐days or more when the system was incorrectly positioned and unable to connect to the Iridium network. We anticipate significant developments in this field and hope that the solutions presented here, and the lessons learned, can be used to inform future advances. New artificial intelligence models and the addition of other sensors such as microphones will expand the system's potential for other, real‐time use cases including real‐time biodiversity monitoring, wild resource management and detecting illegal human activities in protected areas.


  • Open Access

    Quantifying the environmental limits to fire spread in grassy ecosystems

    by Anabelle W. Cardoso, Sally Archibald, William J. Bond, Corli Coetsee, Matthew Forrest, Navashni Govender, David Lehmann, Loïc Makaga, Nokukhanya Mpanza, Josué Edzang Ndong, et al.

    Modeling fire spread as an infection process is intuitive: An ignition lights a patch of fuel, which infects its neighbor, and so on. Infection models produce nonlinear thresholds, whereby fire spreads only when fuel connectivity and infection probability are sufficiently high. These thresholds are fundamental both to managing fire and to theoretical models of fire spread, whereas applied fire models more often apply quasi-empirical approaches. Here, we resolve this tension by quantifying thresholds in fire spread locally, using field data from individual fires ( n = 1,131) in grassy ecosystems across a precipitation gradient (496 to 1,442 mm mean annual precipitation) and evaluating how these scaled regionally (across 533 sites) and across time (1989 to 2012 and 2016 to 2018) using data from Kruger National Park in South Africa. An infection model captured observed patterns in individual fire spread better than competing models. The proportion of the landscape that burned was well described by measurements of grass biomass, fuel moisture, and vapor pressure deficit. Regionally, averaging across variability resulted in quasi-linear patterns. Altogether, results suggest that models aiming to capture fire responses to global change should incorporate nonlinear fire spread thresholds but that linear approximations may sufficiently capture medium-term trends under a stationary climate.


  • Open Access

    Understanding and modelling wildfire regimes: an ecological perspective

    by Sandy P Harrison, I Colin Prentice, Keith J Bloomfield, Ning Dong, Matthias Forkel, Matthew Forrest, Ramesh K Ningthoujam, Adam Pellegrini, Yicheng Shen, Mara Baudena, et al.

    Recent extreme wildfire seasons in several regions have been associated with exceptionally hot, dry conditions, made more probable by climate change. Much research has focused on extreme fire weather and its drivers, but natural wildfire regimes—and their interactions with human activities—are far from being comprehensively understood. There is a lack of clarity about the ‘causes’ of wildfire, and about how ecosystems could be managed for the co-existence of wildfire and people. We present evidence supporting an ecosystem-centred framework for improved understanding and modelling of wildfire. Wildfire has a long geological history and is a pervasive natural process in contemporary plant communities. In some biomes, wildfire would be more frequent without human settlement; in others they would be unchanged or less frequent. A world without fire would have greater forest cover, especially in present-day savannas. Many species would be missing, because fire regimes have co-evolved with plant traits that resist, adapt to or promote wildfire. Certain plant traits are favoured by different fire frequencies, and may be missing in ecosystems that are normally fire-free. For example, post-fire resprouting is more common among woody plants in high-frequency fire regimes than where fire is infrequent. The impact of habitat fragmentation on wildfire crucially depends on whether the ecosystem is fire-adapted. In normally fire-free ecosystems, fragmentation facilitates wildfire starts and is detrimental to biodiversity. In fire-adapted ecosystems, fragmentation inhibits fires from spreading and fire suppression is detrimental to biodiversity. This interpretation explains observed, counterintuitive patterns of spatial correlation between wildfire and potential ignition sources. Lightning correlates positively with burnt area only in open ecosystems with frequent fire. Human population correlates positively with burnt area only in densely forested regions. Models for vegetation-fire interactions must be informed by insights from fire ecology to make credible future projections in a changing climate.


  • Open Access

    Real-time alerts from AI-enabled camera traps using the Iridium satellite network: a case-study in Gabon, Central Africa

    by Robin C. Whytock, Thijs Suijten, Tim van Deursen, Jędrzej Świeżewski, Hervé Mermiaghe, Nazaire Madamba, Narcys Moukoumou, Joeri A. Zwerts, Aurélie Flore Koumba Pambo, Laila Bahaa-El-Din, et al.

    Efforts to preserve, protect, and restore ecosystems are hindered by long delays between data collection and analysis. Threats to ecosystems can go undetected for years or decades as a result. Real-time data can help solve this issue but significant technical barriers exist. For example, automated camera traps are widely used for ecosystem monitoring but it is challenging to transmit images for real-time analysis where there is no reliable cellular or WiFi connectivity. Here, we present our design for a camera trap with integrated artificial intelligence that can send real-time information from anywhere in the world to end-users. We modified an off-the-shelf camera trap (Bushnell™) and customised existing open-source hardware to rapidly create a ‘smart’ camera trap system. Images captured by the camera trap are instantly labelled by an artificial intelligence model and an ‘alert’ containing the image label and other metadata is then delivered to the end-user within minutes over the Iridium satellite network. We present results from testing in the Netherlands, Europe, and from a pilot test in a closed-canopy forest in Gabon, Central Africa. Results show the system can operate for a minimum of three months without intervention when capturing a median of 17.23 images per day. The median time-difference between image capture and receiving an alert was 7.35 minutes. We show that simple approaches such as excluding ‘uncertain’ labels and labelling consecutive series of images with the most frequent class (vote counting) can be used to improve accuracy and interpretation of alerts. We anticipate significant developments in this field over the next five years and hope that the solutions presented here, and the lessons learned, can be used to inform future advances. New artificial intelligence models and the addition of other sensors such as microphones will expand the system’s potential for other, real-time use cases. Potential applications include, but are not limited to, wildlife tourism, real-time biodiversity monitoring, wild resource management and detecting illegal human activities in protected areas.


  • Top Cited Papers
    Open Access

    Robust ecological analysis of camera trap data labelled by a machine learning model

    by Robin C. Whytock, JęDrzej Świeżewski, Joeri A. Zwerts, Tadeusz Bara‐Słupski, Aurélie Flore Koumba Pambo, Marek Rogala, Laila Bahaa‐El‐Din, Kelly Boekee, Stephanie Brittain, Anabelle W. Cardoso, et al.

    Ecological data are collected over vast geographic areas using digital sensors such as camera traps and bioacoustic recorders. Camera traps have become the standard method for surveying many terrestrial mammals and birds, but camera trap arrays often generate millions of images that are time‐consuming to label. This causes significant latency between data collection and subsequent inference, which impedes conservation at a time of ecological crisis. Machine learning algorithms have been developed to improve the speed of labelling camera trap data, but it is uncertain how the outputs of these models can be used in ecological analyses without secondary validation by a human. Here, we present our approach to developing, testing and applying a machine learning model to camera trap data for the purpose of achieving fully automated ecological analyses. As a case‐study, we built a model to classify 26 Central African forest mammal and bird species (or groups). The model generalizes to new spatially and temporally independent data ( n = 227 camera stations, n = 23,868 images), and outperforms humans in several respects (e.g. detecting ‘invisible’ animals). We demonstrate how ecologists can evaluate a machine learning model's precision and accuracy in an ecological context by comparing species richness, activity patterns ( n = 4 species tested) and occupancy ( n =  4 species tested) derived from machine learning labels with the same estimates derived from expert labels. Results show that fully automated species labels can be equivalent to expert labels when calculating species richness, activity patterns ( n = 4 species tested) and estimating occupancy ( n = 3 of 4 species tested) in a large, completely out‐of‐sample test dataset. Simple thresholding using the Softmax values (i.e. excluding ‘uncertain’ labels) improved the model's performance when calculating activity patterns and estimating occupancy but did not improve estimates of species richness. We conclude that, with adequate testing and evaluation in an ecological context, a machine learning model can generate labels for direct use in ecological analyses without the need for manual validation. We provide the user‐community with a multi‐platform, multi‐language graphical user interface that can be used to run our model offline.