Mitchell Lyons

Dr Mitchell Lyons

Role: Research Fellow | Lecturer

Bio: I am a postdoc in the Centre for Ecosystem Science, and my research can be described as a mixture of Ecology, Geography and Statistics. I finished my PhD in 2013, at the University of Queensland, which focused on developing new remote sensing methods for long term monitoring and change detection in terrestrial and marine ecosystems. After this I focused on automated monitoring of seagrass environments using remote sensing and autonomous underwater vehicles (AUVs). On moving to the University of New South Wales, I shifted focus to application of modern statistical and modelling approaches for large scale vegetation classification and mapping problems, with a side interest in drone-acquired image data. I also teach remote sensing in some of the courses in the School of Biological, Earth and Environmental Sciences, as well as programming and statistics in various short courses and workshops (

Technically speaking, my expertise lies in remote sensing and ecological modelling (statistics and machine learning), and I generally take a computational programming (R and Python specifically) approach. Non-technically speaking, I love getting into the bush or into the ocean, love bare feet on grass and I have a penchant for cricket and homebrewing.

Research field keywords: ecology, remote sensing and GIS, ecological modelling, vegetation science, statistical ecology

Publications: see my Google Scholar profile (, and please contact me if you would like a copy of any of my papers.

Code + software: see my github page (, and check out my R packages if you are so inclined:

optimus - model-based clustering diagnostics -

c2c - comparing classification and clustering solutions to eachother -


phone: +61 2 9385 2797 | email: level 5, E26 (biological sciences south) | twitter: @mitchest


Author Date Title Link PDF
Murray et al. 2018 The role of satellite remote sensing in structured ecosystem risk assessments

Abstract: The current set of global conservation targets requires methods for monitoring the changing status of ecosystems. Protocols for ecosystem risk assessment are uniquely suited to this task, providing objective syntheses of a wide range of data to estimate the likelihood of ecosystem collapse. Satellite remote sensing can deliver ecologically relevant, long-term datasets suitable for analysing changes in ecosystem area, structure and function at temporal and spatial scales relevant to risk assessment protocols. However, there is considerable uncertainty about how to select and effectively utilise remotely sensed variables for risk assessment. Here, we review the use of satellite remote sensing for assessing spatial and functional changes of ecosystems, with the aim of providing guidance on the use of these data in ecosystem risk assessment. We suggest that decisions on the use of satellite remote sensing should be made a priori and deductively with the assistance of conceptual ecosystem models that identify the primary indicators representing the dynamics of a focal ecosystem.

View PDF
Lyons et al. 2018 A comparison of resampling methods for remote sensing classification and accuracy assessment

Abstract: Maps that categorise the landscape into discrete units are a cornerstone of many scientific, management and conservation activities. The accuracy of these maps is often the primary piece of information used to make decisions about the mapping process or judge the quality of the final map. Variance is critical information when considering map accuracy, yet commonly reported accuracy metrics often do not provide that information. Various resampling frameworks have been proposed and shown to reconcile this issue, but have had limited uptake. In this paper, we compare the traditional approach of a single split of data into a training set (for classification) and test set (for accuracy assessment), to a resampling framework where the classification and accuracy assessment are repeated many times. Using a relatively simple vegetation mapping example and two common classifiers (maximum likelihood and random forest), we compare variance in mapped area estimates and accuracy assessment metrics (overall accuracy, kappa, user, producer, entropy, purity, quantity/allocation disagreement). Input field data points were repeatedly split into training and test sets via bootstrapping, Monte Carlo cross-validation (67:33 and 80:20 split ratios) and k-fold (5-fold) cross-validation. Additionally, within the cross-validation, we tested four designs: simple random, block hold-out, stratification by class, and stratification by both class and space. A classification was performed for every split of every methodological combination (100’s iterations each), creating sampling distributions for the mapped area of each class and the accuracy metrics. We found that regardless of resampling design, a single split of data into training and test sets results in a large variance in estimates of accuracy and mapped area. In the worst case, overall accuracy varied between ~40–80% in one resampling design, due only to random variation in partitioning into training and test sets. On the other hand, we found that all resampling procedures provided accurate estimates of error, and that they can also provide confidence intervals that are informative about the performance and uncertainty of the classifier. Importantly, we show that these confidence intervals commonly encompassed the magnitudes of increase or decrease in accuracy that are often cited in literature as justification for methodological or sampling design choices. We also show how a resampling approach enables generation of spatially continuous maps of classification uncertainty. Based on our results, we make recommendations about which resampling design to use and how it could be implemented. We also provide a fully worked mapping example, which includes traditional inference of uncertainty from the error matrix and provides examples for presenting the final map and its accuracy.

View PDF
Callaghan et al. 2018 The effects of local and landscape habitat attributes on bird diversity in urban greenspaces


Contrasting trajectories of biodiversity loss and urban expansion make it imperative to understand biodiversity persistence in cities. Size‐, local‐, and landscape‐level habitat factors of greenspaces in cities may be critical for future design and management of urban greenspaces in conserving bird biodiversity. Most current understanding of bird communities in cities has come from disparate analyses of single cities, over relatively short time periods, producing limited understanding of processes and characteristics of bird patterns for improved biodiversity management of the world's cities. We analyzed bird biodiversity in 112 urban greenspaces from 51 cities across eight countries, using eBird, a broadscale citizen science project. Species richness and Shannon diversity were used as response variables, while percent tree cover, percent water cover, and vegetation index were used as habitat predictor variables at both a landscape (5 and 25 km radius) and local‐scale level (specific to an individual greenspace) in the modeling process, retrieved using Google Earth Engine. Area of a greenspace was the most important predictor of bird biodiversity, underlining the critical importance of habitat area as the most important factor for increasing bird biodiversity and mitigating loss from urbanization. Surprisingly, distance from the city center and distance from the coast were not significantly related to bird biodiversity. Landscape‐scale habitat predictors were less related to bird biodiversity than local‐scale habitat predictors. Ultimately, bird biodiversity loss could be mitigated by protecting and developing large greenspaces with varied habitat in the world's cities.


View PDF
Callaghan et al. 2018 A comment on the limitations of UAVS in wildlife research - the example of colonial nesting waterbirds View PDF
Lyons et al. 2018 Bird interactions with drones, from individuals to large colonies


Drones are rapidly becoming a key part of the toolkit for a range of scientific disciplines, as well as a range of management and commercial applications. This presents challenges in the context of how drone use might impact on nearby wildlife, especially birds as they might share the airspace. This paper presents observations (from 97 flight hours) and offers preliminary guidance for drone-monitoring exercises and future research to develop guidelines for safe and effective monitoring with drones. Our study sites spanned a range of arid, semi-arid, dunefield, floodplain, wetland, woodland, forest, coastal heath and urban environments in south-eastern and central Australia. They included a nesting colony of >200 000 Straw-necked Ibis Threskiornis spinicollis, the largest drone-based bird-monitoring exercise to date. We particularly focused on behavioural changes towards drones during the breeding season, interactions with raptors, and effects on birds nesting in large colonies—three areas yet to be explored in published literature. Some aggressive behaviour was encountered from solitary breeding birds, but several large breeding bird colonies were surveyed without such issues. With multi-rotor drones, we observed no incidents that posed a threat to birds, but one raptor attacked and took down a fixed-wing drone. In addition to providing observations of interactions with specific bird species, we detail our procedures for flight planning, safe flying and avoidance of birds, and highlight the need for more research into bird– drone interactions, most notably with respect to territorial breeding birds, safety around large raptors, and the effects of drones on the behaviour of birds in large breeding colonies.

View PDF
Callaghan et al. 2017 Assessing the reliability of avian biodiversity measures of urban greenspaces using eBird citizen science data

ABSTRACT. Urban greenspaces are important areas for biodiversity, serving multiple uses, sometimes including conservation and biodiversity management. Citizen science provides a cheap and potentially effective method of assisting biodiversity management in urban greenspaces. Despite this potential, the minimum amount of citizen science data required to adequately represent a community is largely untested. We used eBird data to test the minimum sampling effort required to be confident in results for three biological metrics, species richness, Shannon diversity, and community composition (Bray-Curtis similarity). For our data, from 30 urban greenspaces in North America, for a 90% threshold level, a minimum mean number of 210, 33, and 58 checklists were necessary for species richness, Shannon diversity, and community composition, respectively. However, when we eliminated those species that were present in fewer than 5% of checklists at a given site, there was a marked decrease in mean minimum number of checklists required (17, 9, and 52, respectively). Depending on the ecological questions of interest, eBird data may be a potentially reliable data source in urban greenspaces. We provide a validation methodology using eBird data, with its associated code in the R statistical environment, to provide confidence for land managers and community groups managing urban greenspaces.

View PDF
Go to top