Lucas N. Joppa
Research Subjects
Research Interests
Lucas N. Joppa has a total of 82 co-authors. Here are the top co-authors by number of shared publications:
- Richard Harper
Lancaster University, Lancaster, United Kingdom
3 shared publications Stephen Emmott
Computational Science Laboratory Microsoft Research 21 Station Road Cambridge CB1 2FB UK
3 shared publicationsAlexander Pfaff
Sanford School of Public Policy, Duke University, PO Box 90239 Durham, NC 27708-0239, USA
2 shared publications- David J. Gavaghan
University of Oxford
2 shared publications - Fei Fang
Carnegie Mellon University
2 shared publications Greg McInerny
Computational Science Laboratory Microsoft Research 21 Station Road Cambridge CB1 2FB UK
2 shared publications
Latest Publications
- Open Access
- 17 July 2019
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 33 (01) , 1401-1408
Deep Reinforcement Learning for Green Security Games with Real-Time Information
Green Security Games (GSGs) have been proposed and applied to optimize patrols conducted by law enforcement agencies in green security domains such as combating poaching, illegal logging and overfishing. However, real-time information such as footprints and agents’ subsequent actions upon receiving the information, e.g., rangers following the footprints to chase the poacher, have been neglected in previous work. To fill the gap, we first propose a new game model GSG-I which augments GSGs with sequential movement and the vital element of real-time information. Second, we design a novel deep reinforcement learning-based algorithm, DeDOL, to compute a patrolling strategy that adapts to the real-time information against a best-responding attacker. DeDOL is built upon the double oracle framework and the policy-space response oracle, solving a restricted game and iteratively adding best response strategies to it through training deep Q-networks. Exploring the game structure, DeDOL uses domain-specific heuristic strategies as initial strategies and constructs several local modes for efficient and parallelized training. To our knowledge, this is the first attempt to use Deep Q-Learning for security games.
- 17 July 2019
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 33 (01) , 1401-1408
Open AccessDeep Reinforcement Learning for Green Security Games with Real-Time Information
Green Security Games (GSGs) have been proposed and applied to optimize patrols conducted by law enforcement agencies in green security domains such as combating poaching, illegal logging and overfishing. However, real-time information such as footprints and agents’ subsequent actions upon receiving the information, e.g., rangers following the footprints to chase the poacher, have been neglected in previous work. To fill the gap, we first propose a new game model GSG-I which augments GSGs with sequential movement and the vital element of real-time information. Second, we design a novel deep reinforcement learning-based algorithm, DeDOL, to compute a patrolling strategy that adapts to the real-time information against a best-responding attacker. DeDOL is built upon the double oracle framework and the policy-space response oracle, solving a restricted game and iteratively adding best response strategies to it through training deep Q-networks. Exploring the game structure, DeDOL uses domain-specific heuristic strategies as initial strategies and constructs several local modes for efficient and parallelized training. To our knowledge, this is the first attempt to use Deep Q-Learning for security games.
- Open Access
- 20 June 2018
- Published by Association for Computing Machinery (ACM)
- p. 1-12
AirSim-W
Increases in poaching levels have led to the use of unmanned aerial vehicles (UAVs or drones) to count animals, locate animals in parks, and even find poachers. Finding poachers is often done at night through the use of long wave thermal infrared cameras mounted on these UAVs. Unfortunately, monitoring the live video stream from the conservation UAVs all night is an arduous task. In order to assist in this monitoring task, new techniques in computer vision have been developed. This work is based on a dataset which took approximately six months to label. However, further improvement in detection and future testing of autonomous flight require not only more labeled training data, but also an environment where algorithms can be safely tested. In order to meet both goals efficiently, we present AirSim-W, a simulation environment that has been designed specifically for the domain of wildlife conservation. This includes (i) creation of an African savanna environment in Unreal Engine, (ii) integration of a new thermal infrared model based on radiometry, (iii) API code expansions to follow objects of interest or fly in zigzag patterns to generate simulated training data, and (iv) demonstrated detection improvement using simulated data generated by AirSim-W. With these additional simulation features, AirSim-W will be directly useful for wildlife conservation research.
- 20 June 2018
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Open AccessAirSim-W
Increases in poaching levels have led to the use of unmanned aerial vehicles (UAVs or drones) to count animals, locate animals in parks, and even find poachers. Finding poachers is often done at night through the use of long wave thermal infrared cameras mounted on these UAVs. Unfortunately, monitoring the live video stream from the conservation UAVs all night is an arduous task. In order to assist in this monitoring task, new techniques in computer vision have been developed. This work is based on a dataset which took approximately six months to label. However, further improvement in detection and future testing of autonomous flight require not only more labeled training data, but also an environment where algorithms can be safely tested. In order to meet both goals efficiently, we present AirSim-W, a simulation environment that has been designed specifically for the domain of wildlife conservation. This includes (i) creation of an African savanna environment in Unreal Engine, (ii) integration of a new thermal infrared model based on radiometry, (iii) API code expansions to follow objects of interest or fly in zigzag patterns to generate simulated training data, and (iv) demonstrated detection improvement using simulated data generated by AirSim-W. With these additional simulation features, AirSim-W will be directly useful for wildlife conservation research.
- Open Access
Evolving Protected-Area Impacts in Mexico: Political Shifts as Suggested by Impact Evaluations
For protected areas (PAs), variation in forest impacts over space—including types of PA—are increasingly well documented, while shifts in impacts over time receive less attention. For Mexico, in the 1990s, PAs effectively were ‘paper parks’. Thus, achieving impacts on the forest would require shifts over time in the politics of PA siting and PA implementation. We rigorously analyze the impacts of Mexican PAs on 2000–2005 loss of natural land cover, using matching to reduce location bias caused by typical land-use economics and politics. We find a 3.2% lower loss, on average, due to PAs. Since politics often vary by type of PA, we also show that in Mexico stricter PAs are closer to cities and have greater impact than mixed-use PAs. These shifts in impacts suggest some potential for PAs to conserve forests.
Open AccessEvolving Protected-Area Impacts in Mexico: Political Shifts as Suggested by Impact Evaluations
For protected areas (PAs), variation in forest impacts over space—including types of PA—are increasingly well documented, while shifts in impacts over time receive less attention. For Mexico, in the 1990s, PAs effectively were ‘paper parks’. Thus, achieving impacts on the forest would require shifts over time in the politics of PA siting and PA implementation. We rigorously analyze the impacts of Mexican PAs on 2000–2005 loss of natural land cover, using matching to reduce location bias caused by typical land-use economics and politics. We find a 3.2% lower loss, on average, due to PAs. Since politics often vary by type of PA, we also show that in Mexico stricter PAs are closer to cities and have greater impact than mixed-use PAs. These shifts in impacts suggest some potential for PAs to conserve forests.
- Open Access
- 22 April 2016
- Published by American Association for the Advancement of Science (AAAS) in Science
- Vol. 352 (6284) , 416-418
Filling in biodiversity threat gaps
Only 5% of global threat data sets meet a “gold standard”
- 22 April 2016
- Published by American Association for the Advancement of Science (AAAS) in Science
- Vol. 352 (6284) , 416-418
Open AccessFilling in biodiversity threat gaps
Only 5% of global threat data sets meet a “gold standard”
- 4 March 2016
- Published by American Association for the Advancement of Science (AAAS) in Science
- Vol. 351 (6277) , 1037
Government: Plan for ecosystem services
- 4 March 2016
- Published by American Association for the Advancement of Science (AAAS) in Science
- Vol. 351 (6277) , 1037
Government: Plan for ecosystem services
- 12 January 2016
- Published by Wiley in Diversity and Distributions
- Vol. 22 (5) , 558-565
Threatened or Data Deficient: assessing the conservation status of poorly known species
Aim: To determine whether extinction risk assessments based on biological collections and using Criterion B of the IUCN Red List Criteria reflect in part an accurate measure of species rarity and thus extinction risk.Location: Madagascar.Methods: We calculate the extent of occurrence (EOO) and area of occupancy (AOO) for orchids using herbarium specimen data. Correlations were made against range, occupancy, extinction risk, number of specimens and the date of description. We calculated the average increase in range per species specimen, correlated this against the date of description and determined significance of the observed EOO accumulation using randomization tests.Results: Significant negative correlations were found between date of description and all measures of range, occupancy and associated Red List Categories and number of specimens, as well as between the average range accumulation per specimen and date of description. Seventy‐five percentage of species’ observed EOO accumulations significantly differed from random. Maximum deviations between observed EOO accumulations and those derived from random sampling were always significantly positive. For most species, this occurred more frequently during the first half of the accumulation sequence.Main conclusions: Species described more recently have smaller ranges and occupancies, fewer specimens and greater perceived extinction risk status. Levels of geographic uniqueness of collections are higher in species described more recently. Awareness of a species range increased faster than random, particularly in the first half of the sampling process, suggesting that newly discovered or yet to be discovered species are rare and likely have a higher risk of extinction. For many species, biological collections represent the sum of our knowledge. While data may be limited, such species should be listed in an appropriate Red List Category in accordance with the IUCN Red List Guidelines rather than as Data Deficient.
- 12 January 2016
- Published by Wiley in Diversity and Distributions
- Vol. 22 (5) , 558-565
Threatened or Data Deficient: assessing the conservation status of poorly known species
Aim: To determine whether extinction risk assessments based on biological collections and using Criterion B of the IUCN Red List Criteria reflect in part an accurate measure of species rarity and thus extinction risk.Location: Madagascar.Methods: We calculate the extent of occurrence (EOO) and area of occupancy (AOO) for orchids using herbarium specimen data. Correlations were made against range, occupancy, extinction risk, number of specimens and the date of description. We calculated the average increase in range per species specimen, correlated this against the date of description and determined significance of the observed EOO accumulation using randomization tests.Results: Significant negative correlations were found between date of description and all measures of range, occupancy and associated Red List Categories and number of specimens, as well as between the average range accumulation per specimen and date of description. Seventy‐five percentage of species’ observed EOO accumulations significantly differed from random. Maximum deviations between observed EOO accumulations and those derived from random sampling were always significantly positive. For most species, this occurred more frequently during the first half of the accumulation sequence.Main conclusions: Species described more recently have smaller ranges and occupancies, fewer specimens and greater perceived extinction risk status. Levels of geographic uniqueness of collections are higher in species described more recently. Awareness of a species range increased faster than random, particularly in the first half of the sampling process, suggesting that newly discovered or yet to be discovered species are rare and likely have a higher risk of extinction. For many species, biological collections represent the sum of our knowledge. While data may be limited, such species should be listed in an appropriate Red List Category in accordance with the IUCN Red List Guidelines rather than as Data Deficient.
- Open Access
- 27 October 2015
- Published by Springer Nature in AMBIO
- Vol. 44 (S4) , 522-526
Technology for nature conservation: An industry perspective
Information age technology has the potential to change the game for conservation by continuously monitoring the pulse of the natural world. Whether or not it will depends on the ability of the conservation sector to build a community of practice, come together to define key technology challenges and work with a wide variety of partners to create, implement, and sustain solutions. I describe why these steps are necessary, outline the latest developments in the field and offer actionable ways forward for conservation agencies, universities, funding bodies, professional societies, and technology corporations to come together to realize the revolution that computational technologies can bring for biodiversity conservation.
- 27 October 2015
- Published by Springer Nature in AMBIO
- Vol. 44 (S4) , 522-526
Open AccessTechnology for nature conservation: An industry perspective
Information age technology has the potential to change the game for conservation by continuously monitoring the pulse of the natural world. Whether or not it will depends on the ability of the conservation sector to build a community of practice, come together to define key technology challenges and work with a wide variety of partners to create, implement, and sustain solutions. I describe why these steps are necessary, outline the latest developments in the field and offer actionable ways forward for conservation agencies, universities, funding bodies, professional societies, and technology corporations to come together to realize the revolution that computational technologies can bring for biodiversity conservation.
- 17 July 2015
- Published by Wiley in Conservation Biology
- Vol. 30 (2) , 362-370
Impact of alternative metrics on estimates of extent of occurrence for extinction risk assessment
In International Union for Conservation of Nature (IUCN) Red List assessments, extent of occurrence (EOO) is a key measure of extinction risk. However, the way assessors estimate EOO from maps of species’ distributions is inconsistent among assessments of different species and among major taxonomic groups. Assessors often estimate EOO from the area of mapped distribution, but these maps often exclude areas that are not habitat in idiosyncratic ways and are not created at the same spatial resolutions. We assessed the impact on extinction risk categories of applying different methods (minimum convex polygon, alpha hull) for estimating EOO for 21,763 species of mammals, birds, and amphibians. Overall, the percentage of threatened species requiring down listing to a lower category of threat (taking into account other Red List criteria under which they qualified) spanned 11–13% for all species combined (14–15% for mammals, 7–8% for birds, and 12–15% for amphibians). These down listings resulted from larger estimates of EOO and depended on the EOO calculation method. Using birds as an example, we found that 14% of threatened and near threatened species could require down listing based on the minimum convex polygon (MCP) approach, an approach that is now recommended by IUCN. Other metrics (such as alpha hull) had marginally smaller impacts. Our results suggest that uniformly applying the MCP approach may lead to a one‐time down listing of hundreds of species but ultimately ensure consistency across assessments and realign the calculation of EOO with the theoretical basis on which the metric was founded.
- 17 July 2015
- Published by Wiley in Conservation Biology
- Vol. 30 (2) , 362-370
Impact of alternative metrics on estimates of extent of occurrence for extinction risk assessment
In International Union for Conservation of Nature (IUCN) Red List assessments, extent of occurrence (EOO) is a key measure of extinction risk. However, the way assessors estimate EOO from maps of species’ distributions is inconsistent among assessments of different species and among major taxonomic groups. Assessors often estimate EOO from the area of mapped distribution, but these maps often exclude areas that are not habitat in idiosyncratic ways and are not created at the same spatial resolutions. We assessed the impact on extinction risk categories of applying different methods (minimum convex polygon, alpha hull) for estimating EOO for 21,763 species of mammals, birds, and amphibians. Overall, the percentage of threatened species requiring down listing to a lower category of threat (taking into account other Red List criteria under which they qualified) spanned 11–13% for all species combined (14–15% for mammals, 7–8% for birds, and 12–15% for amphibians). These down listings resulted from larger estimates of EOO and depended on the EOO calculation method. Using birds as an example, we found that 14% of threatened and near threatened species could require down listing based on the minimum convex polygon (MCP) approach, an approach that is now recommended by IUCN. Other metrics (such as alpha hull) had marginally smaller impacts. Our results suggest that uniformly applying the MCP approach may lead to a one‐time down listing of hundreds of species but ultimately ensure consistency across assessments and realign the calculation of EOO with the theoretical basis on which the metric was founded.
- 20 January 2015
- Published by Wiley in Diversity and Distributions
- Vol. 21 (3) , 258-267
Scientists and software – surveying the species distribution modelling community
Aim: Software use is ubiquitous in the species distribution modelling (SDM) domain; nearly every scientist working on SDM either uses or develops specialist SDM software; however, little is formally known about the prevalence or preference of one software over another. We seek to provide, for the first time, a ‘snapshot’ of SDM users, the methods they use and the questions they answer.Location: Global.Methods: We conducted a survey of over 300 SDM scientists to capture a snapshot of the community and used an extensive literature search of SDM papers in order to investigate the characteristics of the SDM community and its interactions with software developers in terms of co‐authoring research publications.Results: Our results show that those members of the community who develop software and who are directly connected with developers are among the most highly connected and published authors in the field. We further show that the two most popular softwares for SDM lie at opposite ends of the ‘use‐complexity’ continuum.Main conclusion: Given the importance of SDM research in a changing environment, with its increasing use in the policy domain, it is vital to be aware of what software and methodologies are being implemented. Here, we present a snapshot of the SDM community, the software and the methods being used.
- 20 January 2015
- Published by Wiley in Diversity and Distributions
- Vol. 21 (3) , 258-267
Scientists and software – surveying the species distribution modelling community
Aim: Software use is ubiquitous in the species distribution modelling (SDM) domain; nearly every scientist working on SDM either uses or develops specialist SDM software; however, little is formally known about the prevalence or preference of one software over another. We seek to provide, for the first time, a ‘snapshot’ of SDM users, the methods they use and the questions they answer.Location: Global.Methods: We conducted a survey of over 300 SDM scientists to capture a snapshot of the community and used an extensive literature search of SDM papers in order to investigate the characteristics of the SDM community and its interactions with software developers in terms of co‐authoring research publications.Results: Our results show that those members of the community who develop software and who are directly connected with developers are among the most highly connected and published authors in the field. We further show that the two most popular softwares for SDM lie at opposite ends of the ‘use‐complexity’ continuum.Main conclusion: Given the importance of SDM research in a changing environment, with its increasing use in the policy domain, it is vital to be aware of what software and methodologies are being implemented. Here, we present a snapshot of the SDM community, the software and the methods being used.
- 31 October 2014
- Published by Wiley in Conservation Biology
- Vol. 29 (2) , 503-512
Building robust conservation plans
Systematic conservation planning optimizes trade‐offs between biodiversity conservation and human activities by accounting for socioeconomic costs while aiming to achieve prescribed conservation objectives. However, the most cost‐efficient conservation plan can be very dissimilar to any other plan achieving the set of conservation objectives. This is problematic under conditions of implementation uncertainty (e.g., if all or part of the plan becomes unattainable). We determined through simulations of parallel implementation of conservation plans and habitat loss the conditions under which optimal plans have limited chances of implementation and where implementation attempts would fail to meet objectives. We then devised a new, flexible method for identifying conservation priorities and scheduling conservation actions. This method entails generating a number of alternative plans, calculating the similarity in site composition among all plans, and selecting the plan with the highest density of neighboring plans in similarity space. We compared our method with the classic method that maximizes cost efficiency with synthetic and real data sets. When implementation was uncertain—a common reality—our method provided higher likelihood of achieving conservation targets. We found that χ, a measure of the shortfall in objectives achieved by a conservation plan if the plan could not be implemented entirely, was the main factor determining the relative performance of a flexibility enhanced approach to conservation prioritization. Our findings should help planning authorities prioritize conservation efforts in the face of uncertainty about future condition and availability of sites.
- 31 October 2014
- Published by Wiley in Conservation Biology
- Vol. 29 (2) , 503-512
Building robust conservation plans
Systematic conservation planning optimizes trade‐offs between biodiversity conservation and human activities by accounting for socioeconomic costs while aiming to achieve prescribed conservation objectives. However, the most cost‐efficient conservation plan can be very dissimilar to any other plan achieving the set of conservation objectives. This is problematic under conditions of implementation uncertainty (e.g., if all or part of the plan becomes unattainable). We determined through simulations of parallel implementation of conservation plans and habitat loss the conditions under which optimal plans have limited chances of implementation and where implementation attempts would fail to meet objectives. We then devised a new, flexible method for identifying conservation priorities and scheduling conservation actions. This method entails generating a number of alternative plans, calculating the similarity in site composition among all plans, and selecting the plan with the highest density of neighboring plans in similarity space. We compared our method with the classic method that maximizes cost efficiency with synthetic and real data sets. When implementation was uncertain—a common reality—our method provided higher likelihood of achieving conservation targets. We found that χ, a measure of the shortfall in objectives achieved by a conservation plan if the plan could not be implemented entirely, was the main factor determining the relative performance of a flexibility enhanced approach to conservation prioritization. Our findings should help planning authorities prioritize conservation efforts in the face of uncertainty about future condition and availability of sites.