Connectivity conservation planning through deep reinforcement learning

被引:0
|
作者
Equihua, Julian [1 ]
Beckmann, Michael [1 ]
Seppelt, Ralf [1 ,2 ,3 ]
机构
[1] UFZ Helmholtz Ctr Environm Res, Dept Computat Landscape Ecol, Leipzig, Germany
[2] Martin Luther Univ Halle Wittenberg, Inst Geosci & Geog, Halle An Der Saale, Germany
[3] German Ctr Integrat Biodivers Res iDiv, Leipzig, Germany
来源
METHODS IN ECOLOGY AND EVOLUTION | 2024年 / 15卷 / 04期
关键词
connectivity conservation planning; deep reinforcement learning; ecological restoration; machine learning; spatial optimisation; systematic conservation planning; HABITAT PATCHES; INDEXES; OPTIMIZATION; NETWORKS; LEVEL;
D O I
10.1111/2041-210X.14300
中图分类号
Q14 [生态学(生物生态学)];
学科分类号
071012 ; 0713 ;
摘要
The United Nations has declared 2021-2030 the decade on ecosystem restoration with the aim of preventing, stopping and reversing the degradation of the ecosystems of the world, often caused by the fragmentation of natural landscapes. Human activities separate and surround habitats, making them too small to sustain viable animal populations or too far apart to enable foraging and gene flow. Despite the need for strategies to solve fragmentation, it remains unclear how to efficiently reconnect nature. In this paper, we illustrate the potential of deep reinforcement learning (DRL) to tackle the spatial optimisation aspect of connectivity conservation planning. The propensity of spatial optimisation problems to explode in complexity depending on the number of input variables and their states is and will continue to be one of its most serious obstacles. DRL is an emerging class of methods focused on training deep neural networks to solve decision-making tasks and has been used to learn good heuristics for complex optimisation problems. While the potential of DRL to optimise conservation decisions seems huge, only few examples of its application exist. We applied DRL to two real-world raster datasets in a connectivity planning setting, targeting graph-based connectivity indices for optimisation. We show that DRL converges to the known optimums in a small example where the objective is the overall improvement of the Integral Index of Connectivity and the only constraint is the budget. We also show that DRL approximates high-quality solutions on a large example with additional cost and spatial configuration constraints where the more complex Probability of Connectivity Index is targeted. To the best of our knowledge, there is no software that can target this index for optimisation on raster data of this size. DRL can be used to approximate good solutions in complex spatial optimisation problems even when the conservation feature is non-linear like graph-based indices. Furthermore, our methodology decouples the optimisation process and the index calculation, so it can potentially target any other conservation feature implemented in current or future software. Las Naciones Unidas han declarado 2021-2030 la decada para la restauracion ecologica, con el objetivo de prevenir, detener e incluso revertir la degradacion de los ecosistemas del mundo. Esta alteracion es causada a menudo por la fragmentacion de los paisajes naturales. Las actividades humanas dividen y aislan los habitats, haciendolos demasiado pequenos para sustentar poblaciones animales viables o demasiado separados para permitir el forrajeo y el flujo genetico. A pesar de la necesidad de estrategias para resolver la fragmentacion, sigue sin ser claro como reconectar eficazmente a la naturaleza. En este articulo, ilustramos el potencial del Aprendizaje Profundo por Refuerzo (APR) para abordar el aspecto de optimizacion espacial en la planificacion de la conservacion de la conectividad. La propension de los problemas de optimizacion espacial a crecer exponencialmente en complejidad en funcion del numero de variables y sus estados es, y seguira siendo, uno de sus obstaculos mas serios. El APR es una clase de metodos para el entrenamiento de redes neuronales profundas con el fin de resolver tareas de toma de decisiones y se ha utilizado para disenar buenas heuristicas para problemas de optimizacion complejos. Si bien el potencial de el APR para optimizar las decisiones de conservacion parece enorme, actualmente solo existen unos pocos ejemplos de su aplicacion. En este estudio, aplicamos APR a dos rasteres de cobertura del suelo del mundo real en un entorno de planificacion de conectividad, apuntando a la optimizacion de indices de conectividad basados en grafos. Mostramos que APR converge a los optimos conocidos en un ejemplo pequeno donde el objetivo es la mejora del indice Integral de Conectividad y la unica restriccion es el presupuesto. Tambien, mostramos que APR se aproxima a soluciones de alta calidad en un ejemplo mayor, con restricciones adicionales de costos y de configuracion espacial y donde el objetivo es la mejora del indice de Probabilidad de Conectividad. Hasta donde sabemos, no existe ningun software que pueda optimizar este indice sobre datos raster del tamano que nosotros procesamos. El APR puede utilizarse para aproximar buenas soluciones en problemas complejos de optimizacion espacial, incluso cuando el objetivo de conservacion es no lineal, como lo son los indices basados en grafos. Ademas, nuestra metodologia desvincula el proceso de optimizacion y el calculo del indice, por lo que potencialmente puede incorporar cualquier otro objetivo de conservacion implementado en el software actual o futuro.
引用
下载
收藏
页码:779 / 790
页数:12
相关论文
共 50 条
  • [1] Hierarchical Task and Motion Planning through Deep Reinforcement Learning
    Newaz, Abdullah Al Redwan
    Alam, Tauhidul
    2021 FIFTH IEEE INTERNATIONAL CONFERENCE ON ROBOTIC COMPUTING (IRC 2021), 2021, : 100 - 105
  • [2] Deep reinforcement learning for conservation decisions
    Lapeyrolerie, Marcus
    Chapman, Melissa S.
    Norman, Kari E. A.
    Boettiger, Carl
    METHODS IN ECOLOGY AND EVOLUTION, 2022, 13 (11): : 2649 - 2662
  • [3] Path Planning of Multiagent Constrained Formation through Deep Reinforcement Learning
    Sui, Zezhi
    Pu, Zhiqiang
    Yi, Jianqiang
    Tan, Xiangmin
    2018 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), 2018,
  • [4] Network Planning with Deep Reinforcement Learning
    Zhu, Hang
    Gupta, Varun
    Ahuja, Satyajeet Singh
    Tian, Yuandong
    Zhang, Ying
    Jin, Xin
    SIGCOMM '21: PROCEEDINGS OF THE 2021 ACM SIGCOMM 2021 CONFERENCE, 2021, : 258 - 271
  • [5] Ecological Velocity Planning through Signalized Intersections: A Deep Reinforcement Learning Approach
    Pozzi, Andrea
    Bae, Sangjae
    Choi, Yongkeun
    Borrelli, Francesco
    Raimondo, Davide M.
    Moura, Scott
    2020 59TH IEEE CONFERENCE ON DECISION AND CONTROL (CDC), 2020, : 245 - 252
  • [6] Intelligent IoT Connectivity: Deep Reinforcement Learning Approach
    Kwon, Minhae
    Lee, Juhyeon
    Park, Hyunggon
    IEEE SENSORS JOURNAL, 2020, 20 (05) : 2782 - 2791
  • [7] Strategic Workforce Planning with Deep Reinforcement Learning
    Smit, Yannick
    Den Hengst, Floris
    Bhulai, Sandjai
    Mehdad, Ehsan
    MACHINE LEARNING, OPTIMIZATION, AND DATA SCIENCE, LOD 2022, PT II, 2023, 13811 : 108 - 122
  • [8] Multi-Robot Path Planning for Mobile Sensing through Deep Reinforcement Learning
    Wei, Yongyong
    Zheng, Rong
    IEEE CONFERENCE ON COMPUTER COMMUNICATIONS (IEEE INFOCOM 2021), 2021,
  • [9] Advanced planning for autonomous vehicles using reinforcement learning and deep inverse reinforcement learning
    You, Changxi
    Lu, Jianbo
    Filev, Dimitar
    Tsiotras, Panagiotis
    ROBOTICS AND AUTONOMOUS SYSTEMS, 2019, 114 : 1 - 18
  • [10] Reducing the Planning Horizon Through Reinforcement Learning
    Dunbar, Logan
    Rosman, Benjamin
    Cohn, Anthony G.
    Leonetti, Matteo
    MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES, ECML PKDD 2022, PT IV, 2023, 13716 : 68 - 83