Exploring the relative importance of biotic and abiotic factors that alter the self-thinning rule: Insights from individual-based modelling and machine-learning

Ping, Ma; Han, Xiaohui; Yue, Lin; Moore, John R.; Guo, Yaoxin; Yue, Ming

doi:10.1016/j.ecolmodel.2019.01.019

Cited by 14 publications

(4 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In recent years, applications of random forests algorithms in forestry are becoming progressively less restricted than measuring only productivity. Recently, Andrews et al (2018) and Ma et al (2019) have used random forests to identify the drivers of self-thinning and maximum stand density index, which can aid forest managers to better predict and manage stand dynamics. Random forest algorithms are currently also extensively used in forest fire research, including the determination of drivers of burn rate (Boulanger et al 2018), fire severity (Garcia-Llamas et al 2019a, Klauberg et al 2019 or burnt area (Ying et al 2018).…”

Section: Methodological Developments: Statistics Moving Into New Realmsmentioning

confidence: 99%

Modelling Forest Ecosystems: a crossroad between scales, techniques and applications

Blanco

Améztegui

Rodríguez

2020

Ecological Modelling

View full text Add to dashboard Cite

into different modelling approaches to try to understand forest systems, from continental, regional or landscape-scales to individual-base models. Similarly, forest models have been developed at all kind of complexity levels, from very complex, integrative models to simpler models focused on one specific ecophysiological process. Thus, continuous efforts by the forest modelling community have provided a wealth of research and expertise, but at the same time have generated a dispersion and lack of linkages between different modelling approaches and applications.Forest science is integrative by nature: understanding climate, soil and water processes is as important as understanding plant eco-physiology to understand how forests work. Therefore, forest science has evolved during the last centuries to understand how trees and forests grow, and which patterns and processes rule their development through time and space. Following the common scientific method, empirical reductionist approaches tried to isolate and study one ecological factor or variable at a time (Popper 1968), trying to reduce the complexity inherent to ecological processes. On the contrary, theoretical ecological models explore the ecological interactions among the components that exist in a forest, assembling the pieces of evidence generated through traditional field and lab work (Kimmins et al. 2010). In this special issue, we want to show some of the latest developments in forest modelling, understanding forests like working systems in which human intervention has played and is still playing a historical role, and will likely do so in the future.Following the sharp increase in data availability, modelling is now being applied in forest ecosystems at multiple scales, from continental and regional studies on forests and species distribution to leaf scales modelling plague and pathogen infestation in individual branches. Similarly, models are being applied to understand short-time processes such as nutrient transport in cells, to extremely long scales such as genetic drift in tree populations. From the application point of view, models are also being applied from purely ecological issues, such as coexistence and resource competition among tree species, to multidisciplinary socio-economic-ecological issues, such as design of tourist routes inside natural reserves, estimating timber flow to sawmills under climate change scenarios, or engineering concerns such as designing forest harvesting machinery with lower ecological footprint (in energy, rut formation, noise, etc.).When organizing this special issue, we (the guest editors) decided to carry out a quick overview of current trends in forest modelling. To do so, we used Clarivate's Web of Science to carry out a search in the top five journals with "Forestry" as their unique category

show abstract

Section: Methodological Developments: Statistics Moving Into New Realmsmentioning

confidence: 99%

Modelling Forest Ecosystems: a crossroad between scales, techniques and applications

Blanco

Améztegui

Rodríguez

2020

Ecological Modelling

View full text Add to dashboard Cite

show abstract

“…Perry and O'Sullivan [161] studied a hunter-gatherer foraging ABM by using RF to classify emergent narratives of different control parameters, and they showed that ML approaches provide powerful tools to identify emergent narratives from individual trajectories, which is likely most appropriate for the application of the empirically inaccessible cases in the medium-to far-past or future. Ma et al [162] also employed RF to assess the potential effect of biotic and abiotic factors on the simulated selfthinning line, wherein the RF-based model is trained to predict emergent features of the ABM. The ANN can also be used for prediction.…”

Section: Macro Abms/emergence Emulatormentioning

confidence: 99%

Synergistic Integration Between Machine Learning and Agent-Based Modeling: A Multidisciplinary Review

Zhang

Valencia

Chang

2023

IEEE Trans. Neural Netw. Learning Syst.

View full text Add to dashboard Cite

Agent-based modeling (ABM) involves developing models in which agents make adaptive decisions in a changing environment. Machine-learning (ML) based inference models can improve sequential decision-making by learning agents' behavioral patterns. With the aid of ML, this emerging area can extend traditional agent-based schemes that hardcode agents' behavioral rules into an adaptive model. Even though there are plenty of studies that apply ML in ABMs, the generalized applicable scenarios, frameworks, and procedures for implementations are not well addressed. In this article, we provide a comprehensive review of applying ML in ABM based on four major scenarios, i.e., microagent-level situational awareness learning, microagent-level behavior intervention, macro-ABM-level emulator, and sequential decision-making. For these four scenarios, the related algorithms, frameworks, procedures of implementations, and multidisciplinary applications are thoroughly investigated. We also discuss how ML can improve prediction in ABMs by trading off the variance and bias and how ML can improve the sequential decision-making of microagent and macrolevel policymakers via a mechanism of reinforced behavioral intervention. At the end of this article, future perspectives of applying ML in ABMs are discussed with respect to data acquisition and quality issues, the possible solution of solving the convergence problem of reinforcement learning, interpretable ML applications, and bounded rationality of ABM.

show abstract

“…In this sense, machine learning can be used to find the parameterisation producing the best fit between simulations and benchmarking data. Machine learning can also be used to facilitate models development (Recknagel, 2001;Peters et al, 2014;Thessen, 2016;Gobeyn et al, 2019;Ma et al, 2019;Mehta and Pankaj, 2019). For example, Gobeyn et al (2019), proposed evolutionary algorithms, derived from machine learning, to help calibrate models, but also to reduce models complexity.…”

Section: -Challenges and Opportunitiesmentioning

confidence: 99%

“…For example, Gobeyn et al (2019), proposed evolutionary algorithms, derived from machine learning, to help calibrate models, but also to reduce models complexity. Ma et al (2019) To compensate for the increasing demand on computation time, optimization can also be made on the models' development workflow. This encompasses technical tools made to facilitate benchmarking of models, such as the "DGVMTools" R package (Forrest, Scheiter and Steinkamp, unpublished) as well as standardized and streamlined methodologies (LeBauer et al, 2013;Warszawski et al, 2014;Best et al, 2015;Eyring et al, 2016;F.…”

Section: -Challenges and Opportunitiesmentioning

confidence: 99%

Process based modelling of ecosystem structure and dynamics with aDGVM2 : a case study of whole-plant trait trade-offs & shrubs in African savannas

Gaillard¹

View full text Add to dashboard Cite

Shrubs are a characteristic component of savannas, where they coexist with trees and grasses. They are often part of woody encroachment phenomena, which have been observed globally, and the determinant of shrub encroachment cases, which are particularly of concern in African savannas. In response to climate change and land use change, African savannas are vulnerable to biome shifts and shrub encroachment is a process driving and explaining this risk. We contribute to furthering the understanding of shrubs biogeography and ecology by considering the number of stems of woody plants to characterise shrubs phenotype and strategy. We postulate that shrubs are multi-stemmed, compared to single-stemmed trees and integrate this assumption in aDGVM2 (adaptive Dynamic Global Vegetation Model 2). Modelling a trait representing the number of stems of a woody plant implies a trade-off between single-stemmed plants having higher height growth potential and multi-stemmed plants having higher hydraulic capacity but limited height growth. Multi-stemmed individuals, being shorter, are more likely to suffer severe damage from fires than tall single-stemmed trees managing to grow their crown out of the flame zone. We simulate potential vegetation over sub-Saharan Africa at 1° spatial resolution, with aDGVM2 and compare it to simulations without our shrub model turned on. We also test the impact of fire by including or excluding it from our simulations. To assess the accuracy and relevance of our approach, we benchmark our overall model’s performance against multiple satellite derived products of above ground biomass (AGBM), and against specific field measurements of AGBM. We further benchmark our results against vegetation cover type derived from satellite data. We demonstrate that shrubs can be modelled as multi-stemmed woody plants in African savannas based on whole-plant trait trade-off without being predefined as static functional types. Indeed, the addition of our shrub model to aDGVM2 allows for shrubs to emerge dynamically through community assembly processes without a priori categorisation. Our shrub model also improves the simulated vegetation patterns simulated by aDGVM2 in sub-Saharan Africa, particularly in savannas. The simulated pattern of stem number per woody individual broadly follows our assumptions about biogeographic patterns as it is lowest in equatorial African forests and increases in savannas and grasslands as precipitation decreases. Shrubs are more abundant in more water-stressed regions where they have a competitive advantage over trees due to their increased relative water transport potential. However, in arid and hyper-arid regions, further investigations are required. Simulated shrub prevalence is higher in more open and fire prone landscapes, where woody cover and biomass are reduced. Adding shrubs to aDGVM2, while increasing complexity allows for greater simulated diversity. As resilience and resistance of ecosystems have been shown to be influenced by diversity, such model development is necessary to improve our ability to forecast ecosystems responses to changes. However, there are challenges to fully tap this benefit. Assessing the accuracy and relevance of our approach is challenging. Data and simulations are conceptually different which limit the possibility to conclude based on comparison. Benchmarking challenge is exacerbated by the variability existing among satellite derived products and site studies observations. In areas of extremely low biomass and vegetation cover, such as deserts and semi-deserts, the accuracy of our model is more concerning as small differences in absolute values are relatively more important. Categorisation of life-forms shapes our understanding of their ecology and biogeography, thus, consensus about their definition is direly needed. To contribute to this debate, we investigate how vegetation distribution patterns arising from our shrub model inform our understanding of shrub biogeography. First, shrub distribution in trait space (considering stem number), relatively to environmental drivers, concurs with our assumptions. Second, shrub spatial distribution is consistent with our characterisation assumptions. Third, the role of simulated shrubs in an ecosystem supports realistic ecological dynamics. Our model allows for, shrubs to exhibit a specific phenotype, but also a specific life-strategy, which we characterise in terms of persistence strategy (shrubs are mainly resprouters, in contrast to trees, which can be either resprouters or reseeders) and in terms of resource acquisition (rooting strategy) and allocation (carbon investment). Adding stem count as a trait to aDGVM2 increase the range of simulated functional diversity. Our shrub model allows for aDGVM2 to simulate realistic ratio of grass to woody vegetation across sub-saharan Africa. Similarly, it simulates ratio of shrubs to trees consistent with our hypotheses. ...

show abstract

Exploring the relative importance of biotic and abiotic factors that alter the self-thinning rule: Insights from individual-based modelling and machine-learning

Cited by 14 publications

References 53 publications

Modelling Forest Ecosystems: a crossroad between scales, techniques and applications

Modelling Forest Ecosystems: a crossroad between scales, techniques and applications

Synergistic Integration Between Machine Learning and Agent-Based Modeling: A Multidisciplinary Review

Process based modelling of ecosystem structure and dynamics with aDGVM2 : a case study of whole-plant trait trade-offs & shrubs in African savannas

Contact Info

Product

Resources

About