From ed3e0ce100de964beb07cdb74da688f70af0538a Mon Sep 17 00:00:00 2001 From: Tanya Strydom Date: Tue, 12 Nov 2024 10:45:51 +0000 Subject: [PATCH] =?UTF-8?q?=F0=9F=93=9D=20*cracks=20knuckles*=20time=20for?= =?UTF-8?q?=20some=20edits?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- index.qmd | 34 +++++++++++++++++----------------- 1 file changed, 17 insertions(+), 17 deletions(-) diff --git a/index.qmd b/index.qmd index 84fbaba..0d73922 100644 --- a/index.qmd +++ b/index.qmd @@ -87,31 +87,31 @@ The interplay between network representation and network (node and edge) definit **Evolutionary compatibility** -There is compelling evidence that an interaction occurring between two species is the result of their shared (co)evolutionary history [@segarRoleEvolutionShaping2020; @gomezEcologicalInteractionsAre2010; @dallarivaExploringEvolutionarySignature2016] which, in the more proximal sense, is manifested as the 'trait complementarity' between two species [@benadiQuantitativePredictionInteractions2022], whereby one species (the predator) has the 'correct' set of traits that allow it to chase, capture, kill, and consume the other species (the prey). For species pairs where this condition is not met the link is deemed to be forbidden [@jordanoSamplingNetworksEcological2016]; *i.e.,* not physically possible and will always be absent within a network. A network constructed on the basis of evolutionary compatible links is most closely aligned with a metaweb, although it would not be required that the species co-occur (as shown in @fig-process), and arguably makes for a good approximation of the 'Eltonian niche' of species [@soberonGrinnellianEltonianNiches2007]. Finally, one should be aware that it is possible to represent evolutionary compatible interactions as either binary (possible vs forbidden) or as a probability [@banvilleDecipheringProbabilisticSpecies2024], where the probability represents how likely the interaction between two species is to be possible. +There is compelling evidence that an interaction occurring between two species is the result of their shared (co)evolutionary history [@segarRoleEvolutionShaping2020; @gomezEcologicalInteractionsAre2010; @dallarivaExploringEvolutionarySignature2016] which, in the more proximal sense, is manifested as the 'trait complementarity' between two species [@benadiQuantitativePredictionInteractions2022], whereby one species (the predator) has the 'correct' set of traits that allow it to chase, capture, kill, and consume the other species (the prey). For species pairs where this condition is not met the link is deemed to be *forbidden* [@jordanoSamplingNetworksEcological2016]; *i.e.,* not physically possible and will always be absent within a network. A network constructed on the basis of evolutionary compatible links is most closely aligned with a metaweb, although it would not be required that the species co-occur (as shown in @fig-process), and arguably makes for a good approximation of the 'Eltonian niche' of species [@soberonGrinnellianEltonianNiches2007]. Finally, one should be aware that it is possible to represent evolutionary compatible interactions as either binary (possible vs forbidden) or as a probability [@banvilleDecipheringProbabilisticSpecies2024], where the probability represents how likely the interaction between two species is to be possible. **(Co)occurrence** -Although the outright assumption that because two species are co-occurring it must mean that they are interacting is flawed [@blanchetCooccurrenceNotEvidence2020], it is of course impossible for two species to interact (at least in terms of feeding links) if they are not co-occurring in time and space. Thus, although co-occurrence data alone is insufficient to build an accurate and ecologically meaningful representation of *feeding links* it is still a critical process that determines the realisation of feeding links and allows us to constrain a global metaweb to only consider 'realised' communities [@dansereauSpatiallyExplicitPredictions2024] and an understanding of the intersection of species interactions and their co-occurrence is meaningful when one is operating in the space of trying to determine the distribution of a species [@higinoMismatchIUCNRange2023; @pollockUnderstandingCooccurrenceModelling2014], representing something of a fusion of the the Grinnellian and Eltonian niches [@gravelBringingEltonGrinnell2019]. +Although the outright assumption that because two species are co-occurring it must mean that they are interacting is flawed [@blanchetCooccurrenceNotEvidence2020], it is of course impossible for two species to interact (at least in terms of feeding links) if they are not co-occurring in time and space. Thus, although co-occurrence data alone is insufficient to build an accurate and ecologically meaningful representation of *feeding links* it is still a critical process that determines the realisation of feeding links and allows us to constrain a global metaweb to only consider 'realised' communities [@dansereauSpatiallyExplicitPredictions2024] and an understanding of the intersection of species interactions and their co-occurrence (*sensu* a fusion of the the Grinnellian and Eltonian niches niche, @gravelBringingEltonGrinnell2019) is meaningful when one is operating in the space of trying to determine the distribution of a species [@higinoMismatchIUCNRange2023; @pollockUnderstandingCooccurrenceModelling2014]. **Abundance** -The abundance of different the species within the community is thought to influence the realisation of feeding links primarily in two ways. Firstly there is the argument that that structure of networks (and the interactions that they are composed of) are driven *only* by the abundance of the different species and that interactions are not contingent on there being any compatibility (trait matching) between them, *sensu* neutral processes [@canardEmergenceStructuralPatterns2012; @momalTreebasedInferenceSpecies2020]. However, a more ecologically sound assumption would be that the abundance of different prey species will influence the distribution of links in a network [@vazquezUnitingPatternProcess2009], be influencing which prey are targeted or preferred by the predator as abundance influences factors such as the likelihood of two species (individuals) meeting [@poisotSpeciesWhyEcological2015; @banvilleDecipheringProbabilisticSpecies2024], or in the dynamic sense will influence the persistence of viable populations. +The abundance of the different species within the community is thought to influence the realisation of feeding links primarily in two ways. Firstly there is the argument that the structure of networks (and the interactions that they are composed of) are driven *only* by the abundance of the different species and that interactions are not contingent on there being any compatibility (trait matching) between them, *sensu* neutral processes [@canardEmergenceStructuralPatterns2012; @momalTreebasedInferenceSpecies2020]. However, a more ecologically sound assumption would be that the abundance of different prey species will influence the distribution of links in a network [@vazquezUnitingPatternProcess2009], by influencing which prey are targeted or preferred by the predator as abundance influences factors such as the likelihood of two species (individuals) meeting [@poisotSpeciesWhyEcological2015; @banvilleDecipheringProbabilisticSpecies2024], or in the dynamic sense will influence the persistence of viable populations. **Profitability (energetics)** -Ultimately, predator choice is underpinned by the energetic cost-benefit (profitability) of trying to catch, kill, and consume prey (where a predator will optimise energy while minimising handling and search time), and is well described within both optimal foraging [@pykeOptimalForagingTheory1984] and metabolic theory [@brownMetabolicTheoryEcology2004]. The energetic cost of feeding is itself can be deconstructed as the energy content as well as the density (abundance) of prey (as this influences search time) and how these will influence which links are realised [@fig-process], with an argument that body size represents a key trait that may capture and influence these processes [@yodzisBodySizeConsumerResource1992; @whiteRelationshipsBodySize2007]. Additional work on on understanding the energetic cost that the environment imposes on an individual [@cherifEnvironmentRescueCan2024] as well as the way a predator uses the landscape to search for prey [@pawarDimensionalityConsumerSearch2012] is bringing us closer to accounting for the energetic cost of realising feeding links. +Ultimately, predator choice is underpinned by the energetic cost-benefit (profitability) of trying to catch, kill, and consume prey (where a predator will optimise energy intake while minimising handling and search time (energy cost)), and is well described within both optimal foraging [@pykeOptimalForagingTheory1984] and metabolic theory [@brownMetabolicTheoryEcology2004]. The energetic cost of feeding is determined by both the energy content as well as the density (abundance) of prey (as this influences search time), and a predator will opt to select the prey type that will be most profitable. Additional work on on understanding the energetic cost that the environment imposes on an individual [@cherifEnvironmentRescueCan2024] as well as the way a predator uses the landscape to search for prey [@pawarDimensionalityConsumerSearch2012] brings us closer to accounting for the energetic cost of realising feeding links. **Non-trophic interactions** -Perhaps not as intuitive when thinking about the processes that determine feeding links (trophic interactions) is thinking about the role of the ability of non-trophic interactions to modify either the realisation or strength of trophic interactions [@golubskiModifyingModifiersWhat2011; @pilosofMultilayerNatureEcological2017]. Non-trophic interactions can modify interactions either 'directly' *e.g.,* predator *a* outcompetes predator *b* or 'indirectly' *e.g.,* mutualistic/facilitative interactions will alter the fine-scale distribution and abundance of species as well as their persistence [@kefiMoreMealIntegrating2012; @kefiNetworkStructureFood2015; @bucheMultitrophicHigherOrderInteractions2024]. The 'unobservable' nature of non-trophic interactions makes them a challenge to quantify, however their importance in network dynamics [@staniczenkoStructuralDynamicsRobustness2010] as well as cascading effects [*e.g.,* @kamaruDisruptionAntplantMutualism2024] should not be overlooked. +Perhaps not as intuitive when thinking about the processes that determine feeding links is accounting for the ability of non-trophic interactions (such as competition) to modify either the realisation or strength of trophic interactions [@golubskiModifyingModifiersWhat2011; @pilosofMultilayerNatureEcological2017]. Non-trophic interactions can modify interactions either 'directly' *e.g.,* predator *a* outcompetes predator *b* or 'indirectly' *e.g.,* mutualistic/facilitative interactions will alter the fine-scale distribution and abundance of species as well as their persistence [@kefiMoreMealIntegrating2012; @kefiNetworkStructureFood2015; @bucheMultitrophicHigherOrderInteractions2024]. The 'unobservable' nature of non-trophic interactions makes them a challenge to quantify, however their importance in network dynamics [@staniczenkoStructuralDynamicsRobustness2010] as well as cascading effects [*e.g.,* @kamaruDisruptionAntplantMutualism2024] should not be overlooked. ## Contextualising the processes that determine species interactions -It should be self evident that the different processes discussed above are all ultimately going to influence the realisation of interactions as well as the structure of a network, however they are acting at different scales of organisation. Both the **co-occurrence** and the **evolutionary compatibility** are valid at the scale of the species pair of interest, that is the *possibility* of an interaction being present/absent is assessed at the pairwise level and one is left with a 'list' of interactions that are present/absent. Although it is possible to build a network (*i.e.,* metaweb) from this information it is important to be aware that the structure of this network is not constrained by real-world dynamics or conditions, and so just because species are able to interact does not mean that they will [@poisotSpeciesWhyEcological2015]. In order to construct a network who's structure is a closer approximation of reality (localised interactions) one needs to take into consideration the properties of the community as a whole and information about the individuals it is comprised of [@quinteroDownscalingMutualisticNetworks2024], which requires more data at the community scale, such as the abundance of species. +It should be self evident that the different processes discussed above will ultimately influence the realisation of interactions as well as the structure of a network, however they are acting at different scales of organisation. Both the **co-occurrence** and the **evolutionary compatibility** are valid at the scale of the species pair of interest, that is the *possibility* of an interaction being present/absent is assessed at the pairwise level and one is left with a 'list' of interactions that are present/absent. Although it is possible to build a network (*i.e.,* metaweb) from this information it is important to be aware that the structure of this network is not constrained by real-world dynamics or conditions (*i.e.,* the community context), and so just because species are able to interact does not mean that they will [@poisotSpeciesWhyEcological2015]. In order to construct a network who's structure is a closer approximation of reality (localised interactions) one needs to take into consideration the properties of the community as a whole and information about the individuals it is comprised of [@quinteroDownscalingMutualisticNetworks2024], which requires more data at the community scale, such as the abundance of species. # Network construction is nuanced -The act of constructing a 'real world' network will ultimately be delimited by its intended use, however the reality is that the empirical collection of interaction data is both costly and challenging to execute [@jordanoChasingEcologicalInteractions2016; @jordanoSamplingNetworksEcological2016], especially if one wants to capture *all* aspects of the processes discussed in @sec-process (owing to the different time and spatial scales they may be operating at). Thus we often turn to models to either predict networks (be that the interaction between two species, or network structure [@strydomRoadmapPredictingSpecies2021]), or as a means to identify missing interactions (gap fill) within an existing empirical dataset [@bitonInductiveLinkPrediction2024; @stockPairwiseLearningPredicting2021; @dallasPredictingCrypticLinks2017], and so for the purpose of this discussion network construction will be synonymous with using a model as a means to represent or predict a network. That is not to say that there is no need for empirical data collection but rather that using a model for food web prediction (or reconstruction) is a more feasible approach as it allows us to make inferences about interactions that are not happening in the 'observable now' [@strydomRoadmapPredictingSpecies2021], with the added benefit that one is able to build some uncertainty into the resulting network [@banvilleDecipheringProbabilisticSpecies2024]. Additionally different models have different underlying philosophies that allow us to capture one or a few of the processes discussed in @sec-process, and although the delimits and defines what inferences can be made from the resulting network it also allows us to isolate and understand how different processes determine interactions [@stoufferAllEcologicalModels2019; @songRigorousValidationEcological2024]. Here we will introduce the three different types of network representations (metawebs, realised networks, and structural networks), how they link back to (and encode) the different processes determining interactions [@fig-process], and broadly discuss some of the modelling approaches that are used to construct these different network types. This is paralleled by a hypothetical case study (Box 1) where we showcase the utility/applicability of the different network representations in the context of trying to understand the feeding dynamics of a seasonal community. +The act of constructing a 'real world' network will ultimately be delimited by its intended use, however the reality is that the empirical collection of interaction data is both costly and challenging to execute [@jordanoChasingEcologicalInteractions2016; @jordanoSamplingNetworksEcological2016], especially if one wants to capture *all* aspects of the processes discussed in @sec-process (owing to the different time and spatial scales they may be operating at). Thus we often turn to models to either predict networks (be that the interaction between two species, or network structure [@strydomRoadmapPredictingSpecies2021]), or as a means to identify missing interactions (gap fill) within an existing empirical dataset [@bitonInductiveLinkPrediction2024; @stockPairwiseLearningPredicting2021; @dallasPredictingCrypticLinks2017], and so for the purpose of this discussion network construction will be synonymous with using a model as a means to represent or predict a network. That is not to say that there is no need for empirical data collection, but rather that using a model for food web prediction (or reconstruction) is a more feasible approach as it allows us to make inferences about interactions that are not happening in the 'observable now' [@strydomRoadmapPredictingSpecies2021], and has the added benefit that one is able to explicitly account for uncertainty within the network construction process [@banvilleDecipheringProbabilisticSpecies2024]. Most importantly different models have different underlying philosophies, this allows isolate and operate within one (or a few) of the processes discussed in @sec-process, and better sets us up to understand how different processes determine interactions [@stoufferAllEcologicalModels2019; @songRigorousValidationEcological2024]. Here we will introduce the three different types of network representations (metawebs, realised networks, and structural networks), how they link back to (and encode) the different processes determining interactions [@fig-process], and broadly discuss some of the modelling approaches that are used to construct these different network types. This is paralleled by a hypothetical case study (Box 1) where we showcase the utility/applicability of the different network representations in the context of trying to understand the feeding dynamics of a seasonal community. ::: {#box-hypothetical .callout-note} # Box 1 - Why we need to aggregate networks at different scales: A hypothetical case study {.unnumbered} @@ -126,31 +126,29 @@ Although it might seem most prudent to be predicting, constructing, and defining ## Models that predict metawebs (feasible interactions) -This is perhaps the most developed group of models; with a variety of approaches having been developed that typically determine the feasibility of an interaction using the trait compatibility between predator and prey (*i.e.* their evolutionary compatibility) to determine 'feeding rules' [@morales-castillaInferringBioticInteractions2015]. These feeding rules are broadly elucidated in two different ways; mechanistic feeding rules can be explicitly defined and applied to a community [*e.g.,* @shawFrameworkReconstructingAncient2024; @dunneCompilationNetworkAnalyses2008; @roopnarineEcologicalModellingPaleocommunity2017] or they are inferred from a community for which there are interaction data and the 'rules' are then applied to a different community [*e.g.,* @strydomGraphEmbeddingTransfer2023; @pichlerMachineLearningAlgorithms2020; @strydomFoodWebReconstruction2022; @caronAddressingEltonianShortfall2022; @llewelynPredictingPredatorPrey2023; @desjardins-proulxEcologicalInteractionsNetflix2017; @eklofSecondaryExtinctionsFood2013; @cirtwillQuantitativeFrameworkInvestigating2019]. The fundamental difference between these two model groups is that 'mechanistic models' rely on expert knowledge and make explicit assumptions on trait-feeding relationships, whereas the 'pattern finding' models are dependent on existing datasets from which to elucidate feeding rules. These models are useful for determining all feasible interactions for a specific community, and owing to the availability of empirical interaction datasets [*e.g.,* @poelenGlobalBioticInteractions2014; @poisotMangalMakingEcological2016; @grayJoiningDotsAutomated2015], as well as the development of model testing/benchmarking tools [@poisotGuidelinesPredictionSpecies2023], means that these models can be validated and (with relative confidence) be used to construct first draft networks for communities for which we have no interaction data [@strydomFoodWebReconstruction2022], and are valuable not only in data poor regions but also for predicting interactions for 'unobservable' communities *e.g.,* prehistoric networks [@yeakelCollapseEcologicalNetwork2014; @frickeCollapseTerrestrialMammal2022] or future, novel community assemblages. Importantly metawebs are inherently 'static' in the sense that they are *not* able to capture dynamic processes (since the notion of feasibility is all or nothing), however they provide a bigger picture context (*e.g.,* understanding the *entire* diet breadth of a species) and often require little data to construct. +This is perhaps the most developed group of models; with a variety of approaches having been developed that typically determine the feasibility of an interaction using the trait compatibility between predator and prey (*i.e.* their evolutionary compatibility) to determine 'feeding rules' [@morales-castillaInferringBioticInteractions2015]. These feeding rules are broadly elucidated in two different ways; mechanistic feeding rules can be explicitly defined and applied to a community [*e.g.,* @shawFrameworkReconstructingAncient2024; @dunneCompilationNetworkAnalyses2008; @roopnarineEcologicalModellingPaleocommunity2017] or they are inferred from a community for which there are interaction data and the 'rules' are then applied to a different community [*e.g.,* @strydomGraphEmbeddingTransfer2023; @pichlerMachineLearningAlgorithms2020; @strydomFoodWebReconstruction2022; @caronAddressingEltonianShortfall2022; @llewelynPredictingPredatorPrey2023; @desjardins-proulxEcologicalInteractionsNetflix2017; @eklofSecondaryExtinctionsFood2013; @cirtwillQuantitativeFrameworkInvestigating2019]. The fundamental difference between these two model groups is that 'mechanistic models' rely on expert knowledge and make explicit assumptions on trait-feeding relationships, whereas the 'pattern finding' models are dependent on existing datasets from which to elucidate feeding rules. These models are useful for determining all feasible interactions for a specific community, and owing to the availability of empirical interaction datasets [*e.g.,* @poelenGlobalBioticInteractions2014; @poisotMangalMakingEcological2016; @grayJoiningDotsAutomated2015], as well as the development of model testing/benchmarking tools [@poisotGuidelinesPredictionSpecies2023], means that these models can be validated and (with relative confidence) be used to construct first draft networks for communities for which we have no interaction data [@strydomFoodWebReconstruction2022], and are valuable not only in data poor regions but also for predicting interactions for 'unobservable' communities *e.g.,* prehistoric networks [@yeakelCollapseEcologicalNetwork2014; @frickeCollapseTerrestrialMammal2022; @dunhillExtinctionCascadesCommunity2024] or future, novel community assemblages. Importantly metawebs are inherently 'static' in the sense that they are *not* able to capture dynamic processes (since the notion of feasibility is all or nothing), however they provide a bigger picture context (*e.g.,* understanding the *entire* diet breadth of a species) and often require little data to construct. ## Models that predict realised networks (realised interactions) -In order to construct realised networks models need to incorporate *both* the feasibility of interactions (*i.e.,* determine the entire diet breadth of a species) as well as then determine which interactions are realised (*i.e.,* incorporate the 'cost' of interactions). As far as we are aware there is no model that explicitly accounts for both of these 'rules' (although see @olivierExploringTemporalVariability2019) and rather *only* account for processes that determine the realisation of an interaction (*i.e.,* abundance, predator choice, or non-trophic interactions). Although the use of allometry *i.e.,* body size [*e.g.,* @valdovinosBioenergeticFrameworkAboveground2023; @beckermanForagingBiologyPredicts2006] may represent a first step in capturing 'evolutionary compatibility' alongside more energy (predator choice) driven processes we still need to account for other traits that determine feeding compatibility [*e.g.,* @vandewalleArthropodFoodWebs2023 show how incorporating prey defensive properties alongside body size improves predictions]. In terms of constructing realised networks, diet models [@beckermanForagingBiologyPredicts2006; @petcheySizeForagingFood2008] have been used construct networks based on both predator choice (as determined by the handling time, energy content, and predator attack rate) as well as abundance (prey density) and progress has also been made in understanding the compartmentation of energy in networks and how this influences energy acquisition [@woottonModularTheoryTrophic2023; @krauseCompartmentsRevealedFoodweb2003]. As realised networks are are build on the concept of dynamic processes (the abundance of species will always be in flux) these networks are valuable for understanding the behaviour of networks over time or their response to change [@lajaaitiEcologicalNetworksDynamicsJlJulia2024; @delmasSimulationsBiomassDynamics2017; @curtsdotterEcosystemFunctionPredator2019]. However, they are 'costly' to construct (requiring data about the entire community as it is the behaviour of the system that determines the behaviour of the part) and also lack the larger diet niche context afforded by metawebs. +In order to construct realised networks models need to incorporate *both* the feasibility of interactions (*i.e.,* determine the entire diet breadth of a species) as well as then determine which interactions are realised (*i.e.,* incorporate the 'cost' of interactions). As far as we are aware there is no model that explicitly accounts for both of these 'rules' (although see @olivierExploringTemporalVariability2019) and rather *only* account for processes that determine the realisation of an interaction (*i.e.,* abundance, predator choice, or non-trophic interactions). Although the use of allometry *i.e.,* body size [*e.g.,* @valdovinosBioenergeticFrameworkAboveground2023; @beckermanForagingBiologyPredicts2006; @yodzisBodySizeConsumerResource1992; @whiteRelationshipsBodySize2007] may represent a first step in capturing 'evolutionary compatibility' alongside more energy (predator choice) driven processes we still need to account for other traits that determine feeding compatibility [*e.g.,* @vandewalleArthropodFoodWebs2023 show how incorporating prey defensive properties alongside body size improves predictions]. In terms of constructing realised networks, diet models [@beckermanForagingBiologyPredicts2006; @petcheySizeForagingFood2008] have been used construct networks based on both predator choice (as determined by the handling time, energy content, and predator attack rate) as well as abundance (prey density) and progress has also been made in understanding the compartmentation of energy in networks and how this influences energy acquisition [@woottonModularTheoryTrophic2023; @krauseCompartmentsRevealedFoodweb2003]. As realised networks are are build on the concept of dynamic processes (the abundance of species will always be in flux) these networks are valuable for understanding the behaviour of networks over time or their response to change [@lajaaitiEcologicalNetworksDynamicsJlJulia2024; @delmasSimulationsBiomassDynamics2017; @curtsdotterEcosystemFunctionPredator2019]. However, they are 'costly' to construct (requiring data about the entire community as it is the behaviour of the system that determines the behaviour of the part) and also lack the larger diet niche context afforded by metawebs. ## Models that predict structure (interaction agnostic) -Although we identify mechanisms that determine species interactions in @sec-process not all models that are used to predict networks explicitly operate at the 'process' level, but rather represent the *structure* of a network based on a series of *a priori* assumptions as to the distribution of links between species (typically trophic not taxonomic species). These models operate by parametrising an aspect of the network structure, (*e.g.,* the niche model [@williamsSimpleRulesYield2000] makes an assumption as to the expected connectance of the network,although see @allesinaFoodWebModels2009 for a parameter-free model) or alternatively uses structural features of an exiting *realised* network (*e.g.,* stochastic block model, @xieCompletenessCommunityStructure2017). Importantly these structural models do not make species specific predictions (they are usually species agnostic and treat nodes as trophic species) and so cannot be used to determine if an interaction is either possible *or* realised between two species (*i.e.,* one cannot use these models to determine if species $a$ eats species $b$). Although this means this suite of models are unsuitable as tools for predicting species-specific interactions, they have been shown to be sufficient tools to predict the structure of networks [@williamsSuccessItsLimits2008], and provide a data-light (the models often only require species richness) but assumption heavy (the resulting network structure is determined by an assumption of network structure) way to construct a network. +Although we identify mechanisms that determine species interactions in @sec-process not all models that are used to predict networks explicitly operate at the 'process' level, but rather represent the *structure* of a network based on a series of *a priori* assumptions as to the distribution of links between (typically trophic not taxonomic) species. These models operate by parametrising an aspect of the network structure, (*e.g.,* the niche model [@williamsSimpleRulesYield2000] makes an assumption as to the expected connectance of the network,although see @allesinaFoodWebModels2009 for a parameter-free model) or alternatively uses structural features of an exiting *realised* network (*e.g.,* stochastic block model, @xieCompletenessCommunityStructure2017). Importantly these structural models do not make species specific predictions (they are usually species agnostic and treat nodes as trophic species) and so cannot be used to determine if an interaction is either possible *or* realised between two species (*i.e.,* one cannot use these models to determine if species $a$ eats species $b$). Although this means this suite of models are unsuitable as tools for predicting species-specific interactions, they have been shown to be sufficient tools to predict the structure of networks [@williamsSuccessItsLimits2008], and provide a data-light (the models often only require species richness) but assumption heavy (the resulting network structure is determined by an assumption of network structure) way to construct a network. # Making Progress with Networks ## Further development of models and tools -There has been a suite of models that have been developed to predict feeding links, however we are lacking in tools that are explicitly taking into consideration estimating both the feasibility as well as realisation of links, *i.e.,* both interactions and structure simultaneously [@strydomRoadmapPredictingSpecies2021]. This could be addressed either through the development of tools that do both (predict both interactions and structure), or to develop an ensemble modelling approach [@beckerOptimisingPredictiveModels2022; @terryFindingMissingLinks2020] or tools that will allow for the downsampling of metawebs into realised networks [*e.g.,* @roopnarineExtinctionCascadesCatastrophe2006]. Additionally although realised networks are more closely aligned with capturing interaction strength we lack models that allow us to quantify this [@wellsSpeciesInteractionsEstimating2013; @strydomRoadmapPredictingSpecies2021]. In addition to the more intentional development of models we also need to consider the validation of these models, there have been developments and discussions for assessing how well a model recovers pairwise interactions [@strydomRoadmapPredictingSpecies2021; @poisotGuidelinesPredictionSpecies2023], although the rate of false-negatives that may be present in the testing data still present a challenge [@catchenMissingLinkDiscerning2023], and we still lack clear set of guidelines for benchmarking the ability of models to recover structure [@allesinaGeneralModelFood2008]. +There has been a suite of models that have been developed to predict feeding links, however we are lacking in tools that are explicitly taking into consideration estimating both the feasibility as well as realisation of links, *i.e.,* both interactions and structure simultaneously [@strydomRoadmapPredictingSpecies2021]. This could be addressed either through the development of tools that do both (predict both interactions and structure), or to develop an ensemble modelling approach [@beckerOptimisingPredictiveModels2022; @terryFindingMissingLinks2020] or tools that will allow for the downsampling of metawebs into realised networks [*e.g.,* @roopnarineExtinctionCascadesCatastrophe2006]. Additionally, although realised networks are more closely aligned with capturing interaction strength we lack models that allow us to quantify this [@wellsSpeciesInteractionsEstimating2013; @strydomRoadmapPredictingSpecies2021]. In addition to the more intentional development of models we also need to consider the validation of these models, there have been developments and discussions for assessing how well a model recovers pairwise interactions [@strydomRoadmapPredictingSpecies2021; @poisotGuidelinesPredictionSpecies2023], although their are still challenges related to the completeness of the datasets used for validation, specifically the challenge of dealing with false-negatives [@catchenMissingLinkDiscerning2023]. In terms of validating the predicted structure of networks, we still lack clear set of guidelines for benchmarking the ability of models to recover structure [@allesinaGeneralModelFood2008]. ## At what scale should we be predicting and using networks? -We lack an understanding of which processes drive interactions at different scales [@saraviaEcologicalNetworkAssembly2022], as well as to what the appropriate level of aggregation for a 'network' is [@estayEditorialPatternsProcesses2023; @moulatletScalingTrophicSpecialization2024; @saberskiImpactDataResolution2024]. Thus we need an understanding of not only how time and scale influence the interpretation of networks [@moralesEffectSpacePlant2008; @bluthgenEcologyMammalsInteraction2021], but how this is in turn influenced by the type of networks used. Which presents a challenge both in deciding what the appropriate spatial and time scales are for constructing not only a network but also which type of network representation. Space influences both network properties [@galianaSpatialScalingSpecies2018], as well as dynamics [@rooneyLandscapeTheoryFood2008; @fortinNetworkEcologyDynamic2021], and time has implications when it comes to accounting for seasonal turnover in communities [@brimacombeInferredSeasonalInteraction2021; @laenderCarbonTransferHerbivore2010] as well as thinking about co-occurrence, particularly the records that are used to determine co-occurence [@brimacombeApplyingMethodIts2024]. Although multilayer networks may allow us to encode the nuances of space and time [@hutchinsonSeeingForestTrees2019] we still need to understand the implications of *e.g.,* constructing networks that are not at ecologically but rather politically relevant scales [@strydomFoodWebReconstruction2022] and what the implications of this disconnect may be. +The appropriate level of aggregation for a 'network' is an emerging discussion within the field [@estayEditorialPatternsProcesses2023; @moulatletScalingTrophicSpecialization2024; @saberskiImpactDataResolution2024], and perhaps presents the biggest challenge if we want to understand how different processes determine interactions [@saraviaEcologicalNetworkAssembly2022], as well as identify the appropraite networks for different research questions [@fig-future]. Thus we need an understanding of not only how time and scale influence the interpretation of networks [@moralesEffectSpacePlant2008; @bluthgenEcologyMammalsInteraction2021], but how this is in turn influenced by the type of network representations used. Space influences both network properties [@galianaSpatialScalingSpecies2018], as well as dynamics [@rooneyLandscapeTheoryFood2008; @fortinNetworkEcologyDynamic2021], and time has implications when it comes to accounting for seasonal turnover in communities [@brimacombeInferredSeasonalInteraction2021; @laenderCarbonTransferHerbivore2010] as well as thinking about co-occurrence, particularly the records that are used to determine co-occurence [@brimacombeApplyingMethodIts2024]. Although multilayer networks may allow us to encode the nuances of space and time [@hutchinsonSeeingForestTrees2019] we still need to understand the implications of *e.g.,* constructing networks that are not at ecologically but rather politically relevant scales [@strydomFoodWebReconstruction2022] and what the implications of this disconnect may be. -# The future value of networks +## Making use of the different network representations -> developing a dictionary of use... that helps navigate between the levels and assumptions - -It should be clear that there is a high degree of interrelatedness and overlap between the way a network is constructed (modelled or predicted) and the process(es) it captures, these are encoded (embedded) within the network representation and ultimately influences how the network can and should be used [@petcheyFitEfficiencyBiology2011; @berlowGoldilocksFactorFood2008], with different network representations yielding different interpretations of processes [@keyesSynthesisingRelationshipsFood2024]. It is probably both this nuance as well as a lack of clear boundaries and guidelines as to the links between network form and function [although see @delmasAnalysingEcologicalNetworks2019] that has stifled the 'productive use' of networks beyond inventorying the interactions between species. Although, progress with using networks as a means to address questions within larger bodies of ecological theory *e.g.,* invasion biology [@huiHowInvadeEcological2019] and co-existence theory [@garcia-callejasNonrandomInteractionsGuilds2023], has been made we still need to have a discussion on what the appropriate network representation for the task at hand would be. This is highlighted in Box 1, and underscores that we need to evaluate exactly what process a specific network representation captures as well as its suitability for the question of interest. +It should be clear that there is a high degree of interrelatedness and overlap between the way in which a network is constructed (modelled or predicted) and the process(es) that it captures [@fig-process], these are encoded (embedded) within the network representation and ultimately influences how the network can and should be used [@petcheyFitEfficiencyBiology2011; @berlowGoldilocksFactorFood2008], with different network representations yielding different interpretations of processes [@keyesSynthesisingRelationshipsFood2024]. It is probably both this nuance as well as a lack of clear boundaries and guidelines as to the links between network form and function [although see @delmasAnalysingEcologicalNetworks2019] that has stifled the 'productive use' of networks beyond the inventorying the interactions between species. Although progress with using networks as a means to address questions within larger bodies of ecological theory *e.g.,* invasion biology [@huiHowInvadeEcological2019] and co-existence theory [@garcia-callejasNonrandomInteractionsGuilds2023] has been made we still lack explicit guidelines as to what the appropriate network representation for the task at hand would be, and as highlighted in Box 1, underscores the need to evaluate exactly what process a specific network representation captures as well as its suitability for the question of interest. In @fig-future we present a mapping of what we believe are some of the key questions for which interaction networks can be used to the different networks representations that are most suitable, as well as highlight some of the methodological challenges that still need to be improved upon. ## How will novel communities interact? @@ -176,7 +174,9 @@ Specific sub points to consider here is persistence, especially persistence to p | Synthetic networks | Creating ecologically plausible communities for synthetic analyses | Structural networks - data light! | | Practical use | What is both attainable (data constraints) but also of practical use to 'real world' decision making. So moving from theory to applied | ??Regional metawebs?? | -: An informative table +: This table representis an alternative approach to try and think about mapping quations to network representations. + + # References {.unnumbered}