Please enable JavaScript.

Coggle requires JavaScript to display documents.

fluv_vio_defense - Coggle Diagram

- - - - flow velocity is the main environmental factor controlling the abundance and distribution of aquatic macrophytes (Chambers et al., 1991; Baattrup-Pedersen and Riis, 1999; Madsen et al., 2001; Janauer et al., 2010)
        
        reduced plant growth
        
        high drag forces : uprooting, steam breakage, morph
        
        controll of subtrate size
        
        varibale correlation
      - Their distribution and abundance are primarily determined by the hydrologic regime (frequency, duration, and intensity of flood events) (Riis and Biggs, 2003; Franklin et al., 2008),
    - - Unnatural high concentrations of phosphorus, as often occur in agricultural watersheds, can stimulate excessive macrophyte growth (Mainstone and Parr, 2002). However, these factors are still most probably overshadowed by the hydraulic conditions (Barendregt and Bio, 2003; Steffen et al., 2014), as is also indicated by the high accuracy of our model.
        Data
      - light availability (sediments!), nutrient content of riverbed and water (agriculture run off)
  - - - why our models is good
        
        independent validation
        
        With random forests we also applied a distribution modeling technique that is capable of modeling complex interactions among predictor variables and is considered to have one of the greatest discriminatory capacities (Elith et al., 2006; Cutler et al., 2007).
        
        the model is parametric, if parameters are learned or inferred based on the data. A tree in this sense is nonparametric. Of course the tree depth is a parameter of the algorithm, but it is not inherently derived from the data, but rather an input parameter that has to be provided by the user.
        
        true presence / abscene data
        
        If the goal of an analysis is prediction rather than formal explanation of hypotheses, machine learning provides a set of tools that can dramatically improve results.
      - spatial auto correlation
        
        macrophytes in the study area appear in distinct patches (positive spatial autocorrelation). This is not surprising and also no problem when the model is able to explain it (Kühn and Dormann 2012). Our model validation against independent data still showed a high predictive performance, which proves the usefulness of the IFIM approach even without accounting for spatial autocorrelation in the model building process (Araújo and Guisan, 2006)
        
        We tested the error between the observed and predicted macrophyte occurrence (probability) for spatial auto-correlation with the moran’s I index (see L219-223) to see if there is a pattern. All species models showed some degree of spatial autocorrelation in the prediction error. This means our models do not account for all factors causing the species’ distributional pattern.
        
        he BRT model residuals were examined for spatial autocorrelation by calculating Moran’s I for discrete distance classes using a lag of 10 m and testing for significance ([76]). The residuals of six models showed significant spatial autocorrelation, indicating that there was spatial dependence of the observations that could not be accounted for by the fitted model (S4 Table). Therefore, the spatial structure of the data was summarized into a dummy variable describing the position of plots along the river channel (distance from the upstream edge of the study area along main flow path). In addition, this variable reflected the potential dispersal and other effects along the course of the river ([18]). This dummy variable was integrated into the BRT models and its incorporation reduced spatial autocorrelation of five model residuals. However, the residuals of the lichen richness model remained spatially autocorrelated (S4 Table), which should be taken into account when examining the results of this model.
        
        A problem arises with statistical species distribution models (like the suggested GLM) if the residuals (observation minus prediction) show spatial autocorrelation as this violates the assumption of independence and may result in an underestimation of coefficients variances and incorrect parameter estimates (Lennon, 2000). However, both our modelling approaches (deterministic and the non-parametric RF) do not make any assumption on independency of the residuals.
        
        statistical models (regression models) are based on the assumption that the observations for model building are independent - due spatial autocorrelation they are not
      - algorithm
        
        determintic approach
        
        geometric mean
        
        The geometric mean assumes that each environmental variable is equally important, and it can represent the nth root of the product of n individual indices. This approach also implies some compensation, yet as with the product equation, it yields zero suitability for any zero HSI value (Yi, Tang, et al., 2014)
        
        random forest
        
        static model ignore more complex processes
        
        interspecific competition and feedbacks between the plants and the physical environment known as niche construction (Corenblit et al., 2009)
      - uncertainty
        
        Uncertainty in SDMs results both from data deficiencies (e.g., missing covariates, and samples of species occurrences that are small, biased, or lacking absences) and from errors in specification ofthe model (Barry & Elith 2006).
      - niche? fundamental or realized
        
        Fundamental niche is the entire set of conditions under which an animal (population, species) can survive and reproduce itself. Realized niche is the set of conditions actually used by given animal (pop, species), after interactions with other species (predation and especially competition) have been taken into account.
        
        The occurrence or abundance of a species along an environmental gradient often follows Shelford's Law of Tolerance (Shelford, 1911; Odum, 1971): each species thrives best at a particular value (its optimum) and cannot survive when the value is either too low or too high. Each species' occurrence is thus confined to a limited range, its niche. The fundamental niche of a species is determined by physiological processes and cannot normally be observed in the real world because species coexist in communities. What can be observed is the realised niche as modified by competition among species and other intra-ommunity processes. It is the realised niche that is of interest in applied ecology. Species tend to separate their niches, partly so as to minimize competition.
        
        but it looks like myriophyllum aquaticum and potamogetum have really distinct niches, nich sepaeration
      - collinearity
        
        Strong correlation between explanatory variables in a regression model - e.g. When an explanatory variable is nearly a linear combination of other explanatory variables
        
        individual influence of predictors on the response variable becomes difficult to distinguish.
        
        Coefficient estimates become unstable and show extremely large standard errors and insignificant coefficients may be assigned to relevant predictors
        
        Collinearity can be dealt with through apriori variable selection or elimination, or combination of correlated variables into new explanatory terms.
        
        The BRT methodology is well suited for analysing data with multicollinearity issues and performs well in separating the individual effects of collinear predictors ([59]). However, it must be noted that the three most influential site productivity variables, namely litter volume, soil water content and soil organic matter were moderately or strongly correlated (rho between 0.69 and 0.88; S2 Table). Therefore, their individual effects on vegetation may have been difficult to separate even with advanced modelling methods. Moreover, litter volume and soil organic matter were moderately correlated with fluvial disturbance variables (|rho| between 0.51 and 0.69; S2 Table).
        
        can result in overfitting
        
        Lack of model generality is often a result of overfitting
        (Chatfield 1995; Sarle 1995), which can be defined as accepting
        a predictor variable (or a form of a predictor variable, such as
        a squared term or interaction term) that is nominally correlated with the response variable in the data set, but which does
        not represent a relationship that holds generally
      - non-stationarity (the variation in modelled relationships over space)
        
        Geographically weighted regression (GWR), a local variant of traditional (global) regression, has since become the preeminent statistical method to deal with spatial non-stationarity explicitly
        
        A process is called stationary if the statistics that deﬁne it and measured within any subsetaccurately describe the statistics of the entire data
        
        not fulfilled when species change their habitat preference from region to another
  - - - naturalized in Portugal having appeared for the first time more than 10 years ago.
      - It is considered invasive because “it is a naturalized alien plant that produce reproductive offspring in very large numbers”.
      - It was already nearly absent at velocities over 0.1 m/s
  - - - nich construction
- - - - lotic (river)
      - lentic (lakes
    - - benthos (river bes)
      - zoo and phytoplancton
      - invertebrates
        
        no backbone
        
        anthropods (insects)
      - vertebrates
        
        backbone
        
        fish etc
    - - In contrast to species diversity, functional diversity measures the distribution and the range of what organisms do in communities and ecosystems, and thus considers the complementarity and redundancy of co-occurring species (
      - The maintenance of functional diversity can guarantee the continuation of important ecosystem processes and functions despite decreased species numbers
    - - cannocial correspondence analyses
        
        Canonical correspondence analysis is the method that extracts the "best" synthetic gradients from field data on biological communities and environmental features: it forms a linear combination of environmental variables that maximally separates the niches of the species.
        
        The primary result of a CCA is an ordination diagram, i.e. a graph with a coordinate system formed by ordination axes (i.e. the synthetic gradients extracted by CCA
        
        unravel how a multitude of species simultaneously respond to external factors, such as environmental variables, pollutants and management regime
      - p-value
        
        The p value is the evidence against a null hypothesis. The smaller the p-value, the stronger the evidence that you should reject the null hypothesis.
      - principal component analyses
        
        So to sum up, the idea of PCA is simple — reduce the number of variables of a data set, while preserving as much information as possible.
    - - Many reasons could explain why riparian vegetation is excluded from the river elements that
        must be monitored to apply the WFD. Understanding about the interactions between
        vegetation and water/sediment flows that determine the form and dynamics of river channels
        has increased since the 1980s (Dufour et al., 2019). However, some of the most relevant
        results, widely disseminated by Gurnell et al., (2002 a,b), Corenblit et al. (2007, 2009a,b, 2011)
        and other authors, appeared several years after the WFD. Consequently, we argue that the
        WFD was conceived under the traditional geomorphological approach, based on water and
        sediment interactions, without including vegetation as a third essential element of river
        dynamics. Therefore, we also argue that the WFD should be reviewed and updated to
        incorporate knowledge about the influence of riparian vegetation in river hydromorphology
        that has emerged since its approval.
  - - - threshold dependent
        
        sensitivity/specificity
        
        confusion matrix
        
        sensitivity correct identified presences
        
        specificity correct identified absences
        
        how to select threshhold
        
        Transforming the results of species distribution modelling from probabilities of or suitabilities for species occurrence to presences/absences needs a specific threshold.
        
        we used the threshold of occurrence that maximizes the sum of sensitivity and specificity (Cantor et al., 1999; Liu et al., 2005).
        
        kappa
        
        kappa: It measures the extent to which the agreement between
        observed and predicted is higher than that expected by
        chance alone.
        
        Overall accuracy (ao),
        
        Overall accuracy (ao), defined as the probability that a site (either presence or absence) is correctly predicted, is the most common measure used in various disciplines including ecology (Fielding nd Bell 1997).
        
        true Skill
      - AUC
        
        the area under the curve (often referred to as simply the AUC) is equal to the probability that a classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one (assuming 'positive' ranks higher than 'negative').
        
        illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied
        
        The AUC indicates the area under a ROC, ranging from 0 to 1 (so that 0.5 indicates no discrimination and 1.0 perfect classification)
      - validation data
        
        data resampling (split samples, cross-validation, bootstrapping) or, more rarely
        
        Independent validation
      - Where modelers aim to explain patterns or generate hypothe- ses (e.g., in evolutionary biology and classical ecological studies), results are generally assessed using statistical tests of model fit and comparison with existing knowledge. In contrast, when prediction is the aim, evaluation targets predictive ability and current practice usually involves testing predictive performance using data resampling (split samples, cross-validation, bootstrap- ping) or, more rarely, independent data sets. Most
    - - spatial auto correlation
        
        macrophytes in the study area appear in distinct patches (positive spatial autocorrelation). This is not surprising and also no problem when the model is able to explain it (Kühn and Dormann 2012). Our model validation against independent data still showed a high predictive performance, which proves the usefulness of the IFIM approach even without accounting for spatial autocorrelation in the model building process (Araújo and Guisan, 2006)
        
        We tested the error between the observed and predicted macrophyte occurrence (probability) for spatial auto-correlation with the moran’s I index (see L219-223) to see if there is a pattern. All species models showed some degree of spatial autocorrelation in the prediction error. This means our models do not account for all factors causing the species’ distributional pattern.
        
        he BRT model residuals were examined for spatial autocorrelation by calculating Moran’s I for discrete distance classes using a lag of 10 m and testing for significance ([76]). The residuals of six models showed significant spatial autocorrelation, indicating that there was spatial dependence of the observations that could not be accounted for by the fitted model (S4 Table). Therefore, the spatial structure of the data was summarized into a dummy variable describing the position of plots along the river channel (distance from the upstream edge of the study area along main flow path). In addition, this variable reflected the potential dispersal and other effects along the course of the river ([18]). This dummy variable was integrated into the BRT models and its incorporation reduced spatial autocorrelation of five model residuals. However, the residuals of the lichen richness model remained spatially autocorrelated (S4 Table), which should be taken into account when examining the results of this model.
        
        A problem arises with statistical species distribution models (like the suggested GLM) if the residuals (observation minus prediction) show spatial autocorrelation as this violates the assumption of independence and may result in an underestimation of coefficients variances and incorrect parameter estimates (Lennon, 2000). However, both our modelling approaches (deterministic and the non-parametric RF) do not make any assumption on independency of the residuals.
        
        statistical models (regression models) are based on the assumption that the observations for model building are independent - due spatial autocorrelation they are not
        
        n contrast to multiple regression, RF is an algorithmic procedure that makes no a priori assumptions about the relationship between the predictor variables and the response. RF has a reputation for good predictive performance when the data contain a large number predictor variables, and when there are complex nonlinearities and interaction effects in the relationship between the predictors and response variable [44, 48, 49].
        
        Autocorrelation literally means that a variable is correlated with itself.� The simplest definition of autocorrelation states that pairs of subjects that are close to each other are more likely to have values that are more similar, and pairs of subjects far apart
      - uncertainty
        
        Uncertainty in SDMs results both from data deficiencies (e.g., missing covariates, and samples of species occurrences that are small, biased, or lacking absences) and from errors in specification ofthe model (Barry & Elith 2006).
      - collinearity
        
        Strong correlation between explanatory variables in a regression model - e.g. When an explanatory variable is nearly a linear combination of other explanatory variables
        
        individual influence of predictors on the response variable becomes difficult to distinguish.
        
        Coefficient estimates become unstable and show extremely large standard errors and insignificant coefficients may be assigned to relevant predictors
        
        Collinearity can be dealt with through apriori variable selection or elimination, or combination of correlated variables into new explanatory terms.
        
        The BRT methodology is well suited for analysing data with multicollinearity issues and performs well in separating the individual effects of collinear predictors ([59]). However, it must be noted that the three most influential site productivity variables, namely litter volume, soil water content and soil organic matter were moderately or strongly correlated (rho between 0.69 and 0.88; S2 Table). Therefore, their individual effects on vegetation may have been difficult to separate even with advanced modelling methods. Moreover, litter volume and soil organic matter were moderately correlated with fluvial disturbance variables (|rho| between 0.51 and 0.69; S2 Table).
        
        can result in overfitting
        
        Lack of model generality is often a result of overfitting
        (Chatfield 1995; Sarle 1995), which can be defined as accepting
        a predictor variable (or a form of a predictor variable, such as
        a squared term or interaction term) that is nominally correlated with the response variable in the data set, but which does
        not represent a relationship that holds generally
      - non-stationarity (the variation in modelled relationships over space)
        
        Geographically weighted regression (GWR), a local variant of traditional (global) regression, has since become the preeminent statistical method to deal with spatial non-stationarity explicitly
        
        A process is called stationary if the statistics that deﬁne it and measured within any subsetaccurately describe the statistics of the entire data
        
        not fulfilled when species change their habitat preference from region to another
      - overfitting
        
        model is fitted against noise and not the true predictor variables
        
        no transferibility
        
        validation against independent data doesnt work
    - - deterministic approach, univariate
        
        geometric mean
      - why random random forest?
      - static verus dynamic
        
        interspecific competition and feedbacks between the plants and the physical environment known as niche construction (Corenblit et al., 2009)
      - GLM
        
        The first technique we introduce here is called Generalized Linear
        Models (GLM), which represents an extension (generalization) of the classical linear regression method.
      - GAM
        
        Generalised Additive Models (GAMS)
    - - fundamental or realized niche?
        
        Fundamental niche is the entire set of conditions under which an animal (population, species) can survive and reproduce itself. Realized niche is the set of conditions actually used by given animal (pop, species), after interactions with other species (predation and especially competition) have been taken into account.
        
        The occurrence or abundance of a species along an environmental gradient often follows Shelford's Law of Tolerance (Shelford, 1911; Odum, 1971): each species thrives best at a particular value (its optimum) and cannot survive when the value is either too low or too high. Each species' occurrence is thus confined to a limited range, its niche. The fundamental niche of a species is determined by physiological processes and cannot normally be observed in the real world because species coexist in communities. What can be observed is the realised niche as modified by competition among species and other intra-ommunity processes. It is the realised niche that is of interest in applied ecology. Species tend to separate their niches, partly so as to minimize competition.
      - presence/absence or probability
  - - - research in the southern hemisphere and toother biomes than t he temperate and Mediterranean climate
        
        The tropical and subtropical regions where the world’s most biodiverse river basins (the Amazon, Congo, and Mekong)
      - new processes
        
        build a gap model (induvidual tree based model) model with functional types / representative species
        
        account for competition and facilitation processes
        
        since they directly control the dominance of specific species by resource competition and the creation of favorable settlement conditions for other species by facilitation
        
        direct interactions of vegetation with geo-morphodynamic processes
        
        interactions between vegetation and the flow field / sediment transport / river bank accretion
        
        facilitation by ecosystem engineers (e.g. Salicaceae specie, Sparganium erectum) that modify the flow field and trap sediments around (point) bars
      - Testing of model reliability through validation against independent data from different geographic regions or spatially distinct subsets of the region or different time periods.
      - thresholds for plant resitance to develop modes
        
        work by poletti
        
        work by gurnell
    - - olden und poff 2008
        
        Machine learning methods without tears: a primer for ecologists
        
        arguments for random forest
      - tockner and staford 2002
      - Riis, T., and Biggs, B. J. F. (2003). Hydrologic and hydraulic control of macrophyte establishment and performance in streams
      - Tockner, K., and Stanford, J. A. (2002). Riverine flood plains: Present state and future trends. Environ. Conserv. 29, 308–330. doi:10.1017/S037689290200022X.
      - Formann, E., Egger, G., Hauer, C., and Habersack, H. (2014). Dynamic disturbance regime approach in river restoration: Concept development and application. Landsc. Ecol. Eng. 10, 323–337. doi:10.1007/s11355-013-0228-5.
      - Roberts, D. R., Bahn, V., Ciuti, S., Boyce, M. S., Elith, J., Guillera-Arroita, G., et al. (2017). Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography (Cop.). 40, 913–929. doi:10.1111/ecog.02881.
        
        spatial blocking
    - - riverine vegetation
        
        most studies look at either riparian or aquatic
        
        hydrogeomorphological interactions very similar
        
        becomes evident when looking at emergent macrophytes
      - bridging a gap between research and practioner
        
        what is practical for a river manager?
    - - paper 1
      - paper 2
      - paper 3
    - - paper 1
        
        casimir
        
        random forest over multiple regression
  - - - for me the practical aspect was important
    - - Whether the assumption of equilibrium between vegetation and environmental factors inherent of static models can be justified due to the reduction of fluvial dynamics.
      - The underlying research question is how river regulation affects the patterns of aquatic and riparian vegetation and
      - why models for river manageners?
        
        For management, conservation and restoration it is therefore fundamental to understand and predict the spatial pattern of the riverine vegetation (Arthington et al., 2010; Franklin, 2013). Ecological models can help this purpose because they allow the testing of hypothesis regarding the functioning of the ecosystem, the definition of ecological references (null models) for restoration and predictions of the environmental impacts resulting from alternative management scenarios or changes in the environment (Franklin, 2010; Schmolke et al., 2010). Their main advantage when compared to actual physical experiments is that they need less time and financial inputs (Hannon and Ruth, 1998).
- - - - herb phase
      - pioneer phase
        
        ruderal or stress-tolerant strategy hirachy adapted to frequent disturbance and strong hydrological variability (wetting and drying)
      - shrub phase
      - softwood phase
        
        like willows and cottonwoods (‘softwood forest’), such as Populus nigra and Salix alba in
      - transition phase
        
        the species present during this phase are recruited to the understorey in the Early Successional Woodland Phase.
        
        when a a gap is created in the softwood canopy the first shade-tolerant (‘hardwood forest’) understorey species may establish: Fraxinus spp., Quercus spp., Ulmus spp.
      - mature phase
    - - Clements' original concept of a unique deterministic successional trajectory drastically evolved during the 1980s with physical disturbances or stress being identified as the main or secondary drivers of change
        
        the trait approach, more representative of biogeomorphological interactions and feedbacks from
        a functional point of view than the classic taxonomic approach, is the best solution to discriminate between phases of the FBS model.
        
        plant trait guilds
        
        C-S-R Triangle theory (grimme)
        
        stress tolerators
        
        ruderalis
        
        competitors
        
        oldest trait based classification
        
        Riparian fow response guilds are non-taxonomic
        groups of species that have similar morphological and
        physiological trait syndromes and as a consequence, the
        species in these guilds respond to water availability and
        fuvial disturbance in similar ways (Merritt et al. 2010).
        
        co occuring species that share similar traits that suit them to cope flood and dry stress in floodplaingroup ofco-occurringspeciesthatsharesimilartraitsthatsuitthemtointeractwithfloodingandinundationandwateravailabilityindif-ferentways
        
        Trait-based framework
        
        The authors stated that rivers across the world exhibit a huge taxonomical variability, but equivalent environmental constraints apparently led to convergent patterns of adaptive traits across taxa.
        
        a theoretical multiple pathway framework of succession
        
        This is the case of riparian and floodplain areas within river corridors where both fluvial landforms and vegetation are exposed to important variations in hydrosedimentary dynamics, resulting in specific plant strategies operating at different spatiotemporal gradients and scales
        
        The fluvial biogeomorphological succession concept (FBS model hereafter; Corenblit et al., 2007)
        
        Classically, a floristic approach has been used to describe changes in vegetation during succession (Amoros & Wade, 1996). Such an approach is very sensitive to local (habitat) and regional (biogeography) settings and may limit the power of generalization among river systems with different characteristics. Furthermore, species taxonomy does not directly address the underlying mechanisms coupling physical and biological processes. An alternative is to consider plant (functional) traits instead, or in addition to, species composition. During the past two decades, the trait‐based approach has been increasingly and successfully used in community ecology . It is less affected by local variability and may facilitate generalization among systems (Shipley et al., 2016).
  - - - steady verus unsteady flow events
      - fixed topography
    - - supply of moisture, vegetation less dependent on precipitation (Stromberg etal., 1996). In semi-arid regions, very important.
- - - - a lack of natural reference sites for orientation, 90% of river floodplains in Europe and North America are used for agriculture or forestry and no longer harbor natural vegetation communities (Tockner and Stanford, 2002).
    - - florisitc
    - - It has often been thought of as a historic, pre-human reference condition but environmental conditions have changed since pre-human times