Model Tree Approach Research Articles

On Elementary Affective Decisions: To Like Or Not to Like, That Is the Question.

Perhaps the most ubiquitous and basic affective decision of daily life is deciding whether we like or dislike something/somebody, or, in terms of psychological emotion theories, whether the object/subject has positive or negative valence. Indeed, people constantly make such liking decisions within a glimpse and, importantly, often without expecting any obvious benefit or knowing the exact reasons for their judgment. In this paper, we review research on such elementary affective decisions (EADs) that entail no direct overt reward with a special focus on Neurocognitive Poetics and discuss methods and models for investigating the neuronal and cognitive-affective bases of EADs to verbal materials with differing degrees of complexity. In line with evolutionary and appraisal theories of (aesthetic) emotions and data from recent neurocognitive studies, the results of a decision tree modeling approach simulating EADs to single words suggest that a main driving force behind EADs is the extent to which such high-dimensional stimuli are associated with the “basic” emotions joy/happiness and disgust.

Assessing the sensitivity of bovine tuberculosis surveillance in Canada’s cattle population, 2009–2013

The objectives of this study were (1) to estimate the annual sensitivity of Canada’s bTB surveillance system and its three system components (slaughter surveillance, export testing and disease investigation) using a scenario tree modelling approach, and (2) to identify key model parameters that influence the estimates of the surveillance system sensitivity (SSSe). To achieve these objectives, we designed stochastic scenario tree models for three surveillance system components included in the analysis. Demographic data, slaughter data, export testing data, and disease investigation data from 2009 to 2013 were extracted for input into the scenario trees. Sensitivity analysis was conducted to identify key influential parameters on SSSe estimates.The median annual SSSe estimates generated from the study were very high, ranging from 0.95 (95% probability interval [PI]: 0.88–0.98) to 0.97 (95% PI: 0.93–0.99). Median annual sensitivity estimates for the slaughter surveillance component ranged from 0.95 (95% PI: 0.88–0.98) to 0.97 (95% PI: 0.93–0.99). This shows that slaughter surveillance to be the major contributor to overall surveillance system sensitivity with a high probability to detect M. bovis infection if present at a prevalence of 0.00028% or greater during the study period. The export testing and disease investigation components had extremely low component sensitivity estimates—the maximum median sensitivity estimates were 0.02 (95% PI: 0.014–0.023) and 0.0061 (95% PI: 0.0056–0.0066) respectively.The three most influential input parameters on the model’s output (SSSe) were the probability of a granuloma being detected at slaughter inspection, the probability of a granuloma being present in older animals (≥12 months of age), and the probability of a granuloma sample being submitted to the laboratory. Additional studies are required to reduce the levels of uncertainty and variability associated with these three parameters influencing the surveillance system sensitivity.

Effectiveness and Cost Efficiency of Different Surveillance Components for Proving Freedom and Early Detection of Disease: Bluetongue Serotype 8 in Cattle as Case Study for Belgium, France and the Netherlands

Quick detection and recovery of country's freedom status remain a constant challenge in animal health surveillance. The efficacy and cost efficiency of different surveillance components in proving the absence of infection or (early) detection of bluetongue serotype 8 in cattle populations within different countries (the Netherlands, France, Belgium) using surveillance data from years 2006 and 2007 were investigated using an adapted scenario tree model approach. First, surveillance components (sentinel, yearly cross-sectional and passive clinical reporting) within each country were evaluated in terms of efficacy for substantiating freedom of infection. Yearly cross-sectional survey and passive clinical reporting performed well within each country with sensitivity of detection values ranging around 0.99. The sentinel component had a sensitivity of detection around 0.7. Secondly, how effective the components were for (early) detection of bluetongue serotype 8 and whether syndromic surveillance on reproductive performance, milk production and mortality data available from the Netherlands and Belgium could be of added value were evaluated. Epidemic curves were used to estimate the timeliness of detection. Sensitivity analysis revealed that expected within-herd prevalence and number of herds processed were the most influential parameters for proving freedom and early detection. Looking at the assumed direct costs, although total costs were low for sentinel and passive clinical surveillance components, passive clinical surveillance together with syndromic surveillance (based on reproductive performance data) turned out most cost-efficient for the detection of bluetongue serotype 8. To conclude, for emerging or re-emerging vectorborne disease that behaves such as bluetongue serotype 8, it is recommended to use passive clinical and syndromic surveillance as early detection systems for maximum cost efficiency and sensitivity. Once an infection is detected and eradicated, cross-sectional screening for substantiating freedom of infection and sentinel for monitoring the disease evolution are recommended.

Model tree approach for predicting uniaxial compressive strength and Young’s modulus of carbonate rocks

The uniaxial compressive strength (UCS) and Young’s modulus (E) of rock are important parameters for evaluating the strength, deformation, and stability of rock engineering structures. Direct measurement of these parameters is expensive, time-consuming, and even infeasible in some circumstances due to the difficulty involved in obtaining core samples. Recently, soft computing tools have been used to predict UCS and E based on index tests. Most of these tools are not as transparent and easy to use as empirical regression-based models. This study presents another soft computing approach—model trees—for predicting the UCS and E of carbonate rocks. The main advantages of model trees are that they are easier to use than other data learning tools and, more importantly, they represent understandable mathematical rules. In this study, the M5P algorithm was employed to build and evaluate model trees (UCS and E model trees). First, the models were developed in an unpruned form, and then they were pruned to avoid overfitting. The data used to train and test the model trees were collected from quarries in southwestern Turkey. Model trees included Schmidt hammer, effective porosity, dry unit weight, P‐wave velocity, and slake durability index as input variables. When the models were assessed using a number of statistical indices (RMSE, MAE, VAF, and R 2), it was found that unpruned and pruned model trees provide acceptable predictions of UCS and E, although the pruned models are simpler and easier to understand.

Estimation of suspended sediment concentration in the Saint John River using rating curves and a machine learning approach

ABSTRACTSedimentation in navigable waterways and harbours is of concern for many water and port managers. One potential source of variability in sedimentation is the annual sediment load of the river that empties in the harbour. The main objective of this study was to use some of the regularly monitored hydro-meteorological variables to compare estimates of hourly suspended sediment concentration in the Saint John River using a sediment rating curve and a model tree (M5ʹ) with different combinations of predictors. Estimated suspended sediment concentrations were multiplied by measured flows to estimate suspended sediment loads. Best results were obtained using M5ʹ with four predictors, returning an R2 of 0.72 on calibration data and an R2 of 0.46 on validation data. Total load was underestimated by 1.41% for the calibration period and overestimated by 2.38% for the validation period. Overall, the model tree approach is suggested for its relative ease of implementation and constant performance.EDITOR M.C. Acreman; ASSOCIATE EDITOR B. Touaibia

PCN110 - Cost-Minimization Analysis of Capecitabine Versus 5-Fluorouracil for the Treatment of Metastatic Colorectal Cancer: A Decision Tree Model Approach

PCN110 - Cost-Minimization Analysis of Capecitabine Versus 5-Fluorouracil for the Treatment of Metastatic Colorectal Cancer: A Decision Tree Model Approach

Comparison of risk-based versus random sampling in the monitoring of antimicrobial residues in Danish finishing pigs

In Denmark, a monitoring program for residues of antimicrobials in pork is in place involving annual testing of around 20,000 samples from finishing pigs corresponding to 0.1% of the animals slaughtered. Annually, zero to two samples are found above the maximum residue limit. Both authorities and industry have expressed interest in adjusting the monitoring to a risk-based system. The objective of this study was to assess the opportunities and consequences of the monitoring considering: 1) replacing the current bioassay with high-performance liquid chromatography–mass spectrometry (HPLC LC–MS/MS), 2) replacing kidney with muscles as sample matrix, and 3) using indicators to identify high-risk (HR) herds and increase sampling intensity in these herds, lowering sampling in the low-risk (LR) herds, while aiming at continued detection of similar numbers of test-positives at the lowest possible costs.A state-of-the-art stochastic scenario tree modelling approach including economic evaluation of different model outcomes was used. A total of six scenarios were run for penicillin and tetracycline, respectively. Relevant information was obtained through the literature, statistical analysis of existing data as well as consultations with laboratory and slaughterhouse experts. Abattoir recordings of chronic pleuritis were used as an indicator for finishing pig herds (HR=within-herd prevalence>40%). Such risk-based monitoring would have to use muscles and not kidneys, because of logistic challenges in identifying and storing of plucks until testing. However, the bioassay cannot be used on muscle tissue due to low sensitivity for tetracyclines. Different plausible combinations of sample sizes were also modelled.The HPLC LC–MS/MS method detected the same number of cases compared to the bioassay when kidney was used as matrix. HPLC LC–MS/MS has a higher sensitivity when used on muscle but it is almost twice as costly as the bioassay. Risk-based sampling resulted in detection of more residue cases with higher cost-effectiveness than random monitoring. Sampling 7500 HR pigs and 5000 LR pigs resulted in the most cost-effective monitoring among the alternative scenarios. The associated costs would increase by 4%. A scenario involving testing of 5000 HR and 5000 LR animals would result in slightly fewer positives, but 17% savings in costs. The advantages of using HPLC LC–MS/MS compared to the bioassay are a fast response and a high sensitivity for all relevant substances used in pigs. The Danish abattoir companies have implemented a risk-based monitoring similar to the above per January 2016.

Stomatal conductance and intrinsic water use efficiency in the drought year 2003: a case study of European beech

Beech trees were able to cope with the drought of 2003. Harmful water shortage has been avoided by an effective stomatal closure while use of carbon storage pools may have prevented carbon starvation and growth reduction. We applied hydrodynamic modeling together with a tree ring stable isotope approach to identify the physiological responses of beech trees to changing environmental conditions. The drought conditions of the extreme hot and dry summer in 2003 were hypothesized to significantly influence the radial growth of European beech mainly triggered by the stomatal response towards water scarcity leading, in turn, to a decline in carbon assimilation. The functional–structural single tree modeling approach applied, revealed in fact a strong limitation of water use and carbon gain during drought. However, tree ring width data did not show a clear drought response and no differentiation in radial growth during six subsequent years examined (2002–2007) has been observed. Using integrated results from mechanistic carbon–water balance simulations, tree ring carbon and oxygen isotope analysis and tree ring width measurements we postulate that the suggested drought-induced growth decline has been prevented by the remobilization of stored carbohydrates, an early onset in growth and the relatively late occurrence of the severe drought in 2003. Furthermore, we demonstrate that the stomatal response played a significant role in avoiding harmful water tension that would have caused xylem dysfunction. As a result of the combined investigation with physiological measurements (stable isotope approach) and hydrodynamic modeling of stomatal aperture, we could give insights into the physiological control of mature beech tree functioning under drought. We conclude that beech trees have been operating at their hydraulic limits and that the longer or repeated drought periods would have affected the growth considerably.

A model tree approach to forecasting solar irradiance variability

As the penetration of solar power increases, the variable generation from this renewable resource will necessitate solar irradiance forecasts for utility companies to balance the energy grid. In this study, the temporal irradiance variability is calculated by the temporal standard deviation of the Global Horizontal Irradiance (GHI) at eight sites in the Sacramento Valley and the spatial irradiance variability is quantified by the standard deviation across those same sites. Our proposed artificial intelligence forecasting technique is a model tree with a nearest neighbor option to predict the irradiance variability directly. The model tree technique reduces the mean absolute error of the variability prediction between 10% and 55% compared to using climatological average values of the temporal and spatial GHI standard deviation. These forecasts are made at 15-min intervals out to 180-min. A data denial experiment showed that the addition of surface weather observations improved the forecasting skill of the model tree by approximately 10%. These results indicate that the model tree technique can be implemented in real-time to produce solar variability forecasts to aid utility companies in energy grid management.

Contribution of Meat Inspection to the surveillance of poultry health and welfare in the European Union.

In the European Union, Meat Inspection (MI) aims to protect public health by ensuring that minimal hazardous material enters in the food chain. It also contributes to the detection and monitoring of animal diseases and welfare problems but its utility for animal surveillance has been assessed partially for some diseases only. Using the example of poultry production, we propose a complete assessment of MI as a health surveillance system. MI allows a long-term syndromic surveillance of poultry health but its contribution is lowered by a lack of data standardization, analysis and reporting. In addition, the probability of case detection for 20 diseases and welfare conditions was quantified using a scenario tree modelling approach, with input data based on literature and expert opinion. The sensitivity of MI appeared to be very high to detect most of the conditions studied because MI is performed at batch level and applied to a high number of birds per batch.

Neuro-fuzzy model tree approach to virtual sensing of transient diesel soot and NOx emissions

Diesel engine combustion and emission formation are highly nonlinear and thus create a challenge related to engine diagnostics and engine control with emission feedback. This article describes the development of neuro-fuzzy models for prediction of transient NOX and soot emission from a diesel engine. The modeling techniques are motivated by the idea of divide and conquer the input–output space. The complex problem is divided into multiple simpler subproblems, which are then identified using simpler class of models. This article explores two different choices of local models, specifically polynomial and neural networks. The modeling technique is augmented with input relevance algorithm to select the most relevant input regressors. Two algorithms, namely, orthogonal least square and automatic relevance determination, are introduced. The models are data driven, and an advanced experimental setup incorporating a medium duty diesel engine and fast emission analyzers for soot and NOX is used to generate training data. The choice of local models and input relevance algorithm is validated with instantaneous emission recorded during transient schedules different from those used in development. High prediction accuracy, both qualitatively and quantitatively, is demonstrated with low computational cost.

High-quality tree structures modelling using local convolution surface approximation

In this paper, we propose a local convolution surface approximation approach for quickly modelling tree structures with pleasing visual effect. Using our proposed local convolution surface approximation, we present a tree modelling scheme to create the structure of a tree with a single high-quality quad-only mesh. Through combining the strengths of the convolution surfaces, subdivision surfaces and GPU, our tree modelling approach achieves high efficiency and good mesh quality. With our method, we first extract the line skeletons of given tree models by contracting the meshes with the Laplace operator. Then we approximate the original tree mesh with a convolution surface based on the extracted skeletons. Next, we tessellate the tree trunks represented by convolution surfaces into quad-only subdivision surfaces with good edge flow along the skeletal directions. We implement the most time-consuming subdivision and convolution approximation on the GPU with CUDA, and demonstrate applications of our proposed approach in branch editing and tree composition.

Strengths and weaknesses of meat inspection as a contribution to animal health and welfare surveillance

Meat inspection (MI) is one of the most widely implemented and longest running systems of surveillance. It was primarily introduced to identify meat of animals that is not fit for human consumption. Additionally, MI was progressively recognised as a suitable source of data collection and for monitoring a broad spectrum of diseases and conditions concerning animal health and welfare. For Europe, MI tasks are regulated at the European rather than country level and include a set of activities before and after stunning (ante and post mortem inspection) involving visual inspection, palpation and incisions. Over the last decade, the current MI protocol has been challenged because of its low sensitivity for important public health hazards. We aimed to assess the strengths and weaknesses of current MI protocols with primary focus on its utility in the context of animal health – including both notifiable and production diseases – and welfare, i.e. its capacity to detect cases with an aim to quantify the frequency of animal disease and welfare cases. The consequences of an alternative inspection protocol using visual-only inspection were also explored.As a first step, a review of grey and published literature was conducted for a selected number of diseases and welfare conditions in seven species or species groups: swine, poultry, bovines, small ruminants, solipeds and farmed game, represented by red deer, wild boar, rabbits and ostriches. This review highlighted a substantial lack of suitable and accessible published data on the frequency of occurrence of many diseases and conditions affecting food animals in Europe. Additionally, there were very limited data on the detection performance of MI, particularly in relation to specific degrees of severity of clinical signs. Due to the data gaps, a large proportion of input data used in this work was based on expert opinion and general biologic manifestations of the conditions investigated. The probability of case detection was quantified using a scenario tree modelling approach, taking into account the frequency of case presentation and inspection coverage.In general, the performance of MI was highly correlated with the presence of clinical and/or pathological signs in affected animals. Early or subclinical cases were likely to be “non-detectable” at slaughter. Regarding detectable cases, the impact of moving to visual-only inspection was negligible for most notifiable diseases and conditions considered with a few exceptions, primarily detectable cases of tuberculosis. Current MI activities were found to be effective to detect the majority of animal welfare conditions considered by species, predominantly by ante mortem inspection.The effectiveness of MI was also considered for endemic diseases that are not currently subject to systematic control efforts. These included respiratory diseases and parasite infections. It was shown that MI could provide an efficient means of identifying producers in need of animal health advice, provided that information is collected and fed back to veterinarians and livestock farmers. Within an integrated information system, MI could substantially contribute to the control of a considerable range of animal health and welfare issues. Data already collected need to be made available for on-farm decision making. It was also noted that if the slaughter population is strongly affected by international trade, i.e. where a large proportion of animals originate from one country and are slaughtered in another, the usefulness of MI for endemic disease surveillance will be affected by either reduced coverage or bias or both.In conclusion, our results indicate that while ante mortem inspection remains essential for the detection of animal welfare conditions, a move to visual-only post mortem inspection has – for the diseases and conditions considered – negligible negative impact on disease control. However in countries or regions that are not free of TB, special relevance of palpation and cutting of lymph nodes will have to be considered. MI information has considerable potential to inform disease control efforts, but only few countries use it systematically limiting the actual benefit that is achieved by these data. Finally, MI can also provide “back-up” surveillance in a situation where other means of detection fail and may represent the sole means of case detection for certain infections (e.g. liver fluke or cestodes).

Inelastic displacement ratios for soil-structure systems

In this paper, effects of Soil-Structure Interaction (SSI) on Inelastic Displacement Ratios (IDRs) of superstructure are addressed. Four non-dimensional parameters are employed as the crucial parameters which affect the IDRs of soil-structure systems: (1) non-dimensional frequency as the structure-to-soil stiffness ratio; (2) aspect ratio of the superstructure; (3) relative lateral strength of the superstructure; and (4) strain hardening ratio. The soil beneath the superstructure is simulated based on the concept of Cone model. The superstructure is idealized as a nonlinear single-degree-of-freedom (SDOF) oscillator. An in depth sensitivity analysis is conducted to consider the effects of key parameters of soil-structure systems. The results are presented in the form of IDR spectra. The IDR spectra confirm that generally increasing non-dimensional frequency leads to amplification of IDRs. In soil-structure systems, the effect of aspect ratio is dissimilar before and after a threshold period of around 0.65s. Within pre-threshold range, slenderizing superstructure decreases IDR spectra. The trend is reversed for post-threshold range. Increasing strain hardening ratio and relative lateral strength have the same influences on the IDRs of soil-structure system as those of fixed-base structure and give rise to smaller and larger IDRs, respectively. Also, a formula is proposed and respective coefficients are calibrated to obtain IDRs of soil-structure systems using model tree approach. This simple formula can predict the IDRs with acceptable accuracy.

Modelling heavy vehicle car‐following behaviour in congested traffic conditions

SUMMARYThis study develops a car‐following model in which heavy vehicle behaviour is predicted separately from passenger car. Heavy vehicles have different characteristics and manoeuvrability compared with passenger cars. These differences could create problems in freeway operations and safety under congested traffic conditions (level of service E and F) particularly when there is high proportion of heavy vehicles. With increasing numbers of heavy vehicles in the traffic stream, model estimates of the traffic flow could be degrades because existing car‐following models do not differentiate between these vehicles and passenger cars. This study highlighted some of the differences in car‐following behaviour of heavy vehicle and passenger drivers and developed a model considering heavy vehicles. In this model, the local linear model tree approach was used to incorporate human perceptual imperfections into a car‐following model. Three different real world data sets from a stretch of freeway in USA were used in this study. Two of them were used for the training and testing of the model, and one of them was used for evaluation purpose. The performance of the model was compared with a number of existing car‐following models. The results showed that the model, which considers the heavy vehicle type, could predict car‐following behaviour of drivers better than the existing models. Copyright © 2013 John Wiley & Sons, Ltd.

M5 model tree application in daily river flow forecasting in Sohu Stream, Turkey

This study investigate the potential of M5 model tree in predicting daily stream flows in Sohu river located within the municipal borders of Ankara, Turkey. The results of the M5 model tree was compared with support vector machines. Both modelling approaches were used to forecast up to 7-day ahead stream flow. A comparison of correlation coefficient and root mean square value indicates that M5 model tree approach works equally well to the SVM for same day discharge prediction. The M5 model tree also works well up to 7-day ahead discharge forecasting in comparison of SVM with this data set. An advantage of using M5 model tree approach is the availability of simple linear models to predict the discharge as well use of less computational time.

Exploring a Local Linear Model Tree Approach to Car‐Following

AbstractBecause car‐following (CF) models are fundamental to replicating traffic flow they have received considerable attention over the last 50 years. They are in a continuous state of improvement due to their significant role in traffic microsimulations, intelligent transportation systems, and safety engineering models. This article uses the local linear model tree (LOLIMOT) approach to model driver's CF behavior to incorporate human perceptual imperfections into a CF model. This model defines some localities in the input space. These localities are fuzzy and have overlaps with each other. Specific models for each of the localities are then defined and combined in a fuzzy manner to predict the final output. The model was developed using real world dynamic data sets. Three different data sets were used for training, testing, and validating the model. The performance of the model was compared to a number of existing CF models. The results showed very close agreement between the real data and the LOLIMOT outputs.

New Car-Following Model considering Impacts of Multiple Lead Vehicle Types

In the past decade, the development and the application of traffic micro-simulation to replicate real-world traffic behavior have become pervasive among traffic and transport researchers. The modeling of a driver's car-following behavior, which forms the fundamental component of traffic microsimulation, has meanwhile been an important research direction leading to the sophistication of traffic microsimulation. However, recent studies have pointed out that a driver's following behavior varies when the lead vehicle is a passenger car as opposed to a heavy vehicle. Nevertheless, existing models do not precisely address those differences. This oversight could diversely affect the accuracy of traffic microsimulations, particularly with the current trend of an increasing number of heavy vehicles in the traffic stream. A novel car-following model that considered the heterogeneity of lead vehicles was developed. Two types of lead vehicles were considered in this study: passenger cars and heavy vehicles. The model was developed on the basis of the local linear model tree approach. This approach is able to incorporate human perceptual imperfections into a car-following model. The input space is partitioned incrementally, and a linear model is developed for each locality (partition). The final output is calculated by the fuzzy combination of local models according to the validity function of each model. For training and testing purposes, two real-world data sets were obtained from a U.S. freeway under congested traffic conditions. The results showed very close agreement between the real data and the outputs of the proposed model.

Evaluating meteorological comparability in air quality studies: Classification and regression trees for primary pollutants in California's South Coast Air Basin

Meteorology confounds the comparison of air quality data across time and space. This presents challenges, for example, to comparisons of pollutant concentration data obtained with mobile monitoring platforms on different days and/or locations within the same airshed. In part to address this challenge, we employed a classification and regression tree (CART) modeling approach that can serve as a useful and straightforward tool in such air quality studies, to determine the comparability of meteorological conditions between measurement days and locations as well as to compare primary pollutant concentrations corrected by meteorological conditions. Specifically, regression trees were developed to obtain representative concentrations of traffic-related primary air pollutants such as NOx and CO, based on meteorological conditions for 2007–2009 in the California South Coast Air Basin (SoCAB). The resulting regression trees showed strong correlations between the regression classifications developed for different pollutant metrics, such as daily CO and NOx maxima, as well as between monitoring sites. For the SoCAB, the most important meteorological parameters controlling primary pollutant concentrations were the mean surface wind speed, geopotential heights at 925 mbar, the upper air north–south pressure gradient, the daily minimum temperature, relative humidity at 1000 mbar, and vertical stability, in approximate order of importance. The value of developing a regression tree for a single season was also explored by performing CART analysis separately on summer data. Although seasonal classifications were similar to those developed from annual data, the standard deviations of the classification groups were somewhat reduced.

Diatom-inferred decline of macrophyte abundance in lakes of southern Quebec, Canada

Submerged macrophytes play a central role in lake ecosystem functioning; however, their long-term dynamics are poorly understood. We have used the paleolimnological top–bottom approach to reconstruct changes in whole-lake macrophyte biomass between predisturbance and present-day conditions in 37 southern Quebec lakes. Estimates of whole-lake macrophyte biomass were produced using a diatom-based multivariate regression tree model (MRT) and modern analogue approach. Both approaches indicated an overall pattern of declining macrophyte abundance in the region. Based on MRT analysis, 80% of study lakes were classified as historically being in the macrophyte-dominated state, but now only 43% of the lakes are currently in this state. The lakes that shifted MRT group were found to have significantly (p = 0.03) greater building densities in their catchments compared with lakes that did not shift state. These results suggest that human impacts, primarily nutrient inputs and water level fluctuations, have played a role in reducing macrophyte abundance in southern Quebec lakes. Because submerged macrophyte beds help stabilize lake ecosystems and act as a phosphorus sink, a reduction in whole-lake macrophyte biomass could make lake ecosystems more susceptible to eutrophication.