Efficient Loss Landscape Reshaping for Convolutional Neural Networks.

Theoretical and empirical evidence highlights a positive correlation between the flatness of loss landscapes around minima and generalization. However, most current approaches that seek to find flat minima either incur high computational costs or struggle to balance generalization, training stability, and convergence. This work proposes reshaping the loss landscape to induce the optimizer toward flat regions, an approach that has negligible computational costs and does not compromise training stability, convergence, or efficiency. We focus on nonlinear, loss-dependent reshaping functions underpinned by theoretical insights to reshape the loss landscape. To design these functions, we first identify where and how these functions should be applied. With the aid of recently developed tools in stochastic optimization, theoretical analysis shows that steepening the low-loss landscape improves the rate of sharp minimum escape while flattening the high-and ultralow-loss landscapes enhances training stability and optimization performance, respectively. Simulations and experiments reveal that the subtly designed reshaping functions not only induce optimizers to find flat minima and improve generalization performance but also stabilize training, promote optimization, and keep efficiency. Our approach is evaluated on image classification, adversarial robustness, and natural language processing (NLP) tasks and achieves significant improvement in generalization performance with negligible computational cost. We believe that the new perspective introduced in this work will broadly impact the field of deep neural network training. The code is available at https://github.com/LongJin-lab/LLR.

Efficient adaptive learning rate for convolutional neural network based on quadratic interpolation egret swarm optimization algorithm.

Convolutional neural network (CNN) has recently become popular for addressing multi-domain image classification. However, most existing methods frequently suffer from poor performance, especially in performance and convergence for various datasets. Herein, we have proposed an algorithm for multi-domain image classification by introducing a novel adaptive learning rate rule to the conventional CNN. Specifically, we adopt the CNN to extract rich feature representations. Given that the hyperparameters of the learning rate have a positive effect on the prediction error, the Egret Swarm Optimization Algorithm (ESOA) is introduced to update the learning rate, which can jump out of local extrema during exploration. Therefore, combined with quadratic interpolation, the objective function can be approximated by a polynomial, thereby improving its prediction accuracy. To verify the robustness of the proposed algorithm, we conducted comprehensive experiments in five domain public datasets to fulfil the task of image classification. Meanwhile, the highest accuracy rate of 97.15 % was obtained on the test set. The performances of our method on 24 benchmark functions (CEC2017 and CEC2022) are compared with Particle Swarm Optimization (PSO), Genetic Algorithm(GA), Whale Optimization Algorithm(WOA), Catch Fish Optimization Algorithm(CFOA), GOOSE Algorithm(GO) and ESOA. In two benchmark sets, the performance metric values of our algorithm rank no. 1, especially in all unimodal functions in contrast with other baseline algorithms.

Employing hybrid deep learning for near-real-time forecasts of sensor-based algal parameters in a Microcystis bloom-dominated lake.

Harmful cyanobacterial blooms (CyanoHABs) are increasingly impacting the ecosystem of lakes, reservoirs and estuaries globally. The integration of real-time monitoring and deep learning technology has opened up new horizons for early warnings of CyanoHABs. However, unlike traditional methods such as pigment quantification or microscopy counting, the high-frequency data from in-situ fluorometric sensors display unpredictable fluctuations and variability, posing a challenge for predictive models to discern underlying trends within the time-series sequence. This study introduces a hybrid framework for near-real-time CyanoHABs predictions in a cyanobacterium Microcystis-dominated lake - Lake Dianchi, China. The proposed model was validated using hourly Chlorophyll-a (Chl a) concentrations and algal cell densities. Our results demonstrate that applying decomposition-based singular spectrum analysis (SSA) significantly enhances the prediction accuracy of subsequent CyanoHABs models, particularly in the case of temporal convolutional network (TCN). Comparative experiments revealed that the SSA-TCN model outperforms other SSA-based deep learning models for predicting Chl a (R2 = 0.45-0.93, RMSE = 2.29-5.89 µg/L) and algal cell density (R2 = 0.63-0.89, RMSE = 9489.39-16,015.37 cells/mL) at one to four steps ahead predictions. The forecast of bloom intensities achieved a remarkable accuracy of 98.56 % and an average precision rate of 94.04 % ± 0.05 %. In addition, scenarios involving various input combinations of environmental factors demonstrated that water temperature emerged as the most effective driver for CyanoHABs predictions, with a mean RMSE of 2.94 ± 0.12 µg/L, MAE of 1.55 ± 0.09 µg/L, and R2 of 0.83 ± 0.01. Overall, the newly developed approach underscores the potential of a well-designed hybrid deep-learning framework for accurately predicting sensor-based algal parameters. It offers novel perspectives for managing CyanoHABs through online monitoring and artificial intelligence in aquatic ecosystems.

KAGN:knowledge-powered attention and graph convolutional networks for social media rumor detection.

Rumor posts have received substantial attention with the rapid development of online and social media platforms. The automatic detection of rumor from posts has emerged as a major concern for the general public, the government, and social media platforms. Most existing methods focus on the linguistic and semantic aspects of posts content, while ignoring knowledge entities and concepts hidden within the article which facilitate rumor detection. To address these limitations, in this paper, we propose a novel end-to-end attention and graph-based neural network model (KAGN), which incorporates external knowledge from the knowledge graphs to detect rumor. Specifically, given the post's sparse and ambiguous semantics, we identify entity mentions in the post's content and link them to entities and concepts in the knowledge graphs, which serve as complementary semantic information for the post text. To effectively inject external knowledge into textual representations, we develop a knowledge-aware attention mechanism to fuse local knowledge. Additionally, we construct a graph consisting of posts texts, entities, and concepts, which is fed to graph convolutional networks to explore long-range knowledge through graph structure. Our proposed model can therefore detect rumor by combining semantic-level and knowledge-level representations of posts. Extensive experiments on four publicly available real-world datasets show that KAGN outperforms or is comparable to other state-of-the-art methods, and also validate the effectiveness of knowledge.

Modeling and Optimization of Sensitivity and Creep for Multi-Component Sensing Materials.

Pressure sensors urgently need high-performance sensing materials in order to be developed further. Sensitivity and creep are regarded as two key indices for assessing a sensor's performance. For the design and optimization of sensing materials, an accurate estimation of the impact of several parameters on sensitivity and creep is essential. In this study, sensitivity and creep were predicted using the response surface methodology (RSM) and support vector regression (SVR), respectively. The input parameters were the concentrations of nickel (Ni) particles, multiwalled carbon nanotubes (MWCNTs), and multilayer graphene (MLG), as well as the magnetic field intensity (B). According to statistical measures, the SVR model exhibited a greater level of predictability and accuracy. The non-dominated sorting genetic-II algorithm (NSGA-II) was used to generate the Pareto-optimal fronts, and decision-making was used to determine the final optimal solution. With these conditions, the optimized results revealed an improved performance compared to the earlier study, with an average sensitivity of 0.059 kPa-1 in the pressure range of 0-16 kPa and a creep of 0.0325, which showed better sensitivity in a wider range compared to previous work. The theoretical sensitivity and creep were relatively similar to the actual values, with relative deviations of 0.317% and 0.307% after simulation and experimental verification. Future research for transducer performance optimization can make use of the provided methodology because it is representative.

Dynamic Neural Network for Bicriteria Weighted Control of Robot Manipulators.

In recent years, bicriteria optimization schemes for manipulator control have become preferred by researchers, given their satisfactory performance. In this article, a bicriteria weighted (BCW) scheme to remedy joint drift and minimize the infinity norm of joint velocity is proposed. The scheme adopts a novel repetitive motion index that can theoretically decouple the joint error and the position error, which many conventional cyclic motion generation schemes cannot achieve. Subsequently, through transformation, the BCW scheme is converted into a time-varying quadratic programming (QP) problem. Then, a dynamic neural network (DNN) system with a new Fisher-Burmeister function is proposed to address the resulting QP problem. It is proven that the proposed DNN system is free of residual errors, which means that the actual solution is able to converge to the theoretical solution. Another essential feature of the DNN system is that it has a suppression effect on noise. To demonstrate the convergence and robustness of the proposed DNN system, comparative simulations are carried out in nominal and noisy environments. Finally, experiments on Franka Emika Panda are conducted to elucidate the availability of the BCW scheme addressed by the DNN system.

A Kalman-Filter-Incorporated Latent Factor Analysis Model for Temporally Dynamic Sparse Data.

With the rapid development of services computing in the past decade, Quality-of-Service (QoS)-aware selection of Web services has become a hot yet thorny issue. Conducting warming-up tests on a large set of candidate services for QoS evaluation is time consuming and expensive, making it vital to implement accurate QoS-estimators. Existing QoS-estimators barely consider the temporal patterns hidden in QoS data. However, such data are naturally time dependent. For addressing this critical issue, this study presents a Kalman-filter-incorporated latent factor analysis (KLFA)-based QoS-estimator for accurate representation to temporally dynamic QoS data. Its main idea is to make the user latent features (LFs) time dependent, while the service ones time consistent. A novel iterative training scheme is designed, where the user LFs are learned through a Kalman filter for precisely modeling the temporal patterns, and the service ones are alternatively trained via an alternating least squares algorithm for precisely representing the historical QoS data. Empirical studies on large-scale and real Web service QoS datasets demonstrate that the proposed KLFA model significantly outperforms state-of-the-art QoS-estimators in estimation accuracy for dynamic QoS data.

Activated Gradients for Deep Neural Networks.

Deep neural networks often suffer from poor performance or even training failure due to the ill-conditioned problem, the vanishing/exploding gradient problem, and the saddle point problem. In this article, a novel method by acting the gradient activation function (GAF) on the gradient is proposed to handle these challenges. Intuitively, the GAF enlarges the tiny gradients and restricts the large gradient. Theoretically, this article gives conditions that the GAF needs to meet and, on this basis, proves that the GAF alleviates the problems mentioned above. In addition, this article proves that the convergence rate of SGD with the GAF is faster than that without the GAF under some assumptions. Furthermore, experiments on CIFAR, ImageNet, and PASCAL visual object classes confirm the GAF's effectiveness. The experimental results also demonstrate that the proposed method is able to be adopted in various deep neural networks to improve their performance. The source code is publicly available at https://github.com/LongJin-lab/Activated-Gradients-for-Deep-Neural-Networks.

Revealing Physiochemical Factors and Zooplankton Influencing Microcystis Bloom Toxicity in a Large-Shallow Lake Using Bayesian Machine Learning.

Toxic cyanobacterial blooms have become a severe global hazard to human and environmental health. Most studies have focused on the relationships between cyanobacterial composition and cyanotoxins production. Yet, little is known about the environmental conditions influencing the hazard of cyanotoxins. Here, we analysed a unique 22 sites dataset comprising monthly observations of water quality, cyanobacterial genera, zooplankton assemblages, and microcystins (MCs) quota and concentrations in a large-shallow lake. Missing values of MCs were imputed using a non-negative latent factor (NLF) analysis, and the results achieved a promising accuracy. Furthermore, we used the Bayesian additive regression tree (BART) to quantify how Microcystis bloom toxicity responds to relevant physicochemical characteristics and zooplankton assemblages. As expected, the BART model achieved better performance in Microcystis biomass and MCs concentration predictions than some comparative models, including random forest and multiple linear regression. The importance analysis via BART illustrated that the shade index was overall the best predictor of MCs concentrations, implying the predominant effects of light limitations on the MCs content of Microcystis. Variables of greatest significance to the toxicity of Microcystis also included pH and dissolved inorganic nitrogen. However, total phosphorus was found to be a strong predictor of the biomass of total Microcystis and toxic M. aeruginosa. Together with the partial dependence plot, results revealed the positive correlations between protozoa and Microcystis biomass. In contrast, copepods biomass may regulate the MC quota and concentrations. Overall, our observations arouse universal demands for machine-learning strategies to represent nonlinear relationships between harmful algal blooms and environmental covariates.

A Pseudoinversion-Free Method for Weight Updating in Broad Learning System.

Neural networks have evolved into one of the most critical tools in the field of artificial intelligence. As a kind of shallow feedforward neural network, the broad learning system (BLS) uses a training process based on random and pseudoinverse methods, and it does not need to go through a complete training cycle to obtain new parameters when adding nodes. Instead, it performs rapid update iterations on the basis of existing parameters through a series of dynamic update algorithms, which enables BLS to combine high efficiency and accuracy flexibly. The training strategy of BLS is completely different from the existing mainstream neural network training strategy based on the gradient descent algorithm, and the superiority of the former has been proven in many experiments. This article applies an ingenious method of pseudoinversion to the weight updating process in BLS and employs it as an alternative strategy for the dynamic update algorithms in the original BLS. Theoretical analyses and numerical experiments demonstrate the efficiency and effectiveness of BLS aided with this method. The research presented in this article can be regarded as an extended study of the BLS theory, providing an innovative idea and direction for future research on BLS.

An L1-and-L2-Norm-Oriented Latent Factor Model for Recommender Systems.

A recommender system (RS) is highly efficient in filtering people's desired information from high-dimensional and sparse (HiDS) data. To date, a latent factor (LF)-based approach becomes highly popular when implementing a RS. However, current LF models mostly adopt single distance-oriented Loss like an L2 norm-oriented one, which ignores target data's characteristics described by other metrics like an L1 norm-oriented one. To investigate this issue, this article proposes an L1 -and- L2 -norm-oriented LF ( [Formula: see text]) model. It adopts twofold ideas: 1) aggregating L1 norm's robustness and L2 norm's stability to form its Loss and 2) adaptively adjusting weights of L1 and L2 norms in its Loss. By doing so, it achieves fine aggregation effects with L1 norm-oriented Loss 's robustness and L2 norm-oriented Loss 's stability to precisely describe HiDS data with outliers. Experimental results on nine HiDS datasets generated by real systems show that an [Formula: see text] model significantly outperforms state-of-the-art models in prediction accuracy for missing data of an HiDS dataset. Its computational efficiency is also comparable with the most efficient LF models. Hence, it has good potential for addressing HiDS data from real applications.

An α-ß-Divergence-Generalized Recommender for Highly Accurate Predictions of Missing User Preferences.

To quantify user-item preferences, a recommender system (RS) commonly adopts a high-dimensional and sparse (HiDS) matrix. Such a matrix can be represented by a non-negative latent factor analysis model relying on a single latent factor (LF)-dependent, non-negative, and multiplicative update algorithm. However, existing models' representative abilities are limited due to their specialized learning objective. To address this issue, this study proposes an α- ß -divergence-generalized model that enjoys fast convergence. Its ideas are three-fold: 1) generalizing its learning objective with α- ß -divergence to achieve highly accurate representation of HiDS data; 2) incorporating a generalized momentum method into parameter learning for fast convergence; and 3) implementing self-adaptation of controllable hyperparameters for excellent practicability. Empirical studies on six HiDS matrices from real RSs demonstrate that compared with state-of-the-art LF models, the proposed one achieves significant accuracy and efficiency gain to estimate huge missing data in an HiDS matrix.

Dynamical Conventional Neural Network Channel Pruning by Genetic Wavelet Channel Search for Image Classification.

Neural network pruning is critical to alleviating the high computational cost of deep neural networks on resource-limited devices. Conventional network pruning methods compress the network based on the hand-crafted rules with a pre-defined pruning ratio (PR), which fails to consider the variety of channels among different layers, thus, resulting in a sub-optimal pruned model. To alleviate this issue, this study proposes a genetic wavelet channel search (GWCS) based pruning framework, where the pruning process is modeled as a multi-stage genetic optimization procedure. Its main ideas are 2-fold: (1) it encodes all the channels of the pertained network and divide them into multiple searching spaces according to the different functional convolutional layers from concrete to abstract. (2) it develops a wavelet channel aggregation based fitness function to explore the most representative and discriminative channels at each layer and prune the network dynamically. In the experiments, the proposed GWCS is evaluated on CIFAR-10, CIFAR-100, and ImageNet datasets with two kinds of popular deep convolutional neural networks (CNNs) (ResNet and VGGNet). The results demonstrate that GNAS outperforms state-of-the-art pruning algorithms in both accuracy and compression rate. Notably, GNAS reduces more than 73.1% FLOPs by pruning ResNet-32 with even 0.79% accuracy improvement on CIFAR-100.

The Stock Market Model with Delayed Information Impact from a Socioeconomic View.

Finding the critical factor and possible "Newton's laws" in financial markets has been an important issue. However, with the development of information and communication technologies, financial models are becoming more realistic but complex, contradicting the objective law "Greatest truths are the simplest." Therefore, this paper presents an evolutionary model independent of micro features and attempts to discover the most critical factor. In the model, information is the only critical factor, and stock price is the emergence of collective behavior. The statistical properties of the model are significantly similar to the real market. It also explains the correlations of stocks within an industry, which provides a new idea for studying critical factors and core structures in the financial markets.

[The value of growth differentiation factor-15 and extravascular lung water index in severity grading of acute respiratory distress syndrome patients and their prognosis prediction].

OBJECTIVE: To investigate the value of growth differentiation factor-15 (GDF-15) and extravascular lung water index (EVLWI) in severity grading and prognosis prediction of patients with acute respiratory distress syndrome (ARDS). METHODS: Patients with ARDS aged 18-75 years admitted to the department of respiratory intensive care unit (RICU) of Zhengzhou Central Hospital Affiliated to Zhengzhou University from January 2019 to February 2020 were enrolled. All patients were treated with conventional therapies such as mechanical ventilation, anti-infection, stabilization of water, electrolytes and acid-base environment, blood purification and nutritional support according to their conditions. Besides, the pulse-indicated continuous cardiac output (PiCCO) was performed after admission to the department, and EVLWI before treatment and at 24, 48 and 72 hours of treatment were recorded. Serum GDF-15 level was measured by enzyme linked immunosorbent assay (ELISA) during the same period. Patients were classified as mild, moderate, and severe degree according to the 2012 Berlin Definition of ARDS, and EVLWI and GDF-15 levels in patients with different disease levels before and after treatment were compared. In addition, the length of intensive care unit (ICU) stay, ICU mortality, and 28-day mortality of patients with different GDF-15 or EVLWI levels were analyzed comparatively, with the GDF-15 3 458 ng/L and EVLWI 15 mL/kg as the cut point. RESULTS: A total of 82 patients with ARDS were enrolled, including 22 patients with mild ARDS, 28 patients with moderate ARDS, and 32 patients with severe ARDS. The GDF-15 and EVLWI levels in patients with moderate and severe ARDS at each time point before and after treatment were higher than those in patients with mild ARDS. Both GDF-15 and EVLWI levels in patients with severe ARDS were higher than those in the patients with moderate ARDS. The differences were statistically significant at all the time points except for the difference of GDF-15 levels at 24 hours after treatment (ng/L: 3 900.41±546.43 vs. 3 695.66±604.73, P > 0.05). [GDF-15 (ng/L): 3 786.11±441.45 vs. 3 106.83±605.09 before treatment, 3 895.48±558.96 vs. 3 333.29±559.66 at 48 hours, 3 397.33±539.56 vs. 3 047.53±499.57 at 72 hours; EVLWI (mL/kg): 19.06±1.91 vs. 14.31±1.50 before treatment, 18.56±2.23 vs. 13.26±1.69 at 24 hours, 17.23±1.76 vs. 12.45±1.36 at 48 hours, 15.47±1.81 vs. 11.13±2.19 at 72 hours, all P < 0.05]. According to the cut-off value, there were 23 patients with GDF-15 ≥ 3 458 ng/L and GDF-15 < 3 458 ng/L respectively and there were 23 patients with EVLWI ≥ 15 mL/kg and EVLWI < 15 mL/kg respectively. The length of ICU stay and 28-day mortality in patients with high GDF-15 were significantly higher than those in patients with low GDF-15 [length of ICU stay (days): 21.22±2.69 vs. 15.37±3.14, 28-day mortality: 56.5% vs. 21.7%, both P < 0.05]. The length of ICU stay and 28-day mortality in patients with high EVLWI were also significantly higher than those in patients with low EVLWI [length of ICU stay (days): 18.45±2.61 vs. 14.98±2.75, 28-day mortality: 47.8% vs. 17.4%, both P < 0.05]. CONCLUSIONS: To some extent, GDF-15 and EVLWI levels reflect the severity of patients with ARDS, and high GDF-15 and EVLWI levels are significantly associated with poor prognosis in patients with ARDS.

Use statistical machine learning to detect nutrient thresholds in Microcystis blooms and microcystin management.

The frequency of toxin-producing cyanobacterial blooms has increased in recent decades due to nutrient enrichment and climate change. Because Microcystis blooms are related to different environmental conditions, identifying potential nutrient control targets can facilitate water quality managers to reduce the likelihood of microcystins (MCs) risk. However, complex biotic interactions and field data limitations have constrained our understanding of the nutrient-microcystin relationship. This study develops a Bayesian modelling framework with intracellular and extracellular MCs that characterize the relationships between different environmental and biological factors. This model was fit to the across-lake dataset including three bloom-plagued lakes in China and estimated the putative thresholds of total nitrogen (TN) and total phosphorus (TP). The lake-specific nutrient thresholds were estimated using Bayesian updating process. Our results suggested dual N and P reduction in controlling cyanotoxin risks. The total Microcystis biomass can be substantially suppressed by achieving the putative thresholds of TP (0.10 mg/L) in Lakes Taihu and Chaohu, but a stricter TP target (0.05 mg/L) in Dianchi Lake. To maintain MCs concentrations below 1.0 µg/L, the estimated TN threshold in three lakes was 1.8 mg/L, but the effect can be counteracted by the increase of temperature. Overall, the present approach provides an efficient way to integrate empirical knowledge into the data-driven model and is helpful for the management of water resources.

[Clinical path of mechanical ventilation and atomization inhalation therapy].

In recent years, atomization inhalation therapy has been more and more commonly used in patients with mechanical ventilation. However, the establishment of artificial airway has changed the environment and mode of aerosol delivery. Currently, Expert consensus on atomization inhalation during mechanical ventilation has been established to guide clinical practice. However, many ventilators do not support the treatment of aerosol inhalation due to the tedious procedures and numerous drugs. At the same time, the therapeutic effect of atomization inhalation is affected by many factors, such as ventilator mode selection, parameter setting, heating and humidification, using of artificial nose and filter, etc., which often results in poor clinical effects or even damage to the ventilator. In order to standardize the clinical application of mechanical ventilation atomization inhalation technology and avoid many possible problems in operation, the committee members of Respiratory Therapy Group of Severe Medicine Branch of Henan Medical Association discussed and concluded this clinical path, so as to provide clinical reference for the actual operation and drug administration of mechanical ventilation atomization.

Non-Negativity Constrained Missing Data Estimation for High-Dimensional and Sparse Matrices from Industrial Applications.

High-dimensional and sparse (HiDS) matrices are commonly seen in big-data-related industrial applications like recommender systems. Latent factor (LF) models have proven to be accurate and efficient in extracting hidden knowledge from them. However, they mostly fail to fulfill the non-negativity constraints that describe the non-negative nature of many industrial data. Moreover, existing models suffer from slow convergence rate. An alternating-direction-method of multipliers-based non-negative LF (AMNLF) model decomposes the task of non-negative LF analysis on an HiDS matrix into small subtasks, where each task is solved based on the latest solutions to the previously solved ones, thereby achieving fast convergence and high prediction accuracy for its missing data. This paper theoretically analyzes the characteristics of an AMNLF model, and presents detailed empirical studies regarding its performance on nine HiDS matrices from industrial applications currently in use. Therefore, its capability of addressing HiDS matrices is justified in both theory and practice.

Remote examination of the seasonal succession of phytoplankton assemblages from time-varying trends.

The seasonal succession of phytoplankton assemblages is important to ascertain the dynamics of an aquatic ecosystem structure, whereas its occurrence in response to hydrodynamic alterations is not clearly understood. In view of the characteristics of annual water level variation formed by the Three Gorges Dam Project (TGDP), our understanding about how these changes affect phytoplankton structure and dynamics is still very limited due to the shortage of long-term observation data. In this study, we used Huan Jing 1 charge-coupled device images over the past decade to examine the phytoplankton succession dates between cyanobacterial and green algal blooms in the backwater area of the Three Gorges Reservoir (TGR). The results indicated continuous wavelet transform-based peak analysis is an efficiency tool that can illustrate the temporal pattern of phytoplankton succession using satellite-derived chlorophyll ɑ and Cyano-Chlorophyta index thresholds. Water level, air temperature, pH and total nitrogen/total phosphorus ratio were four important factors affecting the decline and rise phase of cyanobacterial blooms in the TGR from 2008 to 2018. Given that the upstream dam operation is likely to alter ecological and environmental conditions in the backwater area, this mechanism, so-called "water-level linkage", could alleviate the persistent period of cyanobacterial and green algal blooms. Remote sensing together with time series analysis provided a useful method to examine the seasonal succession of phytoplankton assemblages in the TGR, and these findings provided strategic insight for the water-quality management in the post-TGDP period.

Application of Bayesian network including Microcystis morphospecies for microcystin risk assessment in three cyanobacterial bloom-plagued lakes, China.

Microcystis spp., which occur as colonies of different sizes under natural conditions, have expanded in temperate and tropical freshwater ecosystems and caused seriously environmental and ecological problems. In the current study, a Bayesian network (BN) framework was developed to access the probability of microcystins (MCs) risk in large shallow eutrophic lakes in China, namely, Taihu Lake, Chaohu Lake, and Dianchi Lake. By means of a knowledge-supported way, physicochemical factors, Microcystis morphospecies, and MCs were integrated into different network structures. The sensitive analysis illustrated that Microcystis aeruginosa biomass was overall the best predictor of MCs risk, and its high biomass relied on the combined condition that water temperature exceeded 24 °C and total phosphorus was above 0.2 mg/L. Simulated scenarios suggested that the probability of hazardous MCs (≥1.0 µg/L) was higher under interactive effect of temperature increase and nutrients (nitrogen and phosphorus) imbalance than that of warming alone. Likewise, data-driven model development using a naïve Bayes classifier and equal frequency discretization resulted in a substantial technical performance (CCI = 0.83, K = 0.60), but the performance significantly decreased when model excluded species-specific biomasses from input variables (CCI = 0.76, K = 0.40). The BN framework provided a useful screening tool to evaluate cyanotoxin in three studied lakes in China, and it can also be used in other lakes suffering from cyanobacterial blooms dominated by Microcystis.