Improvement In Mean Absolute Error Research Articles

Temporal Convolutional Network for Carbon Tax Projection: A Data-Driven Approach

This study introduces a novel application of a temporal convolutional network (TCN) for projecting carbon tax prices, addressing the critical need for accurate forecasting in climate policy. Utilizing data from the World Carbon Pricing Database, we demonstrate that the TCN significantly outperformed traditional time series models in capturing the complex dynamics of carbon pricing. Our model achieved a 31.4% improvement in mean absolute error over ARIMA baselines, with an MAE of 2.43 compared to 3.54 for ARIMA. The TCN model also showed superior performance across different time horizons, demonstrating a 30.0% lower MAE for 1-year projections, and enhanced adaptability to policy changes, with only a 39.8% increase in prediction error after major shifts, compared to ARIMA’s 95.6%. These results underscore the potential of deep learning for enhancing the precision of carbon price projections, thereby supporting more informed and effective climate policy decisions. Our findings have significant implications for policymakers and stakeholders in the realm of carbon pricing and climate change mitigation strategies, offering a powerful tool for navigating the complex landscape of environmental economics.

Accurate crowd counting for intelligent video surveillance systems

The paper presents a novel deep learning approach for crowd counting in intelligent video surveillance systems, addressing the growing need for accurate monitoring of public spaces in urban environments. The demand for precise crowd estimation arises from challenges related to security, public safety, and efficiency in urban areas, particularly during large public events. Existing crowd counting techniques, including feature-based object detection and regression-based methods, face limitations in high-density environments due to occlusions, lighting variations, and diverse human figures. To overcome these challenges, the authors propose a new deep encoder-decoder architecture based on VGG16, which incorporates hierarchical feature extraction with spatial and channel attention mechanisms. This architecture enhances the model’s ability to manage variations in crowd density, leveraging adaptive pooling and dilated convolutions to extract meaningful features from dense crowds. The model’s decoder is further refined to handle sparse and crowded scenes through separate density maps, improving its adaptability and accuracy. Evaluations of the proposed model on benchmark datasets, including Shanghai Tech and UCF CC 50, demonstrate superior performance over state-of-the-art methods, with significant improvements in mean absolute error and mean squared error metrics. The paper emphasizes the importance of addressing environmental variability and scale differences in crowded environments and shows that the proposed model is effective in both sparse and dense crowd conditions. This research contributes to the advancement of intelligent video surveillance systems by providing a more accurate and efficient method for crowd counting, with potential applications in public safety, transportation management, and urban planning.

Graph-enabled spatio-temporal transformer for ionospheric prediction

The ionosphere, crucial for satellite navigation and radio communication, is highly sensitive to solar and geomagnetic activities. Traditional ionospheric prediction models often struggle to capture its complexities. This paper introduces a Graph-Enabled Spatio-temporal transformer (GEST) model, designed to predict the Total electron content (TEC) in the ionosphere with higher accuracy. The GEST model innovatively integrates six dynamic space weather parameters, including the Kp index, and refines grid distances and weights in the Global ionospheric maps (GIM). Leveraging the efficiency of the Transformer architecture in processing spatio-temporal data, the GEST model adapts to varying ionospheric conditions and provides superior precision. Our evaluations, based on data from the Center for orbit determination in Europe (CODE), demonstrate that the GEST model significantly outperforms traditional methods and other intelligent algorithms such as conv-LSTM. It shows marked improvements in Mean absolute error (MAE), Root mean square error (RMSE), Pearson correlation coefficient (CC), and Mean absolute percentage error (MAPE), particularly during periods of intense solar and geomagnetic activity. These advancements highlight the GEST model’s substantial contributions to the field of ionospheric prediction technology, offering a robust tool for enhancing the reliability of satellite-based systems.

Exploring the potential of routine serological markers in predicting neurological outcomes in spinal cord injury

Spinal cord injury (SCI) is a rare condition with a heterogeneous presentation, making the prediction of recovery challenging. However, serological markers have been shown to be associated with severity and long-term recovery following SCI. Therefore, our investigation aimed to assess the feasibility of translating this association into a prediction of the lower extremity motor scores (LEMS) at chronic stage (52 weeks after initial injury) in patients with SCI using routine serological markers.Serological markers, assessed within the initial seven days post-injury in the observational cohort study from the Trauma Hospital Murnau underwent diverse feature engineering approaches. These involved arithmetic measurements such as mean, median, minimum, maximum, and range, as well as considerations of the frequency of marker testing and whether values fell within the normal range. To predict LEMS scores at the chronic stage, eight different regression models (including linear, tree-based, and ensemble models) were used to quantify the predictive value of serological markers relative to a baseline model that relied on the very acute LEMS score and patient age alone.The inclusion of serological markers did not improve the performance of the prediction model. The best-performing approach including serological markers achieved a mean absolute error (MAE) of 6.59 (2.14), which was equivalent to the performance of the baseline model. As an alternative approach, we trained separate models based on the LEMS observed at the very acute stage after injury. Specifically, we considered individuals with an LEMS of 0 or an LEMS exceeding zero separately. This strategy led to a mean improvement in MAE across all cohorts and models, of 1.20 (2.13).We conclude that, in our study, routine serological markers hold limited power for prediction of LEMS. However, the implementation of model stratification by the very acute LEMS markedly enhanced prediction performance. This observation supports the inclusion of clinical knowledge in the modeling of prediction tasks for SCI recovery. Additionally, it lays the path for future research to consider stratified analyses when investigating the predictive power of potential biomarkers.

Graph-Based Traffic Forecasting with the Dynamics of Road Symmetry and Capacity Performance

Symmetry in traffic patterns is a fundamental aspect of intelligent transportation systems, aiming to precisely predict traffic flow in real time despite the complex interplay of spatial and temporal factors. This paper presents a novel method of traffic forecasting that incorporates parameters related to road symmetry into a Graph Convolution Network model. Our model is crafted to dynamically adjust to real-time changes in road conditions, including the presence of symmetric and asymmetric road layouts, which substantially influence traffic flow. We have developed a GCN model that not only accounts for standard traffic flow metrics but also integrates a matrix representing road symmetry. The model undergoes training and validation on the METR-LA dataset, showcasing a significant enhancement in prediction accuracy. In the comparative analysis of state-of-the-art methods, our model demonstrated a significant enhancement in performance, achieving 30.68% improvement in Mean Squared Error (MSE) and a 24.28% improvement in Mean Absolute Error (MAE) over the best-performing method. The implications of our research are profound for urban planners, traffic management systems, and navigation service providers, as it offers a more dependable tool for forecasting traffic conditions, aiding in road design, and refining route planning strategies based on the symmetry of road networks.

Optimising barrier placement for intrusion detection and prevention in WSNs.

This research addresses the pressing challenge of intrusion detection and prevention in Wireless Sensor Networks (WSNs), offering an innovative and comprehensive approach. The research leverages Support Vector Regression (SVR) models to predict the number of barriers necessary for effective intrusion detection and prevention while optimising their strategic placement. The paper employs the Ant Colony Optimization (ACO) algorithm to enhance the precision of barrier placement and resource allocation. The integrated approach combines SVR predictive modelling with ACO-based optimisation, contributing to advancing adaptive security solutions for WSNs. Feature ranking highlights the critical influence of barrier count attributes, and regularisation techniques are applied to enhance model robustness. Importantly, the results reveal substantial percentage improvements in model accuracy metrics: a 4835.71% reduction in Mean Squared Error (MSE) for ACO-SVR1, an 862.08% improvement in Mean Absolute Error (MAE) for ACO-SVR1, and an 86.29% enhancement in R-squared (R2) for ACO-SVR1. ACO-SVR2 has a 2202.85% reduction in MSE, a 733.98% improvement in MAE, and a 54.03% enhancement in R-squared. These considerable improvements verify the method's effectiveness in enhancing WSNs, ensuring reliability and resilience in critical infrastructure. The paper concludes with a performance comparison and emphasises the remarkable efficacy of regularisation. It also underscores the practicality of precise barrier count estimation and optimised barrier placement, enhancing the security and resilience of WSNs against potential threats.

Optimizing Time-Series forecasting using stacked deep learning framework with enhanced adaptive moment estimation and error correction

Time-series forecasting is essential for decision-making activities, and deep learning models have shown promising results in this field. However, existing models face challenges in capturing the complex non-linear patterns and randomness of real-world time-series data. To address this issue, we propose a novel stacked deep learning framework that combines a bi-directional long-short-term memory network, convolutional 1D, and max-pooling-1D layers in each stack. We also introduce an enhanced version of adaptive moment estimation and an error correction method to improve the model's accuracy and convergence speed. We evaluate our model's performance using four real-world datasets (air quality, PM2.5, electricity price, and securities lending) and compare it with existing models based on statistical measures such as mean absolute error, mean squared error, and standard deviation. The results demonstrate that our proposed model outperforms existing models significantly, achieving a ∼ 52 % improvement in mean absolute error for air quality data, a ∼ 36 % improvement in mean squared error for the PM2.5 dataset, and a ∼ 27.5 % improvement in mean squared error for electricity price data. The enhanced adaptive moment estimation and error correction method further improve the model's accuracy and generalization capabilities, making it suitable for multi-domain data sets. Our proposed model has potential applications in various fields, including manufacturing, traffic prediction, weather forecasting, environmental monitoring, finance, and energy management.

Modeling abrupt changes in mine water inflow trends: A CEEMDAN-based multi-model prediction approach

The global impact of coal mining on groundwater resources is increasingly prominent. Constructing waterproof curtains can mitigate the destruction caused by open-pit coal mining on groundwater resources. However, the construction of these curtains alters the hydrogeological conditions, leading to abrupt changes in mine water inflow that are challenging to predict. Traditional models like Long Short-Term Memory Neural Network (LSTM) and Gated Recurrent Unit Neural Network (GRU) demonstrate poor performance when predicting abrupt changes in data trends. This study proposes a multi-model prediction approach and a Self-Attention-based prediction model to address this issue. The proposed Approach, based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), leverages the strengths of multiple models for prediction. The research results found that the proposed Self-Attention model excels at forecasting abrupt shifts in data trends. Comparatively, the multi-model predictions show a 26.12 % improvement in Mean Absolute Error (MAE) and a 33.47 % improvement in Root Mean Square Error (RMSE) when compared to single-model predictions. This study not only unveils the potential of Self-Attention in predicting abrupt trend changes but also integrates it with traditional LSTM and Support Vector Machine (SVM) models. The multi-model prediction framework proposed in this study integrates the strengths of multiple models, achieving higher accuracy than a single model. Accurate prediction of mine water inflow is essential, as it can provide valuable insights for decision-making by coal mining enterprises and local regulatory authorities, offering a scientific basis for effective management.

Prediction of nitrous oxide emission of a municipal wastewater treatment plant using LSTM-based deep learning models.

Accurate assessment of greenhouse gas emissions from wastewater treatment plants is crucial for mitigating climate change. N2O is a potent greenhouse gas that is emitted from wastewater treatment plants during the biological denitrification process. In this study, we developed and evaluated deep learning models for predicting N2O emissions from a WWTP in Switzerland. Six key parameters were selected to obtain the optimal LSTM model by adjusting experimental parameter conditions. The optimal parameter condition was achieved with 150 neurons, the tanh activation function, the RMSprop optimization algorithm, a learning rate of 0.001, no dropout regularization, and a batch size of 128. Under the same conditions, we compared the performance of recurrent neural networks (RNNs) and Long Short-Term Memory (LSTM) networks. We found that LSTM models outperformed RNN models in predicting N2O emissions. The optimal LSTM model achieved a 36% improvement in mean absolute error (MAE), a 19% improvement in root mean squared error (RMSE), and a 6.92% improvement in R2 score compared to the RNN model. Additionally, LSTM models demonstrated better resilience to sudden changes in the target sequence, exhibiting a 9.54% higher percentage of explained variance compared to RNNs. These results highlight the potential of LSTM models for accurate and robust prediction of N2O emissions from wastewater treatment plants, contributing to effective greenhouse gas mitigation strategies.

Significant wave height prediction based on the local-EMD-WaveNet model

This research constructed the innovative Local-EMD-WaveNet, a multi-channel neural network model, specifically designed for the prediction of significant wave height (SWH) at a singular point. It leverages Local Empirical Mode Decomposition (EMD) on significant wave heights in Ghanaian waters, integrating the derived decomposition results with wind speed data. This compiled data is then channeled into the model, which exploits the capabilities of dilated causal convolution to capture and analyze the time-series characteristics integral to future SWH predictions. The model ingeniously embeds EMD within the training process, treating the decomposed sub-waves and wind speed sequences as unique channels along the “depth” dimension. Following the application of dilated causal convolution, these channels are systematically “stacked”. Compared to conventional LSTM and direct data incorporation methods, Local-EMD-WaveNet consistently outperforms, especially in long-term predictions. The model exhibited significant improvements in Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) during 48 and 56 time-step predictions, marking increases of 27.3% and 23.5%, respectively, outshining both WaveNet and LSTM. Particularly in situations with larger wave height variations, Local-EMD-WaveNet accurately captures waveforms' peaks and troughs. These results validate Local-EMD-WaveNet as a reliable wave forecasting tool with considerable potential in ocean engineering and maritime applications.

Physics-constrained cooperative learning-based reference models for smart management of chillers considering extrapolation scenarios

Smart management of building energy devices, including their optimal control and fault detection technology, is of great significance to building energy conservation. The core of smart management is the development of reference models for target energy device. However, existing reference models show poor extrapolation ability when the operation conditions of online data are outside the scope of training data. To tackle this problem, here we propose a novel physics-constrained cooperative learning framework to train multiple reference models in a cooperative manner in order to improve their extrapolation ability. The general idea of cooperative learning is to constrain the output of different reference models on unknown operation conditions such that the physical inconsistent loss is minimized. In this study, two novel physical inconsistent losses, including energy conservation inconsistent loss and mass conservation inconsistent loss, are designed for seven output reference variables of chiller, forming the physics-constrained cooperative neural networks (PCNNs). Comprehensive data experiments are conducted to compare the model performance of PCNNs with other machine learning models, including Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN). The experimental results showed that the PCNNs outperformed the other models under extrapolation scenarios, showing a lager performance improvement of mean absolute error (MAE) and root mean squared error (RMSE) metrics with 26.94% and 23.49%, respectively. The proposed physics-constrained cooperative learning framework might provide a new perspective for the development of reference models in building energy system.

Dual Mach–Zehnder Interferometers With Hierarchical Clustering Analysis Method Applied for Positioning the Intrusion

The method of hierarchical clustering is for the first time employed to locate the intrusion-induced disturbance on the sensing fibers of a dual Mach–Zehnder interferometer (DMZI). Such an intrusion-induced disturbance is located by finding the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${x}$ </tex-math></inline-formula> coordinate of the centroid of the largest cluster on the Euclidean plane through hierarchical clustering with an appropriate linkage criterion employed for determining the distance between two observations. We compare average linkage and complete linkage criteria in the clustering analysis to see which one provides better locating accuracy. In the clustering analysis, the number of clusters is set to be 3–8 in finding the location of disturbance. To reduce the locating error, we also use differential signals here in the clustering analysis. Twelve intrusion events are simulated by knocking the sensing fibers to induce disturbances at a given location. The location of disturbance is determined through the clustering analysis for each intrusion event. The mean of the absolute values of locating errors [mean absolute error (MAE)] for the 12 intrusion events is then estimated. The experimental results in this study demonstrate a maximum MAE of 11.55 m in locating an intrusion with average linkage criterion employed for five-cluster analysis. Also, the MAE could be 3.55 m smaller by using the differential signals for clustering analysis, compared with the case when directly detected signals are used for clustering analysis. The results also confirm that the average linkage criterion provides only a small amount of improvement in MAE over complete linkage criterion.

A hybrid deep learning model for regional O3 and NO2 concentrations prediction based on spatiotemporal dependencies in air quality monitoring network

Short-term prediction of urban air quality is critical to pollution management and public health. However, existing studies have failed to make full use of the spatiotemporal correlations or topological relationships among air quality monitoring networks (AQMN), and hence exhibit low precision in regional prediction tasks. With this consideration, we proposed a novel deep learning-based hybrid model of Res-GCN-BiLSTM combining the residual neural network (ResNet), graph convolutional network (GCN), and bidirectional long short-term memory (BiLSTM), for predicting short-term regional NO2 and O3 concentrations. Auto-correlation analysis and cluster analysis were first utilized to reveal the inherent temporal and spatial properties respectively. They demonstrated that there existed temporal daily periodicity and spatial similarity in AQMN. Then the identified spatiotemporal properties were sufficiently leveraged, and monitoring network topological information, as well as auxiliary pollutants and meteorology were also adaptively integrated into the model. The hourly observed data from 51 air quality monitoring stations and meteorological data in Shanghai were employed to evaluate it. Results show that the Res-GCN-BiLSTM model was better adapted to the pollutant characteristics and improved the prediction accuracy, with nearly 11% and 17% improvements in mean absolute error for NO2 and O3, respectively compared to the best performing baseline model. Among the three types of monitoring stations, traffic monitoring stations performed the best for O3, but the worst for NO2, mainly due to the impacts of intensive traffic emissions and the titration reaction. These findings illustrate that the hybrid architecture is more suitable for regional pollutant concentration.

A hybrid deep learning framework integrating feature selection and transfer learning for multi-step global horizontal irradiation forecasting

The randomness and volatility of solar irradiance pose a challenge to efficient solar energy development and utilization across the world, which increases the necessity of developing an efficient solar irradiance forecasting model. However, previous studies rarely emphasized the importance of transfer learning and nonlinear feature selection, especially for the studies associated with newly-built photovoltaic plants. Moreover, multi-step ahead forecasting research is limited, despite its significance to dispatching efficiency improvement of photovoltaic system. Aiming to address the research gaps, a novel hybrid deep learning framework (HDLF), consisting of the modules of feature selection, feature convolution, forecasting, self-attention, transfer learning, performance evaluation, and performance analysis, is newly proposed in this paper to perform multi-step ahead solar irradiation forecasting. Specifically, the HDLF is pre-trained based on the studied datasets of global horizontal irradiation in the source domains, and the abstract information obtained from pre-training is transferred to the HDLF of the target domain to enhance its performance. To validate the effectiveness of the proposed HDLF, two simulation experiments are carried out based on the datasets from California, USA, with a resolution of 1 h. The corresponding experimental results from the modules of performance evaluation and performance analysis indicate that the maximum improvements in mean absolute error reach 80.19% and 35.67% in two experiments, respectively, thereby confirming the superiority and feasibility of the HDLF.

Generating a long-term (2003−2020) hourly 0.25° global PM2.5 dataset via spatiotemporal downscaling of CAMS with deep learning (DeepCAMS)

Generating a long-term high-spatiotemporal resolution global PM2.5 dataset is of great significance for environmental management to mitigate the air pollution concerns worldwide. However, the current long-term (2003–2020) global reanalysis dataset Copernicus Atmosphere Monitoring Service (CAMS) reanalysis has drawbacks in fine-scale research due to its coarse spatiotemporal resolution (0.75°, 3-h). Hence, this paper developed a deep learning-based framework (DeepCAMS) to downscale CAMS PM2.5 product on the spatiotemporal dimension for resolution enhancement. The nonlinear statistical downscaling from low-resolution (LR) to high-resolution (HR) data can be learned from the high quality (0.25°, hourly) but short-term (2018–2020) Goddard Earth Observing System composition forecast (GEOS-CF) system PM2.5 product. Compared to the conventional spatiotemporal interpolation methods, simulation validations on GEOS-CF demonstrate that DeepCAMS is capable of producing accurate temporal variations with an improvement of Root-Mean-Squared Error (RMSE) of 0.84 (4.46 to 5.30) ug/m3 and spatial details with an improvement of Mean Absolute Error (MAE) of 0.16 (0.34 to 0.50) ug/m3. The real validations on CAMS reflect convincing spatial consistency and temporal continuity at both regional and global scales. Furthermore, the proposed dataset is validated with OpenAQ air quality data from 2017 to 2019, and the in-situ validations illustrate that the DeepCAMS maintains the consistent precision (R: 0.597) as the original CAMS (R: 0.593) while tripling the spatiotemporal resolution. The proposed dataset will be available at https://doi.org/10.5281/zenodo.6381600.

Parameter Identification of Lithium Battery Model Based on Chaotic Quantum Sparrow Search Algorithm

An accurate battery model is of great importance for battery state estimation. This study considers the parameter identification of a fractional-order model (FOM) of the battery, which can more realistically describe the reaction process of the cell and provide more precise predictions. Firstly, an improved sparrow search algorithm combined with the Tent chaotic mapping, quantum behavior strategy and Gaussian variation is proposed to regulate the early population quality, enhance its global search ability and avoid trapping into local optima. The effectiveness and superiority are verified by comparing the proposed chaotic quantum sparrow search algorithm (CQSSA) with the particle swarm optimization (PSO), genetic algorithm (GA), grey wolf optimization algorithm (GWO), Dingo optimization algorithm (DOA) and sparrow search algorithm (SSA) on benchmark functions. Secondly, the parameters of the FOM battery model are identified using six algorithms under the hybrid pulse power characterization (HPPC) test. Compared with SSA, CQSSA has 4.3%, 5.9% and 11.5% improvement in mean absolute error (MAE), root mean square error (RMSE) and maximum absolute error (MaAE), respectively. Furthermore, these parameters are used in the pulsed discharge test (PULSE) and urban dynamometer driving schedule (UDDS) test to verify the adaptability of the proposed algorithm. Simulation results show that the model parameters identified by the CQSSA algorithm perform well in terms of the MAE, RMSE and MaAE of the terminal voltages under all three different tests, demonstrating the high accuracy and good adaptability of the proposed algorithm.

Truncated tensor Schatten [formula omitted]-norm based approach for spatiotemporal traffic data imputation with complicated missing patterns

Rapid advances in sensor, wireless communication, cloud computing and data science have brought unprecedented amount of data to assist transportation engineers and researchers in making better decisions. However, traffic data in reality often has corrupted or incomplete values due to detector and communication malfunctions. Data imputation is thus required to ensure the effectiveness of downstream data-driven applications. To this end, numerous tensor-based methods treating the imputation problem as the low-rank tensor completion (LRTC) have been attempted in previous works. To tackle rank minimization, which is at the core of the LRTC, most of aforementioned methods utilize the tensor nuclear norm (NN) as a convex surrogate for the minimization. However, the over-relaxation issue in NN refrains it from desirable performance in practice. In this paper, we define an innovative nonconvex truncated Schatten p-norm for tensors (TSpN) to approximate tensor rank and impute missing spatiotemporal traffic data under the LRTC framework. We model traffic data into a third-order tensor structure of time intervals×locations (sensors)×days and introduce four complicated missing patterns, including random missing and three fiber-like missing cases according to the tensor mode-n fibers. Despite nonconvexity of the objective function in our model, we derive the global optimal solutions by integrating the alternating direction method of multipliers (ADMM) with generalized soft-thresholding (GST). In addition, we design a truncation rate decay strategy to deal with varying missing rate scenarios. Comprehensive experiments are finally conducted using four different types of real-world spatiotemporal traffic datasets, which demonstrate that the proposed LRTC-TSpN method performs well under various missing cases even with high rates of data loss, meanwhile outperforming other state-of-the-art tensor-based imputation models in almost all scenarios. In general, the performance of the proposed method achieves up to a 52% improvement in mean absolute error (MAE) compared to baseline methods even when the missing ratio is as high as 90%. Moreover, both theoretical and empirical evidences indicate that our method has robust and efficient convergence performances.

An automated machine learning methodology for the improved prediction of reference evapotranspiration using minimal input parameters

AbstractAn accurate and minimal meteorological parameter‐based reference evapotranspiration (ETo) computational model is highly essential to design a cost‐effective and automated solution for precise water allocation in various agricultural and hydrological applications. Automated machine learning (AutoML), a substantially progressive machine learning (ML) paradigm, expedites the data science applications in these domains by automating and alleviating the tedious tasks of developing ML pipelines, particularly in non‐linear hydrological process modelling. This paper proposes an AutoML solution for daily EToprediction in a limited input parameter scenario, for the first time in EToestimation research. Two AutoML frameworks, AutoGluon‐Tabular (AGT) and H2O AutoML were implemented using daily meteorological data of a humid tropical climatic region of Kerala, India and evaluated its performance with the radiation‐based empirical methods and traditional ML methods. The developed AutoML models imitated the input parameter combinations of radiation‐based empirical models such as Priestley‐Taylor, Makkink, Turc and Ritchie, with the Penman‐Monteith EToserving as the target for modelling process. An improvement of mean absolute error from 0.119 to 0.078 mm/day, mean squared error from 0.035 to 0.017 mm/day, root mean squared error from 0.186 to 0.129 mm/day, root mean squared logarithmic error from 0.078 to 0.026 mm/day, and coefficient of determination (R2) from 0.971 to 0.985 observed in the AutoML model, AGT relative to the empirical model, Turc describes its enhanced performance capability. For demonstrating the superiority of AutoML to traditional ML techniques, another set of AutoML models were also developed based on California Irrigation Management Information System datasets and evaluated them with existing study results in literature. The overall result analysis demonstrated the superiority of AGT model in EToprediction in all the weather stations considered in this study. The results of the study confirm that AutoML techniques can be applied to any hydrological dataset and provide an effective automated solution for hydrological process modelling.

On the estimation of boundary layer heights: a machine learning approach

Abstract. The planetary boundary layer height (zi) is a key parameter used in atmospheric models for estimating the exchange of heat, momentum, and moisture between the surface and the free troposphere. Near-surface atmospheric and subsurface properties (such as soil temperature, relative humidity, etc.) are known to have an impact on zi. Nevertheless, precise relationships between these surface properties and zi are less well known and not easily discernible from the multi-year dataset. Machine learning approaches, such as random forest (RF), which use a multi-regression framework, help to decipher some of the physical processes linking surface-based characteristics to zi. In this study, a 4-year dataset from 2016 to 2019 at the Southern Great Plains site is used to develop and test a machine learning framework for estimating zi. Parameters derived from Doppler lidars are used in combination with over 20 different surface meteorological measurements as inputs to a RF model. The model is trained using radiosonde-derived zi values spanning the period from 2016 through 2018 and then evaluated using data from 2019. Results from 2019 showed significantly better agreement with the radiosonde compared to estimates derived from a thresholding technique using Doppler lidars only. Noteworthy improvements in daytime zi estimates were observed using the RF model, with a 50 % improvement in mean absolute error and an R2 of greater than 85 % compared to the Tucker method zi. We also explore the effect of zi uncertainty on convective velocity scaling and present preliminary comparisons between the RF model and zi estimates derived from atmospheric models.

Are RGB-based salient object detection methods unsuitable for light field data?

Considering the significant progress made on RGB-based deep salient object detection (SOD) methods, this paper seeks to bridge the gap between those 2D methods and 4D light field data, instead of implementing specific 4D methods. We observe that the performance of 2D methods changes dramatically with the input refocusing on different depths. This paper attempts to make the 2D methods available for light field SOD by learning to select the best single image from the 4D tensor. Given a 2D method, a deep model is proposed to explicitly compare pairs of SOD results on one light field sample. Moreover, a comparator module is designed to integrate the features from a pair, which provides more discriminative representations to classify. Experiments over 13 latest 2D methods and 2 datasets demonstrate the proposed method can bring about 24.0% and 5.3% average improvement of mean absolute error and F-measure, and outperform state-of-the-art 4D methods by a large margin.