Error Metrics Research Articles

Convolutional neural network-based wind pressure prediction on low-rise buildings

A new model of wind pressure prediction for low-rise buildings is built based on deep convolutional neural network (CNN), and the model is trained and tested by the aerodynamic data from international open database. The prediction accuracy is compared with that of the literature, and then comprehensive error analysis on all 335 predicted taps under 3 untrained wind directions are carried out for mean and RMS pressure coefficients by various error metrics. The study shows that the distributions of mean and RMS coefficients predicted by the CNN model are closer to the experimental results in comparison with those predicted by the traditional artificial neural network (ANN) model and the common deep neural network (DNN) model without the convolution and pooling layers. The relative error on the corner tap is 13.5%, 4.3% and 0.1% by the ANN, DNN, and CNN models respectively. The mean-square-error (MSE) of the corner bay and the whole roof are also eminently reduced by the CNN model. By introducing the convolutional and pooling layers, more prediction errors are confined within 5% and basically no errors exceed 20%. Consequently, the CNN algorithm can be applied in solving various wind pressure prediction problems on buildings in the future work.

Daily Streamflow Forecasting Using Networks of Real-Time Monitoring Stations and Hybrid Machine Learning Methods

Considering the increased risk of urban flooding and drought due to global climate change and rapid urbanization, the imperative for more accurate methods for streamflow forecasting has intensified. This study introduces a pioneering approach leveraging the available network of real-time monitoring stations and advanced machine learning algorithms that can accurately simulate spatial–temporal problems. The Spatio-Temporal Attention Gated Recurrent Unit (STA-GRU) model is renowned for its computational efficacy in forecasting streamflow events with a forecast horizon of 7 days. The novel integration of the groundwater level, precipitation, and river discharge as predictive variables offers a holistic view of the hydrological cycle, enhancing the model’s accuracy. Our findings reveal that for a 7-day forecasting period, the STA-GRU model demonstrates superior performance, with a notable improvement in mean absolute percentage error (MAPE) values and R-square (R2) alongside reductions in the root mean squared error (RMSE) and mean absolute error (MAE) metrics, underscoring the model’s generalizability and reliability. Comparative analysis with seven conventional deep learning models, including the Long Short-Term Memory (LSTM), the Convolutional Neural Network LSTM (CNNLSTM), the Convolutional LSTM (ConvLSTM), the Spatio-Temporal Attention LSTM (STA-LSTM), the Gated Recurrent Unit (GRU), the Convolutional Neural Network GRU (CNNGRU), and the STA-GRU, confirms the superior predictive power of the STA-LSTM and STA-GRU models when faced with long-term prediction. This research marks a significant shift towards an integrated network of real-time monitoring stations with advanced deep-learning algorithms for streamflow forecasting, emphasizing the importance of spatially and temporally encompassing streamflow variability within an urban watershed’s stream network.

Bergen metrics: composite error metrics for assessing performance of climate models using EURO-CORDEX simulations

Abstract. Error metrics are useful for evaluating model performance and have been used extensively in climate change studies. Despite the abundance of error metrics in the literature, most studies use only one or two metrics. Since each metric evaluates a specific aspect of the relationship between the reference data and model data, restricting the comparison to just one or two metrics limits the range of insights derived from the analysis. This study proposes a new framework and composite error metrics called Bergen metrics to summarize the overall performance of climate models and to ease interpretation of results from multiple error metrics. The framework of Bergen metrics are based on the p norm, and the first norm is selected to evaluate the climate models. The framework includes the application of a non-parametric clustering technique to multiple error metrics to reduce the number of error metrics with minimum information loss. An example of Bergen metrics is provided through its application to the large ensemble of regional climate simulations available from the EURO-CORDEX initiative. This study calculates 38 different error metrics to assess the performance of 89 regional climate simulations of precipitation and temperature over Europe. The non-parametric clustering technique is applied to these 38 metrics to reduce the number of metrics to be used in Bergen metrics for eight different sub-regions in Europe. These provide useful information about the performance of the error metrics in different regions. Results show it is possible to observe contradictory behaviour among error metrics when examining a single model. Therefore, the study also underscores the significance of employing multiple error metrics depending on the specific use case to achieve a thorough understanding of the model behaviour.

A Chebyshev neural network-based numerical scheme to solve distributed-order fractional differential equations

This study aims to develop a first-order Chebyshev neural network-based technique for solving ordinary and partial distributed-order fractional differential equations. The neural network is used as a trial solution to construct the loss function. The loss function is utilized to train the neural network via an extreme learning machine and obtain the solution. The novelty of this work is developing and implementing a neural network-based framework for distributed-order fractional differential equations via an extreme learning machine. The proposed method is validated on several test problems. The error metrics utilized in the study include the absolute error and the L2 error. A comparison with other previously available approaches is presented. Also, we provide the computation time of the method.

Investigation of Corrosion Inhibition Efficiency of Pyridine-Quinoline Compounds through Machine Learning

Corrosion in materials is a significant concern for the industrial and academic fields because corrosion causes enormous losses in various fields such as the economy, environment, society, industry, security, safety, and others. Currently, material damage control using organic compounds has become a popular field of study. Pyridine and quinoline stand out as corrosion inhibitors among a myriad of organic compounds because they are non-toxic, inexpensive, and effective in a variety of corrosive environments. Experimental investigations in developing various candidate potential inhibitor compounds are time and resource-intensive. In this work, we use a quantitative structure-property relationship (QSPR)-based machine learning (ML) approach to investigate support vector machine (SVR), random forest (RF), and k-nearest neighbors (KNN) algorithms as predictive models of inhibition performance. (Inhibition efficiency) corrosion of pyridine-quinoline derivative compounds as corrosion inhibitors on iron. We found that the RF model showed the best predictive ability based on the coefficient of determination (R2) and root mean squared error (RMSE) metrics. Overall, our study provides new insights regarding the ML model in predicting corrosion inhibition on iron surfaces.

A Machine Learning Model for Evaluation of the Corrosion Inhibition Capacity of Quinoxaline Compounds

Investigating potential corrosion inhibitors via empirical research is a labor- and resource-intensive process. In this work, we evaluated various linear and non-linear algorithms as predictive models for corrosion inhibition efficiency (CIE) values using a machine learning (ML) paradigm based on the quantitative structure-property relationship (QSPR) model. In the quinoxaline compound dataset, our analysis showed that the XGBoost model performed the best predictor of other ensemble-based models. The coefficient of determination (R2), mean absolute percentage error (MAPE), and root mean squared error (RMSE) metrics were used to objectively assess this superiority. To sum up, our study offers a fresh viewpoint on the effectiveness of machine learning algorithms in determining the ability of organic compounds like quinoxaline to suppress corrosion on iron surfaces.

Development of a Machine Learning Model to Predict the Corrosion Inhibition Ability of Benzimidazole Compounds

The purpose of this study is to use quantitative structure-property relationship (QSPR)-based machine learning (ML) to examine the corrosion inhibition capabilities of benzimidazole compounds. The primary difficulty in ML development is creating a model with a high degree of precision so that the predictions are correct and pertinent to the material's actual attributes. We assess the comparison between the extra trees regressor (EXT) as an ensemble model and the decision tree regressor (DT) as a basic model. It was discovered that the EXT model had better predictive performance in predicting the corrosion inhibition performance of benzimidazole compounds based on the coefficient of determination (R2) and root mean square error (RMSE) metrics compared DT model. This method provides a fresh viewpoint on the capacity of ML models to forecast potent corrosion inhibitors.

Isolating pollen signals from laser diode aerosol Optical Particle Counter (OPC) data through positive matrix factorization (PMF) and Unmix receptor models

Pollen, a significant natural bioaerosol and allergen for sensitized individuals, is expected to increase in prevalence due to climate change. Mitigating allergy symptoms involves avoiding pollen exposure and pre-medication, emphasizing the importance of real-time knowledge of localized ambient air pollen concentrations. Laser diode Optical Particle Counters (OPCs) are commonly used for monitoring particle number concentrations in ambient air. This study explores the hypothesis that OPCs can monitor pollen but may struggle to distinguish them from other particles. We aimed to isolate the pollen signal from collective particle number concentrations using source apportionment models, specifically Positive Matrix Factorization (PMF) and Unmix, applied to multiple bin OPC data. The pollen signals isolated using PMF show slightly better correlation values than those isolated using Unmix. PMF-derived pollen signals exhibit strong correlations with Holoptelea (r = 0.64) and total pollen (r = 0.54) concentrations, while a moderate correlation is observed with Poaceae (r = 0.47). Exclusion of low pollen events strengthens correlations for Holoptelea and Poaceae to very strong (r = 0.87) and strong (r = 0.67), respectively. Although both model types effectively isolate the pollen signal, metrics suggest that Unmix has the potential for more accurate predictions of both moderate and extreme pollen events simultaneously. The Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Relative Root Mean Square Error (RRMSE) metrics for Holoptelea are 46.2 grains m−3, 72.4 grains m−3, and 15.3; for Poaceae, 3.9 grains m−3, 4.9 grains m−3, and 13.0; and for total pollen, 43.5 grains m−3, 72.1 grains m−3, and 14.1. This study represents a significant development in the use of source apportionment models and ambient OPCs for real-time pollen monitoring, offering a cost-effective alternative to conventional automated pollen sensors. Despite challenges, the proposed methodology provides a practical and accessible solution for pollen monitoring, contributing to the advancement of bioaerosol monitoring technologies.

Comparison of Different Machine Learning Approaches for Forecasting Stock Prices

Predicting stock prices is a crucial task that has significant implications for investment decisions, business strategies, and financial market stability. Accurate predictions can help investors make informed decisions, capture opportunities, and minimize risks. Understanding financial markets, economic data, company-specific elements, and a variety of statistical and analytical approaches are all necessary for making accurate stock price predictions. This paper employs three machine learning methods to forecast stock price data from a three-year time series dataset. Stock prediction plays a fundamental role in investment decisions, and accurate predictions are crucial. While various methods for predicting stock trends exist, different stock datasets may require different prediction models. Using three years of stock data, including gold and bitcoin, from the 2022 MCM Problem C, this study applies Support Vector Machines (SVM), Long Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN) to make predictions. Bayesian hyperparameter optimization is employed, and a comparative analysis is conducted. Ultimately, the results indicate that CNN outperforms the other methods, exhibiting relatively low error metrics and a high R-squared value, with no apparent signs of overfitting.

Snow depth in high-resolution regional climate model simulations over southern Germany – suitable for extremes and impact-related research?

Abstract. Snow dynamics play a critical role in the climate system, as they affect the water cycle, ecosystems, and society. In climate modelling, the representation of the amount and extent of snow on the land surface is crucial for simulating the mass and energy balance of the climate system. Here, we evaluate simulations of daily snow depths against 83 station observations in southern Germany in an elevation range of 150 to 1000 m over the time period 1987–2018. Two simulations stem from high-resolution regional climate models – the Weather Research & Forecasting (WRF) model at 1.5 km resolution and the COnsortium for Small scale MOdelling model in CLimate Mode (COSMO-CLM; abbreviated to CCLM hereafter) at 3 km resolution. Additionally, the hydrometeorological snow model Alpine MUltiscale Numerical Distributed Simulation ENgine (AMUNDSEN) is run at point scale at the locations of the climate stations, based on the atmospheric output of CCLM. To complement the comparison, the ERA5-Land dataset (9 km), a state-of-the-art reanalysis land-surface product, is also compared. All four simulations are driven by the atmospheric boundary conditions of ERA5. Due to an overestimation of the snow albedo, the WRF simulation features a cold bias of 1.2 °C, leading to the slight overestimation of the snow depth in low-lying areas, whereas the snow depth is underestimated at snow-rich stations. The number of snow days (days with a snow depth above 1 cm) is reproduced well. The WRF simulation can recreate extreme snow depths, i.e. annual maxima of the snow depth, their timings, and inter-station differences, and thereby shows the best performance of all models. The CCLM reproduces the climatic conditions with very low bias and error metrics. However, all snow-related assessments show a strong systematic underestimation, which we relate to deficiencies in the snow module of the land-surface model. When driving AMUNDSEN with the atmospheric output of the CCLM, the results show a slight tendency to overestimate snow depth and number of snow days, especially in the northern parts of the study area. Snow depth extremes are reproduced well. For ERA5-Land (ERA5L), the coarser spatial resolution leads to larger differences between the model elevation and the station elevation, which contributes to a significant correlation of climatic biases with the elevation bias. In addition, the mean snow depth and number of snow days are strongly overestimated, with conditions that are too snowy in the late winter. Extreme snow depth conditions are reproduced well in the low-lying areas, whereas strong deviations occur with more complex topography. In sum, due to the high spatial resolution of convection-permitting climate models, they show the potential to reproduce the winter climate (temperature and precipitation) in southern Germany. However, different sources of uncertainties, i.e. the spatial resolution, the snow albedo parametrisation, and other parametrisations within the snow model, prevent their further use in a straightforward manner for impact research. Hence, careful evaluation is needed before any impact-related interpretation of the simulations, such as in the context of climate change research.

Prediction of soil organic carbon using machine learning techniques and geospatial data for sustainable agriculture

Sustainable agriculture revolves around soil organic carbon (SOC), which is essential for numerous soil functions and ecological attributes. Farmers are interested in conserving and adding additional soil organic carbon to certain fields in order to improve soil health and productivity. The relationship between soil and environment that has been discovered and standardized throughout time has enhanced the progress of digital soil-mapping techniques; therefore, a variety of machine learning techniques are used to predict soil properties. Studies are thriving at how effectively each machine learning method maps and predicts SOC, especially at high spatial resolutions. To predict SOC of soil at a 30 m resolution, four machine learning models—Random Forest, Support Vector Machine, Adaptive Boosting, and k-Nearest Neighbour were used. For model evaluation, two error metrics, namely R2 and RMSE have been used. The findings demonstrated that the calibration and validation sets’ descriptive statistics sufficiently resembled the entire set of data. The range of the calculated SOC content was 0.06 to 1.76 %. According to the findings of the study, Random Forest showed good results for both cases, i.e. evaluation using cross validation and without cross validation. Using cross validation, RF confirmed highest R2 as 0.5278 and lowest RMSE as 0.1683 for calibration dataset while without cross validation it showed R2 as 0.8612 and lowest RMSE as 0.0912 for calibration dataset. The generated soil maps will help farmers adopt precise knowledge for decisions that will increase farm productivity and provide food security through the sustainable use of nutrients and the agricultural environment.

Real space iterative reconstruction for vector tomography (RESIRE-V)

Tomography has had an important impact on the physical, biological, and medical sciences. To date, most tomographic applications have been focused on 3D scalar reconstructions. However, in some crucial applications, vector tomography is required to reconstruct 3D vector fields such as the electric and magnetic fields. Over the years, several vector tomography methods have been developed. Here, we present the mathematical foundation and algorithmic implementation of REal Space Iterative REconstruction for Vector tomography, termed RESIRE-V. RESIRE-V uses multiple tilt series of projections and iterates between the projections and a 3D reconstruction. Each iteration consists of a forward step using the Radon transform and a backward step using its transpose, then updates the object via gradient descent. Incorporating with a 3D support constraint, the algorithm iteratively minimizes an error metric, defined as the difference between the measured and calculated projections. The algorithm can also be used to refine the tilt angles and further improve the 3D reconstruction. To validate RESIRE-V, we first apply it to a simulated data set of the 3D magnetization vector field, consisting of two orthogonal tilt series, each with a missing wedge. Our quantitative analysis shows that the three components of the reconstructed magnetization vector field agree well with the ground-truth counterparts. We then use RESIRE-V to reconstruct the 3D magnetization vector field of a ferromagnetic meta-lattice consisting of three tilt series. Our 3D vector reconstruction reveals the existence of topological magnetic defects with positive and negative charges. We expect that RESIRE-V can be incorporated into different imaging modalities as a general vector tomography method. To make the algorithm accessible to a broad user community, we have made our RESIRE-V MATLAB source codes and the data freely available at https://github.com/minhpham0309/RESIRE-V.

An Efficient and Interpretable Stacked Model for Wind Speed Estimation Based on Ensemble Learning Algorithms

Wind energy has gained tremendous attention recently due to its abundance potential and to obviate the adverse impacts of fossil fuel consumption. However, the efficient development and operation of wind energy systems require precise and accurate wind speed prediction. To fulfill this aim, machine learning algorithms have shown its prowess in accurately predicting the wind speed and this study proposes an interpretable stacked ensemble with ridge regressor utilizing ML algorithms such as catboost (CB), gradient boost (GB), multilayer perceptron (MLP), and random forest (RF) as a base model. The base models’ estimations are integrated by applying a ridge regressor to enhance the accuracy of wind speed estimation. The dataset consists of 87 600 samples from which 80% samples are used for training. The proposed approach provides interpretability of the prediction results through Shapley additive explanations and feature importance analysis. The result of the proposed method is calculated by comparing the estimation outcomes with each model such as CB, GB, MLP, and RF in terms of error metrics such as root mean squared error (RMSE) and mean absolute error (MAE). The proposed model outperforms the others due to a reduction of error in RMSE by 19.89% and MAE by 24% respectively.

Validation and Safety Profile of a Novel, Noninvasive Fiducial Attachment for Stereotactic Robotic-Guided Stereoelectroencephalography: A Case Series.

We developed, tested, and validated a novel, noninvasive, Leksell G frame-based fiducial attachment, for use in stereotactic registration for stereoelectroencephalography (sEEG). Use of the device increased the number of fixed reference points available for registration, while obviating the need for additional scalp incisions. We report here on our experience and safety profile of using the device. We collected registration data using the fiducial device across 25 adult and pediatric patients with epilepsy consecutively undergoing robotic-guided sEEG for invasive epilepsy monitoring, treated between May 2022 and July 2023. ROSA One Brain was used for trajectory planning and electrode implantation. Postoperative clinical and radiographic data were computed and quantified, including mean registration error for all patients. Entry point, target point (TP), and angular errors were measured. Descriptive statistics and correlation coefficients for error were calculated. Twenty-five patients underwent robotic-guided sEEG implantation (11 patients, bilateral; 10 patients, left unilateral; 4 patients, right). The mean number of electrodes per patient was 18 ± 3. The average mean registration error was 0.77 ± 0.11 mm. All patients were implanted with Ad-Tech depth electrodes. No clinically relevant complications were reported. Analysis of trajectory error was performed on 446 electrodes. The median entry point error was 1.03 mm (IQR 0.69-1.54). The median TP error was 2.26 mm (IQR 1.63-2.93). The mean angular error was 0.03 radians (IQR 0.02-0.05). There was no significant correlation between root mean square error and lead error. Root mean square error did not appreciably change over time, nor were there any significant changes in average angular, entry point, or TP error metrics. A novel, noninvasive, Leksell G frame-based fiducial attachment was developed, tested, and validated, facilitating O-arm-based stereotactic registration for sEEG. This simple innovation maintained an excellent accuracy and safety profile for sEEG procedures in epilepsy patients, with the added advantages of providing additional reference points for stereotactic registration, without requiring additional scalp incisions.

Indirect tuning of a complementary orientation filter using velocity data and a genetic algorithm

In this paper, the accuracy of inertial sensor orientation relative to the level frame is improved through optimal tuning of a complementary filter by a genetic algorithm. While constant filter gains have been used elsewhere, these may introduce errors under dynamic motions when gyroscopes should be trusted more than accelerometers. Optimal gains are prescribed by a Mamdani fuzzy rule base whose membership functions are found using a genetic algorithm and experimental data. Furthermore, model fitness is not based directly on orientation but the error between estimated and ground truth velocities. This paper has three interrelated novel elements. The main novelty is the indirect tuning method, which is simple, low-cost and requires a single camera and inertial sensor. The method is shown to increase tracking accuracy compared with popular baseline filters. Secondary novel elements are the bespoke genetic algorithm and the time agnostic velocity error metric. The contributions from this work can help improve the localization accuracy of assets and human personnel. This research has a direct impact in command and control by improving situational awareness and the ability to direct assets to safe locations using safer routes. This results in increasing safety in applications such as firefighting and battlespace.

A truncated Gaussian distribution based multi-scale segment-wise fusion transformer model for multi-step commodity price forecasting

Accurately forecasting commodity price trends is crucial for producers, market participants, and related enterprises to make informed decisions regarding production planning and scheduling. However, achieving high accuracy in multi-step forecasting poses significant challenges due to the unique financial characteristics inherent in commodities. Thus, this paper proposes a novel truncated Gaussian distribution based multi-scale segment-wise fusion Transformer for multi-step commodity price forecasting. First, a multi-scale segment-wise fusion module, which capture the time dependencies from different time granularity, is designed to describe the time-varying trend characteristics of commodity prices. Second, considering the characteristics of price range fluctuation and truncation, a truncated Gaussian distribution is introduced to describe price uncertainty. Last, to evaluate the proposed method’s effectiveness, extensive experiments are conducted using real data on energy chemical product prices. The experimental results demonstrate that the proposed method accurately captures price change trends and effectively estimates price uncertainty. Compared to the widely adopted Autoformer, our approach achieves approximately 30% reductions in both root mean square error (RMSE) and mean absolute error (MAE) metrics. Additionally, it exhibits certain advantages over the current state-of-the-art (SOTA). In the 20-step and 60-step multi-step prediction tasks, the proposed method achieves RMSE values of 91.18 and 142.94, respectively, surpassing the current SOTA. The introduced research framework provides valuable insights for decision-makers engaged in analyzing and forecasting commodity markets. The code is available on https://github.com/dean-ob/TGD-MSSF.

Evaluation of satellite precipitation products for real-time extreme river flow modeling in data scarce regions

Abstract. Inadequacy of spatio-temporal hydro-climatic data limits the efficacy of hazard monitoring and disaster risk reduction activities in disaster-prone areas. Various satellite missions are recently providing climate data, but prior evaluation and enhancement of these data are necessary for a reliable application. In this study, we conducted performance evaluation and enhancement of three real-time satellite precipitation products (SPPs) (GSMaP, GPM-IMERG, and PERSIANN) for flood modeling in the Blue Nile basin. The bias correction improved the original SPPs, with the largest improvement being for factors generated from 10 d mean data. Flood event hydrograph indicated satisfactory results of error metrics on the devastating flood event of 2012. Employing reliable physical–based distributed hydrologic models provided longer lead time and high-accuracy flood simulation. Furthermore, the results indicate that integrating available initial observed precipitation data improved the efficiency of SPPs simulation, and hence are applicable in operational flood monitoring.

Optimizing hydrogen evolution prediction: A unified approach using random forests, lightGBM, and Bagging Regressor ensemble model

Hydrogen, as a clean and versatile energy carrier, plays a pivotal role in addressing global energy challenges and transitioning towards sustainable energy systems. This study explores the convergence of machine learning (ML) for photocatalytic hydrogen evolution from sucrose solution using perovskite-type catalysts, namely LaFeO3 (LFO) and graphene-supported LaFeO3 (GLFO). This study pioneers the practical application of ML techniques, including Random Forests, LightGBM, and Bagging Regressor, to predict hydrogen yields in the presence of these photocatalysts. LFO and GLFO underwent a thorough characterization study to validate their successful preparation. Noteworthy, the highest hydrogen yield from the sucrose model solution was achieved using GLFO as 3.52 mmol/gcat. The optimum reaction conditions were experimentally found to be pH = 5.25, 0.15 g/L of catalyst amount, and 7.5 mM of HPC (hydrogen peroxide concentration). A pivotal contribution of this research lies in the practical application of ML models, culminating in the development of an ensemble model. This collaborative approach not only achieved an overall R2 of 0.92 but also demonstrated exceptional precision, as reflected in remarkably low error metrics. The mean squared logarithmic error (MSLE) was 0.0032, and the mean absolute error (MAE) was 0.049, underscoring the effectiveness of integrating diverse ML algorithms. This study advances both the understanding of photocatalytic hydrogen evolution and the practical implementation of ML in predicting intricate chemical reactions.

Machine learning-based predictive model for thermal comfort and energy optimization in smart buildings

In the current context of energy transition and increasing climate change, optimizing building performance has become a critical objective. Efficient energy use and occupant comfort are paramount considerations in building design and operation. To address these challenges, this study introduces a predictive model leveraging Machine Learning (ML) algorithms. The model aims to predict thermal comfort levels and optimize energy consumption in Heating, Ventilation, and Air Conditioning (HVAC) systems. Four distinct ML algorithms Support Vector Machine (SVM), Artificial Neural Network (ANN), Random Forest (RF), and EXtreme Gradient Boosting (XGBOOST) are employed for this purpose. Data for the model is collected using a network of Raspberry Pi boards equipped with multiple sensors. Performance evaluation of the ML algorithms is conducted using statistical error metrics, including, Root Mean Square Error (RMSE), Mean Square Error (MSE), Mean Absolute Error (MAE), and coefficient of determination (R2). Results reveal that the RF and XGBOOST algorithms exhibit superior performance, achieving accuracies of 96.7 % and 9.64 % respectively. In contrast, the SVM algorithm demonstrates inferior performance with a R2 of 81.1 %. These findings underscore the predictive capability of the RF and XGBOOST model in forecasting Predicted Mean Vote (PMV) values. The proposed model holds promise for enhancing occupant thermal comfort in buildings while simultaneously optimizing energy consumption in HVAC systems. Further research could explore the practical applications of these findings in building design and operation.

Stage and discharge prediction from documentary time-lapse imagery

Imagery from fixed, ground-based cameras is rich in qualitative and quantitative information that can improve stream discharge monitoring. For instance, time-lapse imagery may be valuable for filling data gaps when sensors fail and/or during lapses in funding for monitoring programs. In this study, we used a large image archive (>40,000 images from 2012 to 2019) from a fixed, ground-based camera that is part of a documentary watershed imaging project (https://plattebasintimelapse.com/). Scalar image features were extracted from daylight images taken at one-hour intervals. The image features were fused with United States Geological Survey stage and discharge data as response variables from the site. Predictions of stage and discharge for simulated year-long data gaps (2015, 2016, and 2017 water years) were generated from Multi-layer Perceptron, Random Forest Regression, and Support Vector Regression models. A Kalman filter was applied to the predictions to remove noise. Error metrics were calculated, including Nash-Sutcliffe Efficiency (NSE) and an alternative threshold-based performance metric that accounted for seasonal runoff. NSE for the year-long gap predictions ranged from 0.63 to 0.90 for discharge and 0.47 to 0.90 for stage, with greater errors in 2016 when stream discharge during the gap period greatly exceeded discharge during the training periods. Importantly, and in contrast to gap-filling methods that do not use imagery, the high discharge conditions in 2016 could be visually (qualitatively) verified from the image data. Half-year test sets were created for 2016 to include higher discharges in the training sets, thus improving model performance. While additional machine learning algorithms and tuning parameters for selected models should be tested further, this study demonstrates the potential value of ground-based time-lapse images for filling large gaps in hydrologic time series data. Cameras dedicated for hydrologic sensing, including nighttime imagery, could further improve results.