Figure - available from: Atmosphere
This content is subject to copyright.
Architecture of the proposed LSTM-based temperature prediction model with a missing data refinement function incorporated.
Source publication
In this paper, we propose a new temperature prediction model based on deep learning by using real observed weather data. To this end, a huge amount of model training data is needed, but these data should not be defective. However, there is a limitation in collecting weather data since it is not possible to measure data that have been missed. Thus,...
Similar publications
With the development of data mining technology, educational data mining (EDM) has gained increasing amounts of attention. Research on massive open online courses (MOOCs) is an important area of EDM. Previous studies found that assignment-related behaviors in MOOCs (such as the completed number of assignments) can affect student achievement. However...
Citations
... Deep learning models, especially neural networks, have proven considerable promise in temperature prediction. Long Short-Term Memory (LSTM) networks can capture long-term dependencies in time-series data due to its widespread adoption (Park et al., 2019;Karunakar et al., 2023;G et al., 2024;Waqas and Humphries, 2024). For instance, LSTM models have effectively improved forecast accuracy over short and extended periods by incorporating missing data refinement processes. ...
Accurate temperature prediction is critical in diverse areas, such as agriculture, disaster management, and urban planning, where understanding climatic patterns is essential. This study explores the application of advanced deep-learning models for temperature forecasting, focusing on the model’s ability to establish complex relationships and temporal dependencies within climatic data. This study evaluates the performance of various deep-learning models for temperature prediction using environmental data. Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional LSTM (BiLSTM), and Bidirectional GRU (BiGRU) models were developed and compared. The models were trained on meteorological data from Ranchi, India, spanning 2014-2024. Performance was assessed using Root Mean Square Error (RMSE), loss function analysis, and statistical significance testing. Results indicate that bidirectional architectures (BiLSTM and BiGRU) consistently outperformed unidirectional models. BiLSTM achieved the lowest RMSE and most balanced loss values across training, validation and test sets. The BiLSTM model performed well by 39.19.7% in RMSE and 15.36% in test loss. From the statistical analysis, BiLSTM is the best performer compared with BiGRU, with a negative t-statistic (-29.65) and a very low p-value (0.00000771), indicating a statistically significant difference.
... The empirical findings utilizing various assessment criteria indicate that the suggested hybrid model effectively operates with regards to energy usage. Furthermore, the LSTM model has been employed in various meteorological prediction applications such as temperature forecasting [27], air quality forecasting [28], wind speed forecasting [29], and water quality forecasting [30]. While the LSTM resolves many issues present in RNN, its learning process is considered challenging due to the dead region effect caused by typical sigmoid and hyperbolic tangent activation functions. ...
Long short-term memory (LSTM) networks are critical in predicting periodic time series data on energy consumption, as many other forecasting methods do not take into account periodicity. Despite the effective forecasting capabilities of LSTM networks in predicting periodic energy consumption data, they are hindered by the dead region effect, which is caused by the sigmoid and hyperbolic tangent activation functions. These functions control the flow of information and determine which data is suitable for updating and learning within specific boundaries, but they also create unused regions that impact the accuracy and efficiency of the learning process in LSTM networks. To address this issue, this study introduces an Internet of Things (IoT) energy consumption forecasting model based on an improved long short-term memory (ILSTM) approach. This model aims to overcome the dead region problem and enhance the accuracy and learning process of traditional LSTM networks. The study collected actual energy consumption data from a residential building using a CT (SCT-013-030) sensor and ESP8266 NodeMCU real model with the Thingspek cloud platform for data processing. Additionally, a storage data recycling (SDR) technique is utilized to address data clustering shortages and fill missing information. The ILSTM forecasting model was assessed using various evaluation metrics including mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE). Additionally, comparisons were made between the throughput, latency, and bill information of the proposed ILSTM forecasting model and the ARIMA, DBN Regression, and conventional LSTM (CLSTM) forecasting models. The evaluation demonstrated that the ILSTM network outperformed the CLSTM network, showing improvements of 61.697% in MAE, 59.248% in MSE, and 50.537% in RMSE. Furthermore, the ILSTM network exhibited lower throughput values for varying energy consumption data compared to the CLSTM, and demonstrated reduced latency compared to ARIMA, DBN Regression, and CLSTM by 40.1, 21.1, and 13.5 cycles, respectively. Lastly, the results revealed that the ILSTM network provided more accurate energy consumption forecasting and bill estimation than the CLSTM.
... Additionally, the CNN model with hourly input data exhibited superior performance compared to the MLP and LSTM models. In [14], LSTM model was applied an to forecast air temperature at three locations in South Korea, utilizing inputs such as wind speed, air temperature, and humidity. The LSTM model, consisting of four layers, accurately predicted air temperature for both short-term (6, 12, and 24 h ahead) and long-term (14 days in advance) periods. ...
... Radhika and Shashi used the Support Vector Machines (SVMs) to forecast atmospheric temperature, which got better prediction performance than the traditional MLP algorithm [8]. In recent year, Park and Kim successfully forecasted atmospheric temperature with defective dataset by using Long Short-Term Memory Neural Network (LSTM) [9]. Chu et al. forecasted atmospheric temperature by using recurrent neural networks only with image data [10]. ...
As global temperatures increased by 1.1 Celsius degrees, there has unprecedented shifts in climate systems. With the rising impact of global warming, exploring the warming trend helps to better understand and maintain the local environment and economy. This study focuses on predicting atmospheric temperature in the Sacramento area using ARIMA and ETS models. The research uses the temperature data from Sacramento Airport's Automated Surface Observing System (ASOS) and explores the prediction performance of ARIMA, ETS, and ARIMAX models in predicting daily average temperatures. The results indicate the ARIMAX (1,1,1) model is the most suitable for forecasting this temperature data with the lowest AIC and RMSE values. However, there are still challenges, in particular the dependence of the ARIMAX model on future exogenous variables, which leads to the forecast outcomes less accurate. To enable this ARIMAX models have better prediction performance, incorporating more exogenous variables is a potential solution. Therefore, the methods discussed in this paper provides ideas for the atmospheric temperature forecasting and points out the direction for further research.
... In developing countries, weather observations often include missing values, especially in diverse regions like Nepal, where the plains, hills, and Himalayas create distinct climatic zones within short distances. These gaps, caused by equipment malfunctions or human error, necessitate weather imputation, which estimates missing data using techniques such as graph neural networks [19][20][21], regression trees, artificial neural networks [22], rain gauge estimation [23], spatial-correlation approaches [24], statistical imputation with quality control measures [25], and machine learning [26,27]. Recent methods like ARIMA [20], LSTM [19], BiLSTM [28], sequential imputation [29], dynamic time warping [30], and edge cutting [31] have further advanced this field. ...
... These gaps, caused by equipment malfunctions or human error, necessitate weather imputation, which estimates missing data using techniques such as graph neural networks [19][20][21], regression trees, artificial neural networks [22], rain gauge estimation [23], spatial-correlation approaches [24], statistical imputation with quality control measures [25], and machine learning [26,27]. Recent methods like ARIMA [20], LSTM [19], BiLSTM [28], sequential imputation [29], dynamic time warping [30], and edge cutting [31] have further advanced this field. Weather imputation is crucial for maintaining continuous, reliable datasets and supporting decision-making in weather-sensitive activities [45]. ...
In regions like Nepal, characterized by diverse geography, missing weather data poses a significant challenge for traditional imputation methods. These methods often struggle to capture the complexities of dynamic environments adequately. To overcome this challenge, our study explores the application of graph neural networks for weather prediction in data-scarce environments. Our approach entails the development of specialized models tailored to accommodate the non-Euclidean topology inherent in weather data. This framework encompasses preprocessing, graph representation, feature selection, and imputation techniques to predict missing atmospheric variables. The adaptability of our models to intricate geography ensures more precise representations of weather conditions. Our research demonstrates the efficacy of these models through rigorous testing on a substantial dataset spanning four decades since 1981. By harnessing state-of-the-art graph neural network technology, our study aims to address existing gaps in weather data prediction, leading to improved historical weather forecasting accuracy. Ultimately, this advancement contributes to enhanced meteorological understanding and forecasting precision in data-scarce regions.
... This trend was confirmed by the ROC curves for RF and DT that had higher TPR (>90%) and lower FPR (<10%) for RF with imputation methods compared to CC (Supplement). Findings by Tiwaskar et al. [36] confirm this improvement in machine learning models' performance with imputation techniques in a study where they tested RF models with various levels of missing values. Performance improvement was observed not only for ensemble models but also for others. ...
... The LI imputation method relies mainly on time-dependent missing-value imputation instead of inter-attribute correlations employed by other imputation techniques [35]. For this reason, LI is incredibly efficient for time series and has been reported to improve the performance of neural network-based classifiers in other studies [36,37]. This is less of an issue for ensemble models that work by segregating data into similar packets small enough to identify their inherent patterns in the terminal nodes. ...
... The MLP model performed moderately with imputed data, even without resampling. This finding aligns with reports that the method improves the performance of neural network-based models; hence, it could be applied to these ML models without resampling [36,37]. The same behavior was observed for SI and MI data fitted to DT and RF models, resulting in better performance than CCs. ...
Missing data and class imbalance hinder the accurate prediction of rare events such as dairy mastitis. Resampling and imputation are employed to handle these problems. These methods are often used arbitrarily, despite their profound impact on prediction due to changes caused to the data structure. We hypothesize that their use affects the performance of ML models fitted to automated milking systems (AMSs) data for mastitis prediction. We compare three imputations—simple imputer (SI), multiple imputer (MICE) and linear interpolation (LI)—and three resampling techniques: Synthetic Minority Oversampling Technique (SMOTE), Support Vector Machine SMOTE (SVMSMOTE) and SMOTE with Edited Nearest Neighbors (SMOTEEN). The classifiers were logistic regression (LR), multilayer perceptron (MLP), decision tree (DT) and random forest (RF). We evaluated them with various metrics and compared models with the kappa score. A complete case analysis fitted the RF (0.78) better than other models, for which SI performed best. The DT, RF, and MLP performed better with SVMSMOTE. The RF, DT and MLP had the overall best performance, contributed by imputation or resampling (SMOTE and SVMSMOTE). We recommend carefully selecting resampling and imputation techniques and comparing them with complete cases before deciding on the preprocessing approach used to test AMS data with ML models.
... The model uses the Rectified Linear Unit (ReLU) activation function, which sets any negative values to 0. ReLU is chosen for its computational efficiency and ability to introduce a simple nonlinear transformation. Additionally, several research studies on forecasting air temperatures using deep learning neural networks have employed the ReLU activation function due to its effectiveness in capturing complex patterns while maintaining simplicity in computation [39]. To mitigate overfitting in the MLP model, a regularization technique known as dropout is implemented. ...
With growing concerns over climate change, accurately predicting temperature trends is crucial for informed decision-making and policy development. In this study, we perform a comprehensive comparative analysis of four advanced time series forecasting models—Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing (ETS), Multilayer Perceptron (MLP), and Gaussian Processes (GP)—to assess changes in minimum and maximum temperatures across four key regions in the United States. Our analysis includes hyperparameter optimization for each model to ensure peak performance. The results indicate that the MLP model outperforms the other models in terms of accuracy for temperature forecasting. Utilizing this best-performing model, we conduct temperature projections to evaluate the hypothesis that the rates of change in temperatures are greater than zero. Our findings confirm a positive rate of change in both maximum and minimum temperatures, suggesting a consistent upward trend over time. This research underscores the critical importance of refining time series forecasting models to address the challenges posed by climate change and supporting the development of effective strategies to mitigate the impacts of rising temperatures. The insights gained from this work emphasize the need for continuous advancement in predictive modeling techniques to better understand and respond to the dynamics of climate change.
... The results proved that by applying the spatial information the estimation accuracy of model boosted noticeably. Park et al. (2019) estimated AT via the lost data refinement method on the basis of LSTM neural network. The results confirmed that the recommended LSTM-refinement model accomplished the least RMSE values for 6, 12, and 24 h temperature forecast and also for 7-and 14-day temperature forecast in comparison with other DNN-based methods with either no refinement or linear interpolation. ...
The precise predicting of air temperature has a significant influence in many sectors such as agriculture, industry, modeling environmental processes. In this work, to predict the mean daily time series air temperature in Muğla city (ATm), Turkey, initially, five different layer structures of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) deep learning-based neural network models through the seq2seq regression forecast module are developed. Then, based on performance evaluation metrics, an optimal DL-based layer network structure designed is chosen to hybridize with the wavelet transform (WT) algorithm (i.e., WT-DNN model) to enhance the estimation capability. In this direction, among potential meteorological variables considered, the average daily sunshine duration (SSD) (hours), total global solar radiation (TGSR) (kw. hour/m²), and total global insolation intensity (TGSI) (watt/m²) from Jan 2014 to Dec 2019 are picked as the most effective input variables through correlation analysis to predict ATm. To thwart overfitting and underfitting problems, different algorithm tuning along with trial-and-error procedures through diverse types of hyper-parameters are performed. Consistent with the performance evaluation standards, comparison plots, and Total Learnable Parameters (TLP) value, the state-of-the-art and unique proposed hybrid WT-(LSTM × GRU) model (i.e., hybrid WT with the coupled version of LSTM and GRU models via Multiplication layer ()) is confirmed as the best model developed. This hybrid model under the ideal hyper-parameters resulted in an R² = 0.94, an RMSE = 0.56 (°C), an MBE = -0.5 (°C), AICc = -382.01, and a running time of 376 (s) in 2000 iterations. Nonetheless, the standard single LSTM layer network model as benchmark model resulted in an R² = 0.63, an RMSE = 4.69 (°C), an MBE = -0.89 (°C), AICc = 1021.8, and a running time of 186 (s) in 2000 iterations.
... The method relies mostly on time dependent missing value imputation as opposed to inter-attribute correlations employed by other imputation techniques [34]. For this reason LI is especially e cient for time series and has been reported to improve the performance of neural network based classi er in other studies [35,36]. ...
... The MLP model had fair performance with linear interpolation imputation without resampling. This is in line with reports that the method improves the performance of neural network based models, hence could be applied to these types of ML models without resampling [35,36]. The same behavior was observed for simple and multiple imputated data tted to decision tree models that resulted in good performance (kappa = 0.465 and 0.455 respectively). ...
Missing data and class imbalance represent a hindrance to accurate prediction of rare events such as mastitis (udder inflammation). Various methods are susceptible to handle the problem, however, little is known about their individual and combined effects on the performance of ML models fitted to AMS (automated milking system) data for mastitis prediction. We apply imputation and resampling to improve performance metrics of classifiers (logistic regression, stochastic gradient descent, multilayer perceptron, decision tree and random forest). Three imputation methods: simple imputer (SI), multiple imputer (MICE) and linear interpolation (LI) were compared to complete cases. Three resampling procedures: synthetic minority oversampling technique (SOMTE), Support Vector Machine SMOTE and SMOTE with Edited Nearest Neighbours were compared. We evaluated different techniques by calculating precision, recall, F1 Score and compared models based on kappa score. Both imputation and resampling techniques improved models performance. Complete case analysis suited the Stochastic Gradient Descent (SGD) Classifier better than resampling or imputation (kappa=0.280). The Logistic regression (LR) performed better with SVMSMOTE rand no imputation (kappa= 0.218). The Random Forest (RF), Decision Tree (DT) and Multilayer Perceptron (MLP) performed better than SGD and LR and handled well class imbalance and missing values without preprocessing. We propose careful selection of the technique to handle class imbalance and missing value prior to subjecting data to ML model is crucial to attain best ML model performance.
... The LDAPS, one of the numerical weather forecasting systems operated by the Korea Meteorological Administration (KMA), provided the boundary and initial conditions for the CFD model simulations every hour during the target period. The LDAPS is based on the unified model [41,42] of the Met Office, UK, and uses an Arakawa C-grid [43] for the horizontal-vertical grid system and a Charney-Phillips grid staggering [44] for the vertical grid system. The LDAPS has a horizontal resolution of 1.5 km and consists of 744 grids in the east-west direction and 928 grids in the north-south direction. ...
Urban areas consist of various land cover types, with a high proportion of artificial surfaces among them. This leads to unfavorable thermal environments in urban areas. Continuous research on the thermal environment, specifically on the sensible heat flux (Qh), has been conducted. However, previous research has faced temporal, spatial, and resolution limitations when it comes to detailed analysis of sensible heat flux in urban areas. Therefore, in this study, a computational fluid dynamics (CFD) model combined with the LDAPS and the VUCM was developed to simulate Qh at one-hour intervals over a 1-month period in an urban area with various land cover types. Model validation was performed by comparing it with measurements, confirming the suitability of the model for simulating Qh. The land cover was categorized into five types: building, road, bare land, grassland, and tree areas. Qh exhibited distinct patterns depending on the land cover type. When averaging the Qh distribution over the target period, buildings, roads, and bare land areas showed a predominance of upward Qh values, while grassland and tree areas displayed dominant downward Qh values. Additionally, even within the same land cover types, slight Qh variations were identified based on their surroundings. The averaged Qh value for building areas was the highest at 36.79 W m⁻², while that for tree areas was −3.04 W m⁻². Moreover, during the target period, the time-averaged Qh showed that building, road, and bare land areas peaked at 14 LST, while grassland and tree areas exhibited very low Qh values. Notably, buildings reached a maximum Qh of 103.30 W m⁻² but dropped to a minimum of 1.14 W m⁻² at 5 LST.