Reduction In Mean Absolute Error Research Articles

Visual deep learning with physics constraints for local scour evolution prediction at monopiles

Local scour threatens the safety of marine structures, necessitating the precise prediction of scour evolution around these structures. A visually oriented deep learning model, called Disentangled Physics-constrained Prediction (DPP), was proposed in this study to predict scour evolution at monopiles reliably. It integrates scouring physics with advanced video prediction techniques through a two-branch architecture. The Physics-constrained Recurrent Module (PRModule) branch leverages Recurrent Neural Networks (RNNs) for temporal differentiation, ensuring accurate prediction of scouring-related physical information. Meanwhile, the Convolutional Long-Short-Term Memory (ConvLSTM) branch captures spatial and temporal dynamics in scouring videos, focusing on the prediction of residual features. DPP outperformed three baseline models in predicting the scour evolution at monopiles. Across three scouring scenarios, DPP achieved a 14.2% decrease in Root Mean Squared Error, a 14.7% reduction in Mean Absolute Error, and an 8.1% increase in Structural Similarity on average, compared to the best-performing baseline model. The predicted scouring frames are found to agree well with the true frames, demonstrating DPP's potential as a valuable tool to protect marine infrastructures.

Forecasting price in a new hybrid neural network model with machine learning

A key aspect of asset investment and risk management is the study of forecasting stock prices. We investigate the machine learning stock price prediction in a new hybrid neural network model and put forth a forecasting method based on machine learning, composite data preprocessing method and the proposed new neural network model. To address the challenge of predicting stock prices in the face of market complexity and noise, we use the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm and the Savitzky–Golay (SG) filter to de-noise and enhance the data, and employ a neural network with three convolutional layers and a long short-term memory (LSTM) layer that enables it to capture complex temporal patterns in the data. We propose a new hybrid neural network prediction model (CEEMDAN-S-C-LSTM) and adopt a machine learning approach to compare it with the benchmark model using CSI 300 index data. The empirical results validate the effectiveness of the frequency decomposition algorithm and the convolutional layer, and demonstrate that our proposed model outperforms the benchmark model. Compared to the best benchmark model, the CEEMDAN-S-C-LSTM model proposed in this study demonstrates a significant improvement in performance. Specifically, it shows a 45.33% reduction in mean absolute error (MAE), 43.44% reduction in root mean square error (RMSE), 45.01% reduction in mean absolute percentage error (MAPE), and 3.90% improvement in coefficient of determination (R2). The study also explores the effect of different numbers of convolution layers and SG filters on the hybrid model. Our research expands on the use of neural networks and machine learning, offering a novel technical approach to making investment decisions and managing risks in financial systems.

An Improved Fusion-Based Semantic Similarity Measure for Effective Collaborative Filtering Recommendations

Semantic-enhanced recommendation systems are promising approaches to overcome the sparsity and cold-start problems, which are hard to handle using the conventional collaborative filtering (CF) approaches. Further research is needed to effectively integrate ontologies into collaborative filtering recommender systems. This paper proposes an ontology-based semantic similarity measure to evaluate similarities between items and eventually generate accurate recommendations. The proposed semantic similarity measure termed fusion-based semantic similarity takes into account the semantics of ontological instances (i.e. items) inferred from a specific domain ontology, which is determined by analyzing the hierarchical relationships among the instances, as well as the features of the instances and their relationships to other instances. The new measure comprehensively captures the semantic knowledge associated with instances by exploiting all possible shared semantics between instances in a given domain ontology. Furthermore, this paper proposes a new semantic-enhanced hybrid recommendation approach as a result of combining the new semantic similarity measure with the standard item-based CF to enhance the quality of generated recommendations. In order to assess the effectiveness of our semantic-enhanced hybrid collaborative filtering method, a series of experiments were conducted to compare the performance of the proposed approach against well-established benchmark techniques. The reported experimental results consistently emphasize its superiority, demonstrating enhanced predictive abilities and a notable improvement in the quality of recommendations. More specifically, the proposed approach achieved notable 6% reduction in Mean Absolute Error (MAE) in certain cases, outperforming other benchmark techniques. Additionally, this study highlights the potential of using semantic-based similarity to enhance the performance of recommendation systems. Such enhancements address challenges within collaborative filtering, potentially leading to advancements in recommendation system design and optimization.

Transfer learning model for cash-instrument prediction adopting a Transformer derivative

Investors aiming for high market returns must accurately predict the prices of various cash instruments. However, making accurate predictions is challenging due to the complex cyclic and trending characteristic of markets, characterized by high volatility and unpredictable fluctuations. Furthermore, many studies overlook how interactions between different markets affect price movements. To address these problems, this research introduces a deep transfer-learning approach derived from the Transformer model, named the rotary-positional encoding autocorrelation Transformer (RAT). Unlike traditional methods, the RAT employs autocorrelation instead of self-attention to more effectively capture periodic features, while rotary-positional encoding preserves both the absolute and relative positioning within sequences to enhance trend understanding. Through transfer learning, the RAT model extracts deep features from a source domain and applies them to a target domain, demonstrating superior performance over LSTM, CNN-LSTM, gated recurrent units (GRUs), and Transformer models in multi-day predictions across 12 cash-instrument datasets. It achieved a substantial increase in accuracy, with a 35.83% reduction in mean squared error (MSE), a 23.95% reduction in mean absolute error (MAE), and a 32.63% increase in the coefficient of determination (R2). This study validates the RAT model’s effectiveness in predicting financial instrument prices.

A Hybrid GARCH and Deep Learning Method for Volatility Prediction

Volatility prediction plays a vital role in financial data. The time series movements of stock prices are commonly characterized as highly nonlinear and volatile. This study is aimed at enhancing the accuracy of return volatility forecasts for stock prices by investigating the prediction of their price volatility through the integration of diverse models. Thus, the study integrated four powerful methods: seasonal autoregressive (AR) integrated moving average (MA), generalized AR conditional heteroskedasticity (ARCH) family models, convolutional neural network (CNN), and bidirectional long short‐term memory (LSTM) network. The hybrid model was developed using the residuals generated by the seasonal AR integrated MA model as input for the generalized ARCH model. Following this, the estimated volatility obtained was utilized as an input feature for both the hybrid CNNs and bidirectional LSTM models. The model’s forecasting performance was assessed using key evaluation metrics, including mean absolute error (MAE) and root mean squared error (RMSE). Compared to other hybrid models, our new proposed hybrid model demonstrates an average reduction in MAE and RMSE of 60.35% and 60.61%, respectively. The experimental results show that the model proposed in this study has good performance and accuracy in predicting the volatility of stock prices. These findings offer valuable insights for financial data analysis and risk management strategies.

An Improved VMD–EEMD–LSTM Time Series Hybrid Prediction Model for Sea Surface Height Derived from Satellite Altimetry Data

Changes in sea level exhibit nonlinearity, nonstationarity, and multivariable characteristics, making traditional time series forecasting methods less effective in producing satisfactory results. To enhance the accuracy of sea level change predictions, this study introduced an improved variational mode decomposition and ensemble empirical mode decomposition–long short-term memory hybrid model (VMD–EEMD–LSTM). This model decomposes satellite altimetry data from near the Dutch coast using VMD, resulting in components of the intrinsic mode functions (IMFs) with various frequencies, along with a residual sequence. EEMD further dissects the residual sequence obtained from VMD into second-order components. These IMFs decomposed by VMD and EEMD are utilized as features in the LSTM model for making predictions, culminating in the final forecasted results. The experimental results, obtained through a comparative analysis of six sets of Dutch coastal sea surface height data, confirm the excellent accuracy of the hybrid model proposed (root mean square error (RMSE) = 47.2 mm, mean absolute error (MAE) = 33.3 mm, coefficient of determination (R2) = 0.9). Compared to the VMD-LSTM model, the average decrease in RMSE was 58.7%, the average reduction in MAE was 60.0%, and the average increase in R2 was 49.9%. In comparison to the EEMD-LSTM model, the average decrease in RMSE was 27.0%, the average decrease in MAE was 28.0%, and the average increase in R2 was 6.5%. The VMD–EEMD–LSTM model exhibited significantly improved predictive performance. The model proposed in this study demonstrates a notable enhancement in global mean sea lever (GMSL) forecasting accuracy during testing along the Dutch coast.

Lapse rate-adjusted bias correction for CMIP6 GCM precipitation data: An application to the Monsoon Asia Region.

Bias correction (BC) of General Circulation Models (GCMs) variables is a common practice when it is being used for climate impact assessment studies at regional scales. The present study proposes a bias correction method (LR-Reg) that first adjusts the original GCM precipitation for local lapse rate corrections and later bias corrects the lapse rate-adjusted GCMs precipitation data with linear regression coefficients. We evaluated LR-Reg BC method in comparison to Linear Scaling (LS) and Quantile Mapping (QMap) BC methods, and NASA's downscaled NEX data for Monsoon Asia region. This study used Coupled Model Intercomparison Project Phase 6 (CMIP6)-based MIROC6 GCM precipitation with historical and projected shared socio-economic pathways (SSP) scenarios (SSP245 and SSP585) datasets. The BC comparison results show that the relative percentage reduction in mean absolute error (MAE) values of LR-Reg over LS-BC was up to 10-30% while this relative reduction in MAE values of LR-Reg was 30-50% over QMap-BC and 75-100% over NASA's NEX-data. The future projected precipitation over Monsoon Asia during dry season shows more decreased precipitation by up to 100% mostly in the south Asia while during wet season shows more increased precipitation by up to 50% mostly in the northeastern China and in the Himalayan belts with respect to the baseline condition (1970-2005). The results on the average precipitation per 0.25 degree increase in latitude analysis shows that the maximums of average monsoon precipitation during baseline period occur at 0 and 25 degree latitudes while the projected monsoon precipitation during both SSP scenarios occurs at 10 and 20 degree latitudes which clearly shows an inward shift in the latitude axis for the projected precipitation in the Monsoon Asia.

Unsupervised CNN-based DIC method for 2D displacement measurement

Digital image correlation (DIC) is a widely used technique for non-contact measurement of deformation. However, traditional DIC methods face challenges in balancing calculation efficiency and the quantity of seed points. Deep learning approaches, particularly supervised learning methods, have shown promise in improving DIC efficiency. However, these methods require high-quality training data, which can be time-consuming to generate ground truth annotations. To address these challenges, we propose an unsupervised convolutional neural network (CNN) based DIC method for 2D displacement measurement. Our approach leverages an encoder-decoder architecture with multi-level feature extraction, a dual-path correlation block, and an attention block to extract informative features from speckle images with varying characteristics. We utilize a speckle image warp model to transform the deformed speckle image to the predicted reference speckle image based on the predicted 2D displacement map. The unsupervised training is achieved by comparing the predicted and original reference speckle images. To optimize the network's parameters, we employ a composite loss function that takes into account both the Mean Squared Error (MSE) and Pearson correlation coefficient. By using unsupervised convolutional neural network (CNN) based DIC method, we eliminate the need for extensive training data annotation, which is a time-consuming process in supervised learning DIC methods. We have conducted several experiments to demonstrate the validity and robustness of our proposed method. The results show a significant reduction in Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) compared to a method proposed by Zhao et al. This indicates that our unsupervised CNN-based DIC approach can achieve accuracy comparable to supervised CNN-based DIC methods. For implementation and evaluation, we provide PyTorch code and datasets, which will be released at the following URL :https://github.com/fead1/DICNet-corr-unsupervised-learning-.

Fusing Sequence and Structural Knowledge by Heterogeneous Models to Accurately and Interpretively Predict Drug-Target Affinity.

Drug-target affinity (DTA) prediction is crucial for understanding molecular interactions and aiding drug discovery and development. While various computational methods have been proposed for DTA prediction, their predictive accuracy remains limited, failing to delve into the structural nuances of interactions. With increasingly accurate and accessible structure prediction of targets, we developed a novel deep learning model, named S2DTA, to accurately predict DTA by fusing sequence features of drug SMILES, targets, and pockets and their corresponding graph structural features using heterogeneous models based on graph and semantic networks. Experimental findings underscored that complex feature representations imparted negligible enhancements to the model's performance. However, the integration of heterogeneous models demonstrably bolstered predictive accuracy. In comparison to three state-of-the-art methodologies, such as DeepDTA, GraphDTA, and DeepDTAF, S2DTA's performance became more evident. It exhibited a 25.2% reduction in mean absolute error (MAE) and a 20.1% decrease in root mean square error (RMSE). Additionally, S2DTA showed some improvements in other crucial metrics, including Pearson Correlation Coefficient (PCC), Spearman, Concordance Index (CI), and R2, with these metrics experiencing increases of 19.6%, 17.5%, 8.1%, and 49.4%, respectively. Finally, we conducted an interpretability analysis on the effectiveness of S2DTA by bidirectional self-attention mechanism. The analysis results supported that S2DTA was an effective and accurate tool for predicting DTA.

Consumption–Production Profile Categorization in Energy Communities

Energy Transition is changing the renewable energy participation in new distributed generation systems like the Local Energy Markets. Due to its inherent intermittent and variable nature, forecasting production and consumption load profiles will be more challenging and demand more complex predictive models. This paper analyzes the production, consumption load profile, and storage headroom% of the Cornwall Local Energy Market, using advanced statistical time series methods to optimize the opportunity market the storage units provide. These models also help the Energy Community storage reserves to meet contract conditions with the Distribution Network Operator. With this more accurate and detailed knowledge, all sites from this Local Energy Market will benefit more from their installation by optimizing their energy consumption, production, and storage. This better accuracy will make the Local Energy Market more fluid and safer, creating a flexible system that will guarantee the technical quality of the product for the whole community. The training of several SARIMAX, Exponential Smoothing, and Temporal Causal models improved the fitness of consumption, production, and headroom% time series. These models properly decomposed the time series in trend, seasonality, and stochastic dynamic components that help us to understand how the Local Energy Market consumes, produces, and stores energy. The model design used all power flows and battery energy storage system state-of-charge site characteristics at daily and hourly granularity levels. All model building follows an analytical methodology detailed step by step. A benchmark between these sequence models and the incumbent forecasting models utilized by the Energy Community shows a better performance measured with model error reduction. The best models present mean squared error reduction between 88.89% and 99.93%, while the mean absolute error reduction goes from 65.73% to 97.08%. These predictive models built at different prediction scales will help the Energy Communities better contribute to the Network Management and optimize their energy and power management performance. In conclusion, the expected outcome of these implementations is a cost-optimal management of the Local Energy Market and its contribution to the needed new Flexibility Electricity System Scheme, extending the adoption of renewable energies.

Ring artifacts correction for computed tomography image using unsupervised contrastive learning

Objective. Computed tomography (CT) is a widely employed imaging technology for disease detection. However, CT images often suffer from ring artifacts, which may result from hardware defects and other factors. These artifacts compromise image quality and impede diagnosis. To address this challenge, we propose a novel method based on dual contrast learning image style transformation network model (DCLGAN) that effectively eliminates ring artifacts from CT images while preserving texture details. Approach. Our method involves simulating ring artifacts on real CT data to generate the uncorrected CT (uCT) data and transforming them into strip artifacts. Subsequently, the DCLGAN synthetic network is applied in the polar coordinate system to remove the strip artifacts and generate a synthetic CT (sCT). We compare the uCT and sCT images to obtain a residual image, which is then filtered to extract the strip artifacts. An inverse polar transformation is performed to obtain the ring artifacts, which are subtracted from the original CT image to produce a corrected image. Main results. To validate the effectiveness of our approach, we tested it using real CT data, simulated data, and cone beam computed tomography images of the patient’s brain. The corrected CT images showed a reduction in mean absolute error by 12.36 Hounsfield units (HU), a decrease in root mean square error by 18.94 HU, an increase in peak signal-to-noise ratio by 3.53 decibels (dB), and an improvement in structural similarity index by 9.24%. Significance. These results demonstrate the efficacy of our method in eliminating ring artifacts and preserving image details, making it a valuable tool for CT imaging.

A hybrid deep learning framework for unsteady periodic flow field reconstruction based on frequency and residual learning

Reconstructing a complete flow field from limited sensor measurement is quite essential for state evaluation, optimization, monitoring, and control of the flow system. Unsteady periodic flow, as a widespread phenomenon in science and engineering, attracts in-depth research over decades. Deep learning has been employed in flow field reconstruction, whereas the accurate estimation for the unsteady flow field with strong nonlinearity is still difficult. To address this problem, we propose a hybrid deep learning framework that incorporates frequency and residual learning to accurately reconstruct an unsteady periodic flow field from limited sensor measurement. First, to extract the frequency features, the historical flow field data is decomposed into different modes with different frequencies named frequency modes via fast Fourier transform (FFT). Next, we construct a hybrid deep neural network framework consisting of an inverse fast Fourier transform (IFFT) block and a residual block. The IFFT block maps sensor measurements to frequency mode temporal coefficients, which are multiplied with frequency modes to recover an IFFT field. Meanwhile, the residual block adaptively generates a residual field to complement the information lost by the IFFT field. Finally, the IFFT field and residual field are combined to produce the final reconstructed flow field. We conduct numerical experiments on the unsteady periodic flow around a cylinder and transonic flow around a NACA0012 airfoil to demonstrate the feasibility and high accuracy of our proposed method. Compared to the widely used proper orthogonal decomposition (POD) and shallow decoder (SD) methods, our approach achieves at least 83.9% and 72.2% reduction in mean absolute error, respectively.

Multi-Level Constrained Intra and Inter Subject Feature Representation for Facial Video Based BVP Signal Measurement.

Facial video-based blood volume pulse (BVP) signal measurement holds great potential for remote health monitoring, while existing methods have issues with convolutional kernel perceptual field constraints. This article proposes an end-to-end multi-level constrained spatiotemporal representation structure for facial video-based BVP signal measurement. First, an intra- and inter-subject feature representation is proposed to strengthen the BVP-related features generation at high, semantic, and shallow levels, respectively. Second, the global-local association is presented to enhance BVP signal period pattern learning, and the global temporal features are introduced into the local spatial convolution of each frame by adaptive kernel weights. Finally, the multi-dimensional fused features are mapped to one-dimensional BVP signals by the task-oriented signal estimator. The experimental results on the publicly available MMSE-HR dataset demonstrate that the proposed structure overperforms state-of-the-art methods (e.g., AutoHR) in BVP signal measurement, with a 20% and 40% reduction in mean absolute error and root mean squared error, respectively. The proposed structure would be a powerful tool for telemedical and non-contact heart health monitoring.

Enhanced Crash Frequency Models Using Surrogate Safety Measures from Connected Vehicle Fleet

As connected vehicle data became available, efforts to employ surrogate safety measures (SSMs) for crash frequency modeling were undertaken. For road safety evaluation, traffic conflicts are quantitatively measured by various SSMs in two dimensions: spatial/temporal proximity (e.g., time-to-collision, TTC) and evasive action (e.g., deceleration rate, DR). However, a single SSM or a single dimension only represents partial images of the true severity of traffic conflicts. Therefore, this study investigates possible enhancements in crash frequency modeling by concurrently using proximity and evasion SSMs. For rear-end crash frequency estimation, five negative binomial regression models and two tree-based models (a regression tree and random forest) were developed. All models were estimated using crashes, traffic volume, and segment length, along with three SSMs (DR, TTC, and modified TTC) extracted from connected vehicle data in Ann Arbor, Michigan. Results show that the multi-SSM model produced a 19.3% reduction in mean absolute error (MAE) compared to the baseline model with no SSM variable, which was significantly higher than those of single-SSM models (4.5−6.9% reductions in MAE). Between all models, the random forest, which is the ensemble machine learning model, produced the highest error reductions (a 44.3% reduction in MAE). These findings show that the concurrent use of proximity and evasion SSMs can yield further enhancements in crash frequency models compared to the singular use of either type of SSM. The proposed modeling method can be used for proactive safety management and assessment using connected vehicle data collected over a short period.

Harmonic Vibrational Frequency Simulation of Pharmaceutical Molecules via a Novel Multi-Molecular Fragment Interception Method.

By means of a computational method based on Density Functional Theory (DFT), using commercially available software, a novel method for simulating equilibrium geometry harmonic vibrational frequencies is proposed. Finasteride, Lamivudine, and Repaglinide were selected as model molecules to study the adaptability of the new method. Three molecular models, namely the single-molecular, central-molecular, and multi-molecular fragment models, were constructed and calculated by Generalized Gradient Approximations (GGAs) with the PBE functional via the Material Studio 8.0 program. Theoretical vibrational frequencies were assigned and compared to the corresponding experimental data. The results indicated that the traditional single-molecular calculation and scaled spectra with scale factor exhibited the worst similarity for all three pharmaceutical molecules among the three models. Furthermore, the central-molecular model with a configuration closer to the empirical structure resulted in a reduction of mean absolute error (MAE) and root mean squared error (RMSE) in all three pharmaceutics, including the hydrogen-bonded functional groups. However, the improvement in computational accuracy for different drug molecules using the central-molecular model for vibrational frequency calculation was unstable. Whereas, the new multi-molecular fragment interception method showed the best agreement with experimental results, exhibiting MAE and RMSE values of 8.21 cm-1 and 18.35 cm-1 for Finasteride, 15.95 cm-1 and 26.46 cm-1 for Lamivudine, and 12.10 cm-1 and 25.82 cm-1 for Repaglinide. Additionally, this work provides comprehensive vibrational frequency calculations and assignments for Finasteride, Lamivudine, and Repaglinide, which have never been thoroughly investigated in previous research.

A multi-terrain feature-based deep convolutional neural network for constructing super-resolution DEMs

Scale conversion between DEMs is an important issue in geomorphometry. There are many mature studies on the generation of low-resolution(LR) DEMs from high-resolution(HR) DEMs. However, as an important and convenient means of obtaining HR DEMs, traditional super resolution (SR) methods have shown insufficient consideration of the terrain features embedded in DEMs. Therefore, this article investigates the combination of terrain features and the use of convolutional neural networks (CNN) to reconstruct HR DEMs, and proposes a multi-terrain feature-based deep CNN for super-resolution(SR) DEMs (MTF-SR). In the experiments, from the perspective of vector and raster terrain features, we fuse raster terrain features in the input and loss functions, and fuse vector terrain features in the optimization of the output of the model. The results show that the MTF-SR model has a 30–50 % reduction in mean absolute error (MAE) compared with interpolation methods, has the lowest slope and aspect error and has a 10 to 30 % improvement in streamline matching rate (SMR). These results point to the advantages of the method in overall elevation accuracy and the preservation of terrain features.

Learning under concept drift and non-stationary noise: Introduction of the concept of persistence

Learning from noisy data is a challenging task especially when the system under consideration has a non-stationary nature. The source of the noise is often assumed to be stationary, however the severity or characteristics of noise may also be time-varying, which causes multiple sources of drift in the collected data. This study introduces a novel adaptive learning rate approach to improve learning when the observations from a non-stationary system is altered by an also non-stationary source of noise. As an example to this approach, we propose Persistence Aware Robust Learner (PeARL), an online learning method that utilizes a novel concept called persistence, which is a local noisiness estimation metric to measure the correspondence of a signal to discrete white noise. Making use of this metric, PeARL is able to adaptively adjust the learning rate for each observation during learning to reduce the effect of noise. With this level of control on the learning rate, noisy instances have less disruptive effect on the maintained estimate. We experimentally evaluate PeARL on (a) systematically generated synthetic data and (b) real-world data, including accelerometer readings collected from people (HASC2010corpus) and current measurements from electric motors collected within the scope of EU-funded research project iRel40. The experiments reveal a favorable region of noise rate, in which the proposed method achieves up to 40% reduction in mean absolute error (MAE).

A Method for Merging Multi-Source Daily Satellite Precipitation Datasets and Gauge Observations over Poyang Lake Basin, China

Obtaining precipitation estimates with high resolution and high accuracy is critically important for regional meteorological, hydrological, and other applications. Although satellite precipitation products can provide precipitation fields at various scales, their applications are limited by the relatively coarse spatial resolution and low accuracy. In this study, we propose a multi-source merging approach for generating accurate and high-resolution precipitation fields on a daily time scale. Specifically, a random effects eigenvector spatial filtering (RESF) method was first applied to downscale satellite precipitation datasets. The RESF method, together with Kriging, was then applied to merge the downscaled satellite precipitation products with station observations. The results were compared against observations and a data fusion dataset, the Multi-Source Weighted-Ensemble Precipitation (MSWEP). It was shown that the estimates of the proposed method significantly outperformed the individual satellite precipitation product, reducing the average value of mean absolute error (MAE) by 52%, root mean square error (RMSE) by 63%, and improving the mean value of Kling–Gupta efficiency (KGE) by 157%, respectively. Daily precipitation estimates exhibited similar spatial patterns to the MSWEP products, and were more accurate in almost all cases, with a 42% reduction in MAE, 46% reduction in RMSE, and 79% improvement in KGE. The proposed approach provides a promising solution to generate accurate daily precipitation fields with high spatial resolution.

Simultaneous multi-crop land suitability prediction from remote sensing data using semi-supervised learning

Land suitability models for Canada are currently based on single-crop inventories and expert opinion. We present a data-driven multi-layer perceptron that simultaneously predicts the land suitability of several crops in Canada, including barley, peas, spring wheat, canola, oats, and soy. Available crop yields from 2013–2020 are downscaled to the farm level by masking the district level crop yield data to focus only on areas where crops are cultivated and leveraging soil-climate-landscape variables obtained from Google Earth Engine for crop yield prediction. This new semi-supervised learning approach can accommodate data from different spatial resolutions and enables training with unlabelled data. The incorporation of a crop indicator function further allows for the training of a multi-crop model that can capture the interdependences and correlations between various crops, thereby leading to more accurate predictions. Through k-fold cross-validation, we show that compared to the single crop models, our multi-crop model could produce up to a 2.82 fold reduction in mean absolute error for any particular crop. We found that barley, oats, and mixed grains were more tolerant to soil-climate-landscape variations and could be grown in many regions of Canada, while non-grain crops were more sensitive to environmental factors. Predicted crop suitability was associated with a region’s growing season length, which supports climate change projections that regions of northern Canada will become more suitable for agricultural use. The proposed multi-crop model could facilitate assessment of the suitability of northern lands for crop cultivation and be incorporated into cost-benefit analyses.

A Stacking Ensemble Model of Various Machine Learning Models for Daily Runoff Forecasting

Improving the accuracy and stability of daily runoff prediction is crucial for effective water resource management and flood control. This study proposed a novel stacking ensemble learning model based on attention mechanism for the daily runoff prediction. The proposed model has a two-layer structure with the base model and the meta model. Three machine learning models, namely random forest (RF), adaptive boosting (AdaBoost), and extreme gradient boosting (XGB) are used as the base models. The attention mechanism is used as the meta model to integrate the output of the base model to obtain predictions. The proposed model is applied to predict the daily inflow to Fuchun River Reservoir in the Qiantang River basin. The results show that the proposed model outperforms the base models and other ensemble models in terms of prediction accuracy. Compared with the XGB and weighted averaging ensemble (WAE) models, the proposed model has a 10.22% and 8.54% increase in Nash–Sutcliffe efficiency (NSE), an 18.52% and 16.38% reduction in root mean square error (RMSE), a 28.17% and 18.66% reduction in mean absolute error (MAE), and a 4.54% and 4.19% increase in correlation coefficient (r). The proposed model significantly outperforms the base model and simple stacking model indicated by both the Friedman test and the Nemenyi test. Thus, the proposed model can produce reasonable and accurate prediction of the reservoir inflow, which is of great strategic significance and application value in formulating the rational allocation and optimal operation of water resources and improving the breadth and depth of hydrological forecasting integrated services.