Reply on AC2

Data-driven models have been recently suggested to surrogate computationally expensive hydrodynamic models to map flood hazards. However, most studies focused on developing models for the same area or the same precipitation event. It is hence not obvious how transferable the models are in space. This study evaluates the performance of a convolutional neural network (CNN) based on the U-Net architecture and the random forest (RF) algorithm to predict flood water depth, the models' transferability in space and performance improvement using transfer learning techniques. We used three study areas in Berlin to train, validate and test the models. The results showed that (1) the RF models outperformed the CNN models for predictions within the training domain, presumable at the cost of overfitting; (2) the CNN models had significantly higher potential than the RF models to generalize beyond the training domain; and (3) the CNN models could better benefit from transfer learning technique to boost their performance outside training domains than RF models.

This paper compares the performance of five different supervised machine learning (SL) algorithms to identify the practical model for tracking copper using solution cathode glow discharge optical emission spectroscopy (SCGD-OES). A total of 4500 SCGD-OES spectra were collected, corresponding to 15 different copper concentrations ranging from 10.3 mg l−1 to 59.6 mg l−1, for training and validating five models: K-nearest neighbors (KNN), random forest (RF), artificial neural network (ANN), convolutional neural network (CNN), and recurrent neural network (RNN). Then, performance of these models were evaluated based on accuracy, precision of predictions and training time. In terms of accuracy, the relative error (RE) was 3% for the CNN model and 4% for the ANN model. Conversely, the RE increased from 21% and 43% for the RF and RNN models to 103% for the KNN model due to the unpredictability of Cu concentration outside the training range. Regarding precision, the relative standard deviation (RSD) values were similar across all models, with the lowest RSD being 3% for the CNN model and the highest being 5% for the RF model. The training time indicated that the KNN and RF models provided the best learning speed, while the CNN model had the slowest learning speed, taking 5 times longer than the ANN model. The accuracy of the ANN and CNN models is in agreement with inductively coupled plasma-optical emission spectroscopy (ICP-OES), with a average difference not exceeding 6% when tracking six different Cu concentrations, both inside and outside the training range. Additionally, this research investigated the weight distribution in the hidden layers of the ANN, CNN, and RNN models and feature importances in the RF model to enhance understanding of the internal structure of models and assist analysts in selecting the practical model for tracking heavy metals based on SCGD-OES.

Precise and timely leaf area index (LAI) estimation for winter wheat is crucial for precision agriculture. The emergence of high-resolution unmanned aerial vehicle (UAV) data and machine learning techniques offers a revolutionary approach for fine-scale estimation of wheat LAI at the low cost. While machine learning has proven valuable for LAI estimation, there are still model limitations and variations that impede accurate and efficient LAI inversion. This study explores the potential of classical machine learning models and deep learning model for estimating winter wheat LAI using multispectral images acquired by drones. Initially, the texture features and vegetation indices served as inputs for the partial least squares regression (PLSR) model and random forest (RF) model. Then, the ground-measured LAI data were combined to invert winter wheat LAI. In contrast, this study also employed a convolutional neural network (CNN) model that solely utilizes the cropped original image for LAI estimation. The results show that vegetation indices outperform the texture features in terms of correlation analysis with LAI and estimation accuracy. However, the highest accuracy is achieved by combining both vegetation indices and texture features to invert LAI in both conventional machine learning methods. Among the three models, the CNN approach yielded the highest LAI estimation accuracy (R 2 = 0.83), followed by the RF model (R 2 = 0.82), with the PLSR model exhibited the lowest accuracy (R 2 = 0.78). The spatial distribution and values of the estimated results for the RF and CNN models are similar, whereas the PLSR model differs significantly from the first two models. This study achieves rapid and accurate winter wheat LAI estimation using classical machine learning and deep learning methods. The findings can serve as a reference for real-time wheat growth monitoring and field management practices.

Although many machine learning methods have been widely used to predict PM2.5 concentrations, these single or hybrid methods still have some shortcomings. This study integrated the advantages of convolutional neural network (CNN) feature extraction and the regression ability of random forest (RF) to propose a novel CNN-RF ensemble framework for PM2.5 concentration modeling. The observational data from 13 monitoring stations in Kaohsiung in 2021 were selected for model training and testing. First, CNN was implemented to extract key meteorological and pollution data. Subsequently, the RF algorithm was employed to train the model with five input factors, namely the extracted features from the CNN and spatiotemporal factors, including the day of the year, the hour of the day, latitude, and longitude. Independent observations from two stations were used to evaluate the models. The findings demonstrated that the proposed CNN-RF model had better modeling capability compared with the independent CNN and RF models: the average improvements in root mean square error (RMSE) and mean absolute error (MAE) ranged from 8.10% to 11.11%, respectively. In addition, the proposed CNN-RF hybrid model has fewer excess residuals at thresholds of 10 μg/m3, 20 μg/m3, and 30 μg/m3. The results revealed that the proposed CNN-RF ensemble framework is a stable, reliable, and accurate method that can generate superior results compared with the single CNN and RF methods. The proposed method could be a valuable reference for readers and may inspire researchers to develop even more effective methods for air pollution modeling. This research has important implications for air pollution research, data analysis, model estimation, and machine learning.

Noninvasive assessment of skin structure using hyperspectral images has been intensively studied in recent years. Due to the high computational cost of the classical methods, such as the inverse Monte Carlo (IMC), much research has been done with the aim of using machine learning (ML) methods to reduce the time required for estimating parameters. This study aims to evaluate the accuracy and the estimation speed of the ML methods for this purpose and compare them to the traditionally used inverse adding-doubling (IAD) algorithm. We trained three models – an artificial neural network (ANN), a 1D convolutional neural network (CNN), and a random forests (RF) model – to predict seven skin parameters. The models were trained on simulated data computed using the adding-doubling algorithm. To improve predictive performance, we introduced a stacked dynamic weighting (SDW) model combining the predictions of all three individually trained models. SDW model was trained by using only a handful of real-world spectra on top of the ANN, CNN and RF models that were trained using simulated data. Models were evaluated based on the estimated parameters’ mean absolute error (MAE), considering the surface inclination angle and comparing skin spectra with spectra fitted by the IAD algorithm. On simulated data, the lowest MAE was achieved by the RF model (0.0030), while the SDW model achieved the lowest MAE on in vivo measured spectra (0.0113). The shortest time to estimate parameters for a single spectrum was 93.05 μs. Results suggest that ML algorithms can produce accurate estimates of human skin optical parameters in near real-time.

Excessive withdrawal of groundwater for agricultural irrigation can cause seawater intrusion into coastal aquifers. Such a case will in turn results in deterioration of irrigation water quality. Determination of irrigation water quality with traditional methods is a time-consuming and costly process. However, machine learning algorithms can be useful tools for modeling and estimating groundwater quality used for irrigation water purposes. In this study, TDS, PS, SAR, and Cl parameters of groundwater were estimated with models based on EC and pH variables. For this purpose, prediction performances of two different deep learning methods (convolutional neural network (CNN) and deep neural network (DNN)) and two different classical machine learning (Random Forest (RF) and extreme gradient boosting (XGBoost)) methods were compared. In addition, predictive uncertainty of the models was determined by quantile regression (QR) analysis. Performance criteria and results of uncertainty analysis revealed that CNN (in testing phase, NSE = 0.95 for TDS, NSE = 0.96 for PS, NSE = 0.67 for SAR and NSE = 0.93 for CI) and DNN (in testing phase, NSE = 0.91 for TDS, NSE = 0.91 for PS, NSE = 0.57 for SAR and NSE = 0.94 for Cl) models had quite a close performance in estimation of TDS, PS, SAR, and Cl parameters and higher than the other two classical machine learning methods. As a result, the CNN model can be considered the best performing model in estimating all quality parameters due to the highest NSE and lowest RMSE values. In addition, the Taylor diagram showed that the values estimated using the CNN model had the highest correlation with the measured data. It was determined that the model with the lowest uncertainty based on the PICP statistics was DNN, followed by the CNN model. However, the CNN model has predicted outliers more accurately. Present findings proved that deep learning models could offer efficient tools for predicting irrigation water quality parameters.

The main goal of this research was to accurately diagnose diseases in pepper plants using convolutional neural network (CNN) models and visual parameters features. Thus, we aimed to build a computing system that can utilise visual cues to identify plant diseases with almost 100% reliability. A total of 2,200 images were trained and tested on a dataset, which was divided into healthy leaves and those with bacterial ailments. The images were preprocessed, resized to 300x300 pixels, and then segmented to extract the relevant traits from them. This CNN model, in a word, is accurate because it detects diseases with 99.65% accuracy. If we want to speak of a high accuracy rate, that's the CNN model. It learns to identify the intricate patterns and features in pepper plants that are associated with diseases. CNN can make revolutionary changes to the identification of plant diseases and help prevent crop damage through timely treatments. Automatic identification of the disease makes life easier for both agriculturists and scientists. The results underscore the importance of image preprocessing, segmentation, and feature extraction in improving the ability of CNN models to locate diseases. Additionally, the performance of the CNN model was compared with other machine learning models such as KNN, Decision Tree, ANN, and Random Forest. It was found that the developed CNN model can predict the response with greater accuracy than the others. Modern high-tech precision farming systems might utilise proposed CNN models for plant disease diagnosis.

The Random Forests (RF) algorithm is a machine learning method that has recently been demonstrated as a viable technique for data-driven predictive modeling of mineral prospectivity, and thus, it is instructive to further examine its usefulness in this particular field. A case study was carried out using data from Catanduanes Island (Philippines) to investigate further (a) if RF modeling can be used for data-driven modeling of mineral prospectivity in areas with few (i.e., <20) mineral occurrences and (b) if RF modeling can handle predictor variables with missing values. We found that RF modeling outperforms evidential belief (EB) modeling of prospectivity for hydrothermal Au–Cu deposits in Catanduanes Island, where 17 hydrothermal Au–Cu prospects are known to exist. Moreover, just like EB modeling, RF modeling allows analysis of the spatial relationships between known prospects and individual layers of predictor data. Furthermore, RF modeling can handle missing values in predictor data through an RF-based imputation technique whereas in EB modeling, missing values are simply represented by maximum uncertainty. Therefore, the RF algorithm is a potentially useful method for data-driven predictive modeling of mineral prospectivity in regions with few (i.e., <20) occurrences of mineral deposits of the type sought. However, further testing of the method in other regions with few mineral occurrences is warranted to fully determine its usefulness in data-driven predictive modeling of mineral prospectivity.

Random forest predictive modeling of mineral prospectivity with small number of prospects and data with missing values in Abra (Philippines)

BackgroundSurgical site infection (SSI) is one of the most common types of health care–associated infections. It increases mortality, prolongs hospital length of stay, and raises health care costs. Many institutions developed risk assessment models for SSI to help surgeons preoperatively identify high-risk patients and guide clinical intervention. However, most of these models had low accuracies.ObjectiveWe aimed to provide a solution in the form of an Artificial intelligence–based Multimodal Risk Assessment Model for Surgical site infection (AMRAMS) for inpatients undergoing operations, using routinely collected clinical data. We internally and externally validated the discriminations of the models, which combined various machine learning and natural language processing techniques, and compared them with the National Nosocomial Infections Surveillance (NNIS) risk index.MethodsWe retrieved inpatient records between January 1, 2014, and June 30, 2019, from the electronic medical record (EMR) system of Rui Jin Hospital, Luwan Branch, Shanghai, China. We used data from before July 1, 2018, as the development set for internal validation and the remaining data as the test set for external validation. We included patient demographics, preoperative lab results, and free-text preoperative notes as our features. We used word-embedding techniques to encode text information, and we trained the LASSO (least absolute shrinkage and selection operator) model, random forest model, gradient boosting decision tree (GBDT) model, convolutional neural network (CNN) model, and self-attention network model using the combined data. Surgeons manually scored the NNIS risk index values.ResultsFor internal bootstrapping validation, CNN yielded the highest mean area under the receiver operating characteristic curve (AUROC) of 0.889 (95% CI 0.886-0.892), and the paired-sample t test revealed statistically significant advantages as compared with other models (P<.001). The self-attention network yielded the second-highest mean AUROC of 0.882 (95% CI 0.878-0.886), but the AUROC was only numerically higher than the AUROC of the third-best model, GBDT with text embeddings (mean AUROC 0.881, 95% CI 0.878-0.884, P=.47). The AUROCs of LASSO, random forest, and GBDT models using text embeddings were statistically higher than the AUROCs of models not using text embeddings (P<.001). For external validation, the self-attention network yielded the highest AUROC of 0.879. CNN was the second-best model (AUROC 0.878), and GBDT with text embeddings was the third-best model (AUROC 0.872). The NNIS risk index scored by surgeons had an AUROC of 0.651.ConclusionsOur AMRAMS based on EMR data and deep learning methods—CNN and self-attention network—had significant advantages in terms of accuracy compared with other conventional machine learning methods and the NNIS risk index. Moreover, the semantic embeddings of preoperative notes improved the model performance further. Our models could replace the NNIS risk index to provide personalized guidance for the preoperative intervention of SSIs. Through this case, we offered an easy-to-implement solution for building multimodal RAMs for other similar scenarios.

Particulate phosphorus (PP) is a major contributor to lakes developing eutrophic. It also serves as one of the most significant sources of phosphorus for primary productivity, serving as a possible source of soluble reactive phosphorus, and contributing a sizable amount of the total phosphorus (TP), so monitoring the spatial and temporal variability of PP is crucial for understanding eutrophication in water bodies. This study aims to propose an algorithm to accurately predict the PP concentration and to assess the performance of the model. Considering Chaohu Lake as a case study, we proposed a random forest algorithm based on the convolutional neural network (CNN‐RF) to investigate the spatial and temporal patterns of PP concentration in the lake using HJ‐2 hyperspectral satellite images. The applicability of backpropagation (BP) neural network, random forest (RF), convolutional neural network (CNN), and CNN‐RF models for remote sensing inversion of PP concentration is assessed through model comparison. The results indicate that the CNN‐RF model has the best prediction performance, with a coefficient of determination (R2) of 0.91, a residual prediction deviation (RPD) of 3.42, a root mean square error (RMSE) of 0.0155 mg/L, and a percentage bias (PBIAS) of 5.01%, which is better than the BP model by 42%, 106%, 51%, and 81%, the RF model by 25%, 77%, 44%, and 59%, and the CNN model by 9.6%, 42%, 30%, and 49%, respectively, and 59% better than the CNN model. The inversion findings suggest that the PP of Chaohu Lake is substantially high in summer and autumn but comparatively low in spring and winter. The spatial distribution demonstrates an arrangement of highest in the western part of the lake, second in the eastern region, and lowest in the central, with obvious spatial and temporal distribution characteristics, and the trend is closely related to the quarterly mean air temperature and mean precipitation. This indicates that based on the differences in phosphorescence scattering signals of different morphologies in water bodies, the use of hyperspectral remote sensing and the CNN‐RF model can effectively extract PP spatiotemporal information, strengthen the learning capability of multiscale characteristics, and contribute to the improvement of the precision of estimating PP concentration, which could provide an innovative approach for determining the degree of eutrophication of lake water bodies.

Predictive modelling of mineral prospectivity using satellite remote sensing and machine learning algorithms

Understanding the learning mechanism of convolutional neural networks in spectral analysis

To predict the potential geographic distribution of Oncomelania hupensis in Yunnan Province using random forest (RF) and maximum entropy (MaxEnt) models, so as to provide insights into O. hupensis surveillance and control in Yunnan Province. The O. hupensis snail survey data in Yunnan Province from 2015 to 2016 were collected and converted into O. hupensis snail distribution site data. Data of 22 environmental variables in Yunnan Province were collected, including twelve climate variables (annual potential evapotranspiration, annual mean ground surface temperature, annual precipitation, annual mean air pressure, annual mean relative humidity, annual sunshine duration, annual mean air temperature, annual mean wind speed, ≥ 0 ℃ annual accumulated temperature, ≥ 10 ℃ annual accumulated temperature, aridity and index of moisture), eight geographical variables (normalized difference vegetation index, landform type, land use type, altitude, soil type, soil textureclay content, soil texture-sand content and soil texture-silt content) and two population and economic variables (gross domestic product and population). Variables were screened with Pearson correlation test and variance inflation factor (VIF) test. The RF and MaxEnt models and the ensemble model were created using the biomod2 package of the software R 4.2.1, and the potential distribution of O. hupensis snails after 2016 was predicted in Yunnan Province. The predictive effects of models were evaluated through cross-validation and independent tests, and the area under the receiver operating characteristic curve (AUC), true skill statistics (TSS) and Kappa statistics were used for model evaluation. In addition, the importance of environmental variables was analyzed, the contribution of environmental variables output by the models with AUC values of > 0.950 and TSS values of > 0.850 were selected for normalization processing, and the importance percentage of environmental variables was obtained to analyze the importance of environmental variables. Data of 148 O. hupensis snail distribution sites and 15 environmental variables were included in training sets of RF and MaxEnt models, and both RF and MaxEnt models had high predictive performance, with both mean AUC values of > 0.900 and all mean TSS values and Kappa values of > 0.800, and significant differences in the AUC (t = 19.862, P < 0.05), TSS (t = 10.140, P < 0.05) and Kappa values (t = 10.237, P < 0.05) between two models. The AUC, TSS and Kappa values of the ensemble model were 0.996, 0.954 and 0.920, respectively. Independent data verification showed that the AUC, TSS and Kappa values of the RF model and the ensemble model were all 1, which still showed high performance in unknown data modeling, and the MaxEnt model showed poor performance, with TSS and Kappa values of 0 for 24%(24/100) of the modeling results. The modeling results of 79 RF models, 38 MaxEnt models and their ensemble models with AUC values of > 0.950 and TSS values of > 0.850 were included in the evaluation of importance of environmental variables. The importance of annual sunshine duration (SSD) was 32.989%, 37.847% and 46.315% in the RF model, the MaxEnt model and their ensemble model, while the importance of annual mean relative humidity (RHU) was 30.947%, 15.921% and 28.121%, respectively. Important environment variables were concentrated in modeling results of the RF model, dispersed in modeling results of the MaxEnt model, and most concentrated in modeling results of the ensemble model. The potential distribution of O. hupensis snails after 2016 was predicted to be relatively concentrated in Yunnan Province by the RF model and relatively large by the MaxEnt model, and the distribution of O. hupensis snails predicted by the ensemble model was mostly the joint distribution of O. hupensis snails predicted by RF and MaxEnt models. Both RF and MaxEnt models are effective to predict the potential distribution of O. hupensis snails in Yunnan Province, which facilitates targeted O. hupensis snail control.

We aimed to establish a rapid and precise method for identifying and quantifying multi-species vegetable oil (corn oil, olive oil (OLO), soybean oil, and sunflower oil (SUO)) adulterations in camellia oil (CAO), using soft ionization by chemical reaction in transfer–high-resolution mass spectrometry (SICRIT-HRMS) and machine learning methods. The results showed that SICRIT-HRMS could effectively characterize the volatile profiles of pure and adulterated CAO samples, including binary, ternary, quaternary, and quinary adulteration systems. The low m/z region (especially 100–300) exhibited importance to oil classification in multiple feature-selection methods. For qualitative detection, binary classification models based on convolutional neural networks (CNN), Random Forest (RF), and gradient boosting trees (GBT) algorithms showed high accuracies (98.70–100.00%) for identifying CAO adulteration under no dimensionality reduction (NON), principal component analysis (PCA), and uniform manifold approximation and projection (UMAP) strategies. The RF algorithm exhibited relatively high accuracy (96.25–99.45%) in multiclass classification. Moreover, the five models, including CNN, RF, support vector machines (SVM), logistic regression (LR), and GBT, exhibited different performances in distinguishing pure and adulterated CAO. Among 1093 blind oil samples, under NON, PCA, and UMAP: 10, 5, and 67 samples were misclassified by CNN model; 6, 7, and 41 samples were misclassified by RF model; 8, 9, and 82 samples were misclassified by SVM model; 17, 18, and 78 samples were misclassified by LR model; 7, 9, and 43 samples were misclassified by GBT model. For quantitative prediction, the PCA-CNN model performed optimally in predicting adulteration levels in CAO, especially with respect to OLO and SUO, exhibiting a high coefficient of determination for calibration (, 0.9664–0.9974) and coefficient of determination for prediction (, 0.9599–0.9963) values, low root mean square error of calibration (RMSEC, 0.9–5.3%) and root mean square error of prediction (RMSEP, 1.1–5.8%) values, and RPD (5.0–16.3) values greater than 3.0. These results indicate that SICRIT-HRMS combined with machine learning can rapidly and accurately identify and quantify multi-species vegetable oil adulterations in CAO, which provides a reference for developing non-targeted and high-throughput detection methods in edible oil authenticity.