Reply on RC1

<p>Avalanche forecasting implies predicting current and future snow instability in time and space. In Switzerland, avalanche bulletins are issued daily during the winter season to warn the public about the avalanche hazard, described by region with one of five danger levels. Assessing avalanche danger is by large a data-driven, yet experience-based decision-making process. It involves analysing a multitude of data diverse in scale – time and space, and concluding by expert judgment on the avalanche scenario. Numerous statistical models were developed in the past, but rarely applied due to limited usefulness in operational forecasting. Modern machine learning techniques open up new possibilities for developing support tools for operational avalanche forecasting. With this aim, we developed a data-driven approach based on the supervised Random Forest (RF) classifier to automatically predict the danger level for dry-snow avalanche conditions in the Swiss Alps. A large database of more than 20 years of meteorological data and modelled snow stratigraphy data obtained with the numerical snow cover model SNOWPACK were used to train the RF algorithm. We optimized the model and selected the best set of input features that combine meteorological variables and features extracted from the simulated profiles, resampled at the same daily resolution as the forecasts. Our target variable was the regional danger level forecast in the public bulletin. We evaluated the predictive performance of the RF model with an independent test set with data of two winter seasons (2018-2019 and 2019-2020). The test set accuracy was 72 %, which is slightly lower than the accuracy estimate of the public forecasts (about 76 %). Given this uncertainty in our target variable, we trained an optimized RF model on a subset containing so-called verified avalanche danger levels. The test set accuracy then increased to 80 %. During the winter season 2020-2021, both RF models were tested in operational setting and automatically predicted a ‘nowcast’ and a ‘forecast’ in real-time.  In parallel, we also tested a deep recurrent neural network model, which used a 7-days time series with 3-hours time resolution as input and also predicted the avalanche danger level. We present a comparison of the performance of the three models. This is one of the first times that a data-driven approach is tested in real-time as a feasible tool for operational avalanche forecasting.</p>

Modeled snow stratigraphy and instability data are a promising source of information for avalanche forecasting. While instability indices describing the mechanical processes of dry-snow avalanche release have been implemented into snow cover models, there exists no readily applicable method that combines these metrics to predict snow instability. We therefore trained a random forest (RF) classification model to assess snow instability from snow stratigraphy simulated with SNOWPACK. To do so, we manually compared 742 observed snow profiles with their simulated counterparts to select the simulated weak layer corresponding to the observed rutschblock failure layer. We then used the observed stability test result and an estimate of the local avalanche danger to construct a binary target variable (stable vs. unstable) and considered 34 features describing the simulated weak layer and the overlying slab as potential explanatory variables. The final RF classifier aggregates six of these features into the output probability Punstable, corresponding to the mean vote of an ensemble of 400 classification trees. Although the training data only consisted of 146 manual profiles labeled as either unstable or stable, the model classified profiles from an independent validation data set with high reliability (accuracy: 88 %, precision: 96 %, recall: 85 %) using manually predefined weak layers. Model performance was even higher (accuracy: 93 %, precision: 96 %, recall: 92 %), when the weakest layers of the profiles were identified with the maximum of Punstable. Finally, we compared model predictions to observed avalanche activity in the region of Davos for five winter seasons. In 73 % of the days, our model correctly discriminated between avalanche days and non-avalanche days. Overall, the results of our RF classification are very encouraging, suggesting it could be of great value for operational avalanche forecasting.

Modeled snow stratigraphy and instability data are a promising source of information for avalanche forecasting. While instability indices describing the mechanical processes of dry-snow avalanche release have been implemented into snow cover models, there exists no readily applicable method that combines these metrics to predict snow instability. We therefore trained a random forest (RF) classification model to assess snow instability from snow stratigraphy simulated with SNOWPACK. To do so, we manually compared 742 observed snow profiles with their simulated counterparts to select the simulated weak layer corresponding to the observed rutschblock failure layer. We then used the observed stability test result and an estimate of the local avalanche danger to construct a binary target variable (stable vs. unstable) and considered 34 features describing the simulated weak layer and the overlying slab as potential explanatory variables. The final RF classifier aggregates six of these features into the output probability Punstable, corresponding to the mean vote of an ensemble of 400 classification trees. Although the training data only consisted of 146 manual profiles labeled as either unstable or stable, the model classified profiles from an independent validation data set with high reliability (accuracy: 88 %, precision: 96 %, recall: 85 %) using manually predefined weak layers. Model performance was even higher (accuracy: 93 %, precision: 96 %, recall: 92 %), when the weakest layers of the profiles were identified with the maximum of Punstable. Finally, we compared model predictions to observed avalanche activity in the region of Davos for five winter seasons. In 73 % of the days, our model correctly discriminated between avalanche days and non-avalanche days. Overall, the results of our RF classification are very encouraging, suggesting it could be of great value for operational avalanche forecasting.

Abstract. Forecasting avalanche danger at a regional scale is a largely data-driven yet also experience-based decision-making process by human experts. In the case of public avalanche forecasts, this assessment process terminates in an expert judgment concerning summarizing avalanche conditions by using one of five danger levels. This strong simplification of the continuous, multi-dimensional nature of avalanche hazard allows for efficient communication but inevitably leads to a loss of information when summarizing the severity of avalanche hazard. Intending to overcome the discrepancy between determining the final target output in higher resolution while maintaining the well-established standard of assessing and communicating avalanche hazard using the avalanche danger scale, avalanche forecasters at the national avalanche warning service in Switzerland used an approach that combines absolute and relative judgments. First, forecasters make an absolute judgment using the five-level danger scale. In a second step, a relative judgment is made by specifying a sub-level describing the avalanche conditions relative to the chosen danger level. This approach takes into account the human ability to reliably estimate only a certain number of classes. Here, we analyze these (yet unpublished) sub-levels, comparing them with data representing the three contributing factors of avalanche hazard: snowpack stability, the frequency distribution of snowpack stability, and avalanche size. We analyze both data used in operational avalanche forecasting and data independent of the forecast, going back 5 years. Using a sequential analysis, we first establish which data are suitable and in which part of the danger scale they belong by comparing their distributions at consecutive danger levels. In a second step, integrating these findings, we compare the frequency of locations with poor snowpack stability and the number and size of avalanches with the forecast sub-level. Overall, we find good agreement: a higher sub-level is generally related to more locations with poor snowpack stability and more avalanches of larger size. These results suggest that on average avalanche forecasters can make avalanche danger assessments with higher resolution than the five-level danger scale. Our findings are specific to the current forecast set-up in Switzerland. However, we believe that avalanche warning services making a hazard assessment using a similar temporal and spatial scale as currently used in Switzerland should also be able to refine their assessments if (1) relevant data are sufficiently available in time and space and (2) a similar approach combining absolute and relative judgment is used. The sub-levels show a rank-order correlation with data related to the three contributing factors of avalanche hazard. Hence, they increase the predictive value of the forecast, opening the discussion on how this information could be provided to forecast users.

Considering the form diversity of tree species composition in the Bagong Mountain National Forest Park of China, we mapped tree species utilizing Machine Learning Algorithms (support vector machines (SVM) and random forest (RF) classifiers) based on the OHS-2 hyperspectral satellite image by different datasets which combined spectral information and hyperspectral-derived vegetation indices (VIs) for improving tree species classification and explored the best performance of them. To verify the improvement, the results of physically-based spectral classifiers (spectral angle mapper (SAM) and maximum likelihood (ML) classifiers) were applied to compare with the results of machine learning algorithms. The results indicated an overall accuracy of 94.01%, 96.08%, 82.9% and 79.3% for SVM, RF, SAM and ML classifiers of the best performance using different datasets. Highest accuracies resulted from two machine learning algorithms classifiers; SVM and RF compared to SAM and ML classifiers. Although SVM outperformed RF when using all hyperspectral bands and VIs, the overall accuracy of the RF classifier is higher when compared to the SVM classifier using VIs combined selected features. Meanwhile, the RF classifier performed better than SVM after removing the redundancy of spectral data in training samples. Moreover, the machine learning algorithms successfully classified a small number of tree species (Cedrus deodara and Pterocarya stenoptera C. DC.) in the study area, but the physical spectroscopy-based method failed to classify these species. Such integration strategy improved the effectiveness of enhancing the accuracy of tree species classification and mapping their distribution on broad spatial and temporal scales using machine learning algorithms and hyperspectral imagery.

Abstract. In recent years, physics-based snowpack models combined with machine-learning techniques have gained momentum in public avalanche forecasting. When integrated with spatial interpolation methods, these approaches enable fully model-driven predictions of snowpack stability or avalanche danger at any location. This raises a key question: are such spatially detailed model predictions sufficiently accurate for operational use? We evaluated the performance of three spatially interpolated model-driven forecasts of snowpack stability and avalanche danger in Switzerland over three winters. As a benchmark, we used the official public avalanche-danger forecasts, specifically focusing on the forecast danger level that includes the sub-levels. We assessed the ability of both model and human forecasts to discriminate between reference distributions of conditions – typically not associated with avalanche activity – and actual avalanche events that were either naturally released or triggered by humans by calculating event ratios as proxies for release probability. Our results show that event ratios clearly increased with higher predicted avalanche probability, lower snowpack stability, or higher forecast sub-level. Overall, both model predictions and human forecasts showed a comparable ability to discriminate between reference and event conditions, with the event ratio increasing exponentially with increasing model-predicted probabilities or forecast sub-levels. However, the human forecasts – which incorporate model output – achieved a small but statistically significant advantage in discriminatory skill. This indicates that while the models alone have not yet reached the full discriminatory power of human forecasters, their performance is already approaching operational usefulness in a setup such as that used in Switzerland. As model quality is expected to improve further in coming years, it is essential to ensure optimal integration into the operational forecasting workflow to realize the full potential of model-based support. Further research should explore how to implement this effectively, how to integrate real-time avalanche occurrence data into model prediction pipelines, and how to validate increasingly high-resolution avalanche forecasts.

Abstract. Random Forest is a supervised learning algorithm developed from decision trees with the application of boostrap aggregating (bagging). This method grows trees from decision trees to produce a forest or the best model called the random forest model. Tree growth is done with randomly selected data with returns through the bagging process. Random forest is considered to provide better performance results for diabetes data among other supervised learning methods, because random forest and has the lowest error rate compared to other methods. Random forest is also an important technique for medical data classification, especially for diagnosing diabetics. In this study, classification was carried out using Pima Indian Diabetes data, which is an American tribe that lives in Arizona and Mexico. Classification analysis was carried out using an algorithm to see the level of accuracy in random forest classification on Pima Indian diabetes data. The results show that the accuracy value of random forest classification is 74.78%, this value is in the accuracy category at the fair classification level. In this random forest classification, there are three main variables that become importance variables, namely glucose then BMI, and age.
 Abstract. Random Forest is a supervised learning algorithm developed from decision trees with the application of boostrap aggregating (bagging). This method grows trees from decision trees to produce a forest or the best model called the random forest model. Tree growth is done with randomly selected data with returns through the bagging process. Random forest is considered to provide better performance results for diabetes data among other supervised learning methods, because random forest and has the lowest error rate compared to other methods. Random forest is also an important technique for medical data classification, especially for diagnosing diabetics. In this study, classification was carried out using Pima Indian Diabetes data, which is an American tribe that lives in Arizona and Mexico. Classification analysis was carried out using an algorithm to see the level of accuracy in random forest classification on Pima Indian diabetes data. The results show that the accuracy value of random forest classification is 74.78%, this value is in the accuracy category at the fair classification level. In this random forest classification, there are three main variables that become importance variables, namely glucose then BMI, and age.

Abstract. Recreational activities in snow-covered mountainous terrain in the backcountry account for the vast majority of avalanche accidents. Studies analyzing avalanche risk mostly rely on accident statistics without considering exposure (or the elements at risk), i.e., how many, when and where people are recreating, as data on recreational activity in the winter mountains are scarce. To fill this gap, we explored volunteered geographic information on two social media mountaineering websites – bergportal.ch and camptocamp.org. Based on these data, we present a spatiotemporal pattern of winter backcountry touring activity in the Swiss Alps and compare this with accident statistics. Geographically, activity was concentrated in Alpine regions relatively close to the main Swiss population centers in the west and north. In contrast, accidents occurred equally often in the less-frequented inner-alpine regions. Weekends, weather and avalanche conditions influenced the number of recreationists, while the odds to be involved in a severe avalanche accident did not depend on weekends or weather conditions. However, the likelihood of being involved in an accident increased with increasing avalanche danger level, but also with a more unfavorable snowpack containing persistent weak layers (also referred to as an old snow problem). In fact, the most critical situation for backcountry recreationists and professionals occurred on days and in regions when both the avalanche danger was critical and when the snowpack contained persistent weak layers. The frequently occurring geographical pattern of a more unfavorable snowpack structure also explains the relatively high proportion of accidents in the less-frequented inner-alpine regions. These results have practical implications: avalanche forecasters should clearly communicate the avalanche danger and the avalanche problem to the backcountry user, particularly if persistent weak layers are of concern. Professionals and recreationists, on the other hand, require the expertise to adjust the planning of a tour and their backcountry travel behavior depending on the avalanche danger and the avalanche problem.

The random forest (RF) classifier, which is a combination of tree predictors, is one of the most powerful classification algorithms that has been recently applied for fault detection and diagnosis (FDD) of industrial processes. However, RF is still suffering from some limitations such as the noncorrelation between variables. These limitations are due to the direct use of variables measured at nodes and therefore the only use of static information from the process data. Thus, this article proposes two enhanced RF classifiers, namely the Euclidean distance based reduced kernel RF (RK-RF <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ED</sub> ) and K-means clustering based reduced kernel RF (RK-RF <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Kmeans</sub> ), for FDD. Based on the kernel principal component analysis, the proposed classifiers consist of two main stages: feature extraction and selection, and fault classification. In the first stage, the number of observations in the training data set is reduced using two methods: the first method consists of using the Euclidean distance as dissimilarity metric so that only one measurement is kept in case of redundancy between samples. The second method aims at reducing the amount of the training data based on the K-means clustering technique. Once the characteristics of the process are extracted, the most sensitive features are selected. During the second phase, the selected features are fed to an RF classifier. An emulated grid-connected PV system is used to validate the performance of the proposed RK-RF <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ED</sub> and RK-RF <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Kmeans</sub> classifiers. The presented results confirm the high classification accuracy of the developed techniques with low computation time.

Introduction: Chronic Obstructive Pulmonary Disease (COPD), is a condition caused by damage to theairways or other parts of the lung that blocks airflow and makes it hard to breathe [1]. COPD is the thirdleading cause of death worldwide, and the seventh leading cause of poor health worldwide [2]. Studieshave shown that 20–86% of people with COPD worldwide may be undiagnosed [3]. As there is currentlyno cure for COPD, early detection is the best option. Current Machine Learning (ML) models focus onusing chest images (CT or X-ray scans) to detect COPD; however, the scanning process can be unsafe forpatients with COPD [4].Methods: This study utilized an open data set containing various physical tests of 100 patients withCOPD. To train this model, the Random Forest (RF) classifier was used. The accuracy was then plottedon a graph.Results: The Random Forest classifier was able to achieve an accuracy of 92.41% with a perfect recallvalue of 1.00. This recall value indicates that the Random Forest classifier was able to correctly diagnoseall of the patients with COPD in this dataset.Discussion: This study developed a novel ML model that can accurately provide a diagnosis for COPD.Further studies could use this code with a larger dataset to obtain a higher accuracy. White individualshave been reported to have a higher prevalence of COPD [5]. By developing a dataset that accounts forrace, we will be able to obtain a more accurate diagnosis.

The objective of this study was to develop a machine learning (ML) application to determine the optimal dosage of sodium hypochlorite (NaOCl) to curtail corrosion and odor by H2 S in the headworks of a water resource recovery facility (WRRF) without overly consuming volatile fatty acids (VFAs) that are essential for the enhanced biological phosphorus removal. Given the highly diverse datasets available, three subproblems were formulated, and three cascaded ML modules were developed accordingly. The final ML models, chosen based on performance, were able to predict various targeted variables. More specifically, in Module 1, a recurrent neural network (RNN) was designed to predict wastewater characteristics. In Module 2, a random forest (RF) classifier and a support vector machine (SVM) classifier were built with the information from Module 1 along with other datasets to predict the concentrations of VFAs and H2 S, respectively. Finally, in Module 3, with the information obtained from Module 2, another RF classifier was developed to predict NaOCl dosage to reduce H2 S but keeping VFAs within the target range. These efforts are relevant and informative for WRRFs that are considering developing Intelligent Water Systems to predict the wastewater characteristics to make operational improvements. PRACTITIONER POINTS: A recurrent neural network (RNN) using long short-term memory (LSTM) successfully predicted influent wastewater parameters. A support vector machine classifier predicted hydrogen sulfide (H2 S) with 97.6% accuracy. The concentration of VFAs, an important parameter in EBPR, was predicted using a random forest classifier with 93.4% accuracy. The optimal NaOCl dosage for H2 S control can be predicted with a random forest classifier using H2 S, VFAs, and flow.

<strong class="journal-contentHeaderColor">Abstract.</strong> Accurately predicting the location, timing and size of natural snow avalanches is crucial for local and regional decision-makers, but remains one of the major challenges in avalanche forecasting. So far, forecasts are generally made by human experts, interpreting a variety of data, and drawing on their knowledge and experience. Using avalanche data from the Swiss Alps and one-dimensional physics-based snowpack simulations, we developed a model predicting the probability of dry-snow avalanches occurring in the vicinity of automated weather stations based on the output of a recently developed instability model. This new avalanche day predictor was compared to benchmark models related to the amount of new snow. Evaluation on an independent data set demonstrated the importance of snow stratigraphy for natural avalanche release, as the avalanche day predictor outperformed the benchmark model based on the three-day sum of new snow (F1 scores: 0.71 and 0.65, respectively). The averaged predictions of both models resulted in the best performance (F1 score: 0.75). In a second step, we derived functions describing the probability for certain avalanche size classes. Using the 24-hour new snow height as proxy of avalanche failure depth yielded the best estimator of typical (median) observed avalanche size, while the depth of the deepest weak layer, detected using the instability model, provided the better indicator regarding the largest observed avalanche size. Validation of the avalanche size estimator on an independent data set of avalanche observations confirmed these findings. Furthermore, comparing the predictions of the avalanche day predictors and avalanche size estimators with a 21-year data set of re-analysed regional avalanche danger levels showed increasing probabilities for natural avalanches and increasing avalanche size with increasing danger level. We conclude that these models may be valuable tools to support forecasting the occurrence of natural dry-snow avalanches.

The Local Climate Zone (LCZ) classification is already widely used in urban heat island and other climate studies. The current classification method does not incorporate crucial urban auxiliary GIS data on building height and imperviousness that could significantly improve urban-type LCZ classification utility as well as accuracy. This study utilized a hybrid GIS- and remote sensing imagery-based framework to systematically compare and evaluate different machine and deep learning methods. The Convolution Neural Network (CNN) classifier outperforms in terms of accuracy, but it requires multi-pixel input, which reduces the output’s spatial resolution and creates a tradeoff between accuracy and spatial resolution. The Random Forest (RF) classifier performs best among the single-pixel classifiers. This study also shows that incorporating building height dataset improves the accuracy of the high- and mid-rise classes in the RF classifiers, whereas an imperviousness dataset improves the low-rise classes. The single-pass forward permutation test reveals that both auxiliary datasets dominate the classification accuracy in the RF classifier, while near-infrared and thermal infrared are the dominating features in the CNN classifier. These findings show that the conventional LCZ classification framework used in the World Urban Database and Access Portal Tools (WUDAPT) can be improved by adopting building height and imperviousness information. This framework can be easily applied to different cities to generate LCZ maps for urban models.

Lung cancer is one of the three deadliest diseases in the world and has rapidly developed. Based on this, researchers conducted research to predict the factors that influence lung cancer. One method to identify this is using data mining methods and classification techniques. Researchers used several popular algorithms in classification to make comparisons of the most accurate algorithms for lung cancer classification. The algorithms used include K-Nearest Neighbor, Random Forest Classifier, Logistic Regression, Linear SVM, Naïve Bayes, Decision Tree, Random Forest, Gradient Boosting, Kernel SVM, and MLPClassifier. The researcher used this algorithm because, in the research that the researcher found on the Kaggle platform, he examined the comparison of the algorithm using the breast cancer dataset. In previous studies, their researchers used SVM, which obtained an accuracy of 96.47%, Neural Networks of 97.06%, and Naïve Bayes with an accuracy of 91.18% to study breast cancer. The difference from previous research is that this study uses several existing algorithms in Machine Learning such as K-Nearest Neighbor, Random Forest Classifier, Logistic Regression, Linear SVM, Naïve Bayes, Decision Tree, Random Forest, Gradient Boosting, Kernel SVM, and MLPClassifier. In addition, this research was conducted to see whether the results of the accuracy of the algorithm that the researchers carried out using the lung cancer dataset had different results. The results of this study found that the more accurate algorithms were Random Forest and Gradient Boosting with an accuracy value of 100%, whereas in previous studies, it was the same. Still, Gradient Boosting had a higher accuracy value than Random Forest. Then, based on the data used in this study, the most influencing factors in predicting a diagnosis of lung cancer are obesity and coughing up blood. The results of this study found that the more accurate algorithms were Random Forest and Gradient Boosting with an accuracy value of 100%, whereas in previous studies, it was the same. Still, Gradient Boosting had a higher accuracy value than Random Forest. Then, based on the data used in this study, the most influencing factors in predicting a diagnosis of lung cancer are obesity and coughing up blood. The results of this study found that the more accurate algorithms were Random Forest and Gradient Boosting with an accuracy value of 100%, whereas in previous studies, it was the same. Still, Gradient Boosting had a higher accuracy value than Random Forest. Then, based on the data used in this study, the most influencing factors in predicting a diagnosis of lung cancer are obesity and coughing up blood.

Diabetes claims the lives of thousands each year, and many individuals remain oblivious to their condition until it reaches a critical stage. This study presents a data mining-based approach aimed at enhancing the early detection and prediction of diabetes, utilizing data from the Pima Indian Diabetes dataset. Despite the adaptability of fuzzy C-Means for various data types, the ultimate outcome of the clustering process hinges on the initial placement of cluster centers. Additionally, precision in data clustering is crucial; it can furnish either extensive, well-grouped data for the random forest or limited data, constraining its efficacy. Our principal objective was to enhance the accuracy of fuzzy C-means clustering and the random forest. To boost the model's performance, we incorporated PCA, fuzzy c-means, and the Random Forest approach. Various algorithmic combinations were employed, and the results unequivocally demonstrate that our model surpasses the original outcomes of the Pima Indian Diabetes Dataset in terms of accuracy. The diabetic prediction model achieved a remarkable accuracy of 97.40\% through the utilization of PCA, logistic regression, and K-Means. However, when employing PCA in conjunction with fuzzy C-means and random forests, an even higher accuracy of 98.96\% was attained. Empirical evidence confirms that the implementation of PCA significantly enhanced the accuracy of both the fuzzy C-means clustering approach and the random forest classifier, deviating from previous findings. To improve the model's performance, we used PCA, fuzzy c-means, and the Random Forest approach. Different algorithm combinations were used, and the results clearly show that our model outperforms the original Pima Indian Diabetes Dataset outcomes in terms of accuracy. The diabetic prediction model was improved to 97.40% accuracy using PCA, logistic regression, and K -Means. Using PCA with fuzzy C-means and random forests, however, we achieved an accuracy of 98.96%. Based on empirical evidence, it has been demonstrated that the implementation of PCA improved the accuracy of the fuzzy C-means clustering approach and the random forest classifier. These findings differ from previous findings.