Agglomerative Clustering Algorithm Research Articles

<b>Towards Predicting Recurrence Risk of Differentiated Thyroid Cancer with a Hybrid Machine Learning Model</b>

Thyroid cancers (TC) often recur. This paper tests a novel unsupervised and supervised hybrid machine learning model to predict the recurrence risk (RR) and its score (RRS) in a population of "differentiated thyroid cancer" (DTC) cases as the prognostic measure. The DTC data (383 × 13) are collected from the UCI library. The population is grouped into "high risk of recurrence" (HRR) and "no high risk of recurrence" (NHRR) using the agglomerative clustering algorithm (ACA). Prior, the dataset is log-normalized [0,1], column-wise as a preprocessing step. Log-normalized values of the predictors, their corresponding coefficients, and the constant/intercept are used to construct a multiple linear regression to compute the RRS. Further, RRS values are normalized [0,1] using a log-sigmoidal function and termed "RRS_norm". RRS_norms closer to the average RRS_norms of HRR and NHRR determine the predicted group. The model’s performance is measured with a confusion matrix, and RRS_norm results are matched with the RR labeled within the dataset. The result shows that ACA can correctly cluster the dataset into HRR and NHRR by 63.4%. Based on the coefficient values, predictors such as "Age", "Gender", "Smoking", "History of smoking", "History of Radiotherapy", "Adenopathy", and "Tumor staging" which comprise 53.84% of the total number of predictors show a positive correlation with "recurrence". However, while matching the RRS_norms with the actual RRs, a 21.68% mismatch is observed, which mandates investigations with other DTC datasets. Received: 26 September 2024 | Revised: 18 November 2024 | Accepted: 29 November 2024 Conflicts of Interest The author declares that he has no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/joebeachcapital/differentiated-thyroid-cancer-recurrence, reference number [14]. Author Contribution Statement Subhagata Chattopadhyay: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.

Text Clustering with Pre-Trained Models: BERT, RoBERTa, ALBERT and MPNet

Text clustering is the process of collecting similar sentences in texts of variable size in the same group. Text clustering methods are an important area used for data analysis and for extracting information from these data. Many studies have been carried out in this area using different approaches and methods. In this study, the results of BERT (Bidirectional encoder representations from transformers), ROBERTa (Robustly optimized BERT pretraining approach), ALBERT(A lite BERT) and MPNet (Masked and permuted pre-training for language understanding) models, which are pre-trained models, and the TF-IDF (Term frequency-inverse document frequency) method, which is traditional statistical feature extraction, were compared while performing text representation. After the feature extraction stage, performance measurements were made by clustering with K-means, BIRCH (Balanced iterative reducing and clustering using hierarchies), Agglomerative clustering and Mini-batch K-means algorithms. When the measurements are evaluated, it has been reported that the pre-trained models give superior clustering results compared to the classical models.

An Automatic Movement Monitoring Method for Group-Housed Pigs.

Continuous movement monitoring helps quickly identify pig abnormalities, enabling immediate action to enhance pig welfare. However, continuous and precise monitoring of daily pig movement on farms remains challenging. We present an approach to automatically and precisely monitor the movement of group-housed pigs. The instance segmentation model YOLOv8m-seg was applied to detect the presence of pigs. We then applied a spatial moment algorithm to quantitatively summarize each detected pig's contour as a corresponding center point. The agglomerative clustering (AC) algorithm was subsequently used to gather the pig center points of a single frame into one point representing the group-housed pigs' position, and the movement volume was obtained by calculating the displacements of the clustered group-housed pigs' center points of consecutive frames. We employed the method to monitor the movement of group-housed pigs from April to July 2023; more than 1500 h of top-down pig videos were recorded by a surveillance camera. The F1 scores of the trained YOLOv8m-seg model during training were greater than 90% across most confidence levels, and the model achieved an mAP50-95 of 0.96. The AC algorithm performs with an average extraction time of less than 1 millisecond; this method can run efficiently on commodity hardware.

Exploring Customer Segmentation in E-Commerce using RFM Analysis with Clustering Techniques

The proliferation of big data and the growth of e-commerce have intensified the challenges associated with extracting actionable data for personalised recommendations and decision-making. With data-driven marketing strategies, understanding and predicting customer behaviour has become paramount for maintaining competitive advantage. This study leverages business analytics tools, focusing on Recency, Frequency, and Monetary (RFM) Analysis, alongside K-Means and Hierarchical (Agglomerative) Clustering algorithms, to segment customer transactional data. Data normalisation, a critical step for accurate clustering, was performed using log transformation and the Power Transformer technique with the Yeo-Johnson parameter, the latter proving more effective for handling both positively and negatively skewed data, enhancing data normalisation and suitability for analysis. This study reveals that RFM Analysis with Hierarchical Clustering outperforms K-Means Clustering, achieving a Silhouette Score of 0.47 and a Calinski–Harabasz Index of 3787.1, indicating a more accurate identification of customer segments. RFM Analysis alone generated eight clusters, while integrating RFM Analysis with both Hierarchical Clustering and K-Means generated three similar-sized clusters with interchanged labels. These metrics highlight the proficiency of Hierarchical Clustering in identifying unique customer segments and customising marketing strategies. The findings indicate that the RFM-Hierarchical Clustering approach enhances segmentation precision and facilitates more refined and effective marketing strategies.

Revisiting historical trends in the Eastern Boundary Upwelling Systems with a machine learning method

Eastern boundary upwelling systems (EBUS) host very productive marine ecosystems that provide services to many surrounding countries. The impact of global warming on their functioning is debated due to limited long-term observations, climate model uncertainties, and significant natural variability. This study utilizes the usefulness of a machine learning technique to document long-term variability in upwelling systems from 1993 to 2019, focusing on high-frequency synoptic upwelling events. Because the latter are modulated by the general atmospheric and oceanic circulation, it is hypothesized that changes in their statistics can reflect fluctuations and provide insights into the long-term variability of EBUS. A two-step approach using Self-Organizing Maps (SOM) and Hierarchical Agglomerative Clustering (HAC) algorithms was employed. These algorithms were applied to sets of upwelling events to characterize signatures in sea-level pressure, meridional wind, shortwave radiation, sea-surface temperature (SST), and Ekman pumping based on dominant spatial patterns. Results indicated that the dominant spatial pattern, accounting for 56%-75% of total variance, representing the seasonal pattern, due to the marked seasonality in along-shore wind activity. Findings showed that, except for the Canary-Iberian region, upwelling events have become longer in spring and more intense in summer. Southern Hemisphere systems (Humboldt and Benguela) had a higher occurrence of upwelling events in summer (up to 0.022 Events/km²) compared to spring (<0.016 Events/km²), contrasting with Northern Hemisphere systems (<0.012 Events/km²). Furthermore, long-term changes in dominant spatial patterns were examined by dividing the time period in approximately two equally periods, to compare past changes (1993-2006) with relatively new changes (2007-2019), revealing shifts in key variables. These included poleward shifts in subtropical high-pressure systems (SHPS), increased upwelling-favorable winds, and SST drops towards higher latitudes. The Humboldt Current System (HumCS) exhibited a distinctive spring-to-summer pattern, with mid-latitude meridional wind weakening and concurrent SST decreases. Finally, a comparison of upwelling centers within EBUS, focusing on changes in pressure and temperature gradients, meridional wind, mixed-layer depth, zonal Ekman transport, and Ekman pumping, found no evidence supporting Bakun’s hypothesis. Temporal changes in these metrics varied within and across EBUS, suggesting differential impacts and responses in different locations.

Adapting differential molecular representation with hierarchical prompts for multi-label property prediction.

Accurate prediction of molecular properties is crucial in drug discovery. Traditional methods often overlook that real-world molecules typically exhibit multiple property labels with complex correlations. To this end, we propose a novel framework, HiPM, which stands for Hierarchical Prompted Molecular representation learning framework. HiPM leverages task-aware prompts to enhance the differential expression of tasks in molecular representations and mitigate negative transfer caused by conflicts in individual task information. Our framework comprises two core components: the Molecular Representation Encoder (MRE) and the Task-Aware Prompter (TAP). MRE employs a hierarchical message-passing network architecture to capture molecular features at both the atom and motif levels. Meanwhile, TAP utilizes agglomerative hierarchical clustering algorithm to construct a prompt tree that reflects task affinity and distinctiveness, enabling the model to consider multi-granular correlation information among tasks, thereby effectively handling the complexity of multi-label property prediction. Extensive experiments demonstrate that HiPM achieves state-of-the-art performance across various multi-label datasets, offering a novel perspective on multi-label molecular representation learning.

ALGORITHME DE CLUSTERING AGGLOMÉRATIF BASÉ SUR MOJO AVEC STRATÉGIE EDMT POUR LES RÉSEAUX DE CAPTEURS SANS FIL À GRANDE ÉCHELLE

The wireless sensor network (WSN) is a distributed sensor network that monitors and stores environmental data wirelessly by connecting dispersed sensor nodes. Since wireless sensor nodes rely on batteries for energy, energy consumption, and limitations are considered fundamental problems. A novel multi-objective jellyfish optimization based on energy, degree, distance, mobility, and time parameters (MOJO-ED2MT) technique has been proposed to overcome these challenges. Three phases are involved in the proposed method: selection of cluster heads, compression of data, and routing of the data. In this first phase, a fuzzy agglomerative clustering algorithm is employed to choose an optimal dual cluster head from inter-cluster and intra-cluster. In the second phase, a neighborhood indexing sequence (NIS) algorithm can compress the number of bits in the data before it is transmitted. In the third phase, jellyfish optimization selects the shortest path based on multi-objective parameters. The simulation analysis and result statistics show that the suggested MOJO-E MT approach performs better than the state-of-the-art algorithms across various performance measures. The proposed MOJO-E MT framework achieves 11.5, 15.4 %, and 17.99 % more network lifetime than EOR-iABC, C3HA, and ML-AEFA algorithms.

Multilayer Annotation Strategy AnnoSePS: Disentangling the Intricate Structure of Selenium-Containing Polysaccharides Based on Preferential Fragmentation Patterns.

Precision mapping of selenium at structural and position levels poses significant challenges in selenium-containing polysaccharide identification. Due to the absence of reference spectra, database-centric approaches are still limited in the discovery of selenium binding sites and distinction among different isomeric structures. A multilayer annotation strategy, AnnoSePS, is proposed for achieving the identification of seleno-substituent and the unbiased profiling of polysaccharides. Applying Snoop-triggered multiple reaction monitoring (Snoop-MRM) identified multidimensional monosaccharides in selenium-containing polysaccharides. Galactose, galacturonic acid, and glucose were the predominant monosaccharides with a molar ratio of 25.19, 19.45, and 11.72, respectively. Selenium present in seleno-rhamnose was found to substitute the hydroxyl group located at C-1 positions through the formation of a Se-H bond. Ions C6H9O3Se-, C6H7O3Se-, C5H5O3Se-, C4H5O2Se-, C3H5O2Se-, C2H3O2Se-, and CHOSe- were defined as the characteristic fragments of seleno-rhamnose. The agglomerative hierarchical clustering algorithm is applied to group spectra from each run based on the characteristic information. Preferential fragmentation patterns in mass spectrometry are revealed by training a probabilistic model. A list of candidate oligosaccharides is generated by step-by-step browsing through the transition pairs for all reference spectra and applying the transitions (addition, insertion, removal, and substitution) to reference structures. Combining time course analyses revealed the linkage composition of selenium-containing oligosaccharides. Glycosidic linkages were annotated based on a synthesis-driven approach. T-Galactose (16.67 ± 5.23%) and T-Galacturonic acid (11.54 ± 4.66%) were the predominant linkage residues. As the database-independent mapping strategy, AnnoSePS makes it possible to comprehensively interrogate spectral data and dissect the fine structure of selenium-containing polysaccharides.

Evaluating visitors’ experience in museum: Comparing artificial intelligence and multi-partitioned analysis

Analysing visitors' behaviour in a museum or in a cultural site is a crucial element to manage spaces and artworks arrangement as well as improving the visit experience. This paper presents the preliminary results of the ARTEMISIA project, exploiting Artificial Intelligence (AI) techniques to study, design and develop a methodology to interpret visitors' behaviour within a museum context, namely the Museum of Rome in Palazzo Braschi (Rome, Italy). The aim is to combine literature on users' experience (UX) analysis with experimental data coming from the visitor anonymous tracking out of motion sensors (users' stand-still positions, viewpoint direction, movements), merging approaches of different research domains. Through the use of agglomerative hierarchical clustering algorithms, four categories of visitors were identified, then associated to user profiles emerged by UX evaluations. Such analysis may lead to new forms of visitors profiling and to the development of a new generation of customised applications in public and private contexts. Identifying and predicting users’ patterns with respect to museum halls arrangement may also be useful to suggest improvement in the museum spaces and exhibitions (new indications, updated storytelling or changes in thematic configuration).© 2023 Elsevier Ltd. All rights reserved.

Business Models of Ukrainian Banks: the Impact of the Revolution of Dignity, the COVID–19 Pandemic, and Russia’s Military Aggression

The purpose of the article is to identify key banking activity models in Ukraine and show how they changed under the influence of the Revolution of Dignity, the COVID–19 pandemic, and the imposition of martial law. The proposed method uses cluster analysis of the main indicators for banking activity (concerning assets, liabilities and income) based on Ward’s agglomerative hierarchical clustering algorithm and the Tau index in the NbClust package as the criterion for evaluating the optimal number of clusters. The research covers all Ukrainian banks, spanning the period 2013–2022. In 2014 (after the Revolution of Dignity), the actions of the National Bank of Ukraine (NBU) had a positive impact on the models of Ukrainian banks: there was no opaque model of banks and banking models became more transparent and more resilient to financial shocks. Between 2021 and 2022, five banking models were identified: universal banks, wholesale funding banks, corporate‑investment banks, retail banks, and commission banks. The negative impact of COVID–19 on the Ukrainian economy was reflected in all banking models by the following: (1) a significant increase in the role of securities in assets, which was caused by a decrease in lending due to an increase in their riskiness; (2) the use of central bank funds to liabilities management, which was evidence of a deterioration in financing conditions in the deposit market. The following main changes in Ukraine’s banking system at the end of 2022 (during the war) were identified: (1) a decrease in the number of banks that mainly use funds from the NBU to support their activities; (2) the closure of only four banks by the NBU (including two with majority shareholders from Russia) during the year; (3) an increase in the share of the non‑government debt securities portfolio in banks’ assets; (4) declines in ROE and ROA for all banking models during the war, and three banking models became unprofitable; (5) the significantly higher share of non‑performing loans in the credit portfolio for all banking models, especially for the retail banking model. The value‑added of this research is the analysis of how banking models in Ukraine have changed at the micro level (reflected in the financial statements) inter alia in wartime.

Weighted double Q-learning based eco-driving control for intelligent connected plug-in hybrid electric vehicle platoon with incorporation of driving style recognition

Driving style poses significant impacts on eco-driving performance of vehicles, especially for those with hybrid powertrains. By incorporating driving style recognition, an efficient velocity planning and energy management strategy is developed for a platoon of intelligent connected plug-in hybrid electric vehicles (PHEVs). Firstly, a high-fidelity driving style classification and recognition model is established based on the agglomerative hierarchical clustering algorithm, and then the support vector machine algorithm is employed to recognize the driving style. Next, a multi-objective velocity planning problem is formulated with the consideration of the fuel economy, driving adaptability and comfort, following performance, and driving safety optimization as the optimization objective, and is then solved based on orthogonal collocation direct transcription and the interior-point methods. Finally, an energy management strategy incorporating model predictive control and weighted double Q-learning algorithm is built. The simulation results demonstrated that the velocity planning algorithm incorporating driving styles achieved preferable adaptability and comfort. The proposed strategy can achieve up to most 98.88 % energy economy of the stochastic dynamic programming for the following PHEVs, and reduce the overall fuel consumption by 27.39 % for the platoon, comparing to that without driving style incorporation.

Cluster-based oversampling with area extraction from representative points for class imbalance learning

Class imbalance learning is challenging in various domains where training datasets exhibit disproportionate samples in a specific class. Resampling methods have been used to adjust the class distribution, but they often have limitations for small disjunct minority subsets. This paper introduces AROSS, an adaptive cluster-based oversampling approach that addresses these limitations. AROSS utilizes an optimized agglomerative clustering algorithm with the Cophenetic Correlation Coefficient and the Bayesian Information Criterion to identify representative areas of the minority class. Safe and half-safe areas are obtained using an incremental k-Nearest Neighbor strategy, and oversampling is performed with a truncated hyperspherical Gaussian distribution. Experimental evaluations on 70 binary datasets demonstrate the effectiveness of AROSS in improving class imbalance learning performance, making it a promising solution for mitigating class imbalance challenges, especially for small disjunct minority subsets.

Robotic Multi-Boll Cotton Harvester System Integration and Performance Evaluation

Several studies on robotic cotton harvesters have designed their end-effectors and harvesting algorithms based on the approach of harvesting a single cotton boll at a time. These robotic cotton harvesting systems often have slow harvesting times per boll due to limited computational speed and the extended time taken by actuators to approach and retract for picking individual cotton bolls. This study modified the design of the previous version of the end-effector with the aim of improving the picking ratio and picking time per boll. This study designed and fabricated a pullback reel to pull the cotton plants backward while the rover harvested and moved down the row. Additionally, a YOLOv4 cotton detection model and hierarchical agglomerative clustering algorithm were implemented to detect cotton bolls and cluster them. A harvesting algorithm was then developed to harvest the cotton bolls in clusters. The modified end-effector, pullback reel, vacuum conveying system, cotton detection model, clustering algorithm, and straight-line path planning algorithm were integrated into a small red rover, and both lab and field tests were conducted. In lab tests, the robot achieved a picking ratio of 57.1% with an average picking time of 2.5 s per boll. In field tests, picking ratio was 56.0%, and it took an average of 3.0 s per boll. Although there was no improvement in the lab setting over the previous design, the robot’s field performance was significantly better, with a 16% higher picking ratio and a 46% reduction in picking time per boll compared to the previous end-effector version tested in 2022.

Identifying definite patterns of unmet needs in patients with multiple sclerosis using unsupervised machine learning

IntroductionPeople with multiple sclerosis (PwMS) exhibit a spectrum of needs that extend beyond solely disease-related determinants. Investigating unmet needs from the patient perspective may address daily difficulties and optimize care. Our aim was to identify patterns of unmet needs among PwMS and their determinants.MethodsWe conducted a cross-sectional multicentre study. Data were collected through an anonymous, self-administered online form. To cluster PwMS according to their main unmet needs, we performed agglomerative hierarchical clustering algorithm. Principal component analysis (PCA) was applied to visualize cluster distribution. Pairwise comparisons were used to evaluate demographics and clinical distribution among clusters.ResultsOut of 1764 mailed questionnaires, we received 690 responses. Access to primary care was the main contributor to the overall unmet need burden. Four patterns were identified: cluster C1, ‘information-seekers with few unmet needs’; cluster C2, ‘high unmet needs’; cluster C3, ‘socially and assistance-dependent’; cluster C4, ‘self-sufficient with few unmet needs’. PCA identified two main components in determining the patterns: the ‘public sphere’ (access to information and care) and the ‘private sphere’ (need for assistance and social life). Older age, lower education, longer disease duration and higher disability characterized clusters with more unmet needs in the private sphere. However, demographic and clinical factors failed in explaining the four identified patterns.ConclusionOur study identified four unmet need patterns among PwMS, emphasizing the importance of personalized care. While clinical and demographic factors provide some insight, additional variables warrant further investigation to fully understand unmet needs in PwMS.

Functional-Coefficient Quantile Regression for Panel Data with Latent Group Structure

This article considers estimating functional-coefficient models in panel quantile regression with individual effects, allowing the cross-sectional and temporal dependence for large panel observations. A latent group structure is imposed on the heterogeneous quantile regression models so that the number of nonparametric functional coefficients to be estimated can be reduced considerably. With the preliminary local linear quantile estimates of the subject-specific functional coefficients, a classic agglomerative clustering algorithm is used to estimate the unknown group structure and an easy-to-implement ratio criterion is proposed to determine the group number. The estimated group number and structure are shown to be consistent. Furthermore, a post-grouping local linear smoothing method is introduced to estimate the group-specific functional coefficients, and the relevant asymptotic normal distribution theory is derived with a normalization rate comparable to that in the literature. The developed methodologies and theory are verified through a simulation study and showcased with an application to house price data from U.K. local authority districts, which reveals different homogeneity structures at different quantile levels.

Mineralogical and elemental composition of the Middle Miocene coal seams from the Alpu coalfield (Eskişehir, Central Türkiye): Insights from syngenetic zeolite formation

This study focuses on determining mineralogical and elemental compositions of coal seams (to the upwards, D, C, B, A, and S0) within seven coal exploration wells from the Alpu coalfield (Eskişehir, Central Türkiye). Furthermore, the special goal of the study is a comparative analysis of the relations between the elements by using agglomerative hierarchical clustering algorithm with different linkage methods as well as different similarity measures. Clay minerals and quartz are commonly detected as abundant to dominant phases, while natural zeolite formations were detected in the studied seams C, B, A, and S0. The SEM-EDX data shows that clinoptilolites in zeolite minerals were observed within the organic matter, while crystalline and non-crystalline analcime minerals along with syngenetic authigenic rhomboid K-feldspars were only detected in the seam A from one studied well. The existence of some micron-sized minerals such as apatite, monazite, and Ti-oxides within the smectite matrix and the measurable amount of Ti in smectite imply that alteration of epiclastic and contemporaneous volcanic inputs was developed under weak acidic to neutral conditions during peat accumulation. The lack of natural zeolite and carbonate minerals in the seam D could be an indicator of weak acidic to neutral conditions and semi-closed hydrogeological conditions. Nevertheless, the alkalinity of mire water, water table, and hydrogeological regime seem to be variable during the accumulation of precursor peats of seams C, B, A, and S0. In turn, alteration of volcanic inputs was observed under neutral to weak alkaline conditions and semi-closed to closed hydrogeological regime. Hence, syngenetic authigenic, micron-sized clinoptilolites were formed. Moreover, the existence of authigenic rhomboid K-feldspars and syngenetic authigenic analcimes in certain exploration well could suggest local increases on dissolved Na+ concentrations, alkalinity, and water table. Except for volcanogenic origin for minerals, accessory micron-sized minerals, like chromite, pentlandite, and allanite grains presumably originated from clastic influxes of ophiolitic rocks in the basement into palaeomires. The variations in mire water chemistry and clastic-influx source area could also control the elemental enrichments in the studied seams. Epiclastic and contemporaneous volcanic inputs into palaeomires seem to control enrichments of Li, B, Sc, and Ti in coal samples, while clastic influx from ophiolitic rocks into palaeomires caused to enrichments of Cr, V and Ni. Furthermore, the liberated Ba, Sr, and As ions from the alteration of epiclastic and contemporaneous volcanic inputs are absorbed by syngenetic zeolite minerals, while anoxic conditions in the palaeomires resulted in precipitation of Sr-barite and As-bearing pyrite grains during peat accumulation and/or early diagenetic stages. Overall, the differences in water chemistry of mire water, epiclastic and contemporaneous volcanic inputs, and clastic influx from the adjacent areas also caused several elemental enrichments and variations in mineralogical compositions of the Middle Miocene coal seams in the Alpu coalfield.

Precipitation Vertical Structure Characterization: A Feature-Based Approach

Abstract The three-dimensional (3D) structure of precipitation systems is highly dependent on hydrometeor formation processes and microphysics. This study aims to characterize distinct vertical profiles of precipitation regimes by relying on the availability of a high-quality, spatially dense radar network and its capability to observe the 3D structure of the storms. A deep-learning-based framework, coupled with unsupervised clustering methods, is developed to identify types of precipitation structures irrespective of their physical properties. A 6-month period of 3D reflectivity profiles from the Multi-Radar Multi-Sensor (MRMS) network is used to identify different regimes and investigate their properties with respect to the underlying environmental conditions. Dominant features retrieved from radar reflectivity profiles using convolutional neural-network-based autoencoders are employed to identify similar-looking vertical structures using coupled k-means and agglomerative clustering algorithms. The k-means method identifies distinct groups, while the agglomerative clustering visualizes intercluster relationships. The framework identifies 18 clusters that can be broadly combined into five groups of varied echo-top heights. The 18 clusters demonstrate variability with respect to structural features and precipitation rate/type, implying that profiles in each group belong to a physically different precipitation regime. An independent analysis of the regime properties is conducted by matching the MRMS reflectivity profiles with environmental parameters derived from the High-Resolution Rapid Refresh model forecasts. The distribution of the environmental variables confirms cluster-specific feature properties, confirming the physics-based regime separation across the clusters and their dependence on the vertical structure. The identified precipitation regimes can assist in developing physics-guided retrievals and studying precipitation regimes. Significance Statement This study proposes a systematic model to identify precipitation profiles of distinct vertical structures and evaluate their dependence on environmental conditions. The model was developed using ground-based radar observations; however, there is potential to extend this model to reflectivity profiles from both ground- and satellite-based sensors. In addition, the identified precipitation regime clusters could be a proxy for the vertical structure of precipitation systems and assist in determining the structural variability within traditional precipitation type classification (e.g., convective versus stratiform). Moreover, identifying the precipitation regimes could also be used to improve satellite-based precipitation retrievals. Finally, a better understanding of precipitation structure would also help improve the initialization of climate models.

TeraHAC: Hierarchical Agglomerative Clustering of Trillion-Edge Graphs

We introduce TeraHAC, a (1+ε)-approximate hierarchical agglomerative clustering (HAC) algorithm which scales to trillion-edge graphs. Our algorithm is based on a new approach to computing (1+ε)-approximate HAC, which is a novel combination of the nearest-neighbor chain algorithm and the notion of (1+ε)-approximate HAC. Our approach allows us to partition the graph among multiple machines and make significant progress in computing the clustering within each partition before any communication with other partitions is needed. We evaluate TeraHAC on a number of real-world and synthetic graphs of up to 8 trillion edges. We show that TeraHAC requires over 100x fewer rounds compared to previously known approaches for computing HAC. It is up to 8.3x faster than SCC, the state-of-the-art distributed algorithm for hierarchical clustering, while achieving 1.16x higher quality. In fact, TeraHAC essentially retains the quality of the celebrated HAC algorithm while significantly improving the running time.

Spatial pattern and driving factors of cropland ecosystem services in a major grain-producing region: A production-living-ecology perspective

Exploring the spatial patterns and driving factors of cropland ecosystem services and the production-living-ecology are essential to implementing spatial zoning management and optimization, especially in major grain-producing regions. In this study, we first developed an evaluation system for evaluating the production-living-ecology of cropland ecosystem services in the Jianghan Plain (JHP). We then revealed the spatial patterns of various cropland ecosystem services and integrated production-living-ecology index based on multi-source data. Additionally, we employed the regionalization with dynamically constrained agglomerative clustering and partitioning algorithm (REDCAP) to delineate agricultural function zones. Finally, we used redundancy analysis to reveal the driving factors of each functional zone. Results indicated that (1) high habitat quality, soil retention, and carbon storage services exhibited spatial similarity in 2020, with the spatial pattern of high in the west and low in the east, water conservation services showed an opposite distribution pattern, while culture and recreation services of high value were concentrated in the northern part, and high grain production was observed in the southern region. Overall, the production-living-ecology index displayed a north–south spatial distribution, with higher values in the north and lower values in the south. (2) based on the spatial zoning results of the production-living-ecology index, the six functional zones were identified, including zones of the production cropland, ecology cropland, ecology and living cropland, production and ecology cropland, production and living cropland and production-living-ecology cropland. (3) the production-living-ecology index and various ecosystem services were mainly influenced by population density, night light and evaporation in the JHP, and trade-offs were observed between the production function, and other functions as well as the production-living-ecology index in each functional zone of cropland ecosystems. The findings hold significant implications for the sustainable development of major grain-producing regions.

Type of Refugee Accommodation and Health of Residents: A Cross-Sectional, Population-Based Cluster Analysis in South-West Germany.

Objectives: Few studies have assessed whether refugees' health is associated with accommodation characteristics. We aimed to devise a typology of refugee accommodation based on variables on the accommodation and its physical context before assessing its association with health in multivariate analyses. Methods: We performed a cluster analysis based on a hierarchal, agglomerative clustering algorithm using Euclidean Distance and Ward's method. We analysed accommodation clusters based on number of inhabitants, degree of housing deterioration, urbanity of location (urban/rural distinction), and remoteness (walking distance to shops, medical or administrative services). In total, we analysed health and accommodation data of 412 refugees and asylum seekers from 58 different accommodation facilities in the federal state of Baden-Württemberg in the south-west of Germany. Results: Accommodations with a moderate occupation, lowest levels of deterioration, and a central urban location showed the best health outcomes in terms of subjective general health status, depression, and generalized anxiety disorder (GAD). Associations were strongest for GAD and weakest for depression. Conclusion: Our findings inform policymakers on layout and location of refugee collective accommodation centres.