Multistage Model Research Articles

Integrated neuronetwork modeling of EEG for diagnostic disorders of brain activity

The article discusses the structured multistage modeling of EEG by means of applied mathematics to customize the input parameter space for neural network prediction. Also, the approaches and models are analyzed for their accuracy in determining the relevant signal features and adaptability to real data. The activity of potentials during brain activity is a biological process that depends on many factors and hides a space of parameters, the search for which and their definition can open us up to a new perspective on the nature and activity of the human brain. A rational way of obtaining data on brain activity is a non-invasive electroencephalogram, which registers the potential difference on the electrodes relative to the base. For further processing of the received data, it is necessary to remove noise and artifacts from them. In this work, a stan-dard algorithm of frequency filtering and filtering from noise caused by the power grid is used. After processing the data, an overview of mathematical models is offered, which with a certain degree of accuracy try to simulate the behavior of the signal or the peak moments of certain features. Added to this is the use of the LSTM model to predict the further behavior of the signal with the preliminary introduction of chaos into the model due to the modified activation function (Gaussian noise) and the input modeling of the weights.

The Formation of Cavansite and Pentagonite in the Wagholi Quarries, Pune, India

Formation conditions of dimorphic minerals cavansite and pentagonite were previously based on theoretical assumptions. In doing so, the associations with other minerals, especially zeolites, that actually occur in nature, were disregarded or incorrectly taken into account. As a result, formation conditions were assumed that are not consistent with those for the associated minerals in the Deccan Volcanic Province. This relates in particular to overestimated high pressure and temperature values, as well as chronological processes of alteration. Long-term field studies and evaluations of numerous samples led to the conclusion that cavansite and pentagonite formed under temperature (approx. 120 °C to 200 °C) and pressure (0.01–0.03 GPa) conditions that are relevant for the associated low-temperature zeolites. Integration of geological and petrographic conditions, as well as crystallization sequences enabled the presentation of a multi-stage mineralization model. It is also explained that, contrary to the original assumption, characteristic pentagonite fivelings are not formed from five, but from six individuals.

Multistage deep learning methods for automating radiographic sharp score prediction in rheumatoid arthritis

The Sharp-van der Heijde score (SvH) is crucial for assessing joint damage in rheumatoid arthritis (RA) through radiographic images. However, manual scoring is time-consuming and subject to variability. This study proposes a multistage deep learning model to predict the Overall Sharp Score (OSS) from hand X-ray images. The framework involves four stages: image preprocessing, hand segmentation with UNet, joint identification via YOLOv7, and OSS prediction utilizing a custom Vision Transformer (ViT). Evaluation metrics included Intersection over Union (IoU), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Huber loss, and Intraclass Correlation Coefficient (ICC). The model was trained using stratified group 3-fold cross-validation on a dataset of 679 patients and tested externally on 291 subjects. The joint identification model achieved 99% accuracy. The ViT model achieved the best OSS prediction for patients with Sharp scores < 50. It achieved a Huber loss of 4.9, an RMSE of 9.73, and an MAE of 5.35, demonstrating a strong correlation with expert scores (ICC = 0.702, P < 0.001). This study is the first to apply a ViT for OSS prediction in RA. It presents an efficient and automated alternative for overall damage assessment. This approach may reduce reliance on manual scoring.

Analyzing COVID-19 progression with Markov multistage models: insights from a Korean cohort

BackgroundUnderstanding the progression and recovery process of COVID-19 is crucial for guiding public health strategies and developing targeted interventions. This longitudinal cohort study aims to elucidate the dynamics of COVID-19 severity progression and evaluate the impact of underlying health conditions on these transitions, providing critical insights for more effective disease management.MethodsData from 4549 COVID-19 patients admitted to Seoul National University Boramae Medical Center between February 5th, 2020, and October 30th, 2021, were analyzed using a 5-state continuous-time Markov multistate model. The model estimated instantaneous transition rates between different levels of COVID-19 severity, predicted probabilities of state transitions, and determined hazard ratios associated with underlying comorbidities.ResultsThe analysis revealed that most patients stabilized in their initial state, with 72.2% of patients with moderate symptoms remaining moderate. Patients with hypertension had a 67.6% higher risk of progressing from moderate to severe, while those with diabetes had an 89.9% higher risk of deteriorating from severe to critical. Although transition rates to death were low early in hospitalization, these comorbidities significantly increased the likelihood of worsening conditions.ConclusionThis study highlights the utility of continuous-time Markov multistate models in assessing COVID-19 severity progression among hospitalized patients. The findings indicate that patients are more likely to recover than to experience worsening conditions. However, hypertension and diabetes significantly increase the risk of severe outcomes, underscoring the importance of managing these conditions in COVID-19 patients.

Ultra-short-term Forecasting Study of Power Load in Mega Steel Industry Based on Multi-stage Modeling

Background: In the large-scale steel industry, significant power load variability, especially during processes like steel smelting, poses challenges to power system safety. Although there is an abundance of research and patents related to load forecasting, studies and patents specifically addressing large industrial load forecasting are sparse. Hence, accurate ultra-short-term load forecasting becomes particularly crucial. Objective: This study proposes an innovative method for ultra-short-term load forecasting to improve prediction accuracy during peak periods and mitigate risks in high-load conditions. Methods: We introduce an LSTM-XGBoost model enhanced by a random forest network and an improved grey wolf optimization algorithm (IGWO) for feature selection and parameter optimization, respectively. Results: Compared to other advanced models, our method demonstrates superior performance across key indicators such as MAPE (1.93%), RMSE (220.81), and R2 coefficient (0.99), and the prediction error is lower during both peak and off-peak periods. For instance, the proposed model achieved a MAPE improvement of over 25% compared to traditional models. Validation with data from multiple time periods confirms the model's accuracy and robustness. Conclusion: The proposed forecasting method effectively tackles load fluctuations in the steel industry, supporting safe and economical power system operations. Future research will aim to further improve peak identification accuracy and enable continuous adaptive learning.

Exploring experimental models of prostate cancer in chemoprevention: Oxidative stress as a key pathway to translational research.

Prostate cancer (PCa) is the second most common cancer in men globally. Its growth is driven by oxidative stress associated with inflammation, aging, and environmental factors, including diet and lifestyle. These factors contribute to multiple stages of PCa progression, including progression to castration-resistant prostate cancer (CRPC). Therefore, oxidative stress represents an intriguing target for PCa chemoprevention and treatment. In vivo experimental models are crucial for understanding the mechanisms of PCa development, validating chemopreventive and therapeutic approaches, and translating preclinical results into clinical applications. We established a transgenic rat for adenocarcinoma of the prostate (TRAP) model, a transgenic rat that efficiently develops androgen-dependent adenocarcinoma, pathologically and biologically mimicking human PCa progression, to clarify the mechanisms of tumor progression, including the involvement of oxidative stress, and established a system for screening the chemopreventive effects of agents against PCa. Additionally, we derived a CRPC model from the TRAP model and developed a distant metastasis model, providing a comprehensive multistage rat model of prostate carcinogenesis. This review presents findings on the molecular mechanisms of PCa and the chemopreventive effects of natural compounds with antioxidant properties, such as polyphenols. We additionally described the potential for repositioning existing drugs with antiandrogenic activity for PCa chemoprevention.

Recurrent Flow Update Model Using Image Pyramid Structure for 4K Video Frame Interpolation

Video frame interpolation (VFI) is a task that generates intermediate frames from two consecutive frames. Previous studies have employed two main approaches to extract the necessary information from both frames: pixel-level synthesis and flow-based methods. However, when synthesizing high-resolution videos using VFI, each approach has its limitations. Pixel-level synthesis based on the transformer architecture requires high complexity to achieve 4K video results. In the case of flow-based methods, forward warping can produce holes where pixels are not allocated, while backward warping approaches struggle to obtain accurate backward flow. Additionally, there are challenges during the training stage; previous works have often generated suboptimal results by training multi-stage model architectures separately. To address these issues, we propose a Recurrent Flow Update (RFU) model trained in an end-to-end manner. We introduce a global flow update module that leverages global information to mitigate the weaknesses of forward flow and gradually correct errors. We demonstrate the effectiveness of our method through several ablation studies. Our approach achieves state-of-the-art performance not only on the XTest and Davis datasets, which have 4K resolution, but also on the SNU-FILM dataset, which features large motions at low resolution.

Analytical Multistage Thermal Resistance Model for NSFET Self-heating Effects

—As semiconductor technology nodes continue to scale down to 3 nm, the self-heating effect in Gate-All-Around Nanosheet Field-Effect Transistors (GAA-NSFETs) has become a significant concern. This issue arises due to the complex interaction between device dimensions, material properties, and thermal management, which can lead to performance degradation and reliability challenges in advanced transistor designs. This paper aims to investigate the self-heating phenomenon in three-stacked nanosheet FETs and to develop a novel thermal resistance model that accurately captures the thermal behavior of these devices. The goal is to create a reliable framework for analyzing and mitigating self-heating effects in nanosheet-based transistors. We employed TCAD and SPICE simulations to analyze the self-heating effect in nanosheet FETs. A new multi-stage thermal resistance model (TRM), incorporating both thermal resistance (Rₜₕ) and thermal capacitance (Cₜₕ), was developed within the Berkeley Short-channel IGFET Model-Common MultiGate (BSIM-CMG) framework. Model accuracy was ensured by fitting the simulated ID-VG curves to experimental data, followed by parameter extraction and calibration based on self-heating evaluations. The proposed multi-stage Rₜₕ model demonstrated strong agreement with the simulation results, providing an accurate representation of the thermal behavior in three-stacked nanosheet FETs. This model offers a robust tool for analyzing self-heating effects in advanced nanosheet devices and can be used to guide the design and optimization of future low-power, high-performance transistors.

A multi-scale multi-stage model for spray freezing of binary solutions

Spray freezing has found extensive applications in drying processes, drug delivery, and mine ventilation. The technique involves atomizing a liquefied material into a cold medium, where the resulting droplets solidify. This work presents a multi-scale model to study the spray freezing of binary mixtures. Specifically, the freezing model of a binary solution is coupled with a spray-droplet dynamics model. The spray freezing of an aqueous sucrose solution is analyzed using this framework, and parametric studies are conducted to examine the effects of droplet size, reactor length, ambient temperature, and liquid flowrate on the system performance. The results show that increasing the reactor length from 1.25 m to 4.25 m leads to a 79 % increase in the final concentration. Lowering the ambient temperature from −15 °C to −25° yields a 2.63-times bigger solid-to-liquid fraction. Additionally, increasing the flowrate from 0.0125 kg/s to 0.0625 kg/s results in a 323 % increase in heat rates.

Steam-Alternating CO2/Viscosity Reducer Huff and Puff for Improving Heavy Oil Recovery: A Case of Multi-Stage Series Sandpack Model with Expanded Sizes

Aiming at the challenges of rapid heat dissipation, limited swept efficiency, and a rapid water cut increase in steam huff and puff development in heavy oil reservoirs, an alternating steam and CO2/viscosity reducer huff and puff method for IOR was proposed. In this work, the effect of CO2 on the physical properties of heavy oil was evaluated, and the optimal concentration of viscosity reducer for synergistic interaction between CO2 and the viscosity reducer was determined. Next, novel huff and puff simulation experiments by three sandpack models of different sizes in series were analyzed. Then, the IOR difference between the pure steam huff and puff experiments and the steam-alternating CO2/viscosity reducer huff and puff were compared. Finally, the CO2 storage rate was obtained based on the principle of the conservation of matter. The results show that the optimal viscosity reducer concentration, 0.8 wt%, can achieve a 98.5% reduction after combining CO2. The steam-alternating CO2/viscosity reducer huff and puff reached about 45 cm at 80 °C in the fifth cycle due to the CO2/viscosity reducer effects. CO2/viscosity reducer huff and puff significantly reduces water cut during cold production, with an ultimate IOR 15.89% higher than pure steam huff and puff. The viscosity reducer alleviates heavy oil blockages, and CO2 decreases oil viscosity and enhances elastic repulsion energy. The highest CO2 storage rate of 76.8% occurs in the initial stage, declining to 15.2% by the sixth cycle, indicating carbon sequestration potential. These findings suggest that steam-alternating CO2/viscosity reducer huff and puff improves heavy oil reservoir development and provides theoretical guidance for optimizing steam huff and puff processes.

Enhance the Concrete Crack Classification Based on a Novel Multi-Stage YOLOV10-ViT Framework.

Early identification of concrete cracks and multi-class detection can help to avoid future deformation or collapse in concrete structures. Available traditional detection and methodologies require enormous effort and time. To overcome such difficulties, current vision-based deep learning models can effectively detect and classify various concrete cracks. This study introduces a novel multi-stage deep learning framework for crack detection and type classification. First, the recently developed YOLOV10 model is trained to detect possible defective regions in concrete images. After that, a modified vision transformer (ViT) model is trained to classify concrete images into three main types: normal, simple cracks, and multi-branched cracks. The evaluation process includes feeding concrete test images into the trained YOLOV10 model, identifying the possible defect regions, and finally delivering the detected regions into the trained ViT model, which decides the appropriate crack type of those detected regions. Experiments are conducted using the individual ViT model and the proposed multi-stage framework. To improve the generation ability, multi-source datasets of concrete structures are used. For the classification part, a concrete crack dataset consisting of 12,000 images of three classes is utilized, while for the detection part, a dataset composed of various materials from historical buildings containing 1116 concrete images with their corresponding bounding boxes, is utilized. Results prove that the proposed multi-stage model accurately classifies crack types with 90.67% precision, 90.03% recall, and 90.34% F1-score. The results also show that the proposed model outperforms the individual classification model by 10.9%, 19.99%, and 19.2% for precision, recall, and F1-score, respectively. The proposed multi-stage YOLOV10-ViT model can be integrated into the construction systems which are based on crack materials to obtain early warning of possible future deformation in concrete structures.

Relieving Overexposure in Information Diffusion Through a Budget Multi-stage Allocation

When information dissemination campaigns on Online Social Networking platforms are too aggressive, this can easily cause information overexposure. Overexposure can break through the psychological and physiological limits that the audience can tolerate, making it difficult for the audience to obtain reasonable cognitive concepts. For example, the overexposure of information referring to an object in a promotional campaign can lead to inflated expectations in individuals. Building on this, we introduce two indicators for individuals’ expectations and actual utility of the object and design a multi-stage triggered mechanism for seed individuals to explore the relieving overexposure problem in information diffusion. We build a multi-stage information diffusion model and characterize the evolution of individual expectations. We verify the hardness result of the relieving overexposure problem by budget multi-stage allocation, and the non-submodularity and non-monotonicity of the objective function. Addressing the non-monotonic and non-submodular set function, we provide a direct influence-oriented algorithm with a greedy approach. Extensive experiments are performed on four real networks to explore how model parameters and network properties affect the effects of multi-stage triggered strategies for seed individuals. Using the experiments, we found that the seed individual multi-stage incremental triggered strategy of dissemination campaign of information referring to an object shows better performance, and the lower the actual utility of the specific object, the more accurate the promotion strategy needs to be developed.

Development of multistage crop yield estimation model using machine learning and deep learning techniques.

In this research paper, machine learning techniques were applied to a multivariate meteorological time series data for estimating the wheat yield of five districts of Punjab. Wheat yield data and weather parameters over 34years were collected from the study area and the model was developed using stepwise multi-linear regression (SMLR), artificial neural network (ANN), support vector regression (SVR), random forest (RF) and deep neural network (DNN) techniques. Wheat yield estimation was done at the tillering, flowering, and grain-filling stage of the crop by considering weather variables from 46 to 4th, 46 to 8th, and 46 to 11th standard meteorological week. Weighted and unweighted Meteorological variables and yield data were used to train, test, and validate the models in R software. The evaluation results showed a consistent and promising performance of RF, SVR, and DNN models for all five districts with an overall MAPE and nRMSE value of less than 6% during validation at all three growth stages. These models exhibited outstanding performance during validation for the Faridkot, Ferozpur, and Gurdaspur districts. Based on accuracy parameters MAPE, RMSE, nRMSE, and percentage deviation, the RF model was found better followed by SVR and DNN models and, hence can be used for district-level wheat crop yield estimation at different crop growth stages.

A rolling horizon heuristic approach for a multi-stage stochastic waste collection problem

In this paper we present a multi-stage stochastic optimization model to solve an inventory routing problem for the collection of recyclable municipal waste. The objective is the maximization of the total expected profit of the waste collection company. The decisions are related to the selection of the bins to be visited and the corresponding routing plan in a predefined time horizon. Stochasticity in waste accumulation is modeled through scenario trees generated via conditional density estimation and dynamic stochastic approximation techniques. The proposed formulation is solved through a rolling horizon approach, providing a rigorous worst-case analysis on its performance. Extensive computational experiments are carried out on small- and large-sized instances based on real data provided by a large Portuguese waste collection company. The impact of stochasticity on waste generation is examined through stochastic measures, showing the importance of adopting a stochastic model over a deterministic formulation when addressing a waste collection problem. The performance of the rolling horizon approach is evaluated, demonstrating that this heuristic provides cost-effective solutions in short computational time. Managerial insights related to different geographical configurations of the instances and varying levels of uncertainty are finally discussed.

A comprehensive multi-stage decision-making model for supplier selection and order allocation approach in the digital economy

The increasingly serious environmental issues and fierce competition caused by globalization have brought pressure on supply chain managers who seek to allocate multiple purchase demands comprehensively, highlighting the significance of supplier assessment considering sustainability and technique. Moreover, many multi-criteria decision-making (MCDM) methods fail to quantify the risk preference of decision-makers (DMs) when conducting the supplier assessment process. Indeed, a hybrid supplier selection and order allocation model that integrates such requirements is yet to be proposed. Thus, this work develops a comprehensive decision-making model that constructs a deep learning model to forecast the potential demand and addresses the sustainable supplier selection based on cumulative prospect theory (CPT) and multi-material order allocation problem simultaneously. The proposed order allocation model is solved by the second generation of adaptive geometry estimation based many-objective evolutionary algorithm, with the technique of order preference similarity to the ideal solution used to filter out the best Pareto solution for DMs as the reference. Through implementing an illustrative case study of a leading Chinese engineering machinery manufacturer followed by a sensitivity analysis, the relatively strong applicability and scalability of the proposed model and methods are demonstrated. The results show that introducing Weibull distribution to estimate the theoretical obsolescence rate of historically sold accessories can result in higher demand prediction accuracy for consumable mechanical accessories. Integrating CPT into the MCDM framework allows us to evaluate suppliers more comprehensively by capturing the effect of DMs’ risk preferences and gain or loss sensitivity.

An integrated multistage ensemble machine learning model for fraudulent transaction detection

Fraudulent transactions continue to pose a concern for financial institutions and organizations, necessitating the development of effective detection tools. Identification and prevention of fraudulent transactions depend heavily on the detection of credit card fraud. Even though instances of credit card fraud are uncommon, they can nonetheless cause significant financial losses because of the high cost of fraudulent transactions. When fraud is discovered early on, investigators can act quickly to stop additional losses. But because the investigation process takes a while, there are only so many warnings that can be looked through in detail in a given day. Thus, a fraud detection model’s main goal is to minimize false alarms and missed fraud situations while producing accurate alerts. To improve fraud identification, we provide in this study an integrated multistage ensemble Machine Learning (IMEML) model that incorporates various multistage ensemble models intelligently, such as Ensemble Independent Classifier (EIC), Ensemble Bagging Classifier (EBC), and Ensemble ML Classifier (EMC). In order to overcome the problem of data imbalance, we use a number of methods-including Instant Hardness Threshold with EMC (IHT+EMC), Cluster Centroids (CC), and Randon Under Sampler (RUS)-that go beyond traditional methods. We run our studies on a 284,807-transaction credit card dataset that is made available to the public. The accuracy rates of 99.94%, 99.91%, 99.14%, 99.52%, and perfect 100% for accuracy, precision, recall, f1-score, and AUC score, respectively, are achieved by the suggested model, demonstrating remarkable performance scores. For real-world fraud detection applications, the EIBMC model sets a new benchmark for identifying fraudulent transactions in high-frequency scenarios by outperforming cutting-edge techniques.

Modeling information flow in a computer processor with a multi-stage queuing model

In this paper, we introduce a nonlinear stochastic model to describe the propagation of information inside a computer processor. In this model, a computational task is divided into stages, and information can flow from one stage to another. The model is formulated as a spatially-extended, continuous-time Markov chain where space represents different stages. This model is equivalent to a spatially-extended version of the M/M/s queue. The main modeling feature is the throttling function which describes the processor slowdown when the amount of information falls below a certain threshold. We derive the stationary distribution for this stochastic model and develop a closure for a deterministic ODE system that approximates the evolution of the mean and variance of the stochastic model. We demonstrate the validity of the closure with numerical simulations.

Optimal operation of the smart electrical network considering energy management of demand side

The smart electrical grid represents a significant advancement in generating, distributing, and consuming electricity. This sophisticated system integrates modern technology and communication tools to enhance energy management efficiency and improve demand costuming within the power network. In this paper, optimal operation of the electrical network with energy management and demand response program (DRP) is implemented. The implementation of the optimal operation is done via multi-stage and multi-objective functions modeling. The DRP modeling is done in first stage to optimal management of consumption in demand side. In second stage, operating cost, emission, power losses and voltage profile are optimized as multi-objective functions modeling with attention to optimal management of consumption in demand side. The solving optimal operation of the electrical network is carried out by using elephant herding optimization (EHO). This problem is implemented on 33-bus test system with hybrid energy resources. Finally, DRP leads to reducing costs, emissions and losses and improving voltage profile in proposed electrical network. Hence, operation costs, emission, power losses, and voltage deviation with the participation of DRP are minimized by 39.15%, 9.94%, 33.35%, and 30.73%, respectively. On the other side, voltage stability is enhanced by 3.66% without considering DRP.

Analysis of Lung Cancer Incidence in Non-Hispanic Black and White Americans using a Multistage Carcinogenesis Model.

There are complex and paradoxical patterns in lung cancer incidence by race/ethnicity and gender; compared to non-Hispanic White (NHW) males, non-Hispanic Black (NHB) males smoke fewer cigarettes per day and less frequently but have higher lung cancer rates. Similarly, NHB females are less likely to smoke but have comparable lung cancer rates to NHW females. We use a multistage carcinogenesis model to study the impact of smoking on lung cancer incidence in NHB and NHW individuals in the Multiethnic Cohort Study (MEC). The effects of smoking on the rates of lung tumor initiation, promotion, and malignant conversion, and the incidence of lung cancer in NHB versus NHW adults in the MEC were analyzed using the Two-Stage Clonal Expansion (TSCE) model. Maximum likelihood methods were used to estimate model parameters and assess differences by race/ethnicity, gender, and smoking history. Smoking increased promotion and malignant conversion but did not affect tumor initiation. Non-smoking-related initiation, promotion, and malignant conversion and smoking-related promotion and malignant conversion differed by race/ethnicity and gender. Non-smoking-related initiation and malignant conversion were higher in NHB than NHW individuals, whereas promotion was lower in NHB individuals. Findings suggest that while smoking plays an important role in lung cancer risk, background risk not dependent on smoking also plays a significant and under-recognized role in explaining race/ethnicity differences. Ultimately, the resulting TSCE model will inform race/ethnicity-specific lung cancer natural history models to assess the impact of preventive interventions on US lung cancer outcomes and disparities by race/ethnicity.

High-elevation Qilian Mountains and its inspiration for tectonics and biodiversity during the late Middle Miocene

The accurate paleoelevation reconstruction of the Qilian Mountains is critical to advancing our understanding the integrity of the Tibetan Plateau (TP) uplift model, its deep structural mechanisms, and corresponding connections with climatic, environmental, and biodiversity changes. Recently, the first quantitative reconstruction of the paleomidrange (i.e., average elevation of the basin and mountains) of the northern Tibetan Plateau (NTP) was completed, using innovative palynological paleoaltimetry-TP/TPAP [(Tsuga% + Podocarpus%) / (Tsuga% + Podocarpus% + Abies% + Picea%)] ratios, which revealed a rapid uplift of the NTP from low to high elevations during the late Middle Miocene. Here, we analyzed the Ebotu Fauna pollen record (13–12 Ma) from the Hongyazi Basin, situated within the Qilian Mountains, to directly infer the paleoelevation of this region. The pollen assemblages were predominantly composed of conifers (average of 64.0 %), including Picea, Cedrus, and Pinus of the Pinaceae, as well as Cupressaceae, with broadleaves and steppes taxa each accounting for <15 %. This pollen composition evidently reflects a vegetation type dominated by high-mountain conifers forest, consistent with Middle Miocene pollen assemblages from the Qaidam Basin. Application TP/TPAP ratios yielded a paleomidrange of 3492 ± 87 m at 13–12 Ma. Combined with the known elevation of the Qaidam Basin during this period (1885 ± 566 m), the elevation of the Qilian Mountains was calculated to be 4338 ± 653 m. In turn, an elevation of 2646 ± 740 m was obtained for the Hongyazi Basin. This high-elevation terrain provides evidence to support the multi-stage convective removal model of the TP caused by crustal shortening and thickening. It also led to the formation of a humid ecosystem dominated by conifers forest in the Qilian Mountains, supporting the diversification of mammalian taxa.