Enhancing Clear Air Turbulence Prediction: A Comparative Analysis of Machine Learning Algorithms Using GFS Forecast and ERA-5 Reanalysis Data

Enhancing Clear Air Turbulence Prediction: A Comparative Analysis of Machine Learning Algorithms Using GFS Forecast and ERA-5 Reanalysis Data | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Enhancing Clear Air Turbulence Prediction: A Comparative Analysis of Machine Learning Algorithms Using GFS Forecast and ERA-5 Reanalysis Data Ivan Bitar Fiuza de Mello, Gutemberg Borges França, Haroldo Fraga de Campos Velho This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4379402/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study evaluates twelve categorical machine learning algorithms using performance metrics, including Probability of Detection, False Alarm Rate, and F-measure, to classify Clear Air Turbulence patterns in the meteorological Global Forecast System (GFS), from the National Oceanic and Atmospheric Administration (NOAA) and European Centre for Medium-Range Weather Forecasts (ECMWF) ReAnalysis reanalysis (ERA-5) datasets along the Brazilian air route between the cities São Paulo and Porto Alegre(air-route in Brazil). Two training strategies, cross-validation, and random sample splitting, are employed for evaluating the performances of the algorithms. Results indicate that only the Random Forest algorithm meets the best performance with the adopted criterion for CAT detection. Algorithms trained with ERA-5 reanalysis data perform slightly better. The statistical best algorithm performance values with ERA-5 re-analysis data (GFS forecast data in parentheses) are POD=0.94 (0.87), FAR=0.08 (0.16), and F-measure=0.94 (0.87). As a future direction, it is suggested to investigate the use of regional models with enhanced resolutions in three key domains: horizontal, vertical, and atmospheric pressure. clear air turbulence machine learning forecast. Figures Figure 1 Figure 2 Figure 3 1. Introduction Turbulence presents a significant meteorological threat in aviation, potentially causing structural damage, injuries to passengers and crew, increased fuel consumption, and flight disruptions (Golding 2002 ; Gultepe et al. 2019 ). Airlines incur additional operational expenses due to heightened maintenance needs for aircraft structural components, resulting in reduced productivity and costs associated with treating injured personnel (Prosser et al. 2023 ). Atmospheric turbulence is categorized into various types, including low-level turbulence (further classified into thermal and mechanical), convective turbulence, clear air turbulence (CAT), and mountain-wave turbulence (Lester 1994 ). Clear air turbulence (CAT) is primarily generated by several mechanisms: (a) strong vertical and/or horizontal wind shear linked with jet streams, tropopause folding areas, and thermal inversions (Reiter 1963 ; Dutton and Panofsky 1970 ; Villiers and Van Heerden 2001; Kim and Chun 2010 ; Williams and Joshi 2013 ; Storer et al. 2019 ), Kelvin-Helmholtz instability (Atlas 1970; Venkatesh and Matthew 2013 ); intense anticyclonic (O'sullivan 1993; Knox 2003), and cyclonic (Ellrod and Knox 2010 ) flows; (b) propagation and breaking of atmospheric gravity waves induced by deep convection external to significant cloudiness of the convective cloud and away from the convection (Keller 1990 ; Cornman and Carmichael 1993 ; Kaplan et al. 2005 ; Wolff and Sharman 2008 ; Kim and Chun 2011 ; Meneguz et al. 2016 ); and (c) breaking of orographic atmospheric gravity waves into a statically stable atmosphere at low and/or high levels (Hopkins 1977 ; Villela 1998 ; Wolff and Sharman 2008 ; Kim and Chun 2010 ; Smith 1989 ). Currently, atmospheric turbulence on an aircraft can be assessed and reported during flight through qualitative and/or quantitative methods. Qualitative observations are facilitated by AIREP, where pilots voluntarily report weather phenomena, including turbulence, icing, and other significant weather events, to air traffic control or meteorological authorities globally. Additionally, PIREP serves a similar purpose but is specifically for pilot reports within US airspace (Racine et al. 2019 ). Qualitative measurements entail significant subjectivity regarding the time differences reported by pilots on a time scale of 10 to 100 seconds per kilometer from the actual turbulence location (Cornman et al. 2004 ), as well as individual judgments of each pilot and aircraft type (Sharman et al. 2006 ). Therefore, it is crucial to verify the main aspects of turbulence reports in AIREPS and PIREPS (Schwartz 1996 ). In contrast, quantitative measurements as used by Menegardo-Souza et al. ( 2021 ), such as the vortex dissipation rate (EDR) and/or vertical acceleration of gravity (VRTG), provide precise data suitable for scientific analysis, including the aircraft's geographical coordinates (latitude and longitude), maximum VRTG, altitude, and the time of measurement. Research related to CAT forecasting in South America is still in its beginning and is limited to the work carried out by Santos and Krapp ( 2006 ); Lyra et al. ( 2007 ); Mello ( 2015 ); Menegardo-Souza et al. ( 2021 ), Menegardo-Souza et al. ( 2022 ), and Gomes et al. ( 2022 ). Thus, the aspects that motivated the present study are: a) currently, the Integrated Center for Aeronautical Meteorology (CIMAER), responsible for meteorological forecasts in Brazilian airspace, does not yet have a regionally calibrated CAT prediction model, b) the procedure for determining the dimensions of airspace affected by CAT and the occurrence period is based on pilot diagnosis during flight, either spontaneously or prompted by air traffic control (the latter to remove or maintain CAT alerts), and c) the southern cone of Brazil, representing the Flight Information Region (FIR) under Curitiba's responsibility, is significantly impacted by CAT, as demonstrated by a study conducted by Gomes et al. ( 2022 ), utilizing AIREP records between October 2nd, 2015, and June 4th, 2019, and VRTG data collected by LATAM A320 aircraft from 2018 to 2019. The objective is to create a hybrid CAT prediction model by 1) utilizing the output from a numerical atmospheric model (and alternatively in re-analysis data) to identify and extract features (potential predictors) associated with regional atmospheric turbulence stages, and 2) configuring and training machine learning algorithms to predict the presence or absence of CAT in the atmospheric volume embracing the flight route using these defined predictors. The structure of this work is outlined as follows. Sections 2 and 3 provide descriptions of the study area and the data utilized for training and testing the machine learning algorithms. Section 4 elucidates the procedural steps of the method. Section 5 analyzes the utilized data, showcasing the performances of the algorithms and the contributions of all predictors in identifying atmospheric patterns of CAT and no-CAT in both forecasted and re-analysis data. Section 6 presents the findings and provides recommendations for future developments. 2. Study Area The study area is located in the region delimited between latitudes 021ºS and 033ºS, and longitudes 043ºW and 054ºW (polygon highlighted in Fig. 1 ). This area includes a considerable part of the South Region of Brazil, part of the Southeast Region of the country and the eastern portion of the state of Mato Grosso do Sul, thus centralizing the region between the airports of São Paulo and Porto Alegre, where the main airways that connect the airports of the two cities are located. [ Figure 1 is about here] 3. Data In Table 1, the information, frequency, and period of each data source used are summarized. Below, the purpose of the data is defined as follows: a) VRTG (Vertical Acceleration of Gravity) represents the vertical acceleration experienced due to gravity (g). Under normal conditions without turbulence, VRTG is assumed to be 1g. Table 2 presents the VRTG threshold values for different turbulence severity levels: no turbulence, class 1, class 2, and class 3. During turbulent flights, gravitational forces can vary positively or negatively based on atmospheric conditions, leading to deviations from the standard 1g. VRTG measurements are collected within the monitoring window, starting 10 seconds after takeoff and ending 4 seconds before landing. A new event is initiated after 300 seconds of normal acceleration (private briefing by LATAM airline). It is important to note that the absence of VRTG data on a particular day does not necessarily indicate the absence of turbulence, as it may simply result from the absence of aircraft recording the phenomenon. The VRTG data used in this study were collected by LATAM A320 aircraft from 2018 to 2019 and primarily employed to define the events under investigation and to guide the retrieval of other data sources (e.g., GFS, ERA-5, GOES-16, synoptic charts and atmospheric profiles) as outlined in Table 1. b) GFS Forecast and ERA-5 reanalysis data serve as the primary sources of atmospheric model data utilized to extract features (predictors) for input into the machine learning algorithms during the train-and-test procedure for categorical output, denoted as Y: turbulence (severity of VRTG) or N: o turbulence, according to Table 2. The GFS and ERA-5 provide 3-hour and 1-hour daily data, respectively, covering the area of interest from the surface up to 150 hPa. Following Menegardo-Sousa et al. (2021); Menegardo-Sousa et al. (2022), and Gomes et al. (2022), the geopotential height, vertical velocity (w), zonal (u), and meridional (v) wind components, pressure, temperature, and potential temperature were selected and some of them used to determine the other physical variables as in Table 3. c) GOES-16 satellite data enables the identification of various atmospheric phenomena. Specifically, it can detect cirrus clouds associated with deep convection and/or jet streams (channel 4), atmospheric features at high levels such as gravity waves and jet streams (channel 8), and convective cells (channel 13). In essence, it allows for the analysis of atmospheric scenarios conducive to the formation and/or presence of clear air turbulence; and d) Synoptic charts and atmospheric-sounding profiles facilitate a diagnostic analysis of current synoptic characteristics. 4. Method It consists of the following steps: i. The analysis of the VRTG dataset encompasses both spatio-temporal and severity aspects, spanning from the lowest recorded altitude to the highest. It investigates the airway route between São Paulo to Porto Alegre airports throughout the available data period. ii. Selection of k (where k =20) case studies, comprising 12 cases with VRTG recorded in the defined air-route and 8 cases without clear air turbulence (CAT), determined through data analysis as outlined in Table 1. iii. The GFS and ERA-5 data are selected before and after the VRTG record as in columns 5 (ERA-5) and 6 (GFS) in Table 4, for the k cases chosen. iv. An atmosphere volume representing the air-route from São Paulo to Porto Alegre is discretized by five meridional vertical cross-sections, each with a latitudinal dimension of approximately 4 degrees (equivalent to approximately 444 km). These cross-sections, labeled (1) through (5), begin respectively at the following latitudes and longitudes: 32ºS and 51ºW, -30.35ºS and -49.75ºW, -28.75ºS and -48.75ºW, -27.25º S and 47.75ºW, and 25.50ºS and 46.75ºW, as illustrated in Figure 2b. Potential predictors are then extracted from GFS and ERA-5 data for selected events. The extraction is performed at intervals of 1º latitude within the Zi range of 500-200 hPa, with a 50 hPa interval, resulting in a total of 35 grid points. vi. The WEKA software package (version 3.9.6) was selected to train all available ML categorical algorithms, utilizing the input and output previously defined and employing cross-validation. In the latter, the training dataset is divided into p equal-sized, mutually exclusive subsets, with each subset used for testing and the remaining p -1 subsets used for training (Holmes et al. 1994). The three classical statistics used to assess dichotomous estimation models are derived from the contingency table. These statistics include the Probability of Detection (POD), False Alarm Ratio (FAR), and F-measure where ideal values are 1, 0, and 1, respectively (Wilks 2006). Here, all trained algorithms are considered optimal if their performance in detecting CAT (using GFS or ERA-5 data) falls within the ranges of POD ≥ 0.80, FAR ≤ 0.20, and F-measure ≥ 0.80. vii. The results are analyzed. 5. Results and discussion Given the methodological procedures undertaken, we hereby present and engage in a comprehensive discussion regarding the ensuing results. 5.1. Characterization of VRTG data Figure 2 offers a comprehensive overview of the geographical distribution of the 1883 VRTG registers, as illustrated in Fig. 2a. These registers are further categorized based on severity class, with classes 1, 2, and 3 corresponding to 1704 (90.60%), 149 (7.91%), and 28 (1.48%) occurrences, respectively (Fig. 2c). Additionally, the mentioned figure presents data on flight level (Fig. 2d), monthly analysis (Fig. 2e), and hourly patterns (Fig. 2f) from March 3rd, 2018, to March 31st, 2021. In Fig. 2a, it is apparent that most turbulence records were clustered nearer to the cities of São Paulo and Porto Alegre. This observation is supported by VRTG data analysis, indicating, Fig. 2a, that 83.31% of occurrences took place during descent-climb flight procedures, while 7.34% occurred during a cruise along the route, and 9.36% transpired during the final approach for aircraft landing. Fig. 2e illustrates that the distribution of records is approximately bimodal. The period from September to March averages 204±39 VRTG per month, while the other period, between April and August, shows an average 44.8% lower than the previous period, averaging 91±12 VRTG per month. Figure 2f demonstrates the hourly distribution, revealing a clear correlation with commercial flights, which predominantly occur during daytime hours. It is essential to underscore that the preceding diagnosis is based on the VRTG records of LATAM aircraft during the specified period. [Figure 2 is about here] 5.2. Training machine learning algorithms It is known that the training of any learning algorithm, which involves a trial-and-error task, is costly as it needs to identify the set of predictors (input), here the values of the variables modeled by GFS Forecast and ERA-5 reanalysis, with output (i.e., whether there is or not VRTG class ≥ 1). Gomes et al. (2022) explored five instances of turbulence within the Curitiba FIR and examined data simulated by both GFS and WRF. The latter conducted simulations across three domains with horizontal resolutions of 18 km (90 x 90 points), 6 km (151 x 151 points), and 2 km (253 x 253 points), alongside a vertical profile consisting of 87 levels, with the upper atmospheric pressure set at 50 hPa, following Kim et al. (2010). The comprehensive analysis of AIREP reports indicates that clear air turbulence (CAT) within the Curitiba FIR, the focal area of the study, primarily results from the subtropical jet inducing shear. Consequently, the researchers opted to examine wind components ( u , v , w ), potential temperature (Θ), turbulent kinetic energy (TKE), and turbulence indices, specifically the Richardson (Ri), Brown (B), Ellrod-Endlich (E-1), Ellrod-Knap (E-2), and Ellrod-Knox (E-3) numbers. These indices were particularly scrutinized across three components: wind profile, Θ and TKE, and Ri and w. Despite the limited sample size of five events, they illustrate that the default indices, without local adjustments, were not as effective as other variables (e.g., wind shear) in identifying CAT events in the studied region. Based on the aforementioned information, the predictors outlined in Table 3 are denoted as u , v , w , WS , DWS , VWS , , and TKE. It is hypothesized that the behaviors of these predictors reflect the occurrence of both CAT and non-CAT events within the examined area, utilizing data from GFS and ERA-5. Subsequently, the dataset comprising input (predictor) and output (YES = VRTG ≥ 1 and NO = VRTG = 1 or no-CAT), as detailed in method step-(v), was initially trained using all available categorical algorithms in the WEKA package, as outlined in method step-(vi). Due to their superior performance, twelve algorithms listed in Table 5 were selected, and their efficacy is herein presented through statistical analysis to distinguish patterns of CAT and no-CAT using GFS forecast and ERA-5 reanalysis data. The values of statistics—POD, FAR, and F-measure—used to evaluate the performances of twelve categorical algorithms, as outlined in Table 3. Two strategies were employed: 1. Following step-(vi) of the method, the algorithms underwent training using cross-validation. This method partitions the dataset into p equal-sized, mutually exclusive subsets, with one subset utilized for testing and the remaining p -1 subsets for training. By doing so, this strategy helps mitigate overfitting, ensuring that the trained algorithms capture the underlying data patterns instead of memorizing noise, and prevents biased test samples. 2. The second strategy, referred to as the random sample studied case, involves splitting the data with 70% for training and 30% for testing. Table 6 presents the performance statistics values of the twelve trained algorithms for classifying CAT patterns in GFS forecast and ERA-5 reanalysis data using two defined training and testing strategies. Considering the optimal performance criterion of the algorithm established to detect CAT, i.e., POD ≥ 0.80, FAR ≤ 0.20, and F-measure ≥ 0.80, it is observed that only algorithms 10 (Random Forest) and 9, 10, 11, and 12 (highlighted in Table 5, corresponding to J48, Random Forest, Random Tree, and REP Tree) trained via cross-validation using GFS forecast and ERA-5 reanalysis data, respectively, met the criterion. However, only the Random Forest algorithm was able to adequately classify the CAT pattern in the random sample of 30% of the data using the second strategy. The average (and the best results in parentheses) values of PODs, FARs, and F-measures of the twelve algorithms, obtained through training with cross-validation for GFS and ERA-5 data, respectively, are as follows: 0.74 (0.84) ± 0.08, 0.31 (0.16) ± 0.08, 0.73 (0.87) ± 0.08 for GFS, and 0.78 (0.94) ± 0.10, 0.26 (0.08) ± 0.10, 0.78 (0.94) ± 0.10 for ERA-5. In general, the results of the algorithms trained with ERA-5 reanalysis data are slightly better (approximately 4% in POD, 5% in FAR, and 5% in F-measure) than those with GFS data. It is widely recognized that ERA-5 reanalysis is a comprehensive dataset generated by European Centre for Medium-Range Weather Forecasts (ECMWF). It integrates observational data with numerical models to construct a detailed account of past weather conditions. In this context, the GFS forecast is employed alongside ERA-5 reanalysis to replicate atmospheric patterns. The disparities between the ML algorithms' outcomes may be attributed to the more frequent hourly updates and heightened realism of the atmospheric environment in ERA-5 reanalysis data, in contrast to GFS forecast data, which is updated every 3 hours and carries a certain level of uncertainty. 5.3 Discussion on Predictors importance Figure 3 illustrates the percentage importance of each of the eight predictors (Θ, DWS , v , VWS , WS , and TKE) in training the Random Forest algorithm, excluding one input at a time. The analysis shows that the values of POD and F-measure vary from -0.11% to -7.55%. TKE and WS contribute the least to the PODs and F-measures, around 0.11%, while Θ and DWS are the most representative, with 7.37% and 7.55%, respectively. The FAR values either remain the same or worsen, with the greatest increase occurring when excluding Θ, leading to a 33.7% increase in its value. The higher representativeness of potential temperature in predicting CAT and no-CAT occurrences can be attributed to its ability to allow comparisons of air masses at different pressure levels. This characteristic makes potential temperature an invaluable tool for gaining insights into the vertical structure of the atmosphere. Furthermore, the contribution of predictors related to shear, such as DWS and VWS , aligns with expectations, especially considering the presence of jet-subtropical, as previously mentioned. These predictors collectively contribute significantly, amounting to 9.72% of the predictive power. The surprising aspect lies in the insignificant contribution of TKE. This unexpected result may be attributed to the spatial resolution of the model utilized, which might smooth out the fluctuating component of velocity inherent in the formulation of TKE. This highlights the importance of considering the limitations of the modeling approach and the need for further investigation to better understand the role of TKE in predictive modeling with high-spatial resolution. [Figure 3 is about here] 6. Conclusion The aim was to develop a hybrid CAT prediction model by utilizing output from both GFS forecast and ERA-5 re-analysis data to identify potential predictors linked to regional atmospheric turbulence stages. Subsequently, machine learning algorithms were trained to forecast the presence or absence of CAT in the atmospheric volume based on these designated predictors. The study evaluated twelve categorical algorithms, utilizing performance metrics such as POD, FAR, and F-measure, to effectively classify CAT patterns within the mentioned datasets using two strategies of ML algorithms. The cross-validation method partitions the dataset into subsets for training and testing to prevent overfitting and biased test samples, while the random sample splitting strategy divides the data into 70% for training and 30% for testing to validate algorithm performance. Analysis of the performance statistics values of the trained algorithms revealed that only the Random Forest algorithm met the established optimal performance criterion for CAT detection. Comparative analysis between GFS forecast and ERA-5 reanalysis data indicated that algorithms trained with ERA-5 reanalysis data exhibited slightly better performance metrics on average. This difference in outcomes can be attributed to the more frequent updates and enhanced realism inherent in ERA-5 reanalysis data, compared to the GFS forecast data. Overall, the study underscores the significance of leveraging advanced datasets such as ERA-5 reanalysis in conjunction with machine learning algorithms to accurately predict and classify atmospheric phenomena like CAT, thereby contributing to improved aviation safety and forecasting capabilities. As a suggestion for future work, it is recommended to explore the utilization of regional models with refined resolutions across three domains, encompassing horizontal, vertical, and atmospheric pressure, like the approach proposed by Kim and Chun ( 2010 ). The objective would be to enhance the prediction accuracy of Clear Air Turbulence (CAT) and consequently configure the CAT prediction model within the FIR-Curitiba airspace region. Declarations Author Contributions: All authors contributed to the study's conception and design. Material preparation, data collection, pre-processing, and analysis were performed by I. B F. M made the experiments. G.B.F. wrote the first draft of the manuscript and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Funding The study is funded by the Department of Airspace Control via the Brazilian Organization for Scientific and Technological Development of Airspace Control (CTCEA) with grant number 002-2018/COPPETEC_CTCEA. Additionally, the authors also thank the National Council for Scientific and Technological Development (CNPq, Brazil) for the research grants CNPq: 315349/2023-9 and CNPq: 307439/2021-6, received by authors HFCV and GBF, respectively. And, also, IBFM thanks to the Coordination for the Improvement of Higher Education Personnel (CAPES) for supplying doctoral scholarship. Data Availability Statement: The data presented in this study are available on request from the corresponding author. The data are not publicly available due to be proprietary data of the Department of Airspace Control (DECEA) and LATAM Airlines. Acknowledgments: The authors are grateful to the DECEA and LATAM for providing the data used in this research. Conflicts of Interest: The authors declare no conflict of interest. References Atlas D, Metcalf J, Richter J, Gossard E (1970) The birth of ‘CAT’ and microscale turbulence. Journal of the Atmospheric Sciences, 27(6), 903–913. https://doi.org/10.1175/15200469(1970)0272.0.CO;2 Breiman L (2001) Random forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324 Cornman LB, Carmichael B (1993) Varied research efforts are under way to find means of avoiding air turbulence. ICAO Journal, 48, 10–15 Cornman LB, Meymaris G, Limber M (2004) An update on the FAA Aviation Weather Research Program’s in situ turbulence measurement and reporting system. Preprints, 11th Conf. on Aviation, Range, and Aerospace Meteorology, Hyannis, MA, Amer. Meteor. Soc., P4.3 Dutton JA, Panofsky HA (1970) Clear air turbulence: a mystery may be unfolding. Science (New York, NY), 167(3920), 937–944. https://doi.org/10.1126/science.167.3920.937 Ellrod GP, Knox JA (2010) Improvements to an operational clear-air turbulence diagnostic index by addition of a divergence trend term. Weather & Forecasting, 25(2), 789–798. American meteorological society. https://doi.org/10.1175/2009WAF2222290.1 Golding W (2002) Turbulence and its impact on commercial aviation. Journal of Aviation/Aerospace Education and Research, 11, 19–29. https://doi.org/10.15394/jaaer.2002.1301 Gomes FP, França GB, Pereira Neto AV (2022) Turbulência de Ar Claro no Sudeste do Brasil: Estudo de Casos. Revista Brasileira de Meteorologia. 37 (1). https://doi.org/10.1590/0102-7786360036 Gultepe I, Sharman R, Williams P, Zhou B, Ellrod G, Minnis P, Trier S, Griffin S, Yum SS, Gharabaghi B, Feltz W, Temimi M, Pu Z, Storer L, Kneringer P, Weston MJ, Chuang H-y, Thobois L, Dimri AP, Dietz SJ, França GB, Almeida MV, Neto FLA (2019) A review of high impact weather on aviation meteorology applications. Pure and Applied Geophysics 176(5): 1869–192. https://doi.org/10.1007/s00024-019-02168-6 Holmes G, Donkin A, Witten IH (1994) Weka: A machine learning work- bench. In Proceedings of the Second Australia and New Zealand Conference on Intelligent Information Systems. https://doi.org/10.1109/ANZIIS.1994.396988 Hopkins RH (1977) Forecasting techniques of clear-air turbulence including that associated with mountain waves. Secretariat of the World Meteorological Organization, Geneva, Switzerland. – No. 482, technical note no. 155, WMO Kaplan M et al. (2005) Characterizing the severe turbulence environments associated with commercial aviation accidents. Part 2: Hydrostatic mesoscale numerical simulations of supergradient wind flow and streamwise ageostrophic frontogenesis. Meteorol. Atmos. Phys. 88, 153–173. https://doi.org/10.1007/s00703-004-0079-6 Keller JL (1990) Clear air turbulence as a response to meso- and synoptic-escale dynamic processes. Mon. Wea. Rev., 118, 2228-2242. https://doi.org/10.1175/1520-0493(1990)1182.0.CO;2 Kim J-H, Chun H-Y (2010) A numerical study of clear-air turbulence (CAT) encounters over South Korea on 2 April 2007. Journal of Applied Meteorology & Climatology, 49(12), 2381–2403. American meteorological society. https://doi.org/10.1175/2010JAMC2449.1 Kim J-H, Chun H-Y (2011) Statistics and possible sources of aviation turbulence over South Korea. Journal of Applied Meteorology and Climatology, 50, 311–324. https://doi.org/10.1175/2010JAMC2492.1 Knox JA (2003) Inertial instability. In Encyclopedia of the Atmospheric Sciences, eds. J. R. Holton, J. Pyle, and J. A. Curry, Academic Press, New York, 1004-1013. https://doi.org/10.1016/B0-12-227090-8/00175-5 Kohavi R (1995) The power of decision tables. In N. Lavrac & S. Wrobel (Eds.), Machine Learning: ECML-95, Lecture Notes in Computer Science (pp. 174–189). Berlin, Heidelberg: Springer. Retrieved September 9, 2020, from http://citeseerx.ist.psu.edu/ viewdoc/summary?doi=10.1.1.49.4576 Landwehr N, Hall M, Frank E (2005) Logistic Model Trees. Machine Learning, 59(1), 161–205. https://doi.org/10.1007/ s10994-005-0466-3 Le Cessie S, van Houwelingen JC (1992) Ridge estimators in logistic regression. Journal of the Royal Statistical Society: Series C (Applied Statistics), 41(1), 191–201. https://doi.org/10. 2307/2347628 Lester PF (1994) Turbulence: A new perspective for pilots. Englewood: Jeppesen Sanderson Inc Lyra AA, Chan CS, Dereczynski CP (2007) Indicadores de turbulência a partir de previsões do modelo regional ETA. Revista Brasileira de Meteorologia, 22(2), 161–182. https://doi.org/10.1590/S0102-77862007000200003 Mello IBF (2015) Climatologia e Estudo de Caso da Turbulência de Céu Claro a Partir de Registro de Aeronaves: Análise de dados observacionais e de modelagem (M.Sc). Instituto Nacional de Pesquisas Espaciais (INPE), São José dos Campos. http://mtc-m21b.sid.inpe.br/col/sid.inpe.br/mtc-m21b/2015/03.13.04.06/doc/publicacao.pdf. Retrieved 9 Sept 2020 Menegardo-Souza F, França GB, Menezes WF, Almeida VA (2021) Synoptic patterns of unusual severe turbulence events in the Santiago (Chile)–Mendoza (Argentina) route region in summer in the Southern Hemisphere. Pure & Applied Geophysics. https://doi.org/10.1007/s00024-021-0280 Menegardo-Souza F, França GB, Menezes WF, Almeida VA (2022) In-Flight Turbulence Forecast Model Based on Machine Learning for the Santiago (Chile)–Mendoza (Argentina) Air Route. Pure & Applied Geophysics. https://doi.org/10.1007/s00024-022-03053-5 Meneguz E, Wells H, Turp D (2016) An automated system to quantify aircraft encounters with convectively induced turbulence over Europe and the Northeast Atlantic. Journal of Applied Meteorology and Climatology, 55, 1077–1089. https://doi.org/10.1175/JAMC-D-15-0194.1 O’Sullivan DJ (1993) Inertial instability and inertia–gravity wave generation in the midlatitude winter stratosphere. Pre-prints, Ninth Conf. on Atmospheric and Oceanic Waves and Stability, San Antonio, TX, Amer. Meteor. Soc., 96–97 Prosser MC, Williams PD, Marlton GJ, Harrison RG (2023) Evidence for large increases in clear-air turbulence over the past four decades. Geophysical Research Letters, 50, e2023GL103814. https://doi.org/10.1029/2023GL103814 Racine N, Konkel A, Sollenberger R, Puzen, G, Thomas B, Marshall J (2019) Human Factors Guidance for the Display of Pilot Reports (PIREPs) on Information Display Systems. DOT/FAA/TC-19/29. Atlantic City International Airport, NJ: Federal Aviation Administration William J. Hughes Technical Center Reiter ER (1963) Jet-stream meteorology. Q J R Meteorological Society 90(385):366–367, University of Chicago Press. https://doi.org/10.1002/qj.49709038525 Santos CB, Krapp ARW (2006) Características de turbulência na FIR-CW - Região de Informação de Voo de Curitiba. https://www.redemet.aer.mil.br/old/uploads/2014/04/turbulencia.pdf Schwartz B (1996) The quantitative use of PIREPs in developing aviation weather guidance products. Weather and Forecasting, 11, 372–384. https://doi.org/10.1175/1520-0434(1996)0112.0.CO;2 Sharman R, Tebaldi C, Wiener G, Wolff J (2006) An integrated approach to mid- and upper-level turbulence forecasting. Weather & Forecasting, 21, 268–287. https://doi.org/10.1175/WAF924.1 Smith RB (1989) Mountain-induced stagnation points in hydrostatic flow. Tellus A, 41A(3), 270–274. https://doi.org/10.3402/tellusa.v41i3.11839 Sonawani S, Mukhopadhyay D (2013) A decision tree approach to classify web services using quality parameters. Presented at the international conference on web engineering and application (ICWA). Retrieved September 9, 2020, from https:// arxiv.org/abs/1311.6240 Storer LN, Williams PD, Gill PG (2019) Aviation turbulence: dynamics, forecasting, and response to climate change. Pure and Applied Geophysics, 176(5), 2081–2095. https://doi.org/10.1007/s00024-018-1822-0 Venkatesh TN, Matthew J (2013) The problem of clear air turbulence: Changing perspectives in the understanding of the phenomenon. Indian Academy of Sciences, Sadhana, v.38, parte 4, p.707-722. https://doi.org/10.1007/s12046-013-0161-1 Villela RJ (1998) Turbulência Inesperada. Aero Magazine, Edição 233 Villiers MP, Heerden JV (2001) Clear air turbulence over South Africa. Meteorol. Appl. v.8, p.119-126. https://doi.org/10.1017/S1350482701001104 Williams PD, Joshi MM (2013) Intensification of winter transatlantic aviation turbulence in response to climate change. Nature Climate Change, 3, 644–648. https://doi.org/10.1038/nclimate1866 Wilks DS (2006) Statistical methods in the atmospheric sciences (2nd ed). San Diego, California: Academic Press Witten I, Frank E (2005) Data mining: practical machine learning tools and techniques – 2nd ed. (p 309) Witten I, Frank E, Hall MA, Pal CJ (2016) Data mining: practical machine learning tools and techniques (p 621) Wolff JK, Sharman R (2008) Climatology of upper-level turbulence over the contiguous United States. Journal of Applied Meteorology & Climatology, 47(8), 2198–2214. https://doi.org/ 10.1175/2008JAMC1799.1 American Meteorological Society Tables Tables 1 to 5 are available in the Supplementary Files section Additional Declarations No competing interests reported. Supplementary Files EnhancingClearAirTurbulencePredictiontables.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4379402","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":300824408,"identity":"d8f0492d-e4b8-490b-b588-b75707d8f7ee","order_by":0,"name":"Ivan Bitar Fiuza de Mello","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzElEQVRIiWNgGAWjYBACNgnGBiAlwcMP4iUUEKelEajHQkYSpDPBgBhrJBhA1lTYGBwA8YjRwifd3P7g5x4JHuPzqxM/PDBgkOcXO0DAYTIHGxt7nknwmN14u1kC6DDDmbMTCGiRSGxs4DkA0nJ2A0hLgsFtIrQ0/gFqMZ5xdvMPorU0g2wx4O/dRqQtQL/MlgFqkbjBu80iwUCCsF/kZ7c/+PjmQJ09f//ZzTd/VNjI80sT0IIAEmCVEsQqBwH+A6SoHgWjYBSMgpEEAA0SQpOvQqqlAAAAAElFTkSuQmCC","orcid":"","institution":"Universidade Federal Rural do Rio de Janeiro","correspondingAuthor":true,"prefix":"","firstName":"Ivan","middleName":"Bitar Fiuza","lastName":"de Mello","suffix":""},{"id":300824412,"identity":"6a95f068-4f88-471b-8470-1203dff04370","order_by":1,"name":"Gutemberg Borges França","email":"","orcid":"","institution":"Universidade Federal Rural do Rio de Janeiro","correspondingAuthor":false,"prefix":"","firstName":"Gutemberg","middleName":"Borges","lastName":"França","suffix":""},{"id":300824416,"identity":"ebe0f02a-fa46-47e4-9391-f94a3e860035","order_by":2,"name":"Haroldo Fraga de Campos Velho","email":"","orcid":"","institution":"National Institute for Space Research","correspondingAuthor":false,"prefix":"","firstName":"Haroldo","middleName":"Fraga de Campos","lastName":"Velho","suffix":""}],"badges":[],"createdAt":"2024-05-07 00:55:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4379402/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4379402/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":56449333,"identity":"f81761a6-b238-4f24-b447-7556a744c27d","added_by":"auto","created_at":"2024-05-14 10:17:38","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":181996,"visible":true,"origin":"","legend":"\u003cp\u003eThe highlighted study area spans between coordinates 21ºS and 33ºS, and 43ºW and 54ºW, encompassing the air route between São Paulo and Porto Alegre (Map Source: \u003ca href=\"https://pt-br.topographic-map.com/\"\u003ehttps://pt-br.topographic-map.com/\u003c/a\u003e).records of LATAM aircraft during the specified period.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4379402/v1/d4265fc6a8b526267098643b.png"},{"id":56449632,"identity":"e2a031bb-ba36-4311-8500-f743b77c0f90","added_by":"auto","created_at":"2024-05-14 10:17:54","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":175677,"visible":true,"origin":"","legend":"\u003cp\u003eVRTG incidents along the São Paulo-Porto Alegre Route: (a) Geographical distribution of occurrences (generated with Google Earth), (b) Five meridional cross-sections where predictors are extracted at defined grid points (latitude, longitude, and Z\u003csub\u003ei\u003c/sub\u003e), (c) VRTG registrations versus its severity, (d) relation with flight level intervals, (e) Monthly variations, and (f) Hourly distribution patterns from March 3rd, 2018, to March 31st, 2021. Severity classes 1, 2, and 3 are denoted by the colors green, yellow, and red, respectively.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4379402/v1/3ed9e6ba04063a4cc3aab960.png"},{"id":56449308,"identity":"7b872a9c-98d4-41db-a255-4c0f2b1faf05","added_by":"auto","created_at":"2024-05-14 10:17:33","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":9117,"visible":true,"origin":"","legend":"\u003cp\u003eThe contribution of each input to the Probability of Detection (POD) and F-measure scores using the Random Forest algorithm for predicting the occurrence of CAT and no-CAT in the studied area.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4379402/v1/75e134a10e2d6e7616f22dcd.png"},{"id":57695650,"identity":"a35affa5-67a2-42dc-995f-d25c99570e9f","added_by":"auto","created_at":"2024-06-04 12:35:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":687806,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4379402/v1/2ca003b1-5c9c-4e9f-b529-e52e80b2e993.pdf"},{"id":56449331,"identity":"f75683ac-f3d9-4112-9245-9793d12c8429","added_by":"auto","created_at":"2024-05-14 10:17:38","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":150587,"visible":true,"origin":"","legend":"","description":"","filename":"EnhancingClearAirTurbulencePredictiontables.docx","url":"https://assets-eu.researchsquare.com/files/rs-4379402/v1/497f6d948c99e17cc8446a62.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Enhancing Clear Air Turbulence Prediction: A Comparative Analysis of Machine Learning Algorithms Using GFS Forecast and ERA-5 Reanalysis Data","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eTurbulence presents a significant meteorological threat in aviation, potentially causing structural damage, injuries to passengers and crew, increased fuel consumption, and flight disruptions (Golding \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Gultepe et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Airlines incur additional operational expenses due to heightened maintenance needs for aircraft structural components, resulting in reduced productivity and costs associated with treating injured personnel (Prosser et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Atmospheric turbulence is categorized into various types, including low-level turbulence (further classified into thermal and mechanical), convective turbulence, clear air turbulence (CAT), and mountain-wave turbulence (Lester \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e1994\u003c/span\u003e). Clear air turbulence (CAT) is primarily generated by several mechanisms: (a) strong vertical and/or horizontal wind shear linked with jet streams, tropopause folding areas, and thermal inversions (Reiter \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e1963\u003c/span\u003e; Dutton and Panofsky \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e1970\u003c/span\u003e; Villiers and Van Heerden 2001; Kim and Chun \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Williams and Joshi \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Storer et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), Kelvin-Helmholtz instability (Atlas 1970; Venkatesh and Matthew \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2013\u003c/span\u003e); intense anticyclonic (O'sullivan 1993; Knox 2003), and cyclonic (Ellrod and Knox \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) flows; (b) propagation and breaking of atmospheric gravity waves induced by deep convection external to significant cloudiness of the convective cloud and away from the convection (Keller \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1990\u003c/span\u003e; Cornman and Carmichael \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e1993\u003c/span\u003e; Kaplan et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Wolff and Sharman \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Kim and Chun \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Meneguz et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2016\u003c/span\u003e); and (c) breaking of orographic atmospheric gravity waves into a statically stable atmosphere at low and/or high levels (Hopkins \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e1977\u003c/span\u003e; Villela \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Wolff and Sharman \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Kim and Chun \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Smith \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e1989\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eCurrently, atmospheric turbulence on an aircraft can be assessed and reported during flight through qualitative and/or quantitative methods. Qualitative observations are facilitated by AIREP, where pilots voluntarily report weather phenomena, including turbulence, icing, and other significant weather events, to air traffic control or meteorological authorities globally. Additionally, PIREP serves a similar purpose but is specifically for pilot reports within US airspace (Racine et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Qualitative measurements entail significant subjectivity regarding the time differences reported by pilots on a time scale of 10 to 100 seconds per kilometer from the actual turbulence location (Cornman et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2004\u003c/span\u003e), as well as individual judgments of each pilot and aircraft type (Sharman et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Therefore, it is crucial to verify the main aspects of turbulence reports in AIREPS and PIREPS (Schwartz \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e1996\u003c/span\u003e). In contrast, quantitative measurements as used by Menegardo-Souza et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), such as the vortex dissipation rate (EDR) and/or vertical acceleration of gravity (VRTG), provide precise data suitable for scientific analysis, including the aircraft's geographical coordinates (latitude and longitude), maximum VRTG, altitude, and the time of measurement.\u003c/p\u003e \u003cp\u003eResearch related to CAT forecasting in South America is still in its beginning and is limited to the work carried out by Santos and Krapp (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2006\u003c/span\u003e); Lyra et al. (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2007\u003c/span\u003e); Mello (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2015\u003c/span\u003e); Menegardo-Souza et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), Menegardo-Souza et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and Gomes et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Thus, the aspects that motivated the present study are: a) currently, the Integrated Center for Aeronautical Meteorology (CIMAER), responsible for meteorological forecasts in Brazilian airspace, does not yet have a regionally calibrated CAT prediction model, b) the procedure for determining the dimensions of airspace affected by CAT and the occurrence period is based on pilot diagnosis during flight, either spontaneously or prompted by air traffic control (the latter to remove or maintain CAT alerts), and c) the southern cone of Brazil, representing the Flight Information Region (FIR) under Curitiba's responsibility, is significantly impacted by CAT, as demonstrated by a study conducted by Gomes et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), utilizing AIREP records between October 2nd, 2015, and June 4th, 2019, and VRTG data collected by LATAM A320 aircraft from 2018 to 2019. The objective is to create a hybrid CAT prediction model by 1) utilizing the output from a numerical atmospheric model (and alternatively in re-analysis data) to identify and extract features (potential predictors) associated with regional atmospheric turbulence stages, and 2) configuring and training machine learning algorithms to predict the presence or absence of CAT in the atmospheric volume embracing the flight route using these defined predictors.\u003c/p\u003e \u003cp\u003eThe structure of this work is outlined as follows. Sections \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and \u003cspan refid=\"Sec3\" class=\"InternalRef\"\u003e3\u003c/span\u003e provide descriptions of the study area and the data utilized for training and testing the machine learning algorithms. Section \u003cspan refid=\"Sec4\" class=\"InternalRef\"\u003e4\u003c/span\u003e elucidates the procedural steps of the method. Section \u003cspan refid=\"Sec5\" class=\"InternalRef\"\u003e5\u003c/span\u003e analyzes the utilized data, showcasing the performances of the algorithms and the contributions of all predictors in identifying atmospheric patterns of CAT and no-CAT in both forecasted and re-analysis data. Section \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e6\u003c/span\u003e presents the findings and provides recommendations for future developments.\u003c/p\u003e"},{"header":"2. Study Area","content":"\u003cp\u003eThe study area is located in the region delimited between latitudes 021\u0026ordm;S and 033\u0026ordm;S, and longitudes 043\u0026ordm;W and 054\u0026ordm;W (polygon highlighted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). This area includes a considerable part of the South Region of Brazil, part of the Southeast Region of the country and the eastern portion of the state of Mato Grosso do Sul, thus centralizing the region between the airports of S\u0026atilde;o Paulo and Porto Alegre, where the main airways that connect the airports of the two cities are located.\u003c/p\u003e \u003cp\u003e \u003cb\u003e[\u003c/b\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e \u003cb\u003eis about here]\u003c/b\u003e\u003c/p\u003e"},{"header":"3. Data","content":"\u003cp\u003eIn Table 1, the information, frequency, and period of each data source used are summarized. Below, the purpose of the data is defined as follows:\u003c/p\u003e\n\u003cp\u003ea)\u0026nbsp; \u0026nbsp;VRTG (Vertical Acceleration of Gravity) represents the vertical acceleration experienced due to gravity (g). Under normal conditions without turbulence, VRTG is assumed to be 1g. Table 2 presents the VRTG threshold values for different turbulence severity levels: no turbulence, class 1, class 2, and class 3. During turbulent flights, gravitational forces can vary positively or negatively based on atmospheric conditions, leading to deviations from the standard 1g. VRTG measurements are collected within the monitoring window, starting 10 seconds after takeoff and ending 4 seconds before landing. A new event is initiated after 300 seconds of normal acceleration (private briefing by LATAM airline). It is important to note that the absence of VRTG data on a particular day does not necessarily indicate the absence of turbulence, as it may simply result from the absence of aircraft recording the phenomenon. The VRTG data used in this study were collected by LATAM A320 aircraft from 2018 to 2019 and primarily employed to define the events under investigation and to guide the retrieval of other data sources (e.g., GFS, ERA-5, GOES-16, synoptic charts and\u0026nbsp;atmospheric profiles) as outlined in Table 1.\u003c/p\u003e\n\u003cp\u003eb)\u0026nbsp; \u0026nbsp;\u0026nbsp;GFS Forecast and ERA-5 reanalysis data serve as the primary sources of atmospheric model data utilized to extract features (predictors) for input into the machine learning algorithms during the train-and-test procedure for categorical output, denoted as Y: turbulence (severity of VRTG) or N: o turbulence, according to Table 2. The GFS and ERA-5 provide 3-hour and 1-hour daily data, respectively, covering the area of interest from the surface up to 150 hPa. \u0026nbsp; Following Menegardo-Sousa et al. (2021); Menegardo-Sousa et al. (2022), and Gomes et al. (2022), the geopotential height, vertical velocity (w), zonal (u), and meridional (v) wind components, pressure, temperature, and potential temperature were selected and some of them used to determine the other physical variables as in Table 3.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ec)\u0026nbsp; \u0026nbsp; \u0026nbsp;GOES-16 satellite data enables the identification of various atmospheric phenomena. Specifically, it can detect cirrus clouds associated with deep convection and/or jet streams (channel 4), atmospheric features at high levels such as gravity waves and jet streams (channel 8), and convective cells (channel 13). In essence, it allows for the analysis of atmospheric scenarios conducive to the formation and/or presence of clear air turbulence; and\u003c/p\u003e\n\u003cp\u003ed) \u0026nbsp; \u0026nbsp;Synoptic charts and atmospheric-sounding profiles facilitate a diagnostic analysis of current synoptic characteristics.\u003c/p\u003e"},{"header":"4. Method","content":"\u003cp\u003e\u0026nbsp;It consists of the following steps:\u003c/p\u003e\n\u003cp\u003ei. \u0026nbsp; The analysis of the VRTG dataset encompasses both spatio-temporal and severity aspects, spanning from the lowest recorded altitude to the highest. It investigates the airway route between S\u0026atilde;o Paulo to Porto Alegre airports throughout the available data period.\u003c/p\u003e\n\u003cp\u003eii. \u0026nbsp; \u0026nbsp;Selection of \u003cem\u003ek\u003c/em\u003e (where \u003cem\u003ek\u003c/em\u003e=20) case studies, comprising 12 cases with VRTG recorded in the defined air-route and 8 cases without clear air turbulence (CAT), determined through data analysis as outlined in Table 1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eiii. \u0026nbsp; The GFS and ERA-5 data are selected before and after the VRTG record as in columns 5 (ERA-5) and 6 (GFS) in Table 4, for the k cases chosen.\u003c/p\u003e\n\u003cp\u003eiv. \u0026nbsp;An atmosphere volume representing the air-route from S\u0026atilde;o Paulo to Porto Alegre is discretized by five meridional vertical cross-sections, each with a latitudinal dimension of approximately 4 degrees (equivalent to approximately 444 km). These cross-sections, labeled (1) through (5), begin respectively at the following latitudes and longitudes: 32\u0026ordm;S and 51\u0026ordm;W, -30.35\u0026ordm;S and -49.75\u0026ordm;W, -28.75\u0026ordm;S and -48.75\u0026ordm;W, -27.25\u0026ordm; S and 47.75\u0026ordm;W, and 25.50\u0026ordm;S and 46.75\u0026ordm;W, as illustrated in Figure 2b. Potential predictors are then extracted from GFS and ERA-5 data for selected events. The extraction is performed at intervals of 1\u0026ordm; latitude within the \u003cem\u003e\u003csub\u003eZi\u003c/sub\u003e\u003c/em\u003e\u003csub\u003e\u0026nbsp;\u003c/sub\u003erange of 500-200 hPa, with a 50 hPa interval, resulting in a total of 35 grid points.\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n\u003cp\u003evi. The WEKA software package (version 3.9.6) was selected to train all available ML categorical algorithms, utilizing the input and output previously defined and employing cross-validation. In the latter, the training dataset is divided into \u003cem\u003ep\u003c/em\u003e equal-sized, mutually exclusive subsets, with each subset used for testing and the remaining \u003cem\u003ep\u003c/em\u003e-1 subsets used for training (Holmes et al. 1994). The three classical statistics used to assess dichotomous estimation models are derived from the contingency table. These statistics include the Probability of Detection (POD), False Alarm Ratio (FAR), and F-measure where ideal values are 1, 0, and 1, respectively (Wilks 2006). Here, all trained algorithms are considered optimal if their performance in detecting CAT (using GFS or ERA-5 data) falls within the ranges of POD \u0026ge; 0.80, FAR \u0026le; 0.20, and F-measure \u0026ge; 0.80.\u003c/p\u003e\n\u003cp\u003evii. The results are analyzed.\u003c/p\u003e"},{"header":"5. Results and discussion","content":"\u003cp\u003eGiven the methodological procedures undertaken, we hereby present and engage in a comprehensive discussion regarding the ensuing results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.1.\u003c/strong\u003e\u003cstrong\u003eCharacterization of VRTG data\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 2 offers a comprehensive overview of the geographical distribution of the 1883 VRTG registers, as illustrated in Fig. 2a. These registers are further categorized based on severity class, with classes 1, 2, and 3 corresponding to 1704 (90.60%), 149 (7.91%), and 28 (1.48%) occurrences, respectively (Fig. 2c). Additionally, the mentioned figure presents data on flight level (Fig. 2d), monthly analysis (Fig. 2e), and hourly patterns (Fig. 2f) from March 3rd, 2018, to March 31st, 2021. In Fig. 2a, it is apparent that most turbulence records were clustered nearer to the cities of São Paulo and Porto Alegre. This observation is supported by VRTG data analysis, indicating, Fig. 2a, that 83.31% of occurrences took place during descent-climb flight procedures, while 7.34% occurred during a cruise along the route, and 9.36% transpired during the final approach for aircraft landing.\u003c/p\u003e\n\u003cp\u003eFig. 2e illustrates that the distribution of records is approximately bimodal. The period from September to March averages 204±39 VRTG per month, while the other period, between April and August, shows an average 44.8% lower than the previous period, averaging 91±12 VRTG per month.\u0026nbsp;Figure 2f demonstrates the hourly distribution, revealing a clear correlation with commercial flights, which predominantly occur during daytime hours. It is essential to underscore that the preceding diagnosis is based on the VRTG records of LATAM aircraft during the specified period.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e[Figure 2 is about here]\u003c/p\u003e\n\u003cp\u003e5.2.\u0026nbsp;Training machine learning algorithms\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIt is known that the training of any learning algorithm, which involves a trial-and-error task, is costly as it needs to identify the set of predictors (input), here the values of the variables modeled by GFS Forecast and ERA-5 reanalysis, with output (i.e., whether there is or not VRTG class ≥ 1).\u003c/p\u003e\n\u003cp\u003eGomes et al. (2022) explored five instances of turbulence within the Curitiba FIR and examined data simulated by both GFS and WRF. The latter conducted simulations across three domains with horizontal resolutions of 18 km (90 x 90 points), 6 km (151 x 151 points), and 2 km (253 x 253 points), alongside a vertical profile consisting of 87 levels, with the upper atmospheric pressure set at 50 hPa, following Kim et al. (2010). The comprehensive analysis of AIREP reports indicates that clear air turbulence (CAT) within the Curitiba FIR, the focal area of the study, primarily results from the subtropical jet inducing shear. Consequently, the researchers opted to examine wind components (\u003cem\u003eu\u003c/em\u003e, \u003cem\u003ev\u003c/em\u003e, \u003cem\u003ew\u003c/em\u003e), potential temperature (Θ), turbulent kinetic energy (TKE), and turbulence indices, specifically the Richardson (Ri), Brown (B), Ellrod-Endlich (E-1), Ellrod-Knap (E-2), and Ellrod-Knox (E-3) numbers. These indices were particularly scrutinized across three components: wind profile, Θ and TKE, and Ri and w. Despite the limited sample size of five events, they illustrate that the default indices, without local adjustments, were not as effective as other variables (e.g., wind shear) in identifying CAT events in the studied region.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBased on the aforementioned information, the predictors outlined in Table 3 are denoted as \u003cem\u003eu\u003c/em\u003e, \u003cem\u003ev\u003c/em\u003e, \u003cem\u003ew\u003c/em\u003e, \u003cem\u003eWS\u003c/em\u003e, \u003cem\u003eDWS\u003c/em\u003e, \u003cem\u003eVWS\u003c/em\u003e,\u0026nbsp;\u0026nbsp;, and TKE. It is hypothesized that the behaviors of these predictors reflect the occurrence of both CAT and non-CAT events within the examined area, utilizing data from GFS and ERA-5.\u0026nbsp;Subsequently, the dataset comprising input (predictor) and output (YES = VRTG ≥ 1 and NO = VRTG = 1 or no-CAT), as detailed in method step-(v), was initially trained using all available categorical algorithms in the WEKA package, as outlined in method step-(vi). Due to their superior performance, twelve algorithms listed in Table 5 were selected, and their efficacy is herein presented through statistical analysis to distinguish patterns of CAT and no-CAT using GFS forecast and ERA-5 reanalysis data.\u003c/p\u003e\n\u003cp\u003eThe values of statistics—POD, FAR, and F-measure—used to evaluate the performances of twelve categorical algorithms, as outlined in Table 3. Two strategies were employed:\u003c/p\u003e\n\u003cp\u003e1.\u0026nbsp; \u0026nbsp;\u0026nbsp;Following step-(vi) of the method, the algorithms underwent training using cross-validation. This method partitions the dataset into p equal-sized, mutually exclusive subsets, with one subset utilized for testing and the remaining \u003cem\u003ep\u003c/em\u003e-1 subsets for training. By doing so, this strategy helps mitigate overfitting, ensuring that the trained algorithms capture the underlying data patterns instead of memorizing noise, and prevents biased test samples.\u003c/p\u003e\n\u003cp\u003e2.\u0026nbsp;\u0026nbsp;The second strategy, referred to as the random sample studied case, involves splitting the data with 70% for training and 30% for testing.\u003c/p\u003e\n\u003cp\u003eTable 6 presents the performance statistics values of the twelve trained algorithms for classifying CAT patterns in GFS forecast and ERA-5 reanalysis data using two defined training and testing strategies. Considering the optimal performance criterion of the algorithm established to detect CAT, i.e., POD ≥ 0.80, FAR ≤ 0.20, and F-measure ≥ 0.80, it is observed that only algorithms 10 (Random Forest) and 9, 10, 11, and 12 (highlighted in Table 5, corresponding to J48, Random Forest, Random Tree, and REP Tree) trained via cross-validation using GFS forecast and ERA-5 reanalysis data, respectively, met the criterion. However, only the Random Forest algorithm was able to adequately classify the CAT pattern in the random sample of 30% of the data using the second strategy. The average (and the best results in parentheses) values of PODs, FARs, and F-measures of the twelve algorithms, obtained through training with cross-validation for GFS and ERA-5 data, respectively, are as follows: 0.74 (0.84) ± 0.08, 0.31 (0.16) ± 0.08, 0.73 (0.87) ± 0.08 for GFS, and 0.78 (0.94) ± 0.10, 0.26 (0.08) ± 0.10, 0.78 (0.94) ± 0.10 for ERA-5. In general, the results of the algorithms trained with ERA-5 reanalysis data are slightly better (approximately 4% in POD, 5% in FAR, and 5% in F-measure) than those with GFS data. It is widely recognized that ERA-5 reanalysis is a comprehensive dataset generated by European Centre for Medium-Range Weather Forecasts (ECMWF). It integrates observational data with numerical models to construct a detailed account of past weather conditions. In this context, the GFS forecast is employed alongside ERA-5 reanalysis to replicate atmospheric patterns. The disparities between the ML algorithms' outcomes may be attributed to the more frequent hourly updates and heightened realism of the atmospheric environment in ERA-5 reanalysis data, in contrast to GFS forecast data, which is updated every 3 hours and carries a certain level of uncertainty.\u003c/p\u003e\n\u003cp\u003e5.3 Discussion on Predictors importance\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFigure 3 illustrates the percentage importance of each of the eight predictors (Θ, \u003cem\u003eDWS\u003c/em\u003e, \u003cem\u003ev\u003c/em\u003e, \u003cem\u003eVWS\u003c/em\u003e, \u003cem\u003eWS\u003c/em\u003e, and TKE) in training the Random Forest algorithm, excluding one input at a time. The analysis shows that the values of POD and F-measure vary from -0.11% to -7.55%. TKE and \u003cem\u003eWS\u003c/em\u003e contribute the least to the PODs and F-measures, around 0.11%, while Θ and \u003cem\u003eDWS\u003c/em\u003e are the most representative, with 7.37% and 7.55%, respectively. The FAR values either remain the same or worsen, with the greatest increase occurring when excluding Θ, leading to a 33.7% increase in its value. The higher representativeness of potential temperature in predicting CAT and no-CAT occurrences can be attributed to its ability to allow comparisons of air masses at different pressure levels. This characteristic makes potential temperature an invaluable tool for gaining insights into the vertical structure of the atmosphere. Furthermore, the contribution of predictors related to shear, such as \u003cem\u003eDWS\u003c/em\u003e and \u003cem\u003eVWS\u003c/em\u003e, aligns with expectations, especially considering the presence of jet-subtropical, as previously mentioned. These predictors collectively contribute significantly, amounting to 9.72% of the predictive power. The surprising aspect lies in the insignificant contribution of TKE. This unexpected result may be attributed to the spatial resolution of the model utilized, which might smooth out the fluctuating component of velocity inherent in the formulation of TKE. This highlights the importance\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eof considering the limitations of the modeling approach and the need for further investigation to better understand the role of TKE in predictive modeling with high-spatial resolution.\u003c/p\u003e\n\u003cp\u003e[Figure 3 is about here]\u003c/p\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eThe aim was to develop a hybrid CAT prediction model by utilizing output from both GFS forecast and ERA-5 re-analysis data to identify potential predictors linked to regional atmospheric turbulence stages. Subsequently, machine learning algorithms were trained to forecast the presence or absence of CAT in the atmospheric volume based on these designated predictors. The study evaluated twelve categorical algorithms, utilizing performance metrics such as POD, FAR, and F-measure, to effectively classify CAT patterns within the mentioned datasets using two strategies of ML algorithms. The cross-validation method partitions the dataset into subsets for training and testing to prevent overfitting and biased test samples, while the random sample splitting strategy divides the data into 70% for training and 30% for testing to validate algorithm performance. Analysis of the performance statistics values of the trained algorithms revealed that only the Random Forest algorithm met the established optimal performance criterion for CAT detection.\u003c/p\u003e \u003cp\u003eComparative analysis between GFS forecast and ERA-5 reanalysis data indicated that algorithms trained with ERA-5 reanalysis data exhibited slightly better performance metrics on average. This difference in outcomes can be attributed to the more frequent updates and enhanced realism inherent in ERA-5 reanalysis data, compared to the GFS forecast data.\u003c/p\u003e \u003cp\u003eOverall, the study underscores the significance of leveraging advanced datasets such as ERA-5 reanalysis in conjunction with machine learning algorithms to accurately predict and classify atmospheric phenomena like CAT, thereby contributing to improved aviation safety and forecasting capabilities. As a suggestion for future work, it is recommended to explore the utilization of regional models with refined resolutions across three domains, encompassing horizontal, vertical, and atmospheric pressure, like the approach proposed by Kim and Chun (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). The objective would be to enhance the prediction accuracy of Clear Air Turbulence (CAT) and consequently configure the CAT prediction model within the FIR-Curitiba airspace region.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAuthor Contributions: All authors contributed to the study's conception and design. Material preparation, data collection, pre-processing, and analysis were performed by I. B F. M made the experiments. G.B.F. wrote the first draft of the manuscript and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study is funded by the Department of Airspace Control via the Brazilian Organization for Scientific and Technological Development of Airspace Control (CTCEA) with grant number 002-2018/COPPETEC_CTCEA. Additionally, the authors also thank the National Council for Scientific and Technological Development (CNPq, Brazil) for the research grants CNPq: 315349/2023-9 and CNPq: 307439/2021-6, received by authors HFCV and GBF, respectively. And, also, IBFM thanks to the Coordination for the Improvement of Higher Education Personnel (CAPES)\u0026nbsp;for supplying doctoral scholarship.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement:\u0026nbsp;\u003c/strong\u003eThe data presented in this study are available on request from the corresponding author. The data are not publicly available due to be proprietary data of the Department of Airspace Control (DECEA) and LATAM Airlines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u003c/strong\u003e The authors are grateful to the DECEA and LATAM for providing the data used in this research.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest:\u003c/strong\u003e The authors declare no conflict of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAtlas D, Metcalf J, Richter J, Gossard E (1970) The birth of \u0026lsquo;CAT\u0026rsquo; and microscale turbulence. Journal of the Atmospheric Sciences, 27(6), 903\u0026ndash;913. https://doi.org/10.1175/15200469(1970)027\u0026lt;0903:TBOAMT\u0026gt;2.0.CO;2\u003c/li\u003e\n\u003cli\u003eBreiman L (2001) Random forests. Machine Learning, 45(1), 5\u0026ndash;32. https://doi.org/10.1023/A:1010933404324\u003c/li\u003e\n\u003cli\u003eCornman LB, Carmichael B (1993) Varied research efforts are under way to find means of avoiding air turbulence. ICAO Journal, 48, 10\u0026ndash;15\u003c/li\u003e\n\u003cli\u003eCornman LB, Meymaris G, Limber M (2004) An update on the FAA Aviation Weather Research Program\u0026rsquo;s in situ turbulence measurement and reporting system. Preprints, 11th Conf. on Aviation, Range, and Aerospace Meteorology, Hyannis, MA, Amer. Meteor. Soc., P4.3\u003c/li\u003e\n\u003cli\u003eDutton JA, Panofsky HA (1970) Clear air turbulence: a mystery may be unfolding. Science (New York, \u003c/li\u003e\n\u003cli\u003eNY), 167(3920), 937\u0026ndash;944. https://doi.org/10.1126/science.167.3920.937\u003c/li\u003e\n\u003cli\u003eEllrod GP, Knox JA (2010) Improvements to an operational clear-air turbulence diagnostic index by addition of a divergence trend term. Weather \u0026amp; Forecasting, 25(2), 789\u0026ndash;798. American meteorological society. https://doi.org/10.1175/2009WAF2222290.1\u003c/li\u003e\n\u003cli\u003eGolding W (2002) Turbulence and its impact on commercial aviation. Journal of Aviation/Aerospace Education and Research, 11, 19\u0026ndash;29. https://doi.org/10.15394/jaaer.2002.1301\u003c/li\u003e\n\u003cli\u003eGomes FP, Fran\u0026ccedil;a GB, Pereira Neto AV (2022) Turbul\u0026ecirc;ncia de Ar Claro no Sudeste do Brasil: Estudo de Casos. Revista Brasileira de Meteorologia. 37 (1). https://doi.org/10.1590/0102-7786360036\u003c/li\u003e\n\u003cli\u003eGultepe I, Sharman R, Williams P, Zhou B, Ellrod G, Minnis P, Trier S, Griffin S, Yum SS, Gharabaghi B, Feltz W, Temimi M, Pu Z, Storer L, Kneringer P, Weston MJ, Chuang H-y, Thobois L, Dimri AP, Dietz SJ, Fran\u0026ccedil;a GB, Almeida MV, Neto FLA (2019) A review of high impact weather on aviation meteorology applications. Pure and Applied Geophysics 176(5): 1869\u0026ndash;192. https://doi.org/10.1007/s00024-019-02168-6\u003c/li\u003e\n\u003cli\u003eHolmes G, Donkin A, Witten IH (1994) Weka: A machine learning work- bench. In Proceedings of the Second Australia and New Zealand Conference on Intelligent Information Systems. https://doi.org/10.1109/ANZIIS.1994.396988\u003c/li\u003e\n\u003cli\u003eHopkins RH (1977) Forecasting techniques of clear-air turbulence including that associated with mountain waves. Secretariat of the World Meteorological Organization, Geneva, Switzerland. \u0026ndash; No. 482, technical note no. 155, WMO\u003c/li\u003e\n\u003cli\u003eKaplan M et al. (2005) Characterizing the severe turbulence environments associated with commercial aviation accidents. Part 2: Hydrostatic mesoscale numerical simulations of supergradient wind flow and streamwise ageostrophic frontogenesis. Meteorol. Atmos. Phys. 88, 153\u0026ndash;173. https://doi.org/10.1007/s00703-004-0079-6\u003c/li\u003e\n\u003cli\u003eKeller JL (1990) Clear air turbulence as a response to meso- and synoptic-escale dynamic processes. Mon. Wea. Rev., 118, 2228-2242. https://doi.org/10.1175/1520-0493(1990)118\u0026lt;2228:CATAAR\u0026gt;2.0.CO;2\u003c/li\u003e\n\u003cli\u003eKim J-H, Chun H-Y (2010) A numerical study of clear-air turbulence (CAT) encounters over South Korea on 2 April 2007. Journal of Applied Meteorology \u0026amp; Climatology, 49(12), 2381\u0026ndash;2403. American meteorological society. https://doi.org/10.1175/2010JAMC2449.1\u003c/li\u003e\n\u003cli\u003eKim J-H, Chun H-Y (2011) Statistics and possible sources of aviation turbulence over South Korea. Journal of Applied Meteorology and Climatology, 50, 311\u0026ndash;324. https://doi.org/10.1175/2010JAMC2492.1\u003c/li\u003e\n\u003cli\u003eKnox JA (2003) Inertial instability. In Encyclopedia of the Atmospheric Sciences, eds. J. R. Holton, J. Pyle, and J. A. Curry, Academic Press, New York, 1004-1013. https://doi.org/10.1016/B0-12-227090-8/00175-5\u003c/li\u003e\n\u003cli\u003eKohavi R (1995) The power of decision tables. In N. Lavrac \u0026amp; S. Wrobel (Eds.), Machine Learning: ECML-95, Lecture Notes in Computer Science (pp. 174\u0026ndash;189). Berlin, Heidelberg: Springer. Retrieved September 9, 2020, from http://citeseerx.ist.psu.edu/ viewdoc/summary?doi=10.1.1.49.4576\u003c/li\u003e\n\u003cli\u003eLandwehr N, Hall M, Frank E (2005) Logistic Model Trees. Machine Learning, 59(1), 161\u0026ndash;205. https://doi.org/10.1007/ s10994-005-0466-3\u003c/li\u003e\n\u003cli\u003eLe Cessie S, van Houwelingen JC (1992) Ridge estimators in logistic regression. Journal of the Royal Statistical Society: Series C (Applied Statistics), 41(1), 191\u0026ndash;201. https://doi.org/10. 2307/2347628\u003c/li\u003e\n\u003cli\u003eLester PF (1994) Turbulence: A new perspective for pilots. Englewood: Jeppesen Sanderson Inc\u003c/li\u003e\n\u003cli\u003eLyra AA, Chan CS, Dereczynski CP (2007) Indicadores de turbul\u0026ecirc;ncia a partir de previs\u0026otilde;es do modelo regional ETA. Revista Brasileira de Meteorologia, 22(2), 161\u0026ndash;182. https://doi.org/10.1590/S0102-77862007000200003\u003c/li\u003e\n\u003cli\u003eMello IBF (2015) Climatologia e Estudo de Caso da Turbul\u0026ecirc;ncia de C\u0026eacute;u Claro a Partir de Registro de Aeronaves: An\u0026aacute;lise de dados observacionais e de modelagem (M.Sc). Instituto Nacional de Pesquisas Espaciais (INPE), S\u0026atilde;o Jos\u0026eacute; dos Campos. http://mtc-m21b.sid.inpe.br/col/sid.inpe.br/mtc-m21b/2015/03.13.04.06/doc/publicacao.pdf. Retrieved 9 Sept 2020\u003c/li\u003e\n\u003cli\u003eMenegardo-Souza F, Fran\u0026ccedil;a GB, Menezes WF, Almeida VA (2021) Synoptic patterns of unusual severe turbulence events in the Santiago (Chile)\u0026ndash;Mendoza (Argentina) route region in summer in the Southern Hemisphere. Pure \u0026amp; Applied Geophysics. https://doi.org/10.1007/s00024-021-0280\u003c/li\u003e\n\u003cli\u003eMenegardo-Souza F, Fran\u0026ccedil;a GB, Menezes WF, Almeida VA (2022) In-Flight Turbulence Forecast Model Based on Machine Learning for the Santiago (Chile)\u0026ndash;Mendoza (Argentina) Air Route. Pure \u0026amp; Applied Geophysics. https://doi.org/10.1007/s00024-022-03053-5\u003c/li\u003e\n\u003cli\u003eMeneguz E, Wells H, Turp D (2016) An automated system to quantify aircraft encounters with convectively induced turbulence over Europe and the Northeast Atlantic. Journal of Applied Meteorology and Climatology, 55, 1077\u0026ndash;1089. https://doi.org/10.1175/JAMC-D-15-0194.1\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Sullivan DJ (1993) Inertial instability and inertia\u0026ndash;gravity wave generation in the midlatitude winter stratosphere. Pre-prints, Ninth Conf. on Atmospheric and Oceanic Waves and Stability, San Antonio, TX, Amer. Meteor. Soc., 96\u0026ndash;97\u003c/li\u003e\n\u003cli\u003eProsser MC, Williams PD, Marlton GJ, Harrison RG (2023) Evidence for large increases in clear-air turbulence over the past four decades. Geophysical Research Letters, 50, e2023GL103814. https://doi.org/10.1029/2023GL103814 \u003c/li\u003e\n\u003cli\u003eRacine N, Konkel A, Sollenberger R, Puzen, G, Thomas B, Marshall J (2019) Human Factors Guidance for the Display of Pilot Reports (PIREPs) on Information Display Systems. DOT/FAA/TC-19/29. Atlantic City International Airport, NJ: Federal Aviation Administration William J. Hughes Technical Center\u003c/li\u003e\n\u003cli\u003eReiter ER (1963) Jet-stream meteorology. Q J R Meteorological Society 90(385):366\u0026ndash;367, University of Chicago Press. https://doi.org/10.1002/qj.49709038525\u003c/li\u003e\n\u003cli\u003eSantos CB, Krapp ARW (2006) Caracter\u0026iacute;sticas de turbul\u0026ecirc;ncia na FIR-CW - Regi\u0026atilde;o de Informa\u0026ccedil;\u0026atilde;o de Voo de Curitiba. https://www.redemet.aer.mil.br/old/uploads/2014/04/turbulencia.pdf\u003c/li\u003e\n\u003cli\u003eSchwartz B (1996) The quantitative use of PIREPs in developing aviation weather guidance products. \u003c/li\u003e\n\u003cli\u003eWeather and Forecasting, 11, 372\u0026ndash;384. https://doi.org/10.1175/1520-0434(1996)011\u0026lt;0372:TQUOPI\u0026gt;2.0.CO;2\u003c/li\u003e\n\u003cli\u003eSharman R, Tebaldi C, Wiener G, Wolff J (2006) An integrated approach to mid- and upper-level turbulence forecasting. Weather \u0026amp; Forecasting, 21, 268\u0026ndash;287. https://doi.org/10.1175/WAF924.1\u003c/li\u003e\n\u003cli\u003eSmith RB (1989) Mountain-induced stagnation points in hydrostatic flow. Tellus A, 41A(3), 270\u0026ndash;274. https://doi.org/10.3402/tellusa.v41i3.11839\u003c/li\u003e\n\u003cli\u003eSonawani S, Mukhopadhyay D (2013) A decision tree approach to classify web services using quality parameters. Presented at the international conference on web engineering and application (ICWA). Retrieved September 9, 2020, from https:// arxiv.org/abs/1311.6240\u003c/li\u003e\n\u003cli\u003eStorer LN, Williams PD, Gill PG (2019) Aviation turbulence: dynamics, forecasting, and response to climate change. Pure and Applied Geophysics, 176(5), 2081\u0026ndash;2095. https://doi.org/10.1007/s00024-018-1822-0\u003c/li\u003e\n\u003cli\u003eVenkatesh TN, Matthew J (2013) The problem of clear air turbulence: Changing perspectives in the understanding of the phenomenon. Indian Academy of Sciences, Sadhana, v.38, parte 4, p.707-722. https://doi.org/10.1007/s12046-013-0161-1\u003c/li\u003e\n\u003cli\u003eVillela RJ (1998) Turbul\u0026ecirc;ncia Inesperada. Aero Magazine, Edi\u0026ccedil;\u0026atilde;o 233\u003c/li\u003e\n\u003cli\u003eVilliers MP, Heerden JV (2001) Clear air turbulence over South Africa. Meteorol. Appl. v.8, p.119-126. https://doi.org/10.1017/S1350482701001104\u003c/li\u003e\n\u003cli\u003eWilliams PD, Joshi MM (2013) Intensification of winter transatlantic aviation turbulence in response to climate change. Nature Climate Change, 3, 644\u0026ndash;648. https://doi.org/10.1038/nclimate1866\u003c/li\u003e\n\u003cli\u003eWilks DS (2006) Statistical methods in the atmospheric sciences (2nd ed). San Diego, California: Academic Press\u003c/li\u003e\n\u003cli\u003eWitten I, Frank E (2005) Data mining: practical machine learning tools and techniques \u0026ndash; 2nd ed. (p 309)\u003c/li\u003e\n\u003cli\u003eWitten I, Frank E, Hall MA, Pal CJ (2016) Data mining: practical machine learning tools and techniques (p 621) \u003c/li\u003e\n\u003cli\u003eWolff JK, Sharman R (2008) Climatology of upper-level turbulence over the contiguous United States. Journal of Applied Meteorology \u0026amp; Climatology, 47(8), 2198\u0026ndash;2214. https://doi.org/ 10.1175/2008JAMC1799.1 American Meteorological Society\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 to 5 are available in the Supplementary Files section\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"clear air turbulence, machine learning, forecast. ","lastPublishedDoi":"10.21203/rs.3.rs-4379402/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4379402/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study evaluates twelve categorical machine learning algorithms using performance metrics, including Probability of Detection, False Alarm Rate, and F-measure, to classify Clear Air Turbulence patterns in the meteorological Global Forecast System (GFS), from the National Oceanic and Atmospheric Administration (NOAA) and European Centre for Medium-Range Weather Forecasts (ECMWF) ReAnalysis reanalysis (ERA-5) datasets along the Brazilian air route between the cities São Paulo and Porto Alegre(air-route in Brazil). Two training strategies, cross-validation, and random sample splitting, are employed for evaluating the performances of the algorithms. Results indicate that only the Random Forest algorithm meets the best performance with the adopted criterion for CAT detection. Algorithms trained with ERA-5 reanalysis data perform slightly better. The statistical best algorithm performance values with ERA-5 re-analysis data (GFS forecast data in parentheses) are POD=0.94 (0.87), FAR=0.08 (0.16), and F-measure=0.94 (0.87). As a future direction, it is suggested to investigate the use of regional models with enhanced resolutions in three key domains: horizontal, vertical, and atmospheric pressure.\u003c/p\u003e","manuscriptTitle":"Enhancing Clear Air Turbulence Prediction: A Comparative Analysis of Machine Learning Algorithms Using GFS Forecast and ERA-5 Reanalysis Data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-14 10:16:09","doi":"10.21203/rs.3.rs-4379402/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f07dfa69-2da1-44b3-84e6-41da026ce2d4","owner":[],"postedDate":"May 14th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-06-04T12:27:25+00:00","versionOfRecord":[],"versionCreatedAt":"2024-05-14 10:16:09","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4379402","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4379402","identity":"rs-4379402","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}