Atmospheric turbulence sensing with structured terahertz beams containing orbital angular momentum Authors: Ruiyi Shen, Miranda van Iersel, Yasaman Ghasempour Published: 2026 ABSTRACT Clear-air turbulence poses a persistent hazard to aviation, yet it evades conventional weather radar and remains difficult to detect. Existing optical sensing methods falter under severe turbulence and are extremely vulnerable to misalignment between airborne nodes. Here we introduce an atmospheric sensing scheme built on structured terahertz (THz) beams carrying orbital angular momentum (OAM). We derive and experimentally validate an analytical model that maps turbulence-induced OAM modal crosstalk to the refractive index structure constant of the atmosphere. Our simulations and experiments demonstrate that this THz-OAM approach remains robust under severe turbulence where optical systems saturate, and offers orders-of-magnitude greater misalignment resilience. Furthermore, this scheme also enables localization of the dominant turbulent clusters through understanding the interaction of diverging THz-OAM beams with the turbulent medium in the reciprocal propagation paths and simultaneously monitoring water vapor to identify turbulence precursors. This work establishes structured THz-OAM beams as a robust framework for remote atmospheric turbulence inference. Introduction Atmospheric turbulence, characterized by stochastic, small-scale, vibrant air motions, presents significant challenges to environmental sensing, aerospace engineering, and optical communication systems due to its random and fast-changing nature in both spatial and temporal domains. In commercial aviation, for example, intense turbulence can impose severe aerodynamic loads that can compromise the structural integrity of aircraft and lead to passenger and crew injuries1,2. The standard method for turbulence detection utilizes onboard X-band (8–12 GHz) weather radar, which infers air motion from the Doppler information of backscattered signals from precipitation particles such as raindrops and snowflakes3–5. This information can guide pilots in taking detours from known turbulent atmospheres. A limitation of this technique, however, is that it fails to detect clear-air turbulence, a phenomenon commonly observed at high altitudes in non-precipitating atmospheres, posing a substantial risk6,7. Beyond aviation, accurate profiling of clear-air turbulence is also important for understanding the dynamics of the Earth’s atmosphere, potentially improving numerical weather prediction, and advancing climate studies8–11. Therefore, the development of a robust, real-time system for accurately characterizing turbulence intensity, especially for clear-air turbulence, is of great importance to both scientific research and practical applications. Most existing clear-air turbulence detection techniques onboard aircraft are confined to reactive, in-situ methods, such as wind sensors and pilot reports. These methods are fundamentally limited because they only offer turbulence estimation upon encounter, rather than providing the remote sensing capability required for advanced warning and proactive avoidance. To address this gap, a few preliminary works have demonstrated the possibility of remote turbulence detection in the optical domain12,13. One prominent approach is to use LIDAR, which interrogates the atmosphere with a pulsed infrared laser to measure the backscattered signal from small aerosols and therefore infer turbulence-induced Doppler velocity shifts in the atmosphere14–16. The efficacy of this approach is often constrained by low signal-to-noise ratios, either due to low aerosol concentrations in clear, dry air, or from significant signal attenuation due to poor weather conditions such as fog17–19. Alternatively, atmospheric turbulence fundamentally alters the propagation of electromagnetic waves; therefore, electromagnetic measurements can provide insights into the properties of the turbulent medium. Specifically, turbulent eddies induce random fluctuations in the refractive index of the atmosphere, which in turn impose phase and amplitude aberrations on a propagating wavefront20–22. This principle inspires the use of optical scintillometers, which measure path-integrated turbulence strength by analyzing the intensity fluctuations (i.e., scintillation) of a laser beam13,23,24. However, the short wavelengths intrinsic to the optical regime introduce fundamental limitations. First, under moderate-to-strong turbulence conditions, optical systems suffer from saturation, where the scintillation index plateaus with increasing turbulence strength25,26. While optical systems offer advantages in characterizing weak turbulence, often analyzed along vertical or slant paths for fields including astronomical imaging and free-space optical communication, a broader understanding of turbulence dynamics across moderate to strong regimes is also essential for applications operating along predominantly horizontal paths, such as environmental monitoring and aviation safety, where severe turbulence events can threaten the safe operation of aircraft. Furthermore, optical systems are extremely sensitive to misalignment between the transmitting and receiving nodes, which can usually be caused by mechanical vibrations on mobile platforms. The minute beam wander can introduce intensity fluctuations that overwhelm the turbulence-induced signatures, making the system unreliable for mobile platforms27,28. To overcome these limitations, particularly for horizontal-path turbulence sensing relevant to aviation, we propose and demonstrate a sensing modality based on structured light at terahertz (THz) frequencies. This approach provides two major advantages over conventional optical techniques. First, the THz regime extends the operational dynamic range of turbulence characterization. Since the strengths of turbulence effects scale with wavelength, THz beams are less susceptible to saturation effects under moderate-to-strong turbulence. Additionally, THz beams are more robust to misalignment, which is common for mobile platforms. This is because the phase tilt induced by angular misalignment is proportional to the wavenumber (k = 2π/λ), which is orders of magnitude smaller for THz waves, resulting in more stable and reliable measurements. While shifting to even longer wavelengths, such as millimeter-wave or microwave bands, further prevents saturation and reduces misalignment sensitivity, it introduces extreme beam divergence from an equivalent transmit aperture, making it impossible to be deployed on mobile nodes. In contrast, the THz bands represent an optimal convergence point that escapes the saturation limits of the optical regime, while restricting beam divergence to a manageable scale. Despite offering this ideal physical balance, the shift to the THz frequencies presents its own sensing constraints. A backscattering-based approach analogous to LIDAR in THz frequencies is invisible to small aerosols in the atmosphere. Furthermore, the weaker interaction with turbulent eddies makes forward-propagating intensity-based measurements (i.e., from a transmitter to a separate receiver) less sensitive. Despite the insensitivity in intensity, the phase aberrations imposed by the random refractive index fluctuations persist29–31. Hence, probing the phase front distortions becomes a promising modality for robust THz turbulence sensing. While prior works in optics have explored using structured light to probe turbulence, they typically rely on coherent detection to directly analyze the complex amplitude and phase profiles of the distorted field32–34. Due to the lack of synchronization between bistatic nodes (i.e., when the transmitter and receiver are not co-located) and the stringent hardware requirements, directly measuring the complex received phase profile is impractical. Instead, we employ an alternative indirect approach to bypass the need for coherent field detection. By projecting the distorted phase front onto an orthogonal basis, we transform the problem of profiling phase aberrations into an incoherent measurement of power distribution across discrete modes to infer turbulence-induced phase changes. Specifically, beams carrying orbital angular momentum (OAM), characterized by their helical phase fronts, offer an ideal basis due to the spatial orthogonality in their mode order, l, in a non-turbulent medium35–37. Interestingly, propagation through atmospheric turbulence corrupts the phase front. This disrupts the orthogonality between distinct OAM modes and induces inter-modal interference, or crosstalk, where power from a transmitted mode, lTX, is distributed into other modes, l ≠ lTX38–40. Our key insight is that the magnitude of modal crosstalk is a direct function of the phase variance induced by the turbulence, and hence, provides a quantitative measure of turbulence strength. This framework transforms the complex problem of measuring random phase aberrations across a beam profile into an incoherent measurement of power distribution across discrete modes. Moreover, this technique offers capabilities beyond path-integrated turbulence estimation. Given that the distribution of atmospheric turbulence is not always homogeneous along the horizontal propagation path (e.g., localized patches of turbulence), we demonstrate that the discrepancy between crosstalk measurements from reciprocal propagation paths (i.e., transmitting THz beams from node A to B vs. B to A) can be used to infer the center location of dominant turbulence clusters along the propagation path. As existing works on turbulence profiling exploit complex, longitudinally structured beams that do not apply in the THz regime due to far-field radiation, our method leverages the natural divergence of THz-OAM beams for turbulence localization. While prior work has viewed OAM crosstalk induced by turbulence as a problem for OAM-based communication systems that needs to be mitigated41–43, here we treat this phenomenon as an opportunity for turbulence sensing. In this work, we establish and validate a comprehensive framework for turbulence sensing with THz-OAM beams. We develop a sensing-oriented analytical model that provides a direct quantitative link between the measured modal crosstalk and the refractive index structure constant (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2). This model is rigorously validated through extensive wave-optics simulations and lab experiments with custom-designed, 3D-printed phase plates to emulate the phase aberrations induced by propagation through atmospheric turbulence. Our experimental results demonstrate a root-mean-square error (RMSE) of 2.22 dB in turbulence characterization across weak to severe turbulence strengths. A comprehensive comparative analysis further highlights the fundamental advantages of the THz modality over optical frequencies, demonstrating its ability to quantify severe turbulence in regimes where an analogous optical system saturates and fails, and its superior resilience to platform misalignment. This work opens possibilities for remote atmospheric diagnostics, environmental monitoring, and the fundamental study of turbulence statistics in the THz regime. The proposed turbulence sensing framework establishes structured THz waves as a robust modality for addressing a critical challenge in aviation safety and climate science. Results Principle and modeling of THz-OAM beam for turbulence sensing To explore the use of THz-OAM beams for atmospheric turbulence sensing, we must first understand the concept behind OAM mode orthogonality and the source of modal crosstalk. The fundamental principle of our sensing modality lies in the interaction between the structured wavefront of THz-OAM beams and the random phase aberrations imposed by atmospheric turbulence, as conceptually illustrated for an airborne link in Fig. 1a. In a non-turbulent medium (i.e., free space), the electric field of an OAM beam with mode l can be expressed in the cylindrical coordinates (r, θ, z) as 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E(r,\theta,z)=A(r,z){e}^{il\theta },$$\end{document}E(r,θ,z)=A(r,z)eilθ, where A(r, z) is the complex amplitude profile while the azimuthal term eilθ defines the helical phase front. To form a measurement basis, we consider a beam with a single transmitted mode lTX. At the receiver, the impinging field is analyzed by projecting it onto a basis of detection modes, denoted as lRX for modal decomposition. In the absence of turbulence, these modes are spatially orthogonal to each other, as the integral over the azimuthal coordinates \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {l}_{RX}| {l}_{TX}\rangle \propto {I}_{\theta }=\int_{0}^{2\pi }{e}^{i({l}_{TX}-{l}_{RX})\theta }d\theta$$\end{document}⟨lRX∣lTX⟩∝Iθ=∫02πei(lTX−lRX)θdθ evaluates to 0 for lTX ≠ lRX. However, when the beam propagates through the atmosphere, turbulent eddies accumulate a random, spatially varying phase aberration that can be represented with Δϕ(r, θ), which corrupts the wavefront and breaks the inherent orthogonality. Hence, the azimuthal integral is no longer guaranteed to be zero even when lTX ≠ lRX: 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{\theta }=\int_{0}^{2\pi }{e}^{i[({l}_{TX}-{l}_{RX})\theta+\Delta \phi (r,\theta )]}d\theta \ne 0.$$\end{document}Iθ=∫02πei[(lTX−lRX)θ+Δϕ(r,θ)]dθ≠0.Fig. 1Overview of the terahertz orbital angular momentum (THz-OAM) atmospheric turbulence sensing modality.a Conceptual illustration of the airborne sensing principle. When a structured THz beam carrying OAM propagates through turbulence, its helical wavefront accumulates random phase aberrations, resulting in a distorted OAM phase profile at the receiver. b As a single-mode THz-OAM beam propagates through a turbulent medium, the wavefront is perturbed, resulting in power leakage (crosstalk) to adjacent modes, which is a direct function of the strength of turbulence. The relative crosstalk (the ratio of power in a crosstalk mode to the power in the transmitted mode) is a monotonic function of turbulence strength (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2), offering a direct basis for quantitative estimation. Overview of the terahertz orbital angular momentum (THz-OAM) atmospheric turbulence sensing modality. a Conceptual illustration of the airborne sensing principle. When a structured THz beam carrying OAM propagates through turbulence, its helical wavefront accumulates random phase aberrations, resulting in a distorted OAM phase profile at the receiver. b As a single-mode THz-OAM beam propagates through a turbulent medium, the wavefront is perturbed, resulting in power leakage (crosstalk) to adjacent modes, which is a direct function of the strength of turbulence. The relative crosstalk (the ratio of power in a crosstalk mode to the power in the transmitted mode) is a monotonic function of turbulence strength (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2), offering a direct basis for quantitative estimation. This non-zero integral signifies the redistribution of power from the transmitted mode lTX into other modes, the magnitude of which is a direct function of the statistical properties of the phase aberration Δϕ(r, θ), offering us an opportunity for quantitative evaluation of turbulence strength. As illustrated in Fig. 1b, increased turbulence strength leads to more distortions in both the amplitude and phase of OAM field profiles, and hence corrupts the mode orthogonality. To model the relationship between crosstalk magnitude and turbulence strength, we first analyze the statistical properties of the integral in Eq. (2). We begin by approximating the turbulence-induced phase aberration, Δϕ(r, θ), as a zero-mean, spatially uncorrelated Gaussian random process with a variance denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{\phi }^{2}$$\end{document}σϕ2. While atmospheric turbulence exhibits known spatial correlations, which we model using the Kolmogorov spectrum20,44, this first-order approximation (i.e., assuming spatially independent Gaussian random phase from turbulence) can still provide important insights into the statistical properties of OAM crosstalk. Importantly, we show in Supplementary Note 1 that the crosstalk model derived with the Gaussian phase assumption exhibits statistics closely matching the results obtained from wave-optics simulations based on Kolmogorov phase screens. Under this assumption, the integral in Eq. (2) becomes a complex random variable whose real and imaginary components are also independent, zero-mean Gaussian random variables. Therefore, the magnitude of the integral representing crosstalk amplitude, by definition, follows the Rayleigh distribution, whose expectation has a direct link to the phase variance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{\phi }^{2}$$\end{document}σϕ2 (see Supplementary Note 1 for derivation): 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbb{E}}[| {I}_{\theta }| ]=\frac{\pi }{\sqrt{2}}\sqrt{1-\exp (-{\sigma }_{\phi }^{2})}.$$\end{document}E[∣Iθ∣]=π21−exp(−σϕ2). Equation (3) clearly shows that the average observed crosstalk increases with higher phase variation of the EM signal caused by stronger turbulence. Next, we establish the quantitative link between the phase variance, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{\phi }^{2}$$\end{document}σϕ2, and the physical turbulence strength, defined by the refractive index structure constant, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2. For a beam with effective diameter D propagating a distance Z through homogeneous and isotropic turbulence, classical turbulence theory derived from the Kolmogorov spectrum provides a direct relationship via the Fried parameter, r0, with454\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{\phi }^{2}\propto {\left(\frac{D}{{r}_{0}}\right)}^{5/3},\quad \,{\mbox{where}}\,\quad {r}_{0}={(0.423{k}^{2}{C}_{n}^{2}Z)}^{-3/5},$$\end{document}σϕ2∝Dr05/3,wherer0=(0.423k2Cn2Z)−3/5, and k = 2π/λ is the wavenumber. This relationship leads to a linear dependence where the phase variance grows monotonically with the refractive index structure constant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2: 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{\phi }^{2}\propto {D}^{5/3}{k}^{2}{C}_{n}^{2}Z.$$\end{document}σϕ2∝D5/3k2Cn2Z. However, this linear model describes the variance in the phase front of a hypothetical beam when it goes through a turbulent medium, and it does not account for phase wrapping. Specifically, in weak turbulence regimes (i.e., r0 > > D), the amount of phase variation is small, and hence phase being wrapped to [−π, π] is irrelevant. In contrast, as turbulence strength increases, this leads to a saturation of the effective phase variance, which the infinitely growing linear model in Eq. (5) fails to capture. It is worth noting that in the present context of aviation hazard avoidance, a model that remains robust in the moderate-to-strong turbulence regime (i.e., having r0 comparable or smaller than D) is necessary. To construct an end-to-end analytical model valid across the wide range of turbulence strengths, particularly for the moderate-to-strong turbulence regime, we incorporate an exponential saturation term. This framework mathematically bridges the commonly accepted linear growth of the variance of phase aberration from propagation through weak turbulence with the saturation behavior under strong turbulence (see Supplementary Note 2 for derivation): 6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{\phi }^{2}=\frac{{\pi }^{2}}{3}(1-\exp (-\alpha {C}_{n}^{2})),$$\end{document}σϕ2=π23(1−exp(−αCn2)), where α is a scaling factor that captures the physical propagation parameters such as beam diameter, wavelength, and propagation distance from Eq. (5). By substituting the phase variance model in Eq. (6) into the crosstalk expression from Eq. (3), we derive our final, end-to-end analytical model for the crosstalk amplitude ratio between lTX and lRX, denoted as Rturb: 7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{turb}({l}_{TX},{l}_{RX})=\frac{\beta \pi }{\sqrt{2}}\sqrt{1-\exp \left(-\frac{{\pi }^{2}}{3}\left(1-\exp (-\alpha {C}_{n}^{2})\right)\right)},$$\end{document}Rturb(lTX,lRX)=βπ21−exp−π231−exp(−αCn2), where β is a secondary scaling coefficient that accounts for the radial overlap integral between the specific transmitted and received modes. Although there exist closed-form expressions for OAM modal crosstalk derived in prior works46, they typically assume a constant beam diameter along the propagation path. While valid for optical beams, this assumption does not hold at THz frequencies due to the divergence of THz-OAM beams. Furthermore, prior models often rely on low-order Zernike polynomial approximations of the wavefront as well as linear phase variance growth with increasing turbulence strength, which are insufficient to capture the complex aberrations induced and the phase wrapping effect27,47, especially by strong turbulence. The more generalized framework we present in this work addresses these limitations. While our derivation also begins with a simplifying Gaussian phase assumption to establish the link between crosstalk magnitude and phase variance in Eq. (7), the final model’s robustness is not limited by this initial step. Instead, we treat α and β as semi-empirical calibrating parameters to avoid fixed assumptions about the wavefront aberration and more accurately represent the physical propagation effect of THz-OAM beams in turbulence. This approach implicitly accounts for the path-dependent divergence of THz beams and avoids fixed assumptions about wavefront aberration. Although a first-principles closed-form derivation for α and β is also possible (see Supplementary Note 3), calibrating them against wave-optics simulations ensures superior model accuracy across diverse turbulence regimes (see Supplementary Note 4). Our modeling provides an end-to-end quantitative relationship for the sensing pipeline illustrated in Fig. 1b, connecting the measurable crosstalk directly to the physical turbulence strength. Numerical validation and performance analysis To validate our analytical model and quantify the performance advantages of turbulence sensing at THz frequencies compared to the optical regime, we conduct a series of wave-optics simulations. Our simulations are based on a widely used split-step beam propagation method, where the turbulence-induced refractive fluctuations are represented by a sequence of random phase screens, and the field propagation between adjacent screens is treated as free-space diffraction calculated with the Rayleigh–Sommerfeld diffraction integral48–51. We simulate a 5-km atmospheric path divided into 20 equally spaced discrete segments, each representing a 250-m propagation distance. The first segment is modeled as free-space propagation, allowing the beam to develop from the source. At the beginning of each of the subsequent segments, we apply a random phase screen that follows Kolmogorov turbulence statistics to the beam’s complex electric field. We obtain the final beam profile after sequentially propagating through each segment. We consider a 50-cm transmitting aperture and a receiver aperture that is sufficiently large to capture the expanded OAM beam after propagation. At the receiver, the distorted field is decomposed via an inner product with an ideal OAM basis to calculate the power distribution across different analyzed modes. Our analytical model in Eq. (7) specifically quantifies the interference between any transmitted mode lTX and received mode lRX as a measurable crosstalk magnitude Rturb(lTX, lRX). Therefore, to validate the model, our analysis focuses on quantifying the power leakage between specific (lTX, lRX) pairs. To account for the randomness of each generated turbulence phase screen and ensure statistically robust results, we perform a Monte Carlo analysis by averaging the resulting crosstalk amplitude ratios over 100 independent repetitions under each simulated environment. First, we validate the analytical model presented in Eq. (7) by calibrating it against simulation data at 300 GHz. This frequency was selected as a representative value for the THz regime (corresponding to a 1 mm wavelength) that falls within an atmospheric transmission window. The modal power spectrum for OAM beams propagating through turbulence typically follows a normal distribution, with power decaying exponentially away from the transmitted mode lTX46,52. In Fig. 2a, we show a few example heatmaps of simulated OAM power spectra as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2. These heatmaps visually confirm that for any given \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 value, the received power is centered at lTX and decays toward adjacent modes. In addition, the power distribution systematically broadens with increasing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2, signifying greater crosstalk. For a given transmitted and received mode pair (lTX, lRX), we implement a non-linear least-squares fit, leveraging the calibrating parameters α and β derived from first-principles physical theory as an initial guess (see Supplementary Note 3). As shown in Fig. 2b, the calibrated model exhibits excellent agreement with the simulated data points across a wide range of turbulence strengths (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 ranging from 10−15 to 10−12m−2/3), a regime that specifically describes moderate-to-strong atmospheric turbulence. The calibrated model yields a coefficient of determination (R2) greater than 0.98, and a mean percentage error of below 5%. We further highlight the robustness of this model by showing its consistent performance across different example (lTX, lRX) pairs in Fig. 2b, proving that our model can accurately predict power leakage from any transmitted mode to various non-primary modes. Finally, we establish the model’s generalizability in Supplementary Note 5, where we demonstrate that a single set of fitting parameters can be analytically transformed to predict crosstalk under different physical scenarios (e.g., path length, beam size, etc.). This is practically important since factors such as beam divergence over varying distances or platform constraints may limit the use of certain high-order modes, requiring the system to flexibly switch mode pairs to maintain reliable turbulence estimation without the need for frequent recalibration.Fig. 2Simulation validation of the analytical model.a Heatmaps of simulated Orbital Angular Momentum (OAM) power spectrum as a function of turbulence strength (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2) for three different transmitted and received mode pairs. We consider a 50-cm transmitter aperture propagating across a 5-km path. Power leakage from the transmitted mode increases as the turbulence becomes stronger. b Crosstalk amplitude ratio vs. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 for the crosstalk channels highlighted in (a). The calibrated analytical model (red) exhibits excellent agreement with the simulation data (blue), achieving a mean model error below 5%. Simulation validation of the analytical model. a Heatmaps of simulated Orbital Angular Momentum (OAM) power spectrum as a function of turbulence strength (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2) for three different transmitted and received mode pairs. We consider a 50-cm transmitter aperture propagating across a 5-km path. Power leakage from the transmitted mode increases as the turbulence becomes stronger. b Crosstalk amplitude ratio vs. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 for the crosstalk channels highlighted in (a). The calibrated analytical model (red) exhibits excellent agreement with the simulation data (blue), achieving a mean model error below 5%. Next, we conduct a performance comparison between our THz-OAM framework for turbulence sensing and an analogous system operating at optical wavelengths (e.g., 1550 nm). We first compare the operational dynamic range of the two systems. For this analysis, we focus on the crosstalk between lTX = 1 and lRX = 0 as a representative example (see Supplementary Note 6 for the impact of mode selection). While our model can be generalized to other mode pairs, we discuss in Supplementary Note 6 that using a single, high-crosstalk neighboring mode pair (e.g., ∣lTX − lRX∣ = 1 in this case) is sufficient for turbulence strength estimation, and including additional lRX modes does not improve estimation accuracy. It is well established that saturation results in the failure of optical systems in strong turbulence53,54. This failure originates from the strong wavelength dependence of the phase variance (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{\phi }^{2}\propto {k}^{2}$$\end{document}σϕ2∝k2) as shown in Eq. (5), which causes the optical wavefront to become completely random under strong turbulence. Our crosstalk simulations for (lTX = 1, lRX = 0) confirm this intuition. As shown in Fig. 3a1, the normalized crosstalk for an OAM-based turbulence estimation system at 1550 nm saturates and becomes non-monotonic for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2} > 1{0}^{-15}\,{{{{\rm{m}}}}}^{-2/3}$$\end{document}Cn2>10−15m−2/3, which creates ambiguity in turbulence inference. The THz system in Fig. 3a2, on the other hand, does not suffer from this limitation over the simulated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 range up to 10−11m−2/3 given the same transmitting aperture size. We consider the same aperture size to offer a direct comparison that isolates wavelength-dependent effects. We note that using a more realistic, smaller optical aperture would cause saturation to occur at even weaker turbulence strengths due to increased beam divergence. This difference is further illustrated in Fig. 3b, which plots the model fitting error for both systems as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2. The estimation error for optical frequencies remains low in the weak turbulence regime, but increases rapidly as it enters saturation. In contrast, the THz system maintains low and stable error across all simulated turbulence strengths. We note that practical limitations, such as system noise or imperfect OAM mode generation, may degrade the THz system’s sensitivity in the very weak turbulence regime. However, as our system is designed for the detection of hazardous, severe turbulence, the THz modality’s robust performance in this regime becomes an advantage.Fig. 3Performance comparison between optical (1550 nm) and terahertz (300 GHz) systems.a Comparison of the operational dynamic range for systems at Terahertz (THz) and optical frequencies. The optical system’s crosstalk response saturates for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2} > 1{0}^{-15}$$\end{document}Cn2>10−15 m−2/3. In contrast, the THz response remains monotonic over the entire simulated range. b Normalized model fitting error for both systems. The error for the optical system diverges in the saturation regime, while the THz system’s error remains consistently low. c The effect of a 2.5 μrad angular misalignment on the simulated crosstalk ratio for optical (left) and THz (right) systems. The misalignment induces a significant deviation from the calibrated model in the optical system, while the effect on the THz system is negligible. Such a small tilt completely corrupts the crosstalk behavior for optical systems. Performance comparison between optical (1550 nm) and terahertz (300 GHz) systems. a Comparison of the operational dynamic range for systems at Terahertz (THz) and optical frequencies. The optical system’s crosstalk response saturates for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2} > 1{0}^{-15}$$\end{document}Cn2>10−15 m−2/3. In contrast, the THz response remains monotonic over the entire simulated range. b Normalized model fitting error for both systems. The error for the optical system diverges in the saturation regime, while the THz system’s error remains consistently low. c The effect of a 2.5 μrad angular misalignment on the simulated crosstalk ratio for optical (left) and THz (right) systems. The misalignment induces a significant deviation from the calibrated model in the optical system, while the effect on the THz system is negligible. Such a small tilt completely corrupts the crosstalk behavior for optical systems. In addition to the operating turbulence strength range, another critical performance metric for a system intended for airborne platforms is its resilience to angular misalignment, a factor often overlooked in prior studies that assume perfectly aligned systems55,56. For lateral misalignment, as long as the beam remains within the receiver aperture, the effect is largely independent of wavelength. In contrast, the system’s response to angular misalignment (i.e., tilt) reveals a significant frequency dependence. The fundamental challenge here is that an angular tilt can distort the wavefront, leading to high modal crosstalk that is indistinguishable from turbulence-induced effects. Therefore, such misalignment can severely compromise the accuracy of turbulence sensing from the OAM cross-mode interference patterns. We model the effect of a small angular tilt by applying a linear phase ramp across the beam’s wavefront. At the receiver plane, for misalignment angles θx and θy in the Cartesian coordinates, the received field is modified by adding a new phase term to the ideal electric field profile E(x, y, z) as 8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{tilt}(x,y,z)=E(x,y,z)\exp (ik(x\sin {\theta }_{x}+y\sin {\theta }_{y})).$$\end{document}Etilt(x,y,z)=E(x,y,z)exp(ik(xsinθx+ysinθy)). Note that this tilt-induced phase ramp is directly proportional to the wavenumber k. We simulate the effect of angular tilt on both a 300 GHz THz system and a 1550 nm optical system, assuming the same physical propagation parameters. To ensure a fair comparison, the turbulence regime chosen in Fig. 3c corresponds to comparable Fried parameter (r0) and crosstalk magnitudes, rather than identical \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 values, as the optical system enters the strong saturation regime at turbulence levels that are still weak for the THz systems according to our previous discussion. As shown, angular tilts as small as 2.5 μrad induce significant deviations of the simulated crosstalk from the calibrated model in the optical system that completely overwhelm the turbulence-induced signatures. This artifact is so large that it yields crosstalk magnitudes exceeding the theoretical maximum under a perfectly aligned system, causing a complete breakdown of the sensing framework. The THz system, in contrast, exhibits a negligible impact from the same amount of angular misalignment. While the absolute tilt tolerance necessitates an active stabilization and misalignment mitigation system for any airborne turbulence detection system, the THz system provides an angular error budget much larger than that of the optical system. This significantly relaxes the requirement for real-time adaptive alignment and proves that the THz-OAM modality offers potential for a more robust turbulence sensing system with nodal mobility. Furthermore, while optical systems typically rely on micro-electromechanical mirrors for angular correction, THz systems can leverage fast electronic beamforming via phased arrays or metasurfaces. In addition, lateral offsets between the transmit beam axis and the receiver aperture can also introduce additional modal crosstalk. We evaluate the impact of lateral nodal misalignment in Supplementary Note 8. Importantly, active beam tracking and beamforming strategies can be implemented to maintain both lateral and angular misalignment57–60. Experimental validation To provide a proof-of-concept validation of our sensing modality, we conduct lab-scale experiments using emulated turbulence phase screens (see Methods and Supplementary Note 7 for more details). A general schematic of our setup is shown in Fig. 4a. We generate OAM-carrying beams at 300 GHz using a 3D-printed spiral phase plate configured at lTX = 1 with an aperture of 3.81 cm. To emulate the effect of propagation through turbulence, we introduce two custom-designed, 3D-printed phase plates that impose phase aberrations following Kolmogorov statistics. We place the two plates at 0.1 m and 0.35 m, respectively, along a 0.6-m propagation path. We first aim to directly characterize the effect of these phase screens on the beam’s complex field. We fabricate 10 sets of turbulence phase plates, corresponding to a Fried parameter, r0, ranging from 28 cm down to 1.76 cm (covering weak to strong turbulence strength with D/r0 ∈ [0.21, 3.26] when normalized to the beam diameter D). We first mount the receiver on a 2D translational stage and use it as a coherent field probe, performing a raster scan to capture the electric field intensity and phase across the beam profile. As illustrated in the field heatmaps in Fig. 4b, the helical wavefront and the donut-shaped intensity profile are progressively corrupted as turbulence strength increases. Subsequently, we numerically decompose the measured field data into different OAM modes, which quantitatively confirms that the power leakage from transmitted mode lTX = 1 into adjacent modes becomes progressively stronger with smaller r0. The minor modal impurities and profile distortion visible in the baseline case without turbulence are expected experimental artifacts as detailed in Supplementary Note 7. Overall, this result validates the correlation between turbulence strength and modal crosstalk.Fig. 4Experimental evaluation of terahertz orbital angular momentum (THz-OAM) turbulence sensing.a Lab-scale turbulence emulation setup. A THz-OAM beam is generated with a spiral phase plate and propagates through two 3D-printed phase plates with Kolmogorov statistics. A final phase plate at the receiver performs mode-selective projection for power measurement. To ensure statistically robust results, we average over 100 turbulence realizations, achieved by randomly rotating the two turbulence phase plates. b Measured OAM field profiles and their corresponding numerical mode decomposition at different turbulence strengths (r0). As the turbulence becomes more severe, the intensity and phase profiles exhibit increasing distortion, leading to an increase in modal crosstalk. c Measured crosstalk as a function of r0. The experimental data show excellent agreement with both the simulation and the calibrated model. d The mean estimation error across the tested range. The experimental results demonstrate robust performance, achieving an overall root mean square error of 2.22 dB. The error remains low across the moderate-to-strong turbulence regime, and only increases for very weak turbulence, where the small expected crosstalk becomes prone to system noise and experimental non-idealities. Blue shaded regions indicate the maximum and minimum bounds. e Histograms of the measured signal amplitude for the transmitted mode (lTX = 1) and crosstalk mode (lRX = 0) under weak (r0 = 28 cm) and strong (r0 = 1.8 cm) turbulence. In strong turbulence, the distributions broaden and overlap, leading to potentially higher errors in turbulence estimation. Experimental evaluation of terahertz orbital angular momentum (THz-OAM) turbulence sensing. a Lab-scale turbulence emulation setup. A THz-OAM beam is generated with a spiral phase plate and propagates through two 3D-printed phase plates with Kolmogorov statistics. A final phase plate at the receiver performs mode-selective projection for power measurement. To ensure statistically robust results, we average over 100 turbulence realizations, achieved by randomly rotating the two turbulence phase plates. b Measured OAM field profiles and their corresponding numerical mode decomposition at different turbulence strengths (r0). As the turbulence becomes more severe, the intensity and phase profiles exhibit increasing distortion, leading to an increase in modal crosstalk. c Measured crosstalk as a function of r0. The experimental data show excellent agreement with both the simulation and the calibrated model. d The mean estimation error across the tested range. The experimental results demonstrate robust performance, achieving an overall root mean square error of 2.22 dB. The error remains low across the moderate-to-strong turbulence regime, and only increases for very weak turbulence, where the small expected crosstalk becomes prone to system noise and experimental non-idealities. Blue shaded regions indicate the maximum and minimum bounds. e Histograms of the measured signal amplitude for the transmitted mode (lTX = 1) and crosstalk mode (lRX = 0) under weak (r0 = 28 cm) and strong (r0 = 1.8 cm) turbulence. In strong turbulence, the distributions broaden and overlap, leading to potentially higher errors in turbulence estimation. While direct field profiling is ideal for crosstalk measurement, a practical sensor at THz wavelengths requires a more direct power measurement and cannot rely on the slow raster scans. To validate our turbulence estimation modality, we implement a mode-selective detection scheme. We custom-designed receiver phase plates that perform a conjugate phase transformation, converting a specific OAM mode into a Gaussian beam that can be focused into the receiver for incoherent power measurement. Importantly, without relying on the reconstruction of the full received field, this measurement scheme reduces the overhead and computational complexity in conventional optical OAM sensing. Since our framework targets the measurement of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2, which is a statistical parameter that is typically considered stationary over second-to-minute scales61–63, the rapid processing from our incoherent measurement scheme allows for acquiring more independent turbulence response samples within a short time window. This ensures a robust and real-time statistical estimation of dynamic turbulence responses. Although our previous analysis is generalized to any arbitrary (lTX, lRX) pairs, as discussed in Supplementary Notes 6 and 7, we select a single receiving mode lRX = 0 adjacent to lTX = 1 with the highest crosstalk magnitude to maximize the signal-to-noise ratio of the crosstalk measurement. We designed two corresponding receiving phase plates to measure the power remaining in the transmitted mode and the power redistributed to the adjacent mode, respectively. To generate robust statistics for each turbulence strength, we emulate 100 independent turbulence realizations by randomly rotating each of the phase plates relative to each other. The received amplitudes at lRX = 1 and lRX = 0 are recorded for each turbulence realization, and the average measured signal amplitude between the two modes is calculated as the normalized crosstalk. As shown in Fig. 4c, the experimental data show excellent agreement with both wave-optics simulations and the calibrated analytical model from Eq. (7). Furthermore, we show in Fig. 4d that the mean estimation error for the experimental data remains low across the entire tested range, achieving an overall RMSE of 2.22 dB, while that for the simulation data is 0.64 dB. Interestingly, the measured error is not uniformly distributed across the sensing range, but rather dominated by the error spike in the weakest turbulence regime, which is an expected artifact from this framework. Our THz-OAM-based modality senses turbulence by measuring a crosstalk signal that approaches zero as turbulence vanishes. In this regime, the turbulence-induced crosstalk is comparable to, or even smaller than, a static background crosstalk floor caused by experimental nonidealities (e.g., minor 3D printing fabrication error, imperfect alignment, etc.). A discussion of the accuracy of phase plate fabrication and the impact of an imperfect experimental setup is provided in Supplementary Note 7. As turbulence strength increases, the desired crosstalk signal quickly grows to overwhelm this artifact floor, where the estimation error drops significantly. Because our sensing framework targets the moderate-to-strong turbulence regime where modal crosstalk dominates the hardware imperfections, the static modal impurities have a negligible impact on the overall sensing robustness. In addition, the simulation results predict an increase in the variance of the estimation error (shaded region) in the stronger turbulence regime. We investigate this phenomenon by plotting the measured amplitude distributions in Fig. 4e. Under weak turbulence (r0 = 28 cm), the signal distributions for the transmitted and crosstalk modes are narrow and well-separated. However, under strong turbulence (r0 = 1.8 cm), the distributions broaden and overlap significantly. This higher statistical variance in power measurements due to increased random phase fluctuation explains the higher variance in turbulence estimation, which can potentially be improved with increased averaging. The agreement between experiment, simulation, and theory in our results validates our core sensing principle. While these measurements provide validation of using THz-OAM beams for turbulence sensing, an optical benchmark under the same physical turbulence conditions, aperture size, and propagation distance is not practical in our experimental platform as a direct comparison. However, the saturation of crosstalk of an OAM optical link and the extreme sensitivity to angular misalignment have been explored in the literature46,64. Instead, we conduct a supplementary experiment that evaluates the improved tolerance of OAM modal crosstalk to angular misalignment for our THz systems, as detailed in Supplementary Note 8. While our lab-scale setup utilizes complex benchtop equipment as a proof-of-concept (details discussed in “Methods” and Supplementary Note 7), practical deployment on mobile nodes requires significantly more compact hardware. Since our proposed framework only requires incoherent power measurements, wideband optical sources and physical delay lines connecting the transmitter and receiver are not required. Instead, a more compact system can be implemented using solid-state THz transceivers (e.g., frequency multipliers based on Schottky diodes)65. To realize OAM generation and reception, practical systems may use reconfigurable beamforming hardware, such as analog beamforming arrays or reflect arrays66,67. The availability of such devices allows for practical deployment on standard commercial aircraft. Future distributed or synthetic aperture implementations may further reduce the hardware complexity of large-aperture OAM transmitters and receivers at THz68. Localizing the center of turbulence with THz-OAM beams In previous discussions, we assumed a statistically uniform turbulent medium, for which our method provides a path-integrated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2. In many practical scenarios, however, the turbulence distribution can be inhomogeneous, consisting of localized, dominant clusters69,70. Under these scenarios, a simple path-integrated average is insufficient as it does not offer insights into advanced warning, such as guiding pilots to detour from a strong turbulent patch. Therefore, it is highly desired to localize the center of the dominant turbulence. Although prior work on turbulence localization in the optical regime exists, it typically requires complex systems, such as using pairs of crossed beams that require large and bulky receiving apertures71, or exploiting longitudinally structured beams that cannot be generated at long distances at THz frequencies due to far-field approximations56,72. We demonstrate a method of localizing the center of a dominant turbulent cluster along a horizontal propagation path by leveraging the natural divergence of THz-OAM beams. Specifically, the underlying principle is that the magnitude of turbulence-induced crosstalk is weighted by the beam’s effective diameter, D(z), at location z along the propagation path. To exploit this, as shown in Fig. 5a, we consider two reciprocal propagation paths between two transceivers: one forward path transmitting from node A (position z = 0) and receiving at node B (position z = Z), and a reciprocal backward path from node B to node A. As illustrated, the THz-OAM beam diameter naturally increases with propagation distance (see Supplementary Note 10 for the full relationship). Assuming identical transceivers, the beam diameter is a function of the distance propagated from its source. Therefore, for the forward path (A → B), a turbulent cluster at location z interacts with a beam that has propagated a distance z, while for the reciprocal backward path (B → A), the same cluster interacts with a beam that has propagated a distance Z − z. This asymmetry is most pronounced for turbulence clusters near either node. For example, a turbulent cluster near node B (i.e., at z ≈ Z) interacts with a wide forward beam (which has propagated nearly the full distance Z) but a narrow backward beam (which has just started propagating). This results in stronger crosstalk for the forward propagating path Rturb,FW(lTX, lRX) than that of the backward propagation Rturb,BW(lTX, lRX) between the same transmitted and received mode pair. Interestingly, the ratio of the measured crosstalk from the forward and backward paths Rturb,FW/Rturb,BW can be used to infer the turbulent cluster’s center location.Fig. 5Turbulence localization through forward-backward crosstalk asymmetry.a Overview of turbulence localization framework. The natural divergence of a terahertz orbital angular momentum (THz-OAM) beam causes it to interact differently with a localized turbulent medium on the forward path vs. on the backward path. This results in an asymmetry in the measured crosstalk, which provides a direct mapping to the center location of the turbulent layer. b Simulation results showing that the crosstalk ratio is a robust metric for localization, as it is largely independent of the turbulence strength (top) and scale (bottom). c Experimental validation of the localization principle. The measured forward-backward (FW-BW) crosstalk ratio as a function of the position of the turbulence phase plate shows a clear monotonic trend in agreement with simulation and theoretical prediction from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n,eff}^{2}$$\end{document}Cn,eff2 calculation. d Characterization of the operational regime for localization. We show that as background turbulence strength increases, the medium becomes effectively homogeneous, causing the measured asymmetry to converge, which defines an operational boundary for turbulence localization. Blue shaded regions indicate the maximum and minimum bounds. Turbulence localization through forward-backward crosstalk asymmetry. a Overview of turbulence localization framework. The natural divergence of a terahertz orbital angular momentum (THz-OAM) beam causes it to interact differently with a localized turbulent medium on the forward path vs. on the backward path. This results in an asymmetry in the measured crosstalk, which provides a direct mapping to the center location of the turbulent layer. b Simulation results showing that the crosstalk ratio is a robust metric for localization, as it is largely independent of the turbulence strength (top) and scale (bottom). c Experimental validation of the localization principle. The measured forward-backward (FW-BW) crosstalk ratio as a function of the position of the turbulence phase plate shows a clear monotonic trend in agreement with simulation and theoretical prediction from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n,eff}^{2}$$\end{document}Cn,eff2 calculation. d Characterization of the operational regime for localization. We show that as background turbulence strength increases, the medium becomes effectively homogeneous, causing the measured asymmetry to converge, which defines an operational boundary for turbulence localization. Blue shaded regions indicate the maximum and minimum bounds. We consider a common scenario where a single, dominant chunk of turbulent mass of finite scale is located along the propagation path between two airborne nodes at approximately the same height with a relatively calm background. To model its connection to forward-backward crosstalk asymmetry, we first define an effective turbulence strength, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n,eff}^{2}$$\end{document}Cn,eff2, that accounts for the total path-integrated turbulence effect weighted by the local beam size. This parameter accounts for the distance-dependent beam size D(z) as well as the turbulence strength profile \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}(z)$$\end{document}Cn2(z) to quantify the overall phase aberration experienced along a propagation path with inhomogeneous turbulence distribution. Building upon the relationship in Eq. (4), the total phase variance is the integral over the entire propagation path. Hence, we approximate the effective turbulence strength as 9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n,eff}^{2}=\frac{\int_{0}^{Z}{C}_{n}^{2}(z)D{(z)}^{5/3}dz}{\int_{0}^{Z}D{(z)}^{5/3}dz}.$$\end{document}Cn,eff2=∫0ZCn2(z)D(z)5/3dz∫0ZD(z)5/3dz. Consider a single turbulence cluster with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n,cluster}^{2}$$\end{document}Cn,cluster2 starting at z = Z1 and lasting until z = Z2. As we primarily target local, dominant turbulence localization, if we assume this cluster to be relatively thin (i.e., ΔZ = Z2 − Z1 being small), we can approximate that the integral becomes proportional to the beam diameter at the center of the turbulent cluster zc = (Z1 + Z2)/2. Specifically, the numerator of Eq. (9) becomes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n,cluster}^{2}\Delta Z{D}_{FW}{({z}_{c})}^{5/3}$$\end{document}Cn,cluster2ΔZDFW(zc)5/3 for the forward path. For an identical pair of transceivers at node A and B, the beam diameter profiles for the forward and backward propagating paths can also be considered reciprocal, described by the relation DBW(z) = DFW(Z − z). Therefore, the numerator of Eq. (9) for the backward path can be expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n,cluster}^{2}\Delta Z{D}_{FW}{(Z-{z}_{c})}^{5/3}$$\end{document}Cn,cluster2ΔZDFW(Z−zc)5/3. This approximation reveals that when we compare the forward and backward effective turbulence strengths, the ratio simplifies to a function of only the beam diameter evaluated at the center of the turbulent medium: 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{C}_{n,eff,FW}^{2}}{{C}_{n,eff,BW}^{2}}\approx \frac{{D}_{FW}{({z}_{c})}^{5/3}}{{D}_{FW}{(Z-{z}_{c})}^{5/3}}.$$\end{document}Cn,eff,FW2Cn,eff,BW2≈DFW(zc)5/3DFW(Z−zc)5/3. This difference in effective \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 translates directly into a difference in measured crosstalk. Since the beam diameter DFW(z) under a given aperture, transmitted OAM mode, and frequency is a monotonic function of distance, the ratio between crosstalk at the forward and backward propagation paths, Rturb,FW(lTX, lRX)/Rturb,BW (lTX, lRX), is uniquely dependent on only the cluster’s center location zc. For robust localization, the forward-backward crosstalk ratio needs to be only a function of the center of the turbulent cluster. In Fig. 5b, we show empirically that the crosstalk ratio at a fixed turbulence location is largely independent of its specific strength or thickness, and therefore confirms that it serves as a robust metric for turbulence localization. Next, we experimentally validate our turbulence localization framework. We divide the propagation path from 0.1 m to 0.6 m after a 0.1-m free-space propagation of the THz-OAM beam into five equal segments. We sequentially place a single phase plate with an effective r0 = 5.3 cm at each segment to emulate a local turbulent cluster. For each of the five locations, we measure the crosstalk magnitude between lTX = 1 and lRX = 0 for both the forward and the backward propagating paths. We then calculate the crosstalk ratio Rturb,FW/Rturb,BW between the forward and backward paths for turbulence localization. The results, shown in Fig. 5c, exhibit a clear monotonic trend in agreement with our simulation and theoretical prediction, confirming that the crosstalk discrepancy serves as a robust metric for localizing inhomogeneous turbulence. Importantly, our model theoretically assumes a dominant turbulent cluster on a calm background. In practice, however, the atmosphere always has a non-zero background turbulence. Therefore, we investigate the performance when a dominant cluster is superimposed on this background. As shown in Fig. 5d, when the background turbulence becomes comparable in strength to the dominant cluster, the medium begins to behave as a homogeneous path. Consequently, the forward-backward asymmetry diminishes, and the crosstalk ratio converges to 1. This result offers an insight into the sensitivity threshold of the inhomogeneity required for effective localization. It is also worth noting that our proposed turbulence localization method is not applicable to equivalent optical systems because of their quasi-collimated beams that exhibit negligible beam divergence over comparable distances. Thus, rather than treating it as a limitation, we leverage the natural divergence of THz-OAM beams, transforming a perceived disadvantage into a powerful sensing technique for localizing dominant turbulent clusters (see Supplementary Note 10). While this technique primarily determines the center location of the turbulent cluster rather than its full profile, such information offers important insight for hazardous turbulence avoidance. Further extensions to jointly determine the location and scale of turbulence are also possible (see Supplementary Note 11). Monitoring water vapor concentration Additionally, the THz regime offers potential for simultaneously monitoring water vapor concentration leveraging THz-specific molecular absorption lines. The total refractive index structure constant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 is comprised of contributions from temperature fluctuations, usually referred to as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{T}^{2}$$\end{document}CT2, and humidity fluctuations, denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{q}^{2}$$\end{document}Cq273. Because optical sensors are largely transparent to water vapor, they are primarily sensitive to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{T}^{2}$$\end{document}CT2. The THz regime, however, can disambiguate these contributions. A dual-frequency THz system can employ a primary OAM beam at a transparent frequency with low water absorption (e.g., 300 GHz) to measure the total \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 via crosstalk. Simultaneously, a second probe beam at a stronger water vapor absorption line (e.g., 325 GHz) would experience attenuation directly proportional to the path-integrated water vapor density (see Supplementary Note 12). This information provides opportunities in identifying regions of high humidity and potential convective instability, which are known precursors of severe turbulence, offering a potential for turbulence prediction long before being detectable by weather radar or other turbulence sensors74,75. Discussion In this study, we establish a comprehensive framework for atmospheric turbulence sensing using structured THz beams carrying OAM. Our approach is based on the principle that random phase aberrations introduced by the refractive index fluctuations in turbulence corrupt the THz-OAM wavefront, and, therefore, compromise the orthogonality between independent OAM modes. We develop a physics-guided analytical model that establishes a direct quantitative link between OAM modal crosstalk and turbulence strength, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2, and rigorously validate this model through both wave-optics simulations and lab experiments with emulated turbulence phase screens. Our experimental results demonstrate an accurate turbulence strength estimation with an RMSE of 2.22 dB across a wide range of turbulence conditions. In addition, we comprehensively analyze the fundamental advantages of the THz modality over conventional optical systems. We show that THz-OAM sensing provides an extended dynamic range for turbulence estimation under severe conditions relevant to aviation safety, and demonstrates improved resilience to the angular misalignments common to airborne platforms. Furthermore, we demonstrate through both simulations and experiments that the distinctive characteristics of THz-OAM radiation enable capabilities extending beyond conventional path-integrated turbulence estimation. By leveraging the natural divergence of THz-OAM beams, we show that asymmetries in the forward and return OAM crosstalk patterns can be analyzed to pinpoint the location of the dominant turbulent cluster within an inhomogeneous medium. Interestingly, we also establish that the THz-based turbulence sensing can enable atmospheric water vapor monitoring, which potentially helps turbulence prediction that guides preventative measures. This work reframes crosstalk in THz-OAM beams from a communication impairment into a powerful sensing mechanism, establishing a robust platform for real-time atmospheric turbulence characterization and aviation hazard avoidance. Methods Experimental setup All experimental measurements were performed with a scaled table-top turbulence emulation setup, as shown in Supplementary Note 7. For THz signal generation and detection, we use the T-Ray 5000 THz time-domain spectroscopy system by TeraMetrix. The transmitter and receiver modules are each equipped with a fiber-coupled photoconductive antenna that is excited by ultra-short pulses, providing raw time-domain measurements with a flat frequency response between 100 and 400 GHz. The THz transmitter generates a Gaussian beam that is modulated by a spiral phase plate designed for lTX = 1 to produce a beam carrying OAM for turbulence sensing. The THz-OAM beam propagates over a 0.6-m path, which includes two emulated turbulence phase screens equally dividing the space between 0.1 m and 0.6 m. For initial OAM field characterization, the receiver is mounted on a 2D translational stage and operates as a coherent field probe, performing a raster scan to sample the complex electric field across the beam profile. Although this method allows for full numerical modal decomposition, it is not scalable for a practical THz sensor. Therefore, for the primary turbulence estimation validation, we implement a mode-selective incoherent detection scheme. We design and fabricate two receiver phase plates based on Fresnel optics that combine the functions of OAM mode conjugation and focusing. These plates are designed to convert a specific incoming OAM mode (lTX = 1 for the transmitted mode and lRX = 0 for the adjacent crosstalk mode) into a Gaussian beam, which is then focused onto the broadband THz detector for power measurement. Due to power limitations, our measurements have been limited to a table-top scale. However, this experimental setup is physically equivalent to a much longer propagation path through the atmosphere, a principle established by scaling via the Fresnel number, F = a2/(λZ), where a is the transmitting aperture size and Z is the total path length. For instance, our lab setup, which uses a 3.81 cm transmitter aperture over a 0.6 m path, can be scaled to a 5-km propagation path with an equivalent aperture of approximately 3.48 m in diameter. Under this scenario, the emulated turbulence strengths in our experiment correspond to an equivalent \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{n}^{2}$$\end{document}Cn2 range between 2 × 10−14 and 2 × 10−12m−2/3, representative of typical moderate-to-strong atmospheric turbulence conditions. Turbulence phase plate fabrication To emulate the phase aberrations induced by atmospheric turbulence in a controlled lab environment, we design and fabricate phase plates based on the standard Fourier transform method for generating random phase screens with Kolmogorov statistics. Specifically, we first generate a matrix of complex random numbers with Gaussian statistics in the spatial frequency domain. This is then filtered by multiplying it by the square root of the Kolmogorov power spectral density function, expressed as 11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Phi (\kappa )=0.023{r}_{0}^{-5/3}{\kappa }^{-11/3},$$\end{document}Φ(κ)=0.023r0−5/3κ−11/3, where r0 is the Fried parameter and κ is the angular spatial frequency. An inverse Fourier transform of this filtered spectrum produces a complex field whose phase, ϕ(x, y), represents a single spatial realization of atmospheric turbulence. This continuous 2D phase map is then converted into a physical height profile, h(x, y), for 3D printing using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(x,y)=\frac{\phi (x,y)}{2\pi }\cdot \frac{\lambda }{n-1}$$\end{document}h(x,y)=ϕ(x,y)2π⋅λn−1, where n is the refractive index of the plate’s material. The final structure is fabricated from polylactic acid (PLA), which has a refractive index of approximately 1.62 at 300 GHz, using a standard office 3D printer. Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file