About

The Alipour Research Group at the University of Illinois Urbana-Champaign specializes in advancing Digital Twins that combine sensing, computing, and visualization to create smart and resilient natural and built environments. We achieve this by developing innovative remote sensing and other technologies.

Research topics

• Computer Science
• Machine Learning
• Artificial Intelligence
• Data Mining
• Physics
• Geology
• Data science
• Systems engineering
• Electrical engineering
• Database
• Engineering
• Algorithm
• Telecommunications
• Acoustics
• Optics
• Remote sensing

Selected publications

• GreenScatter: Through-Canopy Soil Moisture Sensing with UAV-Mounted Radar

  arXiv (Cornell University) · 2026-04-10

  preprintOpen access

  Soil moisture is a critical variable for managing irrigation, improving crop yield, and understanding field-scale hydrology. Radars mounted on unmanned aerial vehicles (UAVs) offer a promising means to monitor soil moisture over large fields with flexible, high-resolution coverage. However, during the growing season, canopy scattering and soil reflections become strongly coupled in the radar measurement. These coupled effects vary with crop structure or flight altitude, complicating the retrieval of soil moisture. To overcome this challenge, we present GreenScatter, a physics-based soil moisture retrieval framework for nadir-looking wideband UAV radars. GreenScatter introduces a microwave radiative transfer model that explicitly captures the dominant electromagnetic interactions between vegetation and soil, enabling accurate modeling of coherent ground backscatter through canopy. In parallel, it develops a radar cross-section (RCS) estimation method that transforms time-domain radar signals into calibrated wideband RCS spectra, isolating soil reflections while compensating for hardware and waveform effects. Together, these components enable robust soil moisture estimation through vegetation across varying canopy conditions and UAV configurations. Field experiments across multiple corn and soybean sites demonstrate consistent retrieval with an average volumetric water content (VWC) error of 4.49%.

  PublisherDOI

• GreenScatter: Through-Canopy Soil Moisture Sensing with UAV-Mounted Radar

  2026-05-08

  articleOpen access

  Soil moisture is a critical variable for managing irrigation, improving crop yield, and understanding field-scale hydrology. Radars mounted on unmanned aerial vehicles (UAVs) offer a promising means to monitor soil moisture over large fields with flexible, high-resolution coverage. However, during the growing season, canopy scattering and soil reflections become strongly coupled in the radar measurement. These coupled effects vary with crop structure or flight altitude, complicating the retrieval of soil moisture. To overcome this challenge, we present GreenScatter, a physics-based soil moisture retrieval framework for nadir-looking wideband UAV radars. GreenScatter introduces a microwave radiative transfer model that explicitly captures the dominant electromagnetic interactions between vegetation and soil, enabling accurate modeling of coherent ground backscatter through canopy. In parallel, it develops a radar cross-section (RCS) estimation method that transforms time-domain radar signals into calibrated wideband RCS spectra, isolating soil reflections while compensating for hardware and waveform effects. Together, these components enable robust soil moisture estimation through vegetation across varying canopy conditions and UAV configurations. Field experiments across multiple corn and soybean sites demonstrate consistent retrieval with an average volumetric water content (VWC) error of 4.49%.

  PublisherDOI

• GreenScatter: Through-Canopy Soil Moisture Sensing with UAV-Mounted Radar

  arXiv (Cornell University) · 2026-04-10

  articleOpen access

  Soil moisture is a critical variable for managing irrigation, improving crop yield, and understanding field-scale hydrology. Radars mounted on unmanned aerial vehicles (UAVs) offer a promising means to monitor soil moisture over large fields with flexible, high-resolution coverage. However, during the growing season, canopy scattering and soil reflections become strongly coupled in the radar measurement. These coupled effects vary with crop structure or flight altitude, complicating the retrieval of soil moisture. To overcome this challenge, we present GreenScatter, a physics-based soil moisture retrieval framework for nadir-looking wideband UAV radars. GreenScatter introduces a microwave radiative transfer model that explicitly captures the dominant electromagnetic interactions between vegetation and soil, enabling accurate modeling of coherent ground backscatter through canopy. In parallel, it develops a radar cross-section (RCS) estimation method that transforms time-domain radar signals into calibrated wideband RCS spectra, isolating soil reflections while compensating for hardware and waveform effects. Together, these components enable robust soil moisture estimation through vegetation across varying canopy conditions and UAV configurations. Field experiments across multiple corn and soybean sites demonstrate consistent retrieval with an average volumetric water content (VWC) error of 4.49%.

  PublisherOA PDF

• FUELVISION: A multimodal data fusion and multimodel ensemble algorithm for wildfire fuels mapping

  International Journal of Applied Earth Observation and Geoinformation · 2025-03-12 · 7 citations

  articleOpen access

  Accurate assessment of fuel conditions is a prerequisite for fire ignition and behavior prediction, and risk management. The method proposed herein leverages diverse data sources – including L8 optical imagery, S1 (C-band) Synthetic Aperture Radar (SAR) imagery, PL (L-band) SAR imagery, and terrain features – to capture comprehensive information about fuel types and distributions. An ensemble model was trained to predict landscape-scale fuels – such as the ’Scott and Burgan 40’ – using the as-received Forest Inventory and Analysis (FIA) field survey plot data obtained from the USDA Forest Service. However, this basic approach yielded relatively poor results due to the inadequate amount of training data. Pseudo-labeled and fully synthetic datasets were developed using generative AI approaches to address the limitations of ground truth data availability. These synthetic datasets were used for augmenting the FIA data from California to enhance the robustness and coverage of model training. The use of an ensemble of methods – including deep learning neural networks, decision trees, and gradient boosting – offered a fuel mapping accuracy of nearly 80%. Through extensive experimentation and evaluation, the effectiveness of the proposed approach was validated for regions of the 2021 Dixie and Caldor fires. Comparative analyses against high-resolution data from the National Agriculture Imagery Program (NAIP) and timber harvest maps affirmed the robustness and reliability of the proposed approach, which is capable of near-real-time fuel mapping. • Near real-time wildland fuels mapping algorithm. • Leverage satellite remote sensing data and terrain features. • Overcome challenge posed by imbalanced datasets. • Leverage General Adversarial Networks for synthetic remote sensing data generation.

  PublisherDOI

• Rapid subsurface sensing via Bayesian-optimized FDTD modeling of ground penetrating radar

  Journal of Building Engineering · 2025-03-06 · 4 citations

  articleOpen accessSenior author

  Ground penetrating radars (GPRs) are widely used for non-destructive sub-surface evaluations but require extensive data processing and expert-guided analysis. Modern tools, including machine learning and waveform-inversion, automate interpretation but require accurate GPR modeling. Past modeling approaches primarily focus on full-scale 3D modeling of GPR antennas, which is computationally expensive. To simulate realistic GPR scans, this study proposes a simplified and efficient modeling approach by calibrating GPR antennas in 1D and 2D finite-difference-time-domain (FDTD) models. Intrinsic radar parameters (e.g., center frequency, waveform type, and bistatic separation) in FDTD are calibrated by Bayesian optimization using a medium with known dielectric properties as the calibration reference. The calibrated model is then applied to materials with unknown extrinsic dielectric properties, simplifying the process by separating the calibration of intrinsic radar parameters from material dielectric properties. The method was validated for two real commercial GPRs of different frequencies, and forward simulations with the calibrated models produced A- and B-scans closely matching real GPR data. The calibrated models were also used to analyze the sensitivity of GPR signals to changes in subsurface material properties and to determine depth resolution and maximum penetration depth of a real GPR. Furthermore, Bayesian wave-inversion using the calibrated model could accurately estimate dielectric properties of building materials such as concrete and soil, demonstrating strong correlations (0.89–0.99) with in-situ sensor measurements. Estimation of dielectric properties with the proposed model was about nine times faster than with full-scale 3D models. The proposed method offers an efficient and promising approach for subsurface material characterization applications such as concrete nondestructive testing and soil moisture-sensing. • Simplified model calibration for accurate and realistic simulation of real GPRs. • Use of calibrated models to effectively determine the depth resolution of a GPR. • Accurate forward modeling (A- and B-scan generation) using calibrated models. • Accurate material property estimation using calibrated models and optimization. • Proposed method is significantly faster than full-scale 3D modeling.

  PublisherDOI

• Bridging simulation and reality in subsurface radar-based sensing: physics-guided hierarchical domain adaptation with deep adversarial learning

  arXiv (Cornell University) · 2025-12-19

  preprintOpen accessSenior author

  Accurate estimation of subsurface material properties, such as soil moisture, is critical for wildfire risk assessment and precision agriculture. Ground-penetrating radar (GPR) is a non-destructive geophysical technique widely used to characterize subsurface conditions. Data-driven parameter estimation methods typically require large amounts of labeled training data, which is expensive to obtain from real-world GPR scans under diverse subsurface conditions. A physics-based GPR model using the finite-difference time-domain (FDTD) method can be employed to generate large synthetic datasets through simulations across varying material parameters, which are then utilized to train data-driven models. A key limitation, however, is that simulated data (source domain) and real-world data (target domain) often follow different distributions, which can cause data-driven models trained on simulations to underperform in real-world scenarios. To address this challenge, this study proposes a novel physics-guided hierarchical domain adaptation framework with deep adversarial learning for robust subsurface material property estimation from GPR signals. The proposed framework is systematically evaluated through the laboratory tests for single- and two-layer materials, as well as the field tests for single- and two-layer materials, and is benchmarked against state-of-the-art methods, including the one-dimensional convolutional neural network (1D CNN) and domain adversarial neural network (DANN). The results demonstrate that the proposed framework achieves higher correlation coefficients R and lower Bias between the predicted and measured parameter values, along with smaller standard deviations in the estimations, thereby validating their effectiveness in bridging the domain gap between simulated and real-world radar signals and enabling efficient subsurface material property retrieval.

  PublisherDOI

• Bridging simulation and reality in subsurface radar-based sensing: physics-guided hierarchical domain adaptation with deep adversarial learning

  ArXiv.org · 2025-12-19

  articleOpen accessSenior author

  Accurate estimation of subsurface material properties, such as soil moisture, is critical for wildfire risk assessment and precision agriculture. Ground-penetrating radar (GPR) is a non-destructive geophysical technique widely used to characterize subsurface conditions. Data-driven parameter estimation methods typically require large amounts of labeled training data, which is expensive to obtain from real-world GPR scans under diverse subsurface conditions. A physics-based GPR model using the finite-difference time-domain (FDTD) method can be employed to generate large synthetic datasets through simulations across varying material parameters, which are then utilized to train data-driven models. A key limitation, however, is that simulated data (source domain) and real-world data (target domain) often follow different distributions, which can cause data-driven models trained on simulations to underperform in real-world scenarios. To address this challenge, this study proposes a novel physics-guided hierarchical domain adaptation framework with deep adversarial learning for robust subsurface material property estimation from GPR signals. The proposed framework is systematically evaluated through the laboratory tests for single- and two-layer materials, as well as the field tests for single- and two-layer materials, and is benchmarked against state-of-the-art methods, including the one-dimensional convolutional neural network (1D CNN) and domain adversarial neural network (DANN). The results demonstrate that the proposed framework achieves higher correlation coefficients R and lower Bias between the predicted and measured parameter values, along with smaller standard deviations in the estimations, thereby validating their effectiveness in bridging the domain gap between simulated and real-world radar signals and enabling efficient subsurface material property retrieval.

  PublisherOA PDF

• Wildfire Fuels Mapping through Artificial Intelligence-based Methods: A Review

  Earth-Science Reviews · 2025-02-07 · 5 citations

  reviewOpen access

  Understanding fire behavior is a crucial step in wildfire risk assessment and management. Accurate and near real-time knowledge of the spatio-temporal characteristics of fuels is critical for analyzing pre-fire risk mitigation and managing active-fire emergency response. Geospatial modeling and land cover mapping using remote sensing combined with artificial intelligence techniques can provide fuel information at regional scales with high accuracy and resolution, as evidenced by the extensive recent work in the literature that appeared with increasing volume in the open literature. This paper provides a comprehensive survey of the state-of-the-art in wildfire fuel mapping, focusing on the research frontier of fire fuel models, fuel mapping methods, remote sensing data sources, existing datasets/reference maps, and applicable artificial intelligence techniques. The main findings highlight the increasing research on fire fuel mapping worldwide, with a considerable emphasis on multispectral imagery and the Random Forest classifier for its efficacy with limited data. The majority of these studies concentrate on relatively limited geographical scales spanning a small variety of fuel types, thus leaving a gap in regional and national-scale mapping. Further, this review focuses on identifying the major challenges in wildfire fuel mapping and viable solutions as they relate to (i) ground truth data scarcity, (ii) mapping understory vegetation, (iii) temporal latency, and (iv) lack of uncertainty-aware models. Lastly, this paper identifies potential AI-driven solutions that promise a significant leap in fuel mapping and discusses the latest developments and potential future trends in AI-based fuel mapping applications.

  PublisherDOI

• UAV-Based Remote Sensing of Soil Moisture Across Diverse Land Covers: Validation and Bayesian Uncertainty Characterization

  IEEE Transactions on Geoscience and Remote Sensing · 2025-01-01 · 2 citations

  articleOpen accessSenior author

  High-resolution soil moisture (SM) observations are critical for agricultural monitoring, forestry management, and hazard prediction, yet current satellite passive microwave missions are unable to directly provide retrievals at tens-of-meter spatial scales. Unmanned aerial vehicle (UAV)–mounted microwave radiometry presents a promising alternative, but most evaluations to date have focused on agricultural settings, with limited exploration across other land covers and few efforts to quantify retrieval uncertainty. This study addresses both gaps by evaluating SM retrievals from a drone-based Portable L-band Radiometer (PoLRa) across shrubland, bare soil, and forest strips in Central Illinois, U.S., using a 10-day field campaign in 2024. Controlled UAV flights at altitudes of 10 m, 20 m, and 30 m were performed to generate brightness temperatures (T<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</sub>) at spatial resolutions of 7 m, 14 m, and 21 m. SM retrievals were carried out using multiple tau-omega-based algorithms, including the single channel algorithm (SCA), dual channel algorithm (DCA), and multi-temporal dual-channel algorithm (MT-DCA). A Bayesian inference framework was then applied to provide probabilistic uncertainty characterization for both SM and vegetation optical depth (VOD). Results show that the gridded T<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</sub> distributions consistently capture dry-wet gradients associated with vegetation density variations, and spatial correlations between polarized observations are largely maintained across scales. Validation against <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in situ</i> measurements indicates that PoLRa-derived SM retrievals from the SCA-V and MT-DCA algorithms achieve unbiased root-mean-square errors (ubRMSE) generally below 0.04 m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> across different land covers. Bayesian posterior analyses confirm that reference SM values largely fall within the derived uncertainty intervals, with mean uncertainty ranges around ± 0.02 m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> and ± 0.11 m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> for SCA and DCA-related retrievals. These findings underscore the potential of UAV-mounted PoLRa for high-resolution SM retrieval across varied landscapes and emphasize the need for standardized calibration and uncertainty quantification frameworks to support broader scientific and operational adoption.

  PublisherOA PDFDOI

• Quantum Machine Learning With Limited Data: A Remote Sensing Perspective

  2025-08-03 · 1 citations

  article

  The demand for extensive field data for remote sensing applications has created bottlenecks for classical machine learning algorithms. This paper presents a quantum machine learning model based on a quantum support vector machine (QSVM) to classify Holm Oak trees using PRISMA Hyperspectral Imagery. The performance of the developed quantum machine learning model is validated in terms of dataset size, overall accuracy, number of qubits, training, and predicting speed. The results showed that QSVM outperformed Classic SVM by 5% accuracy with 50 samples at 12 qubits/feature or higher and 20 samples at 16 qubits/feature. QSVM training times were 284 seconds for 50 samples and 54 seconds for 20 samples, while prediction times for 400 pixels were 5243 seconds (50-sample model) and 2845 seconds (20-sample model). Higher accuracy achieved by QML exhibits great potential for limited data studies, with a drawback of slow training/testing speed. Another observation is that the sample size influences QSVM prediction speed in the training process. Therefore, we propose Quantum Machine Learning as a potential solution to tackle ground truth challenges as it can learn from less data.

  PublisherDOI

Frequent coauthors

• Devin K. Harris

  University of Virginia

  86 shared

• Mehrdad Shafiei Dizaji

  University of Massachusetts Lowell

  28 shared

• I. Ab Aziz

  22 shared

• Yuxiang Zhao

  Jiangnan University

  21 shared

• Kaiyuan Wang

  Zhejiang Ocean University

  16 shared

• Adam C. Watts

  16 shared

• Osman E. Ozbulut

  15 shared

• Tianshu Li

  14 shared

Labs

• Alipour Research Group @IllinoisPI

Education

• PhD, Civil and Environmental Engineering

  University of Virginia

  2019

Awards & honors

• Faculty Fellow, National Center for Supercomputing Applicati…