Supath Dhital
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Supath Dhital has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Anupal Baruah
Department of Geography and Environment, University of Alabama, Tuscaloosa, USA
2 shared publications - Dipsikha Devi
Department of Geography and Environment, University of Alabama, Tuscaloosa, USA
2 shared publications - Sagy Cohen
University of Alabama, Tuscaloosa, Box 870322, Tuscaloosa, AL, United States [US
2 shared publications - Yixian Chen
Department of Geography and Environment, University of Alabama, Tuscaloosa, USA
2 shared publications Aakriti Adhikari
Department of Agriculture, University of Agriculture and Forestry
1 shared publicationAnupal Baruah
1 shared publication
Latest Publications
- Open Access
- 24 December 2025
- Published by Wiley
Spatial and Temporal Variability in Global River Slopes from the Surface Water and Ocean Topography (SWOT) Satellite
River water surface slope (WSS) is fundamental for hydrologic, hydraulic, and geomorphic modeling. Unlike channel bed slope, often treated as short-term stationary, WSS varies dynamically as flow interacts with bed heterogeneity. We analyze global spatial and monthly patterns of WSS using the SWOT Level 2 River Single-Pass Vector data (within 2023/08 – 2025/05). Both hemispheres show significant monthly differences: WSS is generally higher in winter and lower in summer, likely due to higher discharge and stage during typical events that flatten the water surface, and low-flow winters where rugged bed topography exerts stronger control. Month-to-month changes are larger during winter than summer. Despite temporal variability, spatial patterns remain strongly tied to topography. Monthly WSS coefficient of variation is strongly negatively correlated with median WSS (a topographic-slope proxy), indicating flatter rivers have higher WSS variability. These results clarify large-scale water-surface dynamics and motivate testing temporal WSS in hydrologic and hydraulic models.
- 24 December 2025
- Published by Wiley
Open AccessSpatial and Temporal Variability in Global River Slopes from the Surface Water and Ocean Topography (SWOT) Satellite
River water surface slope (WSS) is fundamental for hydrologic, hydraulic, and geomorphic modeling. Unlike channel bed slope, often treated as short-term stationary, WSS varies dynamically as flow interacts with bed heterogeneity. We analyze global spatial and monthly patterns of WSS using the SWOT Level 2 River Single-Pass Vector data (within 2023/08 – 2025/05). Both hemispheres show significant monthly differences: WSS is generally higher in winter and lower in summer, likely due to higher discharge and stage during typical events that flatten the water surface, and low-flow winters where rugged bed topography exerts stronger control. Month-to-month changes are larger during winter than summer. Despite temporal variability, spatial patterns remain strongly tied to topography. Monthly WSS coefficient of variation is strongly negatively correlated with median WSS (a topographic-slope proxy), indicating flatter rivers have higher WSS variability. These results clarify large-scale water-surface dynamics and motivate testing temporal WSS in hydrologic and hydraulic models.
- 11 November 2025
- Published by Elsevier in Advances in Water Resources
- Vol. 206
Flooding, nonlinear scaling and Jensen’s inequality: Analyzing changes in inundation extent with river discharge nonstationarity
- 11 November 2025
- Published by Elsevier in Advances in Water Resources
- Vol. 206
Flooding, nonlinear scaling and Jensen’s inequality: Analyzing changes in inundation extent with river discharge nonstationarity
- Open Access
- 3 November 2025
- Published by Wiley
Enhancement of low-fidelity flood inundation mapping through surrogate modeling
Reliable Flood Inundation Mapping (FIM) is a crucial component in flood forecasting and analysis frameworks; however, selecting prediction solvers necessitates balancing accuracy and computational demand. High-fidelity (HF) hydrodynamic models are accurate yet resource-intensive, while Low-Fidelity (LF) terrain-based solvers are efficient but less precise. This study develops a Surrogate Model (SM) that fuses LF-derived FIMs with deep learning to capture nonlinear flood dynamics and improve predictive performance. LF-FIMs are generated using the National Oceanic and Atmospheric Administration, Office of Water Prediction (OWP) Height Above Nearest Drainage (HAND)-based framework. The attention U-Net Convolutional Neural Network SM is trained on LF-FIMs with topographic predictors and physics-based HEC-RAS HF benchmarks from 28 U.S. regions and evaluated across spatial and temporal domains. The SM enhances accuracy and generalization across unseen domains while capturing flood evolution, though challenges persist in certain terrains. Independent evaluation at 20 sites shows the SM improves the Critical Success Index by a median of 14% and Probability of Detection by 11%, while reducing the False Alarm Rate by 13%. Event-scale inference showed 38% improvement in FIM accuracy, with the highest agreement observed at peak inundation. Evaluation with building footprint shows even greater evidence of the SM-FIM, improving flood exposure estimation. Prediction quality was found to be influenced by LF-FIM biases; in low-relief and urban environments. Despite these limitations, the SM adaptively enhances LF-FIMs with significant computational efficiency, and has been integrated into a FIM prediction framework, demonstrating its potential as a scalable, reliable, and efficient component of national flood forecasting systems.
- 3 November 2025
- Published by Wiley
Open AccessEnhancement of low-fidelity flood inundation mapping through surrogate modeling
Reliable Flood Inundation Mapping (FIM) is a crucial component in flood forecasting and analysis frameworks; however, selecting prediction solvers necessitates balancing accuracy and computational demand. High-fidelity (HF) hydrodynamic models are accurate yet resource-intensive, while Low-Fidelity (LF) terrain-based solvers are efficient but less precise. This study develops a Surrogate Model (SM) that fuses LF-derived FIMs with deep learning to capture nonlinear flood dynamics and improve predictive performance. LF-FIMs are generated using the National Oceanic and Atmospheric Administration, Office of Water Prediction (OWP) Height Above Nearest Drainage (HAND)-based framework. The attention U-Net Convolutional Neural Network SM is trained on LF-FIMs with topographic predictors and physics-based HEC-RAS HF benchmarks from 28 U.S. regions and evaluated across spatial and temporal domains. The SM enhances accuracy and generalization across unseen domains while capturing flood evolution, though challenges persist in certain terrains. Independent evaluation at 20 sites shows the SM improves the Critical Success Index by a median of 14% and Probability of Detection by 11%, while reducing the False Alarm Rate by 13%. Event-scale inference showed 38% improvement in FIM accuracy, with the highest agreement observed at peak inundation. Evaluation with building footprint shows even greater evidence of the SM-FIM, improving flood exposure estimation. Prediction quality was found to be influenced by LF-FIM biases; in low-relief and urban environments. Despite these limitations, the SM adaptively enhances LF-FIMs with significant computational efficiency, and has been integrated into a FIM prediction framework, demonstrating its potential as a scalable, reliable, and efficient component of national flood forecasting systems.
- Open Access
- 20 June 2025
- Published by Wiley
Exploring Seasonal Dynamics of Landsat 8 and 9 Data on Vegetation Indices in the Mountainous Regions of Nepal
Earth observation technologies have revolutionized our ability to monitor and assess environmental changes, offering unprecedented insights into the dynamics of ecosystems, land use, and climate patterns. As these technologies evolve, newer satellites like Landsat-9 (L9) offer refined capabilities that demand careful assessment of data quality and reliability improvements. With L9's enhanced radiometric resolution over Landsat-8 (L8), evaluating the impact on data accuracy and temporal resolution is crucial for ensuring consistency in long-term Earth observation and environmental monitoring. This study presents a unique way of evaluating the data consistency between two satellites by examining vegetation indices' seasonal performance and consistency from L8 and L9, focusing on NDVI, NDWI, EVI, and SAVI. Different evaluation metrics, as well as performance over different land use categories, are evaluated. This study shows that L8 generally overestimates vegetation index values compared to L9, except for SAVI, where both satellites perform similarly. L9 exhibits greater sensitivity to vegetation and water features, especially during transitional seasons such as spring and fall. The correlation between indices from both satellites remains robust, with R² values consistently exceeding 0.8, peaking at 0.92 in the fall. In the summer, L9's finer temporal resolution better captures rapid phenological changes, leading to increased variability compared to other seasons. Quantitative metrics indicate that L8 overestimates vegetation indices, with higher average RMSE (~0.05), MAE (~0.04), and a percentage bias of 4% compared to L9. This overestimation persists across most land use classes and seasons, with a notable ~15% negative bias in grassland and shrubland during the summer. L9's enhanced capabilities make it a more reliable tool for monitoring vegetation, particularly water features, in rapidly changing environments. Future research should explore a range of geographical and climatic contexts and incorporate additional indicators and high-resolution data, such as multi-spectral drone imagery, to strengthen validation efforts.
- 20 June 2025
- Published by Wiley
Open AccessExploring Seasonal Dynamics of Landsat 8 and 9 Data on Vegetation Indices in the Mountainous Regions of Nepal
Earth observation technologies have revolutionized our ability to monitor and assess environmental changes, offering unprecedented insights into the dynamics of ecosystems, land use, and climate patterns. As these technologies evolve, newer satellites like Landsat-9 (L9) offer refined capabilities that demand careful assessment of data quality and reliability improvements. With L9's enhanced radiometric resolution over Landsat-8 (L8), evaluating the impact on data accuracy and temporal resolution is crucial for ensuring consistency in long-term Earth observation and environmental monitoring. This study presents a unique way of evaluating the data consistency between two satellites by examining vegetation indices' seasonal performance and consistency from L8 and L9, focusing on NDVI, NDWI, EVI, and SAVI. Different evaluation metrics, as well as performance over different land use categories, are evaluated. This study shows that L8 generally overestimates vegetation index values compared to L9, except for SAVI, where both satellites perform similarly. L9 exhibits greater sensitivity to vegetation and water features, especially during transitional seasons such as spring and fall. The correlation between indices from both satellites remains robust, with R² values consistently exceeding 0.8, peaking at 0.92 in the fall. In the summer, L9's finer temporal resolution better captures rapid phenological changes, leading to increased variability compared to other seasons. Quantitative metrics indicate that L8 overestimates vegetation indices, with higher average RMSE (~0.05), MAE (~0.04), and a percentage bias of 4% compared to L9. This overestimation persists across most land use classes and seasons, with a notable ~15% negative bias in grassland and shrubland during the summer. L9's enhanced capabilities make it a more reliable tool for monitoring vegetation, particularly water features, in rapidly changing environments. Future research should explore a range of geographical and climatic contexts and incorporate additional indicators and high-resolution data, such as multi-spectral drone imagery, to strengthen validation efforts.
- Open Access
- 18 February 2025
- Published by Wiley
Enhancement of the NOAA Flood Inundation Mapping Framework (OWP HAND-FIM) through Surrogate Modeling
Reliable Flood Inundation Mapping (FIM) is a key component in the flood forecasting and analysis framework. The selection of FIM prediction solvers is a trade-off between computational and data costs and accuracy. Physically based (hydraulic) models are considered more accurate but computationally and data expensive, while low complexity terrain-based solvers offer efficiency and scalability at the expense of accuracy. The NOAA Office of Water Predictions (OWP) operational FIM framework is based on the Height Above Nearest Drainage (HAND) approach. In this study, a deep Convolutional neural network (CNN) technique is explored to enhance the OWP HAND-FIM using a surrogate modeling approach. Among various architectures, an encoder-decoder U-Net architecture performs well to classify the flooded area precisely. U-Net's skip connections help to maintain high-resolution spatial information by combining characteristics from multiple Convolutional layers, which improves prediction accuracy, whereas skip connections help reduce vanishing gradient concerns, which improves the model's generalization capabilities. The model is trained using physics-based HEC-RAS simulations and a suite of other input features (digital elevation model, topographical wetness index, land use, land cover, flow accumulations, etc.) over three HUC8-scale watersheds. The results show a considerable improvement in FIM accuracy when using a surrogate CNN model compared to the original OWP HAND-FIM. The transferability of the model is tested in a fourth HUC8 region, showing considerable enhancement in the OWP HAND-FIM after the trained surrogate model is applied. The enhanced FIM was also evaluated using building footprints, resulting in a significant improvement in the surrogate model impact detection accuracy. This demonstrates that CNN holds a generalizing capability by capturing the spatial correlation between pixels. Introducing physical laws to attain behaviors during modeling and increasing the size and diversity of its training data is expected to further enhance the model's transferability and predicting capability.
- 18 February 2025
- Published by Wiley
Open AccessEnhancement of the NOAA Flood Inundation Mapping Framework (OWP HAND-FIM) through Surrogate Modeling
Reliable Flood Inundation Mapping (FIM) is a key component in the flood forecasting and analysis framework. The selection of FIM prediction solvers is a trade-off between computational and data costs and accuracy. Physically based (hydraulic) models are considered more accurate but computationally and data expensive, while low complexity terrain-based solvers offer efficiency and scalability at the expense of accuracy. The NOAA Office of Water Predictions (OWP) operational FIM framework is based on the Height Above Nearest Drainage (HAND) approach. In this study, a deep Convolutional neural network (CNN) technique is explored to enhance the OWP HAND-FIM using a surrogate modeling approach. Among various architectures, an encoder-decoder U-Net architecture performs well to classify the flooded area precisely. U-Net's skip connections help to maintain high-resolution spatial information by combining characteristics from multiple Convolutional layers, which improves prediction accuracy, whereas skip connections help reduce vanishing gradient concerns, which improves the model's generalization capabilities. The model is trained using physics-based HEC-RAS simulations and a suite of other input features (digital elevation model, topographical wetness index, land use, land cover, flow accumulations, etc.) over three HUC8-scale watersheds. The results show a considerable improvement in FIM accuracy when using a surrogate CNN model compared to the original OWP HAND-FIM. The transferability of the model is tested in a fourth HUC8 region, showing considerable enhancement in the OWP HAND-FIM after the trained surrogate model is applied. The enhanced FIM was also evaluated using building footprints, resulting in a significant improvement in the surrogate model impact detection accuracy. This demonstrates that CNN holds a generalizing capability by capturing the spatial correlation between pixels. Introducing physical laws to attain behaviors during modeling and increasing the size and diversity of its training data is expected to further enhance the model's transferability and predicting capability.
- Open Access
- 12 July 2023
- Published by California Digital Library (CDL) in EarthArXiv
Forecasting Weather using Deep Learning from the Meteorological Stations Data : A Study of Different Meteorological Stations in Kaski District, Nepal
Contemporarily, one of the most pressing concerns is reliable and rapid weather forecasting. In Nepal, the Department of Hydrology and Meteorological uses a numerical modeling approach to forecast the weather, which is tardy and requires high end equipment to process the information so deep learning approach will be best alternative. This project aims to forecast the next 2-hour Precipitation and Air Temperature for Pokhara Domestic Airport meteorological station and next day Precipitation, Maximum and Minimum air Temperature forecast for Lumle, Begnas and Lamachaur meteorological station, total of four meteorological stations of the Kaski District, Nepal using Long Short-Term Memory (LSTM): a Recurrent Neural Network (RNN) and deploy the outputs through the web portal. The four hourly parameters: Rainfall, Relative Humidity (R.H), Wind Speed and Air Temperature were used for modeling the airport station forecast whereas Rainfall, Relative Humidity (R.H), Maximum and Minimum Temperature were used for modeling the Begnas and Lumle station forecast and only Precipitation data was used for Lamachaur station. Averaging and linear interpolation techniques were used to fill out the missing values and outliers were detected using Box Plot and replaced with threshold value for each parameter. Stochastic Gradient Descent and Adam optimizer are used to optimize the LSTM model. Among all the models prepared, Root Mean Square Error (RMSE) values range from 0.58 to 4.08 for precipitation model and from 0.16 to 0.82 for air temperature model and Mean Absolute Error (MAE) values range from 0.21 to 2.87 for precipitation model and from 0.12 to 0.64 for air temperature model were the values of the final model that indicates better accuracy for air temperature. The R² values range from 0.89 to 0.99 indicating the train and test data were fitted to the model really well.
- 12 July 2023
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessForecasting Weather using Deep Learning from the Meteorological Stations Data : A Study of Different Meteorological Stations in Kaski District, Nepal
Contemporarily, one of the most pressing concerns is reliable and rapid weather forecasting. In Nepal, the Department of Hydrology and Meteorological uses a numerical modeling approach to forecast the weather, which is tardy and requires high end equipment to process the information so deep learning approach will be best alternative. This project aims to forecast the next 2-hour Precipitation and Air Temperature for Pokhara Domestic Airport meteorological station and next day Precipitation, Maximum and Minimum air Temperature forecast for Lumle, Begnas and Lamachaur meteorological station, total of four meteorological stations of the Kaski District, Nepal using Long Short-Term Memory (LSTM): a Recurrent Neural Network (RNN) and deploy the outputs through the web portal. The four hourly parameters: Rainfall, Relative Humidity (R.H), Wind Speed and Air Temperature were used for modeling the airport station forecast whereas Rainfall, Relative Humidity (R.H), Maximum and Minimum Temperature were used for modeling the Begnas and Lumle station forecast and only Precipitation data was used for Lamachaur station. Averaging and linear interpolation techniques were used to fill out the missing values and outliers were detected using Box Plot and replaced with threshold value for each parameter. Stochastic Gradient Descent and Adam optimizer are used to optimize the LSTM model. Among all the models prepared, Root Mean Square Error (RMSE) values range from 0.58 to 4.08 for precipitation model and from 0.16 to 0.82 for air temperature model and Mean Absolute Error (MAE) values range from 0.21 to 2.87 for precipitation model and from 0.12 to 0.64 for air temperature model were the values of the final model that indicates better accuracy for air temperature. The R² values range from 0.89 to 0.99 indicating the train and test data were fitted to the model really well.