Enhancing Multi-Resolution Land Characteristics Identification Using Deep Neural Networks for Environmental Monitoring
Open Access DepositedAccurate classification of land use and land cover (LULC) is crucial for effective environmental management, urban planning, and climate modeling. The National Land Cover Database (NLCD) in particular plays a pivotal role as the official source of nationwide land cover data to key federal agencies such as the Environmental Protection Agency (EPA), Federal Emergency Management Agency (FEMA), U.S. Geological Survey (USGS), National Park Service (NPS), U.S. Forest Service, and the National Oceanic and Atmospheric Administration (NOAA). Recent advancements in deep learning, especially transformer-based models, have demonstrated significant potential in improving land use and land cover classifications. These methods have achieved accuracies above 90% on specialized, smaller datasets. While effective, the NLCD classification model is currently achieving classification accuracies around 77.5%.This praxis harnessed the capabilities of state-of-the-art deep learning models, specifically convolutional neural networks and vision transformers, to improve the accuracy and efficiency of LULC classification across the expansive and multifaceted NLCD dataset. These types of models are well-suited for capturing global dependencies within the data, which is particularly beneficial for the nuanced differentiation required in land LULC classification tasks. The data was sourced from the Landsat 8 missions at a 30m resolution and paired with 30-500m resolution elevation, ecological, and temporal data. The Landsat imagery covered nine bands across the visible, near infrared, and short-wave infrared spectra. Enhancements included augmenting the bands to include spectral indices, textural features, temporal mechanisms, and spatial context. The outcome of this research was a significant enhancement in the accuracy of LULC classifications, supporting improved environmental policies and land management practices. By applying newly developed image classification techniques to the precise monitoring of land cover changes, this praxis produced a deep neural network that achieved 92.7% accuracy.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Cantrall_gwu_0075A_17634.pdf | 2025-12-12 | Open Access |
|