Land-Use and Land-Cover Classification in Semi-Arid Areas from Medium-Resolution Remote-Sensing Imagery: A Deep...

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Deep Learning Applications

Background:

• Detailed Land-Use and Land-Cover (LULC) information is crucial for urban planning, disaster management, and climate change adaptation.
• Deep Learning (DL) offers a significant advancement in LULC classification.
• Limited research exists on DL for LULC mapping in semi-arid regions, especially comparing Sentinel-2 band combinations with Convolutional Neural Network (CNN) models.

Purpose of the Study:

• To evaluate the effectiveness of different Sentinel-2 band combinations for LULC classification in semi-arid regions using CNN models.
• To assess the transferability of optimized CNN models across different semi-arid study sites.
• To identify optimal Sentinel-2 bands for accurate LULC mapping in challenging semi-arid environments.

Main Methods:

• Trained a CNN model using Sentinel-2 data from a semi-arid site in Pakistan (Gujranwala).
• Compared a four-band (NIR, green, blue, red) composite with a ten-band composite (including SWIR and red-edge bands).
• Applied the pre-trained CNN model to two additional semi-arid sites (Lahore, Faisalabad) to test transferability.

Main Results:

• The proposed CNN architecture was validated.
• The four-band composite CNN model demonstrated significant robustness in semi-arid environments.
• This approach proved effective despite high spectral similarity among land-cover features.

Conclusions:

• The study highlights the efficacy of a simplified four-band Sentinel-2 composite for LULC mapping in semi-arid regions using deep learning.
• Optimized CNN models show good transferability for semi-arid LULC classification.
• The findings support the use of specific Sentinel-2 bands for improved LULC mapping accuracy in challenging environments.