Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A light-weight model to generate NDWI from Sentinel-1

Saleh Sakib Ahmed, Saifur Rahman Jony, Md. Toufikuzzaman, Saifullah Sayed, Rashed Uz Zzaman, Sara Nowreen, M. Sohel Rahman

TL;DR

This work tackles the cloud limitations of Sentinel-2 in NDWI-based water mapping by introducing a lightweight U-Net that generates NDWI directly from Sentinel-1 radar data. Ground-truth NDWI is obtained from cloud-free Sentinel-2 imagery using the standard $NDWI = \frac{Green - NIR}{Green + NIR}$ formulation and scaled to the [0,1] range, with the data prepared as 128×128 patches from a larger 512×512 baseline. The model achieves a high test accuracy of 0.9134, AUC of 0.8656, and a Mean IoU of 0.4139, demonstrating strong pixel-wise water classification and segmentation capabilities, even under clouds and nighttime. This end-to-end SAR-to-NDWI mapping provides a robust baseline for continuous water monitoring and motivates future exploration of alternative architectures and generative approaches.

Abstract

The use of Sentinel-2 images to compute Normalized Difference Water Index (NDWI) has many applications, including water body area detection. However, cloud cover poses significant challenges in this regard, which hampers the effectiveness of Sentinel-2 images in this context. In this paper, we present a deep learning model that can generate NDWI given Sentinel-1 images, thereby overcoming this cloud barrier. We show the effectiveness of our model, where it demonstrates a high accuracy of 0.9134 and an AUC of 0.8656 to predict the NDWI. Additionally, we observe promising results with an R2 score of 0.4984 (for regressing the NDWI values) and a Mean IoU of 0.4139 (for the underlying segmentation task). In conclusion, our model offers a first and robust solution for generating NDWI images directly from Sentinel-1 images and subsequent use for various applications even under challenging conditions such as cloud cover and nighttime.

A light-weight model to generate NDWI from Sentinel-1

TL;DR

This work tackles the cloud limitations of Sentinel-2 in NDWI-based water mapping by introducing a lightweight U-Net that generates NDWI directly from Sentinel-1 radar data. Ground-truth NDWI is obtained from cloud-free Sentinel-2 imagery using the standard formulation and scaled to the [0,1] range, with the data prepared as 128×128 patches from a larger 512×512 baseline. The model achieves a high test accuracy of 0.9134, AUC of 0.8656, and a Mean IoU of 0.4139, demonstrating strong pixel-wise water classification and segmentation capabilities, even under clouds and nighttime. This end-to-end SAR-to-NDWI mapping provides a robust baseline for continuous water monitoring and motivates future exploration of alternative architectures and generative approaches.

Abstract

The use of Sentinel-2 images to compute Normalized Difference Water Index (NDWI) has many applications, including water body area detection. However, cloud cover poses significant challenges in this regard, which hampers the effectiveness of Sentinel-2 images in this context. In this paper, we present a deep learning model that can generate NDWI given Sentinel-1 images, thereby overcoming this cloud barrier. We show the effectiveness of our model, where it demonstrates a high accuracy of 0.9134 and an AUC of 0.8656 to predict the NDWI. Additionally, we observe promising results with an R2 score of 0.4984 (for regressing the NDWI values) and a Mean IoU of 0.4139 (for the underlying segmentation task). In conclusion, our model offers a first and robust solution for generating NDWI images directly from Sentinel-1 images and subsequent use for various applications even under challenging conditions such as cloud cover and nighttime.
Paper Structure (11 sections, 1 equation, 3 figures, 1 table)

This paper contains 11 sections, 1 equation, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Problem Description
  4. Datasets
  5. Data Preprocessing
  6. Otsu Thresholding
  7. Model Architecture
  8. Evaluation Metrics
  9. Code, Environment, and Availability
  10. Results and Discussions
  11. Conclusions

Figures (3)

  • Figure 1: The Figure showcases the procedure of how we train our CloudBreaker model. We use Sentinel-1 images as input and the NDWI of corresponding Sentinel-2 images as ground truth for the U-Net model (only cloud-free images are considered). After training, CloudBreaker becomes capable of directly producing NDWI images of corresponding Sentinel-2 images from given Sentinel-1 images
  • Figure 2: Model was trained using cloud-free Sentinel-1 data and Sentinel-2 images. Finally, we tested it with Sentinel-1 data containing clouds to demonstrate its ability to generate NDWI images of Sentinel-2 when such images are not properly available. The generated NDWI, as well as the two bands of Sentinel-1 data, VV and VH, are shown.
  • Figure 3: (a) The model was trained with Sentinel-1 images and ground truth as Sentinel-2 images. The generated image is on the rightmost, with the Sentinel-2 image in the middle and the hand-labeled image on the leftmost. (b) This image shows how the two bands of Sentinel-1 images, VH and VV, are used to train the model, along with the NDWI of Sentinel-2 images, to generate NDWI for those Sentinel-1 images.