Member-only story
Enhancement of MODIS NIDVI to 10m resolution using U-Net
The normalized difference vegetation index (NDVI), introduced in the 1970s, has powerful applications in land management, food security, and physical models. Acquiring NDVI in both high spatial and temporal resolutions is preferable for these applications. however, the temporal and spatial resolution has a trade-off, making it difficult to use satellite images. To solve this issue, a lot of researchers proposed different architectures of convolutional neural networks (CNN) capable of estimating high-resolution NDVI. One of these studies, titled ‘Downscaling of MODIS NDVI by Using a Convolutional Neural Network-Based Model with Higher Resolution SAR Data’, proposed to estimate 10-m high-resolution NDVI from MODIS (Moderate Resolution Imaging Spectroradiometer) 250-m resolution NDVI by using Sentinel-1 10-m resolution synthetic aperture radar (SAR) data.
I decided to implement the same procedure in my chosen area of interest in Switzerland, with U-Net for 2019.
Datasets
In the first step, I search for the less cloudy (~20%) acquisitions of Sentinel-2 over 2019. Eight (8) acquisitions were selected between January and December. Next, the closed acquisition for Sentinel-1 and MODIS was downloaded. Band 4 and 8 of sentinel-2 were downloaded from the Registry of Open Data on AWS (for more…
