Lecture 3

The AI-based Sentinel-1-Sentinel-2 Data Fusion for Crop monitoring

Description: While optical RS data provide relevant information for cropland monitoring challenges such as cloud cover can hinder the effective use of optical RS data. To cope with this challenge, SAR observations can be fused into optical data to increase time series density.

Our objective is to investigate if we can overcome these limitations by generating a synthetic NDVI. We will do this by integrating Sentinel-1 radar data with the available information from Sentinel-2 optical data.

The central hypothesis is that a synthetic NDVI generated from Sentinel-1 imagery can effectively reduce the temporal gaps between NDVI readings, as Sentinel-1 data offers a higher temporal resolution than Sentinel-2. This could lead to more accurate and continuous monitoring of crop rotations.

Develop skills to:

• Handling big data pre- and processing within Google Earth Engine
• Calculation and comparison of NDVI and RVI indices
• Hands-on use of deep learning models for fusing optical and radar data

Duration: 1,5 hr (0,5 theoretical part, 1h practical exercises)

Lecture Flow (content):

4.1. Data preprocessing and downloading on Google Earth Engine (GEE)Fusion of Sentinel 1, Sentinel 2, and Auxiliary data
4.2. Image Mapping, Train Test Dataset Creation
4.3. Feature Engineering and Data Fusion
4.4. Training a Unet model to create a synthetic NDVI
4.5. Generation of a denser timeseries combining the Sentinel 2 NDVI and Predicted Synthetic NDVI
4.6. References

Outcomes:

• Acquire skills in merging optical and radar data using AI tools at parameter and pixel level