Phase 4
To ensure data quality, we employed algorithmic filtering, removing 97 dark, 124 cloudy, and 187 sea images out of the 463 taken. Following this, we selected the best image for analysis which was taken over Andhra Pradesh, Karnataka and Maharashtra in South India.
- The filtering for the dark images was done by turning the images into grayscale and calculating the average intensity of the pixels; then, with a defined threshold, we removed the dark images.
- The filtering for cloudy images was done by employing the Otsu's method for image thresholding: and algorithmic filtering that, assuming only to classes, calculates the optimal threshold to separate these two classes by minimizing the intra-class variance.
- The filtering for sea images was done by employing a NDVI classifier: a classifier to remove images taken over water that makes use of the NDVI in order to distinguish water pixels; then the algorithm remove images with a water pixel percentage above a chosen threshold.
For the selected region of interest (ROI), we employed a customized GSD algorithm considering Earth's curvature, resulting in improved accuracy of corner coordinate estimation. Our estimation is still subject to some degree of error due to the fact that it does not take into account the rotation of the image.
In this experiment, our initial objective was to explore the relationship with urban areas near the area of interest. However, due to complications arising from the lack of images for VCI application, we decided to gradually proceed with a study of the current NDVI using the image captured from the ISS.
Subsequently, using the Google Earth Engine API, we extracted the average NDVI values (starting from 2019) for the area of interest in order to graphically represent an NDVI evolution chart.
This allowed us to gain valuable insights into the vegetation's condition over a specific time period. Additionally, we calculated the average VCI by utilizing the collected data used for the NDVI evolution representation.