Data Enrichment of Sentinel-2 and Landsat-8 Surface-reflectance Measurements for Agriculture Oriented Services

Since the first attempts to utilize the useful information incorporated in surface reflectance measurements of plants, almost a fifty years ago, which consisted of the introduction of the “simple ratio” index in 1969 and subsequently in 1973 the famous normalized difference vegetation index, there was a need to derive quantities that will mease interpretation of original measurements and improve their usefulness. Accumulated domain knowledge over this long period of time brought a vast diversity of surface- reflectance derived broadband vegetation and spectral indices that were specially designed to fulfill the user needs in characterization of plant health and growth conditions. With the advancement of satellite imaging technology and related data policies in recent years, previously introduced quantities (in the form of spectral indices) gained in value as efficient tools for simple and effective characterization of complex biophysical processes at large scales and with increasing spatial resolution. Although these indices were initially designed with intention to be computationally simple, due to technology constraints, they proved successful in numerous applications. This area of research is still very active and aimed towards improvement of their robustness to environmental factors like soil variability and its properties. Currently available computational power enables design of large data cubes, as aggregating structure that can incorporate an abundance of previously designed and finely tuned spectral indices,which opens new possibilities for their application. Feature extraction workflows in the domain of land cover and land use classification are one of the application areas that can benefit from the enrichment of original measurements through computation of known spectral indices available in the literature. Thus, a significant amount of currently available domain knowledge can be incorporated directly in the feature engineering process. A large number of different spectral indices also contributes to the overall capabilities of a more traditional applications, like visualization of agriculture related processes and easier photo-interpretation through e.g. web based services. In this context we made a comprehensive overview and implemented a processing workflow with more than 40 most significant broadband vegetation and spectral indices, equipped with Matlab and Python programming interfaces. In such a way, original data cubes provided by Sentinel-2 and Landsat-8 multispectral instruments are further enriched,offering enhanced discriminability in tasks such as classification and change detection, as well as improved visual interpretation that is demonstrated through interactive web service