Biophysical parameters from remotely sensed imagery as a tool for monitoring and evaluating ecosystem biodiversity
Today together with Essential Climate Variable, it has started the process of the definition of Essential Biodiversity Variable which can support the monitoring and protection of ecosystem biodiversity.
In this context, the benefits of remotely sensed imagery for biodiversity indicators are related to the synoptic view, regular and repeatable acquisitions, multi-annual time series of observations and cost-effectiveness for remote and inaccessible areas. Moreover, the new satellite sensors (such as the Sentinel family) are notably increasing the remote sensing capabilities. Such large data availability has determined the necessity to develop accurate and robust retrieval methods to improve the estimation of these variables from remotely sensed imagery.
The retrieval process of these parameters from satellite images (optical and radar) is typically a challenging task and it falls in the category of ill-posed problem. This means that beyond the non-linearity of the relationship between input features (sensor measurements) and the target variables (soil moisture, biomass, etc.), more than one combination of soil characteristics may lead to the same electromagnetic response at the sensor. Furthermore, given a scene of interest, each system will provide information on a different aspect of the phenomena at the ground (e.g., the spatial patterns or the temporal dynamic) and could be also affected on different extents by different disturbing factors.
This suggests the importance of a synergic use of multiple available remote sensing systems (from satellite to drone based sensors) for a comprehensive, accurate and robust understanding and monitoring of the natural processes at the ground. On the other side the proper selection of the retrieval approach is a key issue.
In this context, the seminar will present currently available techniques for the retrieval of biophysical parameters from remotely sensed data addressing inversion of physical-based models, parametric and non-parametric approaches such as Bayesian procedure, Neural Networks, Support Vector Regression and Ensemble techniques. Each approach will be presented in specific applications indicating advantages, disadvantages and perspectives for the upcoming missions such as Sentinel 1 and 2. In addition the synergic use of different sensors (optical and radar) will be specifically addressed in the context of the retrieval process.