Subproject 1: Monitoring

Subproject 1 Monitoring establishes a high-resolution, satellite-based monitoring system for past and present atmospheric (precipitation and temperature) and land-cover parameters (woody density), which are relevant for assessing the status of savanna rangelands in the context of IDESSA. The monitoring system will be based on an eco-climatological approach that links operational, multi-source remote-sensing data with in-situ observations using machine learning approaches. The developed models operationally generate time series of climate and woody densities across the study area, which will be implemented into the integrative database (subproject 3) and will serve the rangeland model (subproject 2) as a baseline for short-term scenario computations. Updates of the time series are quasi-continuously processed and presented on an established IDESSA server. The developed system will be made freely available to act as an observatory for local-scale effects of global change on ecosystem processes and management interactions.

To estimate rainfall areas and rainfall rates from Meteosat satellite data, different machine learning algorithms (random forests, neural networks, averaged neural networks, support vector machines) were compared for their suitability. The differences between the model performances were small. However, in average, neural networks performed slightly better than the other models and will therefore be applied in the rainfall retrieval for South Africa.
(source: Meyer, H., K├╝hnlein, M., Appelhans, T., Nauss, T. (2016): Comparison of four machine learning algorithms for their applicability in satellite-based optical rainfall retrievals. Atmospheric Research 169: 424-433).