A scalable gaussian process analysis algorithm for biomass monitoring

摘要: Biomass monitoring is vital for studying the carbon cycle of earth's ecosystem and has several significant implications, especially in context understanding climate change its impacts. Recently, detection methods have been proposed to identify land cover changes temporal profiles (time series) vegetation collected using remote sensing instruments, but do not satisfy one or both two requirements biomass problem, that is, operating online mode handling periodic time series. In this paper, we adapt Gaussian process (GP) regression detect such series an fashion. While GP widely used as a kernel-based learning method classification, their applicability massive spatiotemporal data sets, data, limited owing high computational costs involved. We focus on addressing scalability issues associated with based algorithm. This paper makes contributions. First, propose algorithm demonstrate effectiveness detecting different types Normalized Difference Vegetation Index (NDVI) obtained from study area IA, USA. Second, efficient Toeplitz matrix solution which significantly improves complexity memory method. Specifically, can analyze length t O(t2) while maintaining O(t) footprint, compared O(t3) requirement standard manipulation methods. Third, describe parallel version be simultaneously large number three implementations: threads, Message Passing Interface (MPI), hybrid implementation threads MPI. Experimental results show scales better than multithreaded MPI implementations. The application scalable demonstrated analyzing observation data. implementation, 1536 computing cores, NDVI set Iowa nearly 5 s, serial algorithm, Cholesky decomposition routines, takes days same set. © 2011 Wiley Periodicals, Inc. Statistical Analysis Data Mining 4: 430–445,