Search Results
: Machine learning methods can map properties from one observational dataset to another, even if the underlying measurements or physical fields do not allow for a direct physical comparison. Furthermore, machine learning tools can learn to detect anomalies. Machine learning is therefore well placed to detect and isolate erroneous observations. These capabilities can be applied within data assimilation frameworks, but also for the realization of a digital twin concept for the entire Earth system or for
: Machine learning methods can map properties from one observational dataset to another, even if the underlying measurements or physical fields do not allow for a direct physical comparison. Furthermore, machine learning tools can learn to detect anomalies. Machine learning is therefore well placed to detect and isolate erroneous observations. These capabilities can be applied within data assimilation frameworks, but also for the realization of a digital twin concept for the entire Earth system or for
comes after the unsupervised learning is to test whether the patterns identified by an algorithm correspond to climatologically distinct environments and if so which ones. TDA is an alternative approach to address the lack of labels. With TDA we seek to match imagery to the original four classes identified by Stevens et al. (2020) , yet only require a small number of labeled samples. We map patches of the MODIS imagery into topological space and then investigate whether there are significant
comes after the unsupervised learning is to test whether the patterns identified by an algorithm correspond to climatologically distinct environments and if so which ones. TDA is an alternative approach to address the lack of labels. With TDA we seek to match imagery to the original four classes identified by Stevens et al. (2020) , yet only require a small number of labeled samples. We map patches of the MODIS imagery into topological space and then investigate whether there are significant
