Multiple frequency particle filtering approaches in paleoclimatology

Paleoclimatology makes use of particle filters to reconstruct past climates using climate proxies which come with a wide variety of time resolutions. As existing particle filters have trouble to adapt to these varying timescales, we evaluated five online particle filtering methods for accuracy and resistance to filter degeneracy. The methods were first tested on multivariate setups on Lorenz63, a three dimensional chaotic system. RMSE and correlation analyses were performed. Three out of the five particle filters were then applied to a more complex climate model (LOVECLIM). We opted for a pre-industrial climate reconstruction with both ocean and land proxies located in the upper northern hemisphere. We performed correlation analyses to evaluate reconstructions in different zones on and around the proxy locations. We showed that all investigated methods could be applied to both systems with some limitations.