Using Fixed Rank Filtering to Make Spatio-Temporal Interpolation of MODIS Temperature

Kriging is widely-used for spatial structure exploration and spatial data interpolation. However, most kriging methods are designed for spatial interpolation and not for spatio-temporal data interpolation. After constructing basic functions for a specific experimental area, we made spatio-temporal predictions for MODIS temperature data with Fixed Rank Filtering (FRF). We compared these prediction results with the interpolation results of Fixed Rank Kriging (FRK) and discuss the differences. Experimental results show that when points are evenly distributed in space, the FRK method obtains higher prediction accuracy with results slightly better than FRF. However considering the temporal relevancy of temperature, when data is missing in a larger area, the FRF method shows a capability to comprehensively exploit spatio-temporal information better than other methods for achieving higher quality interpolation results in cases that lack spatial information.