I have spatio-temporal albedo (roughly, the 'reflectivity' of earth's surface) dataset, from NASA's MODIS satellite, for a 130 square kilometer area. The dataset contains raster files in the NetCDF format, with a file for each day, and a grid size of 500 m*500 m. There are a lot of 'NA' values in each file, due to cloud cover, satellite errors etc. Till now, I have simply spatially averaged the albedo data from the dataset to construct a simple time-series. I use this time-series to create a machine-learning based model to predict snow water equivalent.

I want to see if there's a way to include the spatial variability in the dataset, in the time-series. I'm also curious to know what would be the best way to spatially interpolate the data.

  1. Is there a way I can condense the variability, which might be due to factors such as elevation, aspect and slope of the area, into one or more time-series?
  2. I have looked at Principal Component Analysis/Empirical orthogonal functions to do the above. Can such methods be used for spatial averaging?
  3. What would be the best way to spatial interpolate, considering the numerous NA value cells? Is there a way to take into account the elevation, and other factors, into the interpolation?

Any suggestions would be greatly appreciated. Thanks!

Note: I use R for my analysis.

  • $\begingroup$ Spatio-temporal modeling is a huge and quickly evolving subject. Even one book cannot do it justice, but you might start with a look at Cressie & Wikle. The answers to your questions are--in order--what do you mean by "condense?"; maybe; and definitely yes, but the "best way" mentioned in #3 depends on many, many factors. $\endgroup$ – whuber Mar 31 '15 at 22:51
  • $\begingroup$ @whuber Thank you. By condensing, I mean being able to represent the spatial variability of the satellite data in a single time-series (or a small number of series), in that same way that I do now by spatially-averaging. I'm worried that I'm losing the spatial information by simple averaging. $\endgroup$ – small_world Apr 2 '15 at 18:32

I can't comment on 1.) and 2.), but for spatial interpolation with multiple independent variables, a commonly used method is kriging, also called co-kriging if used with covariates.


The method is implemented in several R packages and thus should be easy to implement; a general overview, including worked examples and more advanced methods, is given in chapter 8 of [Bivand, Pebesma, Gómez-Rubio]; Applied Spatial Data Analysis with R; Springer 2013. [Chun & Griffith] Spatial Statistics & Geostatistics; Sage 2013, also gives an overview and worked examples in R, but is a much less accessible text more suited for experts that just want a out-of-the-box code.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.