Hi,
I have 26 sampling locations, relatively well distributed, but not adequately enough to cover the study area. I have also a digital elevation model, from which I can extract much more locations. An important detail is that the temperature data are distributed in low elevations, but no temperature data is available at high elevations. Hence, I want to use the elevation data to sensitize the temperature interpolation process through co-kriging, which works with non collocated points. I prefer to use gstat, but I am not sure: a) what is an optimal number of elevation points that I have to use (I know that I have 26 temperature points, and that I need elevation points at high elevations) b) is there an automatic process tha provides reasonable kriging predictions, such as autokrige, for co-kriging?
Konstantinos Panagiotou
1. Prepare the data for analysis. Collect the temperature data and digital elevation model (DEM) data into a dataset, making sure the data is in a compatible format.
2. Perform exploratory data analysis. Produce scatter plots, histograms, and other visualizations of the data to assess correlations between temperature and elevation.
3. Fit a co-kriging model to the data. Use the temperature data and elevation data to find the best fit for a co-kriging model.
4. Use the model to make predictions. Use the model to predict temperature values at different elevation levels.
5. Validate the predictions. Compare the predicted values to the actual temperature values at different elevation levels to determine the accuracy of the model.
You don't mention the specifics of how large your area of interest is (i.e. a 1 sq km, 10 sq km, 1000 sq km, 10000 sq km, etc.) or what exactly 'low' and 'high' mean, along with the baseline those are stating from, or where in the world it is - which would indicate the orography and general texture of the terrain relief. Perhaps extreme, but in my neighborhood, that can range from 2000 ft to 14,000 ft in 18 miles - 50 miles east that same distance has a range of about a 100ft, not 12,000 ft of elevation. Also you don't mention the chronological aspects and intervals.
Merely creating some sort of interpolated surface between your samples, depending on the above is self referential and artificial. The actual 'surface', again depending on the area of interest, is another 'DEM', grid data from the weather model of the region appropriate for the AOI which captures the general trends in the AOI.
Also, there is another correction, depending on your data source, it is not only the measured temperature on the interpolated surface, but the Density Altitude relationship correction, which will vary across the surface. Even very small variation in pressure, temperature, and altitude can have influences of a thousand feet - this life or death correction is done immediately before every airplane flight.
These aren't the only factors, and whether they matter of not depends on the extents and sampling of your area of interest, and your analytical goals.
- Badges
- Science topic
- Similar topics
- Geoscience
- Cartography
- Mapping