Task #361
open
Test and compare the GAM method on several days (10 days for now) for Oregon
Added by Benoit Parmentier almost 13 years ago.
Updated over 12 years ago.
Activity type:
Coding/analysis
Description
The GAM regression is applied on a set of 10 days spread over a year to test its performance. We expect changes in the prediction fit with the different time of year (e.g. Winter and Summer) so replicating the same method for a set of 10 days will be useful for for the development of the new methods.
A r script will be developed to produce and compare the outputs for the 10 selected days.
Files
- Category set to Climate
- Assignee set to Benoit Parmentier
- Activity type set to Coding/analysis
- Parent task deleted (
#359)
- Status changed from New to In Progress
Beginning of the r code to produce prediction for 10 dates. Fit and validation will be evaluated using AIC, GCV, deviance and RMSE,.
I am reading about interactions and experimenting with the visualization tools.
At this stage, it is still unclear to me what interaction matters (among the variables:lat,lon,ELEV_SRTM, Eastness, Northness). There are also some errors in the input data which resulted in errors in the fit. There is a need to screen the data before performing the GAM...This requires some additional r coding.
- The R code has been updated to include some interaction terms. Examination of 3D surface plot with tmax on the z-axis and the various independent variables (lat, long, ELEV_SRTM,...) reveals different interesting features (see pdf document attached). So far two interactions terms have been included: lat-long and Eastness-Northness. To replicate the ANUSPLINE model a three way interaction term will aslo be included in the GAM model (lat,long, ELEV_SRTM).
Many models have been explored using the GAM methodology. At this stage, more detailed diagnostics is needed so for the past two weeks work has focused on the analysis of:
- patterns in residuals.
- outliers detections and screening of "bad" values.
- examining the contribution of various variables to the models in particular the land cover and LST averages.
Also available in: Atom
PDF
