Hello!
I have run RF in Rattle using my AGB data (300 sample plots) as target and 15 variable (SAVI, NDVI, RDVI, NIR etc.). So I have two basic questions;
1. How I can know which tree, rule and node should I take for the spatial biomass estimation?
2. How I can apply selected rule on my spatial variables in ARCMAP (GIS software)?
I guess there is no straightforward solution to your question. Random forest uses really simple trees (it is based on a little number of variables only, drawn at random, for each node). However, there is a measure of the importance of the variables (type ?importance.randomForest in R.
That might be a first step toward what you are looking for...
Do not hesitate to ask more precision directly to my profile if you want more precision, I am not often on ResearchGate lately...
i wonder wether you would not be more interested in boosted regression trees, rather than random forests. The results from random forests is quite often a bit erratic, and it deals less well with continuous variables than boosts regression trees (use the gbm package). I've never managed to properly implement a RF or GBM model in argGIS. I would rather apply the model to images in R directly using thebraster package. I can share an annotated script that I supply to my students in case you're interested.
Dr. Thomas! Thanks for your suggestions. After struggling with continuous AGB data (target) and NDVI, SAVI, LAI, Elevation etc. (Inputs). Actually I have 300 sample plots.
I found that non of the these Data Mining Models (CART, RF) are producing sufficient accuracy. So I made 6 and 4 classes (target) and now achieved 60% accuracy. But still I am struggling to take the RF rules and apply in arcmap for spatial biomass mapping for entire area. Please do share your code and let me see how far I can go.
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics
- Geographic Information System