In Action! – Module 3 (A3), we will simulate data for the entire field (one simulation per image pixel). The known profiles are used to predict the whole map using linear fitting.
Here, the coarse map will be generated, Python file: 3.1_one_feat_approach_ind_calc.py. The known profiles are used to predict the whole map using linear fitting. The leave-one-out cross validation method (LOO) was used, since there are too few data for the k-fold cross validation method.
As observed in the A1 – Step 6 section, the profiles have similar texture. Since the LOO indicated that the error is relatively higher than the standard deviation regarding texture depth changes on each specific location, the predicted topsoil layer will be used the same for all other deeper layers. Please, note that this is a characteristic specific to this field. It is expected that other fields have different behaviour.
At this stage, the attendant also has the coarse SOC topsoil map. For the other SOC layers, the next step is.
At this stage, the attendant will calculate both the second and third layers for the SOC. The model that will be trained is the exponential model explained on Lesson 3 theory. Although an exponential function is not linear, it is common to call the fitting process “linear”, because there are transforms that it is possible to be done which converts the exponential model into a linear model, allowing the traditional linear fitting to be used in such cases.
The second and third layers are generated in the script 3.2_soc_fit.py.
The script to create MONICA files for the whole field is 3.3.a_run_field_from_maps.py, and to run MONICA, the same script with small changes: 3.3.b_run_field_from_maps.py.