The results of cross validation for LR and RK are shown in Fig. 6. Incorporation
of spatial dependency improves the performance of the
model in two ways, decreasing the error and improving predictive capability.
Using RK the value of R2 increased from 0.61 to 0.69. This moderate
increment also reduced the error in estimatedWSC from 11.01 to
9.77 t ha−1, which represents a RER of 11.3%. In terms of predictive capability,
Fig. 6 graphically shows that RK increases the capability for
predicting high biomass values. At WSC values close to 80 t ha−1, the
value estimated by LR is 60 t ha−1while RK predicts 74 t ha−1, a figure
that is nearer to the observed value.