Ash Electrical Conductivity Prediction Using Univariate and Multivariate Models.

Ash is the most common material in post-fire environments, and is fundamental to plant recover in the immediate period after the fire. The distribution of ash on soil surface protects soils from erosion and is an important source of nutrients. The total availability of nutrients in ash is measured by the Electrical Conductivity (EC) of the slurries. The amount of nutrients dissolved on ash extracts depends on ash mineralogical properties, especially the amount of CaCO3. The presence of major elements, as Calcium (Ca), Magnesium (Mg), Sodium (Na) and Potassium (K) also influence ash EC. The objective of this work is try to predict ash EC using several predictors, as CaCO3, Ca, Mg, Na, K, Sulphur (S), Silica (SiO₂) and Total Phosphorous (TP). A total of 40 ash samples were used in this work. Ash was collected after a medium to high fire severity in Portugal. Since data did not follow the Gaussian distribution, a logarithm was in order to respect normality requirements. Linear and multivariate regressions were used to identify the best EC predictors. Significant regressions were considered at a p<0.05. The results showed that Ca, Mg, K and S explained significantly, 28, 29, 22 and 12% of the EC variability. The remaining variables did not have significant regressions. The variables which showed significant linear regressions were used in the multivariate models. All the multivariate regression models (CaxMg, CaxK, CaxS, MgxK, MgxS, KxS, CaxMgxK and CaxMgxKxS) explained significantly the variability of EC. CaxS was the model with least capacity of explanation (28%), while CaxMgxK and CaxMgxKxS, had the highest capacity of explanation (44%) of EC. However, CaxK explained a total of 43% of EC variability, showing that these two variables can explain the larger part of EC variance.  In all the models, the observed values were significantly correlated (p<0.05) with the predicted, the residuals followed the normal distribution and the errors were very close to 0.