GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.2 GNGTS 2024 injecton tme of 2 s with 4 stacks, a tme delay of 40 ms and a logarithmic sampling of the IP decay curve using 20 gates. We also logged leachate levels in fve diferent piezometers along L2 and L3. We inverted ERT/TDIP data for resistvity and integral chargeability using the VEMI sofware (De Donno G., Cardarelli E.; 2017), which employs the fnite element method for solving the forward problem and a Gauss-Newton inversion algorithm, while the linear approach (Method I afer Oldenburg D. W., Li Y.; 1994) is adopted for chargeability forward modelling. Afer inversion we added to the resistvity ( ρ ) and integral chargeability ( M ) models also the normalized chargeability ( M n ) to emphasize the contributon of surface conductvity, derived by normalizing chargeability by the resistvity. Fig. 1 – Aerial image (a) and plan (b) of the municipal solid waste landfll in Central Italy, with the locaton of the four investgated ERT/IP lines (L1-L4) and of the fve piezometers (P1-P5) To integrate informaton from electrical inverted models, we applied a machine learning-based approach (Piegari E. et al.; 2023), which is based on a cluster analysis in the parameter model space defned by the inverted values of ρ , M and M n . However, instead of the widely-used K- means, we performed the cluster analysis by using the Fuzzy C-Means algorithm (Bezdek J. C.; 2013). This algorithm is a sof clustering approach, which allows data points to belong to multple clusters with a degree of membership µ that is a functon of the distance to the closest centroids. This makes the algorithm more robust to noise and outliers in the data compared to K-means and we can also have an assessment of the reliability of the clustered secton by means of the degree of membership secton. The optmal number of clusters was assessed by the Elbow method applied to the explained variance, while the quality of the results was fnally evaluated by calculatng the Silhouete value (Rousseeuw P. J.; 1987). At the end of the clustering procedure, we achieved a single model that integrates all geoelectrical informaton providing an accurate identfcaton of leachate accumulaton zones through a color- based scale, where dark red color is associated to the fully saturated zones.