MARQUES, R., ZÊZERE, J.L., AMARAL, P., GASPAR, J.L., QUEIROZ, G. (2010) - Empirical Landslides Early-Warning System (ELEWS-Pov) development for the Povoação county (S. Miguel island, Azores) based on the application of the Generalized Extreme Values distribution (GEVd) for rainfall-triggering thresholds assessment. Bateira C., Soares, L., Gomes, A. E Chaminé, H.I. (Eds). Actas / Proceedings do V Congresso Nacional de Geomorfologia. Porto, 8 - 11 Dezembro de 2010. Associação Portuguesa de Geomorfólogos. pp 327-332.

 For the development of the Empirical Landslide Early‐Warning System to Povoação County (ELEWS‐Pov), the role of rainfall on landslide activity was analysed. The process comprised the reconstruction of absolute rainfall for different rainfall durations (D=[1,90] ϵ IR) associated with each one of the 55 landslide events identified from 1976/77 to 2007/08. The critical rainfall combination Intensity(I)‐Duration(D) responsible for each landslide event was assessed using the Generalised Extreme Values distribution (GEVd). For the distribution parameters estimation it was used the Maximum Likelihood (ML) method. Landslide events in the study area are ruled by the function I(D)=136,86 D^‐0,556 (R² =0,94). For D≠1, I values underestimate the rainfall registered on the day of the event (P_D1) increasing this bias with D→90. To overturn this bias it was calculated the correlation function between D and P_D1 given by P_D1(D)=131,81 D^‐0,143. Rainfall‐triggered landslides threshold was defined using the best boundary function (BF) between landslide events and non landslide events that result in the highest ratio between true landslide events truly predicted and false positives for both I‐D and P_D1‐D correlations. For I‐D and P_D1‐D correlations the BF functions correspond to BFI(D)=61,93 D^‐0,4643 and BFP_D1(D)=61,93 D^‐0,0582, respectively. The alert and the alarm of the ELEWS‐Pov are activated for a specific date dx(I(D); P_D1) when the rainfall threshold conjunction dx(I(D)) ≥ BFI(D) ˄ dx(P_D1) ≥ BFP_D1(D)‐c is satisfied using c=0 to the alarm and c=10 for the alert. The threshold conjunction was normalized by the mean annual precipitation (MAP) and applied to 1 meteorological stations and 4 other rain gauges located at Povoação County. All the calculation processes were automated in MATLAB language, for update the functions at the end of each year or at each new landslide event, giving a dynamic behaviour to the rainfall threshold conjunction. A JAVA application was developed to read real‐time rainfall data and automatically calculate if the alarm or the alert conjunctions are satisfied, sending alert and alarm messages by e‐mail and SMS.