Unfavorable hydrological conditions experienced from 2014 to 2019 led to the depletion of main reservoir systems in Brazil, causing an increase of thermal energy dispatch. However, an important share of the observed thermal generation was out of economic merit, commanded by government entities which risk perception relies mainly on experts tacit knowledge. Despite the common sense that storage in reservoirs is intrinsically linked to system security, the metrics employed so far failed to compute the system s real needs in terms of required stored energy in hydroelectric plants. By the end of 2019, ONS proposed a new method to assess the need for additional thermal dispatch the Referential Storage Curve (CREF - Curva Referencial de Armazenamento). However, it fails as a reference for the security of energy supply since it considers very specific assumptions of rivers inflows and thermal generation. Besides, based on its iterative trial and error process, it can result in sub-optimal results of minimum storage levels. This work proposes a new method to evaluate the security of power supply in systems with predominance of hydroelectricity. This method is intended to be an evolution to the CREF method, and it is based on the development of an optimization model that computes the minimum secure levels for hydroelectric plants operation in each month, from a recursive simulation of historical inflow series from 1931 to 2018. In addition, based on the simulation results, reference curves were suggested for the continuous monitoring of the reservoirs operation, with the purpose of subsidizing Brazilian government entities decisions on unorthodox thermal generation dispatch. The monitoring of the proposed reference curves is expected to represent a more robust criterion for decisions on out-of-merit thermal generation in Brazilian power system.