Although the accuracy of the dataset at a particular grid point is dependent on the number of surrounding weather stations available for the interpolation, the computations shown here are based on time series averaged in a large–scale spatial domain, which may minimize the errors in some regions due to the poor data coverage. 

The CRU TS 3.24.01 dataset was chosen for this study because it provides both precipitation and temperature, fields necessary for the computation of the SPEI index, thus reducing errors due to the mixing of different data sets.

The Lagrangian method applied here is not able to calculate E and P separately. It is also limited by the use of the time derivative of moisture, in which unrealistic fluctuations can be interpreted as moisture fluxes, as well as by the limited resolution and uncertainties of the input data. Nevertheless, the random errors may cancel each other out given the large number of particles found in an atmospheric column.

The regions actuating as major moisture sources were identified from the climatological annual means of E-P, which summarise the most relevant moisture sources throughout the year. Such moisture sources are not stationary (Gimeno et al., 2013), and although defining them at a monthly scale would be more realistic, a comparative analysis of the evolution of its contribution throughout the year would not be possible when considering the monthly variation of spatial domain of the sources.

The estimative of the residence time is still a controversial issue, being dependent on the methodologies applied. Although the residence time can vary both spatially and seasonally, this study considers the global 10-d residence time in order to apply the same value for all the regions investigated here and to be consistent with most previous studies applying the same methodology around the world.

SPEI was computed from time series averaged over the RR domain in order to identify the domain-scale episodes affecting each region. However, because each drought event is unique in terms of its temporal and spatial development, a grid point analysis would reveal areas more affected by dry conditions during a particular episode.