Autonomous micro-grids based on solar photovoltaic (PV) are one of the most promising solutions to provide electricity access in many regions worldwide. Different storage/PV capacities can produce the same level of quality service, but an optimal design is typically identified to minimize the levelized cost of electricity. This cost optimization however relies on technical and economic hypothesis that come with large uncertainties and/or spatial disparities. This article explores the sensitivity of the optimal sizing to variations and uncertainties of such parameters. Using data from Heliosat and ERA5, we simulate the solar PV production and identify the least cost configurations for 200 locations in Africa. Our results show that the optimal configuration is highly dependent on the characteristics of the resource, and especially on its co-variability structure with the electric demand on different timescales. It is conversely rather insensitive to cost hypotheses, which allow us to propose simple pre-sizing rules based on the only characteristics of the solar resource and electricity demand. The optimal storage capacity can be estimated from the 75th percentile of the daily nocturnal demand and the optimal PV capacity from the mean demand and the standard deviation of the daily power difference between solar production and demand.