Precipitation is a complex process which is both extremely varying and intermittent over a wide range of space-time scales. At the same time, precipitation is a key variable of urban geosciences. Multifractals have been developed to analyze and simulate across scales this variability and intermittency, while climate networks characterize the spatial dependence patterns and teleconnections at a given scale, typically at the resolution of the data. It is natural to analyze how climate network models behave at different scales. While the ultimate goal is to understand how to downscale climate networks, a preliminary step is to understand how to upscale them having the possibility to test the resulting approach on empirical data. The latter is attempted in this communication using the TRMM 3B42-v7 daily resolution dataset. At first, we define networks at larger space-time scales, by first coarse graining the precipitation data, then applying the same type of climate network definition to these coarse-grained data. The delicate question is how to suitably renormalize the characteristic space-time scales and threshold of similarity definition of time series. We show this can be done using multifractal guidance, especially to preserve the scaling relation between time and space. On the other hand, one can directly coarse grain the network. We conclude this communication by a detailed comparison of both types of upscaling, in particular with the help of classical characteristics of complex networks like the degree centrality, local clustering, closeness centrality, etc.