All results in this paper are based upon a new dataset consisting in 60 Swadesh lists of 207 items, overall 12,420 terms collected during 2018-2019. Each list corresponds to a different variety of Malagasy, which is not simply identified by the name of the ethnicity but also by the precise location where the variety was collected. This is very important since some traditional ethnic groups are a heritage of historical events rather than representing communities with similar habits and dialects. This new dataset is by far the best available, both for dimension and completeness. The varieties are classified both by standard tools, as the trees generated by UPGMA and NJ which privilege genealogy by detecting vertical transmissions, and by a new method which privileges horizontal exchanges. The new method results in a two-dimensional chart of Madagascar which realistically reproduces geography despite being generated only by comparison of words. The landing date of the ancestors of Malagasy is determined about 650 CE. This result is obtained by a straightforward approach based on the comparison of the UPGMA Malagasy family tree with the analogous tree of Romance family of languages for which all dates are well historically attested. We also propose an improved definition of Diversity computed for every locus in Madagascar and not only in places where the dialects were collected. Moreover, Diversity becomes a locally determined quantity as it is usually in biology. Diversity differences point to the South-East coast as the location where the first colonizers landed or, at least, where Malagasy variants started their dispersion. Finally, we find that the dialect spoken by the Mikea, a hunter-gatherer people in the South-West of Madagascar, is not very different from the variants of their neighbours Vezo and Masikoro. Therefore, Mikea unlikely can be linked to eventual aboriginal populations living in Madagascar prior to the main colonization event in 650 CE.