I emailed Roland and watched the video below, here is a summary of all that:
In the video he says that he believes 70% of drought crop loss can be mitigated by an established gm/ccs (6 years old or old). When I asked him for some specific numbers from the field, he he said he prefers to use the average results from successful systems, not the best case scenario. So based on that data, an effective gm/cc system will yield 50% or more of a normal harvest, even during drought years!
Here are the numbers. He sources this information from regions where the farmland was degraded and average corn yields were getting as low as 1 tons/ha by common farming methods. In these areas, established gm/ccs systems were able to routinely increase yields by 3 fold! So they get yields of 3 t/ha.
In drought years, the normal farms were harvesting around 0.3 t/ha, whereas the gm/ccs systems were getting 1.5 t/ha or more on average. So that compared to the normal 3 t/ha on a good rainy season, they are still getting 50% of their harvests.
These results were averaged from at least a dozen different small farms in Mali, Zambia, and Madagascar, where various gm/ccs (a variety of species in Mali, gliricidia in Zambia and a 9-month cowpea in Madagascar) were being used.
If you want more information on this, this video of his presentation at the Nov 2019 ECHO conference is great and of course his book, Restoring the Soil is the best source: