Savanization is an efficient strategy to confront desertification by increasing herbaceous productivity in drylands providing income to local population relying on grazing. Hence, to assess successful savanization herbaceous production must be estimated accurately. The conventional technique uses random sampling, which might misestimate productivity underneath the canopies due to tree-grass interactions. Here we present an improved model to assess biomass production accounting for tree-grass effects using a stratified sampling technique. Our model calculates biomass underneath the canopy in two configurations: (a) a cone shape, accounting for gradual changes along the bole-to-drip line with radiuses representing topographic aspects, and (b) a cylindrical shape, accounting for biomass underneath the canopy not affected by the tree. We tested our model in the Acacia victoriae savannah of Yatir at the Northern Negev drylands, Israel. Results showed that biomass underneath the canopy were up to 3-fold higher than the measured in between trees. Although the total canopied area was only 4.4% of the savannah, biomass underneath canopies constituted 7% of the total savannah production. Thus, conventional sampling might significantly underestimate biomass production in denser savannah. Our model was adjusted to multi-species savannah and different geographic aspects and could be used in drylands systems elsewhere.