Continental flood basalts intruded and erupted millions of km$^3$ of magma over $\sim$ 1-5 Ma. Previous work proposed the presence of large ($>$ 10$^5$-10$^6$ km$^3$) crustal magma reservoirs to feed these eruptions. However, in Paper I, we illustrated that this model is inconsistent with observations, by combining eruptive rate constraints with geochemical and geophysical observations from the Deccan Traps and other CFBs. Here, we use a new mechanical magma reservoir model to calculate the variation of eruptive fluxes (km$^3$/year) and volumes for different magmatic architectures. We find that a single magma reservoir cannot explain the eruptive rate and duration constraints for CFBs. Using a 1D thermal model and characteristic timescales for magma reservoirs, we conclude that CFB eruptions were likely fed by a number of interconnected small-medium ($\sim$ 10$^2$ - 10$^{3.5}$ km$^3$) magma reservoirs. It is unlikely that each individual magma reservoir participated in every eruption, thus permitting the occasional formation of large xenocrysts (e.g., megacrystic plagioclase). This magmatic architecture permits (a) large volume eruptive episodes with 10s to 100s of years duration, and (b) relatively short time-periods separating eruptive episodes (1000s of years) since multiple mechanisms can trigger eruptions (via magma recharge or volatile exsolution, as opposed to long term (10$^5$ - 10$^6$ year) accumulation of buoyancy overpressure); (c) lack of large upper-crustal intrusive bodies in various geophysical datasets. Our new proposed magmatic architecture has significant implications for the tempo of CFB volatile release (CO$_2$ and SO$_2$), potentially helping explain the pre-K-Pg warming associated with Deccan Traps.

Flood basalts are some of the largest magmatic events in Earth history, with intrusion and eruption of millions of km$^3$ of basaltic magma over a short time period ($\sim$ 1-5 Ma). A typical continental flood basalt (CFB) is emplaced in hundreds of individual eruptive episodes lasting decades to centuries with lava flow volumes of 10$^3$- 10$^4$ km$^3$. These large volumes have logically led to CFB models invoking large magma reservoirs ($>$ 10$^5$-10$^6$ km$^3$) within the crust or at Moho depth. Since there are currently no active CFB provinces, we must rely on observations of past CFBs with varying degrees of surface exposure to develop and test models. In the last few decades, significant improvements in geochronological, geochemical, paleomagnetic, volcanological, and paleo-proxy measurements have provided high-resolution constraints on CFB eruptive tempo - the volume, duration, and frequency of individual eruptive episodes. Using the well-studied Deccan Traps as an archetype for CFB systems, we compile multiple lines of evidence - geochronology, eruption tempo, dike spatial distribution, intrusive-extrusive ratio, geochemical variations, and volcanological observations - to assess the viability of previous models. We find that the presence of just a few large crustal magma reservoirs is inconsistent with these constraints. Although observations from the Deccan Traps primarily motivate our model, we discuss constraints from other CFBs to illustrate that this conclusion may be broadly applicable, with important implications for interpreting CFB geochemical datasets as well as the timing and volumes of climate-altering volatile emissions associated with CFBs.