Pre-salt carbonate: Tupi (formerly Lula) microbialites under 2 km of salt

The rocks under the salt

• Stratigraphy. The reservoirs belong to the Guaratiba Group, laid down during and after the rifting that opened the South Atlantic in the Early Cretaceous (Moreira et al. 2007). The main reservoir at Tupi is the Barra Velha Formation, of Aptian age, from the post-rift sag phase (Cruz et al. 2021). Beneath it lies the Itapema Formation, of Barremian to early Aptian age, from the rift phase. Above both lies the Ariri Formation, the Aptian evaporites that seal them.
• The Barra Velha carbonates were deposited in a large alkaline lake. They are built of shrub-like crystals of fibrous calcite and millimetre-sized calcite spherulites, set in or replacing magnesium-silicate clays such as stevensite. They were long called microbialites, but whether microbes built them is disputed: Wright and Barnett (2015) argue that the textures can form without biology, like travertines. The clays were later partly dissolved, which left moulds as well as pores between the shrubs (Wright, GeoExpro).
• The Itapema coquinas are beds of bivalve shells, tens to hundreds of metres thick, with porosity between and inside the shells; they are a second reservoir in several fields.
• Reservoir quality. In the oil zone of the Tupi pilot area the average effective porosity is about 9 %, from a cutoff of 6 % to about 24 %, and the average permeability about 200 mD, varying strongly with rock type. The better rocks sit on structural highs, where the lake's higher-energy deposits accumulated (Cruz et al. 2021).
• The fluid is a light oil of 28 to 30 degrees API with gas-oil ratios of 200 to 300 m³/m³ (Cruz et al. 2021). Carbon dioxide makes up roughly 8 to 15 % of the gas that comes out of the oil (MIT Carbon Capture and Sequestration Technologies database), which shapes the whole development.
• The salt is mostly halite and anhydrite, with layers of carnallite (a potassium-magnesium chloride) and tachyhydrite (a calcium-magnesium chloride), salts that creep much faster than halite and trouble drilling (Fernandes et al. 2020).

Exercise: read Figure 9.5

1. Open the figure as it starts: oil-filled limestone at the Tupi pilot's mean porosity, 9 %, a quarter of its pore space in moulds and vugs, 1 % in microcracks. Plate (a) places the model in the discovery well's column; its $V_P$ is 5506 m/s.
2. In plate (b) follow the dashed level line at that velocity. It meets the all-interparticle curve at 7.8 % porosity and the all-mould curve at 17.3 %. That 9.5-point window is the headline value: what velocity alone cannot tell you about porosity in a carbonate.
3. Set the moulds and vugs to 0 and then to 100 %, with the cracks at zero. At 9 % porosity $V_P$ goes from 5416 to 6127 m/s, 711 m/s from pore shape alone.
4. Switch the fluid to brine. $Z_P$ rises by only 2.6 %: the stiff frame leaves little for the fluid to change. Then raise the microcracks to 5 % and watch plate (c): the gap between Gassmann and the isolated-pore limit grows from 14 to 108 m/s in oil.
5. Switch the rock on the reservoir to anhydrite. The base of the salt turns from a peak, +0.180, to a trough, −0.125. Back on halite, raise the porosity past 22.1 % to see the same contact change polarity on porosity alone.

Why carbonate rock physics is different

The workflows of Parts 5 and 7 were largely built on sandstones, where porosity sits between grains and its shape varies little from rock to rock. Carbonates are made of reactive minerals that dissolve and recrystallise, so their pores come in many kinds: between particles, inside them, moulds of dissolved grains or clays, vugs, micropores and fractures.

• Pore shape sets the stiffness. A rounded mould weakens the mineral frame little; a flat pore or a crack weakens it a lot. At one porosity, carbonate velocities therefore spread widely: Eberli et al. (2003) showed that rocks with moulds and intraframe pores stay fast at high porosity while microporous and cracked rocks are slow, and Weger et al. (2009) quantified the effect of pore structure on velocity. Xu and Payne (2009) model it with three families of spheroidal pores, the approach Figure 9.5 follows.
• The classical transforms leave shape out. Wyllie's time average, $1/V_P = (1-\phi)/V_{\mathrm{ma}} + \phi/V_{\mathrm{fl}}$, averages slownesses and knows nothing of pore shape. The Hashin-Shtrikman bounds hold for any mixture, but for a stiff mineral with soft pores they are far apart, so they bracket rather than predict.
• Gassmann's equation strains. It assumes one homogeneous mineral, pores connected well enough that the fluid pressure evens out during a wave cycle, and no chemical interaction between fluid and frame, so the shear modulus does not change. The Barra Velha mixes calcite, dolomite, silica and magnesium clays; carbonates hold isolated vugs, micropores and cracks; and laboratory measurements show shear moduli that change with brine saturation (Adam, Batzle and Brevik 2006; Baechle et al. 2009). At seismic frequencies Adam, Batzle and Brevik found that Gassmann predicted the bulk modulus of carbonates whose pores are rounded, which is where Figure 9.5 shows it holding.

The consequence for interpretation is direct. An inversion that returns $V_P$ or $Z_P$ brackets porosity only within a window set by the unknown pore types, and the pore types must come from cores, thin sections and logs. Porosity and facies prediction in the pre-salt is therefore a calibrated, probabilistic exercise (Sections 7.4 and 7.5), not a single transform.

Imaging through the salt, and watching the reservoir change

• The salt bends the waves. Halite carries $V_P$ near 4500 m/s, far faster than the post-salt section above it, so ray paths bend sharply at the top of the salt and the subsalt image depends on how well the salt's shape and velocity are known (Section 9.3). The Ariri evaporites are layered: anhydrite is faster than halite and the magnesium salts are slower, so the layering must enter the velocity model and generates interbed multiples.
• The base of the salt can be bright, dim or reversed. Its reflection depends on what sits on each side. Figure 9.5, plate (d), shows halite on a tight, stiff carbonate giving a peak and on a porous or cracked one a weak or reversed reflection, while anhydrite on the same rock gives a trough.
• Time-lapse needed nodes. Feasibility studies predicted that production would change the impedance of Tupi's stiff carbonates by only about 1 to 2 %, near the limit of detection. Petrobras therefore acquired the Tupi Nodes Pilot, the first ocean-bottom-node survey in the ultra-deep Santos Basin: 954 nodes on 36 lines over 111 km², a baseline in 2015 and a monitor in 2017. After interbed-multiple attenuation and least-squares migration the repeatability reached about 2 % NRMS at the pre-salt level, and the differences tracked the water and gas injected (Cruz et al. 2021).
• Carbon dioxide goes back down. The produced carbon dioxide is separated on the FPSOs and reinjected. At Tupi, reinjection began in a pilot in 2011 (MIT Carbon Capture and Sequestration Technologies database), and the pilot area alternates water with gas rich (about 80 %) or poor (about 5 %) in carbon dioxide (Cruz et al. 2021). Across its pre-salt fields Petrobras reported reinjecting 14.2 million tonnes of carbon dioxide in 2024 and 67.9 million tonnes since 2008 (Agência Brasil 2025).

What remains uncertain

• How the Barra Velha formed. Microbial or abiotic, the origin of its shrubs and spherulites changes how facies and their pore types are predicted away from wells.
• Pore types between the wells. Seismic velocity alone leaves a wide porosity window (Figure 9.5); narrowing it needs pore-type information that only cores and logs provide, and its spread between wells is a geological prediction.
• How far Gassmann can be trusted. At seismic frequencies connected pores are a fair assumption for most of the rock; cracks, isolated vugs and micropores, and fluids that react with carbonate, are where it fails, and their share in the reservoir is uncertain.
• The velocity of the salt. The intrasalt layering is mapped only where wells and good images reach, and errors in it move the subsalt image.
• Small time-lapse signals. With impedance changes of a few percent, repeatability and residual multiples decide what a monitor survey can show (Cruz et al. 2021).