Deepwater turbidite QI: Girassol-style fluid prediction

What has been published about the seismic

• High-resolution 3D was the main tool to describe the reservoir and to monitor it (Lefeuvre et al. 2004).
• Gas re-injection: the produced gas went back into the reservoir, under the operator's policy of recycling it, and following that gas was the main reason a high-resolution 4D survey was shot so early in the field's life (Lefeuvre et al. 2004).
• The first monitor was shot in the last two weeks of 2002, over the high-resolution base survey of 1999, after a year of production and about six months of gas injection (Jourdan, Lefeuvre and Dubucq 2006; Gonzalez-Carballo et al. 2006). The 4D was used to refine the flow model and to place the last development wells (Lefeuvre et al. 2004), was inverted for impedance (Lafet et al. 2005; Hubans and Berthet 2007), and was matched with the production history (Roggero et al. 2012).
• What this section does not claim: the AVO class of Girassol's sands, its rock-physics template, the field's own depth below the mudline and the success rate of its wells are not in these sources, so the figure models them instead of quoting them.

The question, and the three legs

An amplitude map shows where the rock is soft. Clean porous sand is softer than shale with brine in it, and softer still with oil, so brightness alone cannot say which it is. A drill decision asks three questions, each answered from a different product:

1. Sand: is there enough reservoir? Net sand from an impedance inversion (Section 7.4).
2. Fluid: does the sand hold oil? The AVO gradient: an oil sand's top lies below the trend of the wet rocks in a crossplot of intercept $A$ against gradient $B$ (Castagna, Swan and Foster 1998; Section 5.5). Its distance below that trend is a fluid factor, in the spirit of Smith and Gidlow (1987) and Fatti et al. (1994), whose weighted stack vanishes for wet rock; Foster, Keys and Lane (2010) show that the distance grows with the fluid's compressibility.
3. Closure: can the trap hold a column? The top-reservoir depth map, each closure followed down to its spill point (Section 3.6).

Two maps that respond to the same property, such as RMS amplitude and impedance, both of which fall for soft rock, are one leg counted twice. The fluid needs the gradient.

The model in Figure 9.1

The figure is a model in Girassol's style, not its data. A 6 by 6 km window holds a 40 m interval of channel fill, levees and two lobes (Part 4) in Part 5's medium shale, draped over three domes under 1,400 m of water and about 2 km of burial, the deep end of Block 17's range, where Part 7's soft-sand frame is calibrated. The sand's frame is Part 7's soft-sand model, its fluids Part 7's brine and gas and the book's 40°API oil (Girassol's is 32°API), substituted by Gassmann's equation; each cell is the Backus average of its sand and shale. Lobe A and the mid-fan closure are full of oil to their spill points; Lobe B, a closure full of sand, was never charged. The products come from the book's own kernels: the intercept and gradient of Section 5.4, the exact Zoeppritz coefficient, an RMS amplitude, net sand read from P-impedance alone, and from half a year after first oil a monitor survey (Section 8.2) while water injectors sweep the flank and gas fills the crest.

Reading the figure

1. RMS amplitude. Put the probe on the channel axis up-dip, outside every closure. It reads 0.033, against 0.047 at the oily crest of Lobe A: wet clean sand is bright too. The map draws the sand, not the fluid.
2. AVO fluid factor. Each cell's top is a point in (b). The wet cells make a cloud along a trend fitted to the survey itself; oil lowers the sand's $V_{\mathrm{P}}$ far more than its $V_{\mathrm{S}}$ and pulls the top below the cloud. The crest of Lobe A lies 2.1$\,\sigma$ below the trend; the wet channel axis lies 0.6$\,\sigma$ above it.
3. Net sand from $I_{\mathrm{P}}$. Calibrated on wet rock, impedance reads oil as extra sand: the whole 40 m interval at the crest of Lobe A, where the model has 25 m, and 11 m too much on average inside the oil. Impedance alone cannot separate the fluid from the lithology; that is why Section 7.4 reads lithology from $V_{\mathrm{P}}/V_{\mathrm{S}}$ as well as from impedance.
4. Top-reservoir depth. The three closures are found from the depth map, each flooded from its crest to its spill point: columns of 62 m over Lobe A, 24 m over the mid-fan and 32 m over Lobe B. The map cannot tell that Lobe B is wet.
5. The three-leg screen. At the default cuts (fluid factor 1.5$\,\sigma$, 15 m of net sand) the screen flags 1.6 km$^2$, of which 99 % holds oil in the model, and finds 83 % of the model's 1.9 km$^2$ of oil. Loosen the fluid cut and it finds more oil and passes more wet rock; tighten both cuts and every flagged cell is oil, but most of the oil is missed. Of Lobe B's 0.6 km$^2$ of sand, 0.06 km$^2$ passes the fluid leg at 1.5$\,\sigma$, by noise alone.
6. The monitor. Water pushed in on the flank replaces oil and raises $I_{\mathrm{P}}$; gas injected at the crest lowers it. A year after first oil, when Girassol's first monitor was shot, the model's changes are small (0.07 km$^2$ swept, 0.07 km$^2$ of gas cap); three years after it the monitor sees 0.5 km$^2$ harden and 0.3 km$^2$ soften by more than 2 %, against 0.8 km$^2$ truly swept and 0.3 km$^2$ of gas cap. Thin sand changes less than the 2 % the figure calls a change, so part of the sweep goes unseen.

What time-lapse added

A survey shot before production cannot show where the injected water and gas go. At Girassol, following the re-injected gas drove the timing of the first monitor, and the 4D images were of very high quality although the fluid changes and pressure effects made them hard to interpret (Lefeuvre et al. 2004). Their uses were to update the reservoir flow model, to choose where the last development wells went (the same source) and to constrain the history match (Roggero et al. 2012); Gonzalez-Carballo et al. (2006) credit the repeated monitoring with reserves added by wells aimed at undrained panels and with better-placed development wells. The surveys were also inverted for impedance, by a 4D stratigraphic inversion (Lafet et al. 2005) used to interpret the production mechanisms (Hubans and Berthet 2007); the change of impedance is the quantity plate (d) reads.

The model holds the pore pressure constant, so only saturation changes. At a real field the pressure moves too: a rise softens the frame as gas does, a fall stiffens it as water does (Section 8.2), and separating the two is much of the work.

Where the workflow carries over, and where it does not

• It carries over to soft, porous clastic reservoirs, where a fluid changes the rock's $V_{\mathrm{P}}$ enough to move its top well off the wet trend, and where closures can be mapped.
• It weakens in stiff, deeply buried or cemented sands, where Gassmann's change of the bulk modulus is small and oil barely moves the top (Section 5.5); in beds thinner than tuning, where top and base interfere (Section 1.7); and wherever rocks other than shale and brine sand lie off the trend, coal and evaporites among them (Section 5.5).
• The trend must be refitted for each interval and survey: its slope depends on the local rocks, so a trend borrowed from elsewhere moves every fluid factor.