How many field plots do you actually need to calibrate a biomass model?

Every technical reviewer has seen the same line in a project design document: "a sufficient number of field plots were established to calibrate the model." Sufficient according to what. Plot count comes from how much above-ground biomass varies across your strata and how tight an error bar the registry will accept, not from a number you can look up in a table.
Why plot count depends on variance, not a rule of thumb
Forest inventory statisticians have been working this problem since before satellite imagery existed, and the math hasn't changed. The standard sample-size formula for a target precision looks something like n = (t × CV / E)², where CV is the coefficient of variation in biomass within a stratum and E is the allowable error as a fraction of the mean. Plug in a CV of 40%, which is typical for a mixed secondary forest with uneven stand age, and a target error of 10% at 90% confidence, and you land somewhere around 40 to 50 plots per stratum. Drop the CV to 20% in a plantation block and the number falls to a dozen or so.
This is why two projects of similar hectarage can report wildly different plot counts and both be defensible. A reviewer checking a design document should be asking for the CV estimate and the target error, not comparing raw plot totals against some other project they've seen.
Stratification does more work than adding plots
Before increasing n, most inventory foresters stratify first. Splitting a landscape into strata by canopy height class, stand age, or forest type usually cuts within-stratum variance faster than brute-forcing more plots into a single pooled sample. A project with five strata and 15 plots each will often out-perform a project with 60 plots scattered across the whole area with no stratification, for the same field effort.
The practical sequence most crews follow: draw strata from existing canopy height or cover data, put a modest plot count in each, calculate the within-stratum CV from the first field season, then decide whether any stratum needs top-up plots before the equations get fit. That second-pass calculation is the step projects skip when they're trying to close out a field season on schedule, and it's usually the step a technical review flags.
Where remote sensing changes the plot math
None of this makes field plots optional. Allometric equations still need ground-truthed stem diameters and species composition to anchor the regression, and a registry isn't going to accept a biomass figure with no destructive or semi-destructive sampling behind it anywhere in the chain. What remote sensing changes is how far those plots have to stretch.
Once a regression is calibrated against a reasonable plot set, a wide-swath biomass layer that regresses canopy structure to above-ground biomass across the whole landscape lets you apply that calibration consistently year over year without re-running a full plot campaign every reporting cycle. The plots still do the calibration work. The imagery carries that calibration across hectares the crew didn't walk, and gives you a refreshed figure for the stock-change years in between full remeasurement.
For a reviewer, the question worth asking isn't just "how many plots" but whether the project documented the CV, the stratification logic, and how the resulting equation gets applied to unsampled area. A project that shows its sample-size math is usually the one that holds up under a second look.
If your landscape layer needs refreshing between plot campaigns, take a look at what Biomass Mapping produces from existing imagery.