This is not as much an error as a warning, it is saying that by sampling from a distribution for a nonlinear transform of a parameter implies sampling from a different distribution for the parameter itself, up to the (log absolute) determinant of the Jacobian. There’s more discussion about this here, here, and here, as well as in the Stan documentation.
If I understand correctly at each time point you have a variable number of observations, so one solution is to use implicit ragged data structures, since Stan doesn’t really support ragged arrays. Nevertheless, in your case since you are summing over biomass levels, an absence of observation is equivalent to zero, so you can also create a matrix-like, 2D-array and pad the absent observations with zeros, so you can just sum over the columns (assuming each day is a column, or rows if you have them transposed).