1. GPS inversion for the source
uz = (1-nu)/pi * dV * d/(r^2+d^2)^1.5
ur = (1-nu)/pi * dV * r/(r^2+d^2)^1.5
ur/uz = r/d -> depth from the arrow slope
chi2 = sum ((model - data)/sigma)^2
For a fixed depth the model is linear in dV, so the best dV follows from least squares and only the depth needs scanning. Noise makes shallow and deep solutions trade off against volume: that is why the chi2 valley is elongated.
2. S-P hypocentre location
t_SP = D (1/Vs - 1/Vp) D = t_SP * Vp*Vs/(Vp - Vs)
Vp = 5.5 km/s, Vs = Vp / (Vp/Vs ratio)
Each station gives a distance, so the event lies on a circle (in this cross-section) of radius D around it. A Levenberg-Marquardt fit finds the point closest to all circles. Pick noise spreads the answer into a cloud (Monte Carlo of 150 re-picks); assuming the wrong Vp/Vs scales every distance and biases the result.
3. DOAS flux
Q = u * integral c(x) dx c(x) = Q/(sqrt(2 pi) sigma u) exp(-x^2/2 sigma^2)
captured fraction of a +-k sigma scan = erf(k/sqrt 2)
A wind error maps one-to-one into a flux error. A scan that does not reach the plume wings misses mass in the way the error function predicts.
Honest simplifications
Axisymmetric point source in a uniform elastic half-space, no topographic correction, independent Gaussian errors, constant velocities, a single Gaussian plume cross-section, cross-section (2D) geometry for the location problem.
Independent numerical checks