This is the oscillator for an ultrasonic misting circuit (ye olde patent US4996502). I think that something like 99% of the Chinese HF misters (the 1.5-2.5MHz ones, not the micronozzle ones) use this topology, with slightly different values and bias schemes.
The idea would be to resonate the TD piezoceramic at its own frequency with a moderate power (10-30W, depending on the design). In my interpretation it should work in this way (correct me if I'm wrong): the whole thing is substantially a common collector Colpitts.
R1, R2, R3, R4 (often there are trimmers or pots there) gives static current for the operating point.
L3, C4 form an RF trap/DC smoothing do avoid losing power thru the supply lines.
TD and C2 are the resonant tank proper, C1 and C3 are the Colpitts capacitor divider. Nominally C1 is 1.8nF and C3 47nF but it varies in the wild. Also I think that they are different to adjust the gain, is that correct? C2 officially is only for protection against a short in TD but I reckon it participates in the oscillation. It's drawn horribly but that's the topology. The 'output' of the oscillator would be the C1-C3-L1 node.
L1 is a quite big choke (22µH to more than 220µH).
C5 and R6 gives some noise filtering and decoupling (but a lot escapes, anyway). R5 limits base current and helps stabilizing the gain (I think).
Now, for the things I have absolutely no idea: L2 is 4-5 air spire for about 100nH. Best thing I can think of is avoid spurious auto-oscillations. C6 (100pF) gives collector feedback but I have no idea of what would be its design function (in fact, it is not strictly needed, it seems to work even without it).
Now, for the question proper: this thing is different from the textbook Colpitts since the tank is anchored to the positive supply. Usually you see the 'external' terminals of TD, C4 and C1 to ground.
I interpreted this in this way: the goal of this circuit is to maximize the power on TD. So they made it the primary branch for current to flow (+ve, TD,C2,R5,Q1,L2,L1), and in fact there is a huge base current flowing (with the whole circuit at 700mA about 600mA are thru the base). Q1 is huge, too, so it works :D
The problem is that the TD transducers have many resonance pairs (they are piezoceramic discs so probably they move in many directions). And on some of them, instead of picking the nominal 1.7MHz, the oscillator locks on a 300-400kHz mode (not good, of course). The usual solution for the Colpitts would be a suitable inductor across C1 to bypass the lower modes (like for overtone crystals), but it can't be done (that would be a DC short).
I was thinking of working on C2 and C3 to steer the response, would that be a good solution? any other ideas or enlightenment about this circuit?