Electronic – Regarding Transmission line

An interconnection should be considered a transmission line if reflections are detrimental to your application. As Leon Heller says, 1/10th wavelength is a good rule-of-thumb.

However, all this depends on the nature of the signal (1/10th of what wavelength?). With sine waves, the wavelength in the cable will be about $\frac{2 X 10^8}{f}$ where f is the frequency, but will vary according to the cable type. The effect of reflections on sine waves is to cause standing waves, so the amplitude will vary along the length of the cable but will always be sinusoidal. With square waves the situation is complicated by harmonics. Although the square-wave may have a low fundamental frequency, the harmonics introduced by the edges will be significant. The effect of reflections in this case is undershoot, overshoot or ringing depending on where the mismatches are (if both $R_S$ & $R_L$ are lower than $Z_0$ for example, you will get ringing).

So do you need to match the source? No. As long as the load is perfectly matched, the wavefront will travel down the transmission line and be fully absorbed by the load. Consider a digital system with 3.3V logic for example. If you terminated both source and load you would only get 1.65V swing at the load. Terminating only at the load is termed parallel termination.

There is also a technique known as series termination in which the source is terminated but the load is left open circuit. In this case, a half-amplitude edge propagates down the line and is reflected back to the source, lifting the voltage to full amplitude as it returns until it is absorbed by the source termination. Along the line, you would see a 'stepped' signal except right at the load where you would see a clean waveform.

RS-485 is good for 100kbps at 1.2km, using two wires for data and two for power. You could probably stretch it to 2km at 1200bps.

Here is a TI application note for a 2-wire power + data solution. It couples the data transceivers through large capacitors, and connects the power source and sinks through normal-mode transformer filters.

1000BASE-LX gigabit Ethernet can run 5km over fiber.

A reflection may take 40us to travel 4km at 10^5 km/s. At 1200bps, you have 400us settling time from the leading edge of a bit to a sampling point a half-bit away.

In the comments, the OP mentions using untwisted, unshielded pairs. This will hurt noise immunity. Consider using parity bits, CRC's, retry protocols, or error correction.