The best way to answer this is with a NEC simulation of a dipole on the ground.
Using a "Pastoral, medium hills and forestation" ground, Conductivity 0.006 S/m Relative Permittivity 13.
18 metres long, 2 mm diameter wire.
It resonates at 7.0 MHz and the impedance is 100 Ohms not 73.
First thing to note is that the pattern is almost completely omnidirectional. The dipole is lying on the X axis:
Here are two elevation cuts - the blue is in the plane of the dipole, green is perpendicular to it.
See how it radiates more off the ends than to the sides. This is an effect of the ground, where the travelling wave on the dipole radiates towards the end. It happens on any dipole over real ground, but the effect is stronger when it's close to the ground like this.
The (relative) low angle radiation is not bad, quite similar to a dipole at a more normal height.
Here are two azimuth cuts, blue at 10 degree elevation, green at 20 degrees.
Finally, as you may have guessed from the graphs, the efficiency of the antenna is about 2.5%, or -16 dB. The peak gain is -9 dBi upward. NEC doesn't compute ground losses directly, but you can request that it integrates the total power over the radiation pattern.
AVERAGE POWER GAIN= 5.3226E-002
SOLID ANGLE USED IN AVERAGING=( 2.0000)*PI STERADIANS.
(there's a factor of two hidden in there, because it's only averaging over the half hemisphere)
For comparison, the same antenna 10 metres above the ground is 75% efficient, just -1 dB.
For QRP transmissions I would say this antenna is wasting too much power. It would effectively take a 100 Watt transmitter and radiate 3 W. Lifting it a few metres above the ground will make a big difference.
For shortwave listening however, -16 dB probably won't matter at all. In most conditions, your receiver will be externally noise limited which means the wanted signal is competing only with the external noise from nearby electronics and distant thunderstorms. A low gain antenna attenuates both signal and noise, the signal to noise ratio stays the same.
Copying from a previous answer, because this table of ground conductivity and permittivity for NEC or other FEM simulations is valuable and should be easy to find online:
Most kinds of NEC can simulate a Somerfeld-Norton ground. Try simulating with a lossy ground and see what happens to the impedance. Keep the wires several radii above the ground for most accurate results. Use this table of earth parameters to select something reasonable for your situation:
Description |
Conductivity (S/m) |
Relative permittivity |
Fresh water |
0.001 |
80 |
Salt water |
5 |
81 |
Pastoral, low hills, rich soil. |
0.0303-0.01 |
14-20 |
Flat country, marshy, densely wooded |
0.0075 |
12 |
Pastoral, medium hills and forestation |
0.006 |
13 |
Pastoral, medium hills and forestation, heavy clay soils |
0.005 |
13 |
Rock soil, steep hills, typically mountainous |
0.002 |
12-14 |
Sandy, dry, flat, coastal cities, industrial areas |
0.001 |
5 |
Cities, heavy industrial areas, high buildings |
0.001 |
3 |
These are from the SuperNEC GUI URM, sadly very hard to find online. The manual is part of the supernec package which you can download.