This is because a sample is supposed to ideally represent the whole population from which it was taken. Taking samples from different location within the population ensures that they are no bias in the sample. Each part of the population should be well represented in the sample so that the conclusions drawn from the sample represent the properties of the whole population.