Ideally, you would define a population, then arrange for a probability-based sample from that population. That approach gives one the best confidence that: (a) the sample is representative of the target population; and (b) sample statistics are not likely to be wildly wrong as estimators of population parameter(s).
I suspect that there is, in fact, such a target population: those who clear Chinese immigration and customs services each month. However, getting that information is likely impossible (unless you happen to be well-placed within the Chinese government!).
So, you'll likely have to define a different target population. Perhaps: (a) people who visit social media sites having to do with China, tourism, or both; (b) people who visit a specific site within China (perhaps a national monument, museum, or other heritage-related site); (c) some list of members of a related enterprise, such as "Friends/Patrons of a museum"; (d) target some specific subset of consumers: Those who shop at store X, or those who buy items online, from Alibaba or some other vendor. Clearly, some population definitions might work better for your study than others. However, you'll need to temper that with what is feasible in your situation.
There are plenty of studies that use "convenience" samples (rather than probability samples). The key with these is to describe them as thoroughly as possible (while respecting anonymity of individual respondents), so that others reading your work can draw their own conclusions as to whether your target population is pertinent for the study.
In sampling, generally more (bigger N) is better. However, representativeness of the target population is key for enhancing generalizability.
In research, probability sampling is a must in order to ensure the randomization of selection (no bias) and representativeness of the population. When these two criteria are met, then only you can make generalization towards the whole population based on the results. However, it is easier said than done especially in the case where you do not have sampling frame where you can obtain random sample.
In the study where there is no sampling frame available, the researcher has to develop the sampling frame himself using the following steps.
1. Identify the "natural heritage sites of the world" and select these destinations.
2. Through social media calling for "Chinese tourists who used to visit those destinations". Invite them to respond to the self-administered questionnaire.
3. Continue your survey for one year or so until it is "exhaustive"
4. From the response you obtained (let say 5000), develop the sampling frame.
5. From this sampling frame you can use the most suitable probability sampling method to select you respondents and their response for your study.