我有一个地方的24小时总降雨量的信息，100年一遇的风暴事件。(来源:https://hdsc.nws.noaa.gov/hdsc/pfds/index.html)。我想利用这些信息创建一个真实的24小时内降雨的时间序列，以10秒为增量。

我怎么才能做到这一点?我可以假设均匀分布，这很天真。关于什么样的分发是合适的，有什么指导方针吗?< / p >

听起来对我来说，真正正确地模拟…这有点像问一个假设的100层建筑将如何建造的时间表。< / p >

There's certainly some patterns to how extreme rainfall events can go: stationary front/stalled tropical cyclone/slow moving chained supercells (perhaps extremely large hail can be a factor in this [though I'm not sure how well that's measured in precipitation totals really])/maximized terrain ascent/flow off very moist water body. And some guidelines to the process as well: there is a maximum convective rainfall rate, and times of the day when certain precipitation events are favored [diurnal seabreeze, nocturnal low-level jet, etc]. You could use all these as a rough guide in your simulation.

Or if you really wanted to go intense on it, you could try gathering a distribution of like the top percentile of rain events, either in general, for the region, or for the site of interest (or from nearby sites with longer datasets if necessary), and try to work off the guidelines of the temporal "shape" and timing of such events. (I can't offhand remember any summary datasets of the temporal distribution of major flash floods, though it's not impossible that at least specific types have some out there if you search the journals?)

You could then either just use a representative "typical" layout given by that data, or instead come up with some complex "simulation" production code for events using randomization along with the probability densities of each hourly value.

But even then, a rare flood event can often be a bit different in style versus more "typical" floods; for example some areas may only see remnant tropical cyclones or strong tropical cyclones like once in a century. So they wouldn't match the observed patterns that well.

The big picture is still that some events can be many hours of strong rain, others can be shorter duration of extreme rain, or even multiple separate rounds of storms. So simulating a flood's rainfall timing... especially at that temporal resolution (I'm not sure standard rain gauges measure to near that time increment; certainly in past recordkeeping generally the base time period has been hourly, though with start and end times to the minute, and potentially datasets obtainable down to nearer 5 minute resolution [ASOS] or perhaps shorter from some mesonets/micronets may be found with enough work. Or radar datasets may be available to use as non-in-situ estimates, though they'll mostly be about 5 minutes resolution as well, except for rarer focused rapid-scan/mobile/phased array datasets).

So you have to weigh how important realism is. If you want to attempt to really get the maximum reality, it may take some serious work. If you want a rough estimate, you could use similar events of the same type/in the region of interest as a fair idea. But if you heard a foot of snow will fall in a storm... will it be concentrated over a few hours of heavy blizzard, or spread of more steady precipitation over the entire day? It can well be either. And that's the case with rainfall events too. So I'd argue there's never going to be a perfect expectation. If your study is on something like flood evolution, probably go towards a worst-case scenario, and focus it on a fairly quick event.