These tables provide various estimated probabilities regarding tropical cyclone signals issued by the Hong Kong Observatory. The track position, intensity and signal issuance data from 1961 to 2014 inclusive were fitted by a statistical model. These probabilities are based on the forecast track issued by our site. NOTE: This information is by no means official or related to the Hong Kong Observatory in any way.

The first table includes in force, issuance and cancellation probabilities for the next 72 hours binned at 6-hour intervals. The first row indicates the number of hours elapsed since the time of observation for the current report; the day of week and clock hour (in HKT, 24-hour format) are provided for easier reference. The next rows are signal-specific: "Signal 1 or above", "Signal 3 or above" and "Signal 8 or above" are currently implemented. Probabilities for a given signal are only displayed if they are not all zero. For each signal there are two rows of numbers: The first row gives the probabilities that the signal (or above) is in force at the time indicated by the column header, and is meant to be interpreted for a particular point in time only (NOT a period of time). What the second row displays depends on whether the signal (or above) is currently in force; if it is not in force, it displays the probabilities that the relevant signal (or above) will be issued for the first time in the 6-hour interval ending at the time indicated by the column header. For example, the number corresponding to the column "+36" indicates the probability that the signal (or above) will not be issued in the next 30 hours, counting from the time of observation, but will be issued between 30 and 36 hours. It should be noted that this probability is smaller than the one for "issuance between 30 and 36 hours" only (not shown in table), but in most cases they will be close, unless the forecast track is such that the chance of multiple issuances is non-negligible. If the signal concerned (or above) is in force, the second row displays the probabilities of cancellation instead. A similar interpretation applies. Different colour schemes are employed for the "in force", "issuance" and "cancellation" rows to prevent confusion.

The second table includes only the issuance and cancellation probabilities for the first 72 hours of the forecast, or fewer if the forecast track is shorter. It gives the probabilities that a particular signal (or above) is issued or cancelled any time after the time of observation and before the end of forecast or 72 hours, whichever is shorter. The number of times the signal is issued or cancelled does not matter here. When the signal (or above) is not currently issued, the "issued" row is relevant; otherwise the "cancelled" row is relevant instead. Mathematically, this number is the sum of the "issued" or "cancelled" row for the relevant signal in the binned table above.

Note that all numbers are rounded to the nearest integer, and so anything less than 0.5% is rounded to 0%, while anything above 99.5% is rounded to 100%.

This feature is currently under trial; in this period, we will continue to provide the existing, more subjective estimate using the descriptors "Low", "Medium", "High" and "Very High". They may not match exactly as the model is being fine-tuned and debugged.

Hong Kong Weather Watch
July 30, 2016