Modeling Uncertainties in Weather

Share on Facebook Tweet about this on Twitter Share on LinkedIn

Here is a common scheduling issue. Offshore oil drilling in Gulf of Mexico can be affected by bad weather and in particular, hurricanes. In the case of hurricanes, drilling is shut down and crews are evacuated to an onshore location. While the severity of hurricanes can vary the impact is still largely the same (shutdown, evacuation, startup) so the main issue is what is the probability that a hurricane will occur. In addition, hurricanes can only occur during a specific period: in the Gulf of Mexico the hurricane season runs from June to November with peak intensity generally in the late summer early fall. In the example below, we can see that the range of non-working days in a particular period can vary year to year. So how do we model this risk?

One of the solutions is to use a time-dependent global risk “Hurricane”. This risk would only affect activities performed during the hurricane season. It would be possible to assign a statistical distribution for the moment of risk. In this case, we might use a normal distribution: as hurricanes are most likely to occur in September/October. However, using a time dependent global risk requires that is has a specific risk impact. For example, a fixed or relative delay, or restart the activity. The problem with this approach is that the impact could vary widely for different activities. Impacts would depend on the actual calendar day when hurricane occurs. For example, if a hurricane occurs during a weekend and holiday, the delay may be shorter.

Another proposed solution to weather modeling is to create a calendar or several calendars that would include non-working days or other restricted schedules that could occur during a hurricane. Once we have developed these “weather” calendars, we then have to assign a probability to how often they will be in effect during a project. For example, we could create a “Hurricane calendar” that has 5 non-working days in September. Through our research we know that there is a 10% chance of a hurricane occurring in this area during September. Therefore, during a Monte Carlo simulation, the “Hurricane” calendar would be used as the project calendar in 10% of the iterations and the standard project calendar in the remaining 90%.

In project scheduling calendars can be assigned to:

  • Complete projects (project calendar)
  • Individual tasks
  • Resources

In each case, it is possible to have a probabilistic calendar. Probabilistic project calendars can be used to model other issues in addition to weather. For example, we can create a resource calendar that models the unavailability of a particular resource. If we know that a particular resource will most likely be absent 20% of the time in June, we can create a calendar to model the non-working time and have it run on 20% probability during Monte Carlo simulations.

In project risk analysis software applications, you don’t need to define probabilistic calendars explicitly. Instead, you could use a weather modeling tool and define number of non-working days and probability of interruptions because of weather. The tool would then create probabilistic calendars automatically. 

To read how weather calendars are implemented in RiskyProject software click here