Factors affecting the value of environmental predictions to the energy sector

Methodology for valuing an environmental prediction

Environmental predictions may allow entities in the energy sector to make better decisions. For example, knowing it will be hot tomorrow allows energy producers to plan for the resulting high demand. However, the forecast only has value if it reduces uncertainty: the forecast of a hot July day will be much less valuable than the forecast of an unexpected May heat wave. The “net benefit” value of the forecast can be quantified by determining the added financial value of these better informed decisions (Mason, [1966]). In addition, entities must be able to integrate weather forecasts into their decision structures in order to benefit from a particular forecast (Stewart et al., [2004]). Of course, for a forecast to be useful it must save more money than it costs to produce, with the caveat that there may be multiple uses for the same forecast.

A variety of approaches have been developed to value EP-like information as reviewed in (Turquoise Technology Solutions [2007]). Early efforts to quantify the value of weather information are synthesized by (Macauley, [2005]) which also sheds light on the cost/benefit decision of purchasing additional pieces of information. The latter decision problem is also described by several governmental bodies in (Danish Ministry of Transport, [2006]; WMO Secretariat, [2007]; NAV Canada, [2002]). Early papers applying a classical decision science approach include (Baquet et al., [1976]), who quantified the value of frost forecasts to orchard operators deciding whether or not to run heaters. A more general theoretical framework for this valuation problem was provided by (Hilton, [1981]) while (Kite-Powell [2005]) applied the theory to an ocean environmental data application.

The expected value is a cornerstone in many of these valuation frameworks. For example, suppose a power company must decide how to bid tomorrow’s production from their various generation assets into tomorrow’s market. Because this process is repeated every day and the possibility of bid leading to a catastrophic loss is very low, it makes sense to measure the effectiveness of a bidding strategy by its average performance, or expected value. In such a situation a slight edge in predicting temperature, hence demand, and hence electricity price, might have a small daily value but a large annual value.

However, the expected value is not always an appropriate metric. Hurricane forecasts to drilling rig operators in the Gulf of Mexico are valued by (Considine et al., [2002]) who suggest that, unless a forecast brings near certainty, it is worth little. Oil rigs will be evacuated even at a slight chance of catastrophic weather conditions because the costs of a missed evacuation dwarf that of an unnecessary evacuation. An incremental improvement in the accuracy of the forecast is unlikely to change this decision.

Electricity demand and temperature

The framework for this case study summarizes the U.S. focused studies of (Teisberg et al., [2005]) and (Hobbs et al., [1999]) and use similar stylized facts about the general behaviour of energy systems. Although these works are not focused on Canada, the similar power systems and climate conditions of the two countries allow useful comparisons.

Because economically meaningful quantities of electricity cannot be stored, generation must match consumption in a dynamic manner. Consumption varies tremendously due to weather (i.e., air conditioning during heat waves and heating during extreme cold) and more regular patterns of use (e.g., lower demand during the night and on weekends or holidays). Electricity is most economically generated in large plants which produce it at a constant rate. Generating electricity at a moment’s notice is much more expensive. For this reason, accurate demand forecasts, which in turn require accurate temperature forecasts are very valuable. This case study further develops the short term supply and demand aspects of this issue and outlines how the economic value of forecasts can be approximated.

Power demand is low at night and high during the day and tends to be lower on weekends and public holidays. Because of heating and cooling requirements, exterior temperature is a key power use driver. Figure 1 shows a plot of Ontario power demand against the temperature at Toronto’s Pearson airport at two different hours of the day. The figure shows that when days are cold, power demand rises when temperature falls, while on hot days power demand rises with temperature. Since Ontario does not have a spatially uniform temperature, this relationship could be improved by use of temperature forecasts at various locations, but the figure is nonetheless instructive. It is also important to note that most of the other power use drivers, such as number of daylight hours and timing of weekends and public holidays are predictable and so do not require sophisticated forecasts

Matching this fluctuating demand requires an understanding of the economic and engineering features of the commonly used generation sources. A typical electrical system includes several generating technologies with different marginal costs of production. These plants are dispatched using solutions of the “unit commitment problem” (Baldick [1995]) that, for a given level of production, use the cheapest sources of power first. Power demand varies over the day, with predictable and unpredictable components, so unit commitment must be performed repeatedly to find the best variable generation mix over the course of the day. The dynamic nature of this process creates a complication since the output of some generation types cannot be adjusted quickly, if at all. The need to take these flexibility characteristics into account can result in plans which no longer use the absolute low-cost producer. For example, nuclear power must typically always be dispatched.

Highly responsive plants are therefore required to enable the variation of these plans. Two main methods can be used for this purpose: spinning reserve and Automatic Generation Control (AGC). A unit is part of the spinning reserve if, though ready to immediately do so, it is not generating any power. Plants running in AGC surrender operational control to the system operator, which operates them to balance system load. The technical and environmental issues around spinning reserve and AGC are reviewed in (NOAA/NESDIS [2002]). AGC units are able to rapidly increase or decrease their output, as commanded, to respond to load changes. Thermal power plants operating in spinning reserves still consume significant amounts of fuel to “keep their boilers hot”. Therefore improvements in temperature (and hence load) forecast accuracy can create both economic and environmental value by reducing the unnecessary use of spinning reserves. In a U.S. study, (Teisberg et al., [2005]) found that accurate 24-hour temperature forecasts are worth US$166 million per year in optimizing the generation mix.

It should be noted that the high value reported here results from the impact of small daily profits repeated day after day. This repeated decision framework means that small statistical improvements in decisions will, over time, contribute to significant value.

In contrast to the simple single variable/single repeated decision setting of this case study, there are situations when multi-factor forecast information may be valuable for tactical, operational and strategic decisions. These are discussed in the following section within the context of hydroelectric power.

Hydrology case study

Hydroelectric generation depends strongly on water inflows and water levels (Zhao & Davison [2009a]). Both flow rates and water levels vary over time, and this variability has a significant impact on the generation capacity of utilities. Indeed, (Hydro Québec, [2006]) identifies hydrological inflow risk as its largest single financial risk factor, surpassing interest rate risk and foreign exchange risk.

This case study quantifies the potential value of sophisticated hydrological EP to hydroelectricity companies across operational and tactical decision scales. The use of the term “potential” signifies the curious absence of EP in current industry practices.

Sea ice environmental predictions for offshore oil and gas production

Not unlike the case of hurricanes affecting oil and gas production in the Gulf of Mexico (Considine et al. [2002]), both drifting sea ice (pack ice) and icebergs can be a threat to offshore oil platforms in northern waters, such as those tapping the Jeanne D’Arc basin located below the Grand Banks, East of Newfoundland (Hibernia, Terra Nova, and White Rose). These three sites respectively produce about 314,000, 150,000 and 100,000 barrels per day of light crude oil (Offshore technology, [2006]). All three locations are at risk not only from icebergs, which given the shallow 80 m depths can scour the ocean floor and damage underwater infrastructure, but also from floating sea ice at thicknesses of 1–2 m. It is interesting to note that different oil platform technology has been selected for these sites.

The Hibernia Oil Platform is a fixed “gravity based structure” (GBS) which is designed to be nearly invulnerable to the encroachment of sea ice or icebergs. This invulnerability comes at the heavy financial cost of $1 billion for the additional reinforcement (Ralph, [2007]). In contrast, the extraction at the Terra Nova and White Rose sites is accomplished by means of a floating production, storage, and offloading vessel (FPSO) that can be disengaged in response to a severe ice threat. In the latter case, the timing of the decision to disengage is critical. If the vessel is moved in response to every threat, production will be delayed at a high cost. On the other hand, if the vessel never leaves, it will be at risk to damage from ice. Coupled with this set of decision processes is a complicated threat management environment which allows sea ice and icebergs to be diverted from the structure using icebreakers and even water cannons. Environmental predictions of the trajectory of floating sea ice and icebergs are an integral part not only of the threat management system but also of the evacuation system.

To obtain insight into the challenges of incorporating forecasts into sea ice/iceberg management for offshore operations, simple expected value decision experiments for the FPSO case are discussed below. However, while the risk to life and limb from floating sea ice is low (Ralph, [2007]; Kirby [2007]) icebergs present significant catastrophic loss risk. The need to heavily weigh such catastrophic events in decisions will decrease the following estimates.

The decision may be framed by assuming that the platform is surrounded by three concentric circles, denoting various threat levels. Each successive breach by an iceberg or sea ice reduces the time remaining to the decision maker and forces an intermediate shutdown so that, if necessary, there is enough time to evacuate.

Assume that the FPSO services an oil well, modelled on White Rose, with a daily production of 100,000 barrels of light crude (sold on the international market), and a production lifetime of 10 years. An ice threat (sea ice or iceberg) arriving at the outer perimeter of a given concentric circle will take a given time to hit the oil platform.

The following analysis is constructed to model how a decision maker would incorporate ice forecast information in FPSO movement decisions. It does not represent detailed industry practice. In the absence of forecasts, the probability of ice breaching a given area is assumed proportional to that area. Additional information about the direction from which ice threats typically come will change this scaling relationship (if ice always comes from the north, the probability of ice breaching a circular region is proportional to its diameter).

For indicative purposes only, assume that the outer circle has a radius three times and the middle circle a radius twice that of the inner circle. The resulting probability of ice breaches must follow in the progression 9:4:1. For the purpose of this analysis, we assume that, during an assumed 30-day ice season, the outer circle is breached 9% of days. Countermeasures involving serious economic loss need only be initiated as the second circle is breached, so this corresponds to an assumption of approximately 1.33 times per year of second circle events and approximately once every 3 years of extremely serious, inner circle breach events.

A threat protocol to safeguard the oil platform workers and the natural environment is instituted so that when the outer circle is breached by ice, the steps of a shutdown and evacuation sequence are initiated. We consider a three-level threat model. When ice is outside the outer concentric circle, a threat level of zero exists. When ice breaches the outermost circle, the FPSO makes ready for sea, incurring a modest cost in fuel burned. If the ice leaves the outermost circle again, this step can be reversed. If, on the other hand, ice continues to breach the second circle, the FPSO must shut down oil production, at a much heavier cost. Even ignoring any petroleum engineering consequences, the result is to defer production of oil from the present to the end of the reservoir’s life. This incurs a known cost of 100,000 x today’s oil price, with a countervailing benefit of 100,000 x the oil price 10 years from now, discounted to the present day at an appropriate interest rate (spread over the London Interbank Offered Rate, LIBOR). If we assume that today’s oil price is US$70 per barrel, discounted at 6%, and that the value of oil grows simply with a 2% inflation rate, the cost of deferring a day’s production is about $2.25 million. Naturally, this result is very sensitive to the underlying oil price assumptions.

Further, assume that the ice leaves the middle circle after a single day (either to breach the inner circle or to retreat to the outer circle), and that if the ice retreats to the outer circle, production can be restarted immediately. If the ice breaches the innermost circle, the FPSO needs to disconnect from the wellhead and steam away. The cost of this is estimated to be an entire week of lost production-to cover the time required to steam to safety, return, and perform the delicate operation of re-connecting to the wellhead.

A mobile escort vessel armed with water cannon and having tug and icebreaking capabilities can provide safety countermeasures. Such vessels cost about $25,000/day (Ralph, [2007]) and are employed to redirect (with cannon or towing) icebergs out of the threat area and to break up floating sea ice.

We assume that employing such a vessel without using forecasts clears the threat at each stage with a fixed probability. Employing a vessel will be nearly worthwhile if its use for a single day reduces the probability of a middle circle breach lasting one day by even 1%. However, the vessel cannot come on a day-by-day basis. It must be engaged monthly at a cost of $750,000. Even though the Hibernia, White Rose, and Terra Nova oil platforms lie within a few hours steaming of one another so perhaps could share a single vessel, the cost of engaging it may not be worthwhile if the probability of an ice-related shutdown is small.

The expected value of environmental prediction information is assessed by carrying through two sets of calculations, one with and one without ice forecasts. The base case computes the expected annual cost of ice to an offshore platform similar to the White Rose without any use of countermeasures such as escort vessels nor any use of environmental prediction. This provides a baseline against which the other strategies can be judged. Four percent of the time the middle circle is breached. A middle circle breach requires a one day production shutdown at a cost of one day’s deferred production, estimated to be $2.25 million. As we assume 30 ice risk days in a year so the total cost of this middle circle breach is 30x0.04x$2.25 = $2.7 million. Conditioned on the middle circle being breached, one-quarter of the time the inner circle is breached next. An inner circle breach requires a shutdown for 7 days (after which it is assumed that the ice has left the outermost threat circle) at a cost of 7 days of lost production. Again, there are 30 ice days in a year so the cost of an inner circle breach is $4.725 million. So the total expected annual cost of ice, with no countermeasures and no use of environmental prediction, is about $7.4 million.

A more realistic variation of the base case assumes that an escort vessel is employed for the 30 days. The cost of the escort vessel is shared between 3 platforms at a daily cost of $25,000 for a total cost of $0.25 million per platform. We assume that 10% of the time, the escort vessel is able to stop a floating ice threat from moving from one threat level to the next higher level. When the escort vessel is present, the middle circle is breached only 4%*(0.9)² = 3.24% = of the time, and the inner circle is breached just (0.9)³x1% = 0.729% of the time. So the expected loss due to ice decreases to $5.6 million. As the escort vessel costs $250,000, the total expected loss with escort vessel is $5.85 million. This represents a savings of $1.55 million relative to the original base case, suggesting that with the crude assumptions employed here, an escort vessel is well worth the money even in the absence of EP.

What is the effect of EP on the latter variation of the base case? In essence, good EP allows the threat regions to be shrunk, reducing the probability of having to engage in costly countermeasures. The better the EP, the more comfortable we can be with nearby ice and the more confident we can be about the escort vessel’s ability to avert threats. Under reasonable assumptions quantifying these effects (Turquoise Technology Solutions [2007]), the net effect is to save $3.98 million, a substantial improvement from the $1.55 million saved by the escort vessel alone or of the $740,000 saved by passive EP alone. We can therefore attribute a significant savings of $3.98 million - $1.55 million = $2.4 million to the incremental benefit of EP in this context.

Discussions with stakeholders (Dewhurst, [2007]) emphasized that weather forecasts were used by stakeholders on an operational level, but mostly to minimize danger due to high winds and high seas rather than to forecast ice movement. Because of the time pressure under which such decisions are taken and their multi-factor nature, it is difficult to isolate the value of a decision framework such as that described here.

The case studies presented here suggest that predictions of environmental variables ranging from temperature to water inflows and levels to the motion of floating ice in the ocean all have value to different participants in the Canadian energy industry. However, a recurring theme arising in our discussions with many stakeholders was that, despite this potential value, the full possibilities of environmental predictions were rarely being used. In the next section of this paper we address the reasons for this apparent failure and present some policy recommendations which could release more value from these forecasts.