Integrating climate forecasts and natural gas supply information into a natural gas purchasing decision

Meteorol. Appl. 7, 211–216 (2000) Integrating climate forecasts and natural gas supply information into a natural gas purchasing decision David Chang...
Author: Antony Sparks
3 downloads 2 Views 184KB Size
Meteorol. Appl. 7, 211–216 (2000)

Integrating climate forecasts and natural gas supply information into a natural gas purchasing decision David Changnon, Michael Ritsche, Karen Elyea, Steve Shelton and Kevin Schramm, Meteorology Program, Department of Geography, Northern Illinois University, DeKalb, IL 60115, USA

This paper illustrates a key lesson related to most uses of long-range climate forecast information, namely that effective weather-related decision-making requires understanding and integration of weather information with other, often complex factors. Northern Illinois University’s heating plant manager and staff meteorologist, along with a group of meteorology students, worked together to assess different types of available information that could be used in an autumn natural gas purchasing decision. Weather information assessed included the impact of ENSO events on winters in northern Illinois and the Climate Prediction Center’s (CPC) long-range climate outlooks. Non-weather factors, such as the cost and available supplies of natural gas prior to the heating season, contribute to the complexity of the natural gas purchase decision. A decision tree was developed and it incorporated three parts: (a) natural gas supply levels, (b) the CPC long-lead climate outlooks for the region, and (c) an ENSO model developed for DeKalb. The results were used to decide in autumn whether to lock in a price or ride the market each winter. The decision tree was tested for the period 1995–99, and returned a cost-effective decision in three of the four winters.

1. Introduction Northern Illinois University (NIU) spends on average approximately $10 million dollars per year for heating and cooling its facilities. Due to the recent deregulation of natural gas markets, users of this commodity now have more flexibility in their gas purchasing decisions. Deregulation allows pricing to be set in direct reflection of current supply and demand. Previous studies and interviews with utility decision-makers point to season long temperature anomalies as having the greatest impact on supply and demand, and therefore, price (Weiss, 1982; Changnon et al., 1995). NIU’s heating plant manager and staff meteorologist are involved in making decisions related to the purchase of natural gas for winter heating. Cold season temperature anomalies account for most fluctuations in natural gas demand and prices; and improved climate forecast modelling has the potential to save millions in the purchase of natural gas (Brown & Murphy, 1987). Because of deregulation the NIU heating plant manager has the option to lock in a lower, fixed price at the beginning of a heating season if factors point to an upcoming season wrought with below average temperatures. In the spring of 1998 a faculty-directed project in the Applications in Climatology course taught at NIU

(Changnon, 1998) developed an ENSO model for the heating plant manager (Changnon et al., 1999). The following year a second group of students met with the university decision-makers and discussed potential limitations with the initial model. Due to the success of the Climate Prediction Center’s (CPC) 1997−98 El Niño-based winter forecast (Barnston et al., 1999) and evidence that anomalous natural gas supplies (either high or low) impacted autumn natural gas prices (Schoonmaker, 1997), the staff meteorologist and heating plant manager wanted the group to determine if the previously developed decision model could be improved by including other relevant information. The research objective was to assess: (a) the previous ENSO model (Changnon et al., 1999) and improve it where possible, (b) long-lead climate forecasts, and (c) natural gas supply and cost information, as tools to improve the natural gas purchasing decision for NIU. Based on an independent evaluation of climate forecasts and natural gas supply information the heating plant manager could then choose the most prudent purchasing plan for the upcoming winter. Importantly, the three tools may not lead to a unanimous purchasing decision. For example, if weather has been mild for a year or more and a surplus supply of natural gas exists, then this factor may make for a low daily fluctuating price (Knox et al., 1985). This circumstance indicates that the heating plant may save money by riding the


D Changnon, M Ritsche, K Elyea, S Shelton and K Schramm daily market although near or below average winter temperatures are expected. This simple example demonstrates the complexity often found in weathersensitive decisions that generally do not rely solely on climate information. A decision tree that incorporated two different climate forecasts and natural gas supply information was developed to enhance decision-making efforts related to natural gas purchases. This tool was tested for the years 1995–99 to determine whether it correctly chose to lock in a price or ride the market fluctuations on a dayto-day basis each winter.

2. Decision tree 2.1. Tree development The decision tree was designed around information from three tools that independently suggest whether to ride the current market or lock in a fixed natural gas price for an upcoming heating season. The first tool, the ENSO model (Changnon et al., 1999), uses ENSO classifications (El Niño, La Niña, and non-ENSO) of prior winters for years 1951–98, the April through September sea-surface temperature (SST) trends (increasing, steady, or decreasing), and the classification of the following winter’s temperature anomaly (cool, average, or warm) for DeKalb, Illinois. The ENSO model was enhanced by adding the following year’s winter ENSO classification, El Niño (E), La Niña (L), or non-ENSO (N) event (Table 1). Adding ENSO classifications for the following winter provided the heating plant manager with additional information about a particular forecast and improved the decision tree outcome. For example, in developing the forecast for the winter of 1998/99, one would have to determine the ENSO classification for the previous winter (1997/98) and the SST trends for the April through September (1998) period. In this case, the 1997/98 winter was classified as an El Niño, followed by decreasing SSTs, and seven of eight previous winters (removing 1997/98) with these characteristics experienced average to cooler than average temperatures (Table 1). In this case, the initial ENSO model (Changnon et al., 1999) would forecast average to cooler than average temperatures for DeKalb. However, by adding the upcoming winter’s ENSO classification (E, L, or N), realizing that La Niña conditions were expected during the 1998/99 winter, and recognizing that the three previous winters when La Niñas followed El Niños experienced warm to average seasonal temperatures, the 1998/99 forecast would be different (the 1998/99 winter verified as warmer than average in DeKalb). In several scenarios shown in Table 1 the addition of the ENSO classification for the upcoming winter improves the forecast certainty. For example, El Niños followed by nonENSO winters are generally average to cooler than average, while non-ENSOs followed by El Niños are


warmer than average in all six cases. In years when there are limited number of cases to predict the future winter temperature anomaly, such as winters following a La Niña winter, the user should understand that this model has limited use and may not be applicable (N/A) for use in the decision tree. The second tool incorporated into the decision tree was the 1995–99 CPC long-lead climate outlooks for the northern Illinois region. Although ENSO relationships are integrated into these forecasts using canonical correlation analysis, other information such as optimal climate normals and predictions generated from complex ocean−atmosphere coupled models are included (Barnston et al., 1994). These forecasts are made from 0.5 to 12.5 months in advance of the particular threeTable 1. ENSO model indicating previous winter ENSO classification, trend in April through September SST following that winter, and next winter’s actual temperature anomaly (warm, average, or cool) experienced in Dekalb, IL, and actual ENSO state (E for El Niño, L for La Niña, and N for non-ENSO). For example, by its location in the table 1996-97E indicates that the 1996–97 winter was classified as a non-ENSO event, the non-ENSO event was followed by increasing SSTs from April through September 1997, the 1997–98 winter verified as warm (above average) in DeKalb, and 1997–98 winter was associated with an El Niño. Previous winter’s ENSO classification

Following winter’s temperature anomaly in DeKalb and ENSO classification Warm


EL NINO Increasing SST Trend Steady SST Trend 1951-52N Decreasing SST Trend 1972-73L 1997-98L NON-ENSO EVENT Increasing SST Trend 1971-72E 1981-82E 1985-86E 1990-91E 1993-94E 1996-97E Steady SST Trend 1952-53N 1959-60N 1966-67L 1979-80N 1989-90N Decreasing SST Trend 1974-75L LA NINA Increasing SST Trend 1988-89N Steady SST Trend 1955-56N Decreasing SST Trend

1965-66N 1969-70L 1987-88L 1991-92N

Cool 1976-77N 1986-87E 1968-69E 1957-58N 1982-83N 1994-95N


1961-62N 1962-63N 1977-78N

1953-54L 1956-57E 1995-96N


1958-59N 1983-84L

1960-61N 1980-81N 1992-93N

1967-68E 1970-71N 1973-74N 1954-55L

1975-76E 1984-85N

Integrating climate forecasts and natural gas supply information month period (December–February), thus providing decision-makers with adequate lead-time to implement the information into decisions. The climate outlook shows the excess likelihood above 33.3% that a region will be above normal (A), normal (N), or below normal (B). When the CPC has insufficient skill in predicting the upcoming period, climatology (CL) is used on the climate outlooks. The user will not emphasize the CPC forecast tool (N/A) when considering a purchasing decision in these CL situations. The third part of the decision tree was the comparison of current natural gas supply to the average natural gas supply on a weekly basis (Figure 1) to determine if actual supply levels were near average, below (i.e. >4% below average for date), or above (i.e. >4% above average for date). These levels were arbitrarily chosen based on discussions with the heating plant manager. The average supply curve was based on averaging weekly natural gas supply information available during the 1993–99 period. The average supply level for 1 October, the purchasing decision date, was 85%. Actual supply levels for the four years included 86% (normal) for 1995/96, 77% (below) for 1996/97, 81% (normal) for 1997/98, and 91% (above) for 1998/99. Supply anomalies, either above or below the average, were expected to have an influence on the price of natural gas. For example, the natural gas supply had a significant impact on prices during the 1998/99 winter. Natural gas supplies were at record high levels due to low usage in the previous warm winter of 1997/98 (Energy Information Administration, 1999). Since natural gas supply was high and demand was low (the 1998/99 winter was warm in many parts of the US), prices remained low throughout the winter and the best decision would be to go with a rolling market price. A natural gas purchasing decision to lock in a price would come when supply levels were found to be near average or below average. Understanding the importance of

this factor in a purchasing decision represented an important learning experience for a group of applied climatologists.

2.2. Tree implementation Implementation of the decision tree involves determining what the three tools suggest for the upcoming winter’s purchasing decision. Flow-charts for each tool were created that lead to a decision to either lock or ride (Figure 2). Each average (normal) condition in the three parts of the tree was given a lock decision, reflecting the conservative nature of the decision tree. In years when each of the three tools provided either a lock or ride decision the three tools were weighted equally in developing a purchasing decision. Those years when the two climate forecast factors disagreed (thus cancelling each other), the purchasing decision was essentially based on the natural gas supply tool. As described earlier, the ENSO model and the CPC outlook each have an additional branch in the flowchart for not applicable (N/A). In cases where one of the two forecast factors was N/A and the other forecast factor differs from the natural gas supply decision, the heating plant manager will have to exercise his/her judgement in determining a purchasing decision. In cases where both climate tools indicated N/A, the weight of the decision will be placed on the natural gas supply tool. In the future, the ENSO and CPC tools will become more valuable as more data and further enhancements are added. Overall, with a larger sample size (more winters) the decision-maker may be able to adjust the weights for each factor based on their success in predicting the correct purchasing decision. An example of how to implement the decision tree can be seen by examining the forecast for the winter of 1997/98. A non-ENSO event occurred in 1996/97 and

Figure 1. Natural gas supply data for the 1995–99 period. Average supply values are represented by the thin solid line, while actual supply values are represented by the heavy solid line.


D Changnon, M Ritsche, K Elyea, S Shelton and K Schramm month and are reported in cost per therm (Figure 3). For each winter (1995/96 through 1998/99) the October price was determined and a dashed line representing that value was drawn through March. Shaded areas above the dashed line indicate periods when the rolling market price would have been higher than the fixed October price, while those shaded areas below the dashed line indicate that the rolling price is lower than the fixed October price. For example, if the decision tree suggested to ride the market, and according to the natural gas pricing information the price was lower than the October price for most of the period from December through March, then the model was considered to have made the correct decision. More weight was given to this part of the heating season (December through March) because 68% of DeKalb’s heating degree days (usage) are accumulated during this period.

2.3. Tree results

Figure 2. Decision tree flow-chart using three tools.

was followed by an increasing trend in summer SSTs during 1997 (Table 1). An El Niño was forecast to occur during the 1997/98 winter. In all five previous El Niños (excluding 1996/97 from the sample) with this development (previous non-ENSO winter and increasing summer SSTs), DeKalb experienced warmer than average temperatures. Looking at the flowchart (Figure 2) for the ENSO model, ride would be chosen because warmer conditions would assume less demand for natural gas and thus lower prices during the winter. The CPC long-lead climate outlook produced in September 1997 (when the purchasing decision is made) predicted a 10% excess likelihood of ‘above average’ temperatures in a large region including DeKalb and northern Illinois, leading to another ride decision. The natural gas supply tool showed that in late September 1997 the supply level was 4% below the 1993–99 average, thus classifying it as ‘normal’ and leading to a decision to lock in a price because average to below average supplies suggest higher future prices as seasonal demand increases. Since two of the three tools suggest riding the market, a decision to ride would be implemented. By riding the market during 1997/98 academic year, the heating plant reduced expenditures for natural gas (Changnon et al., 1999). The purchasing choice for each winter, as determined by the decision tree, was compared to the price of natural gas during the months of December through March. The natural gas price data are averaged for each


The decision tree was tested for four winters (1995/96 through 1998/99) since CPC long-lead outlooks were only available during this period. Although the decision tree may be implemented at any time, it was arbitrarily determined that the purchasing decision would be made near 1 October. Table 2 contains the decision by each part of the decision tree and the overall final decision suggested by equally weighting each tool. Using monthly natural gas price and usage information obtained from the NIU heating plant manager, estimates of the total cost associated with December through March heating were determined using (a) an October fixed price (per therm) and (b) a rolling market price (Table 3). Interestingly, the October fixed price varied greatly over the four-year period including $0.17 per therm in 1995/96, $0.195 per therm in 1996/97, $0.31 per therm in 1997/98, and $0.24 per therm in 1998/99. These costs do not include delivery costs and other fees. This wide range in October prices indicated that a particular decision to go with a fixed price or ride the market could reduce NIU’s total expenditures for natural gas over the heating season. Comparisons of total winter (December through March) expenditures for natural gas indicate that NIU would have saved more than $300,000 going with the fixed October price in 1995/96, more than $600,000 going with the fixed price in 1996/97, more than $250,000 going with the rolling market price in 1997/98, and more than $125,000 going with the rolling market price in 1998/99. Table 4 compares the decision tree purchasing choice and the actual natural gas purchasing choice based on the lowest total winter costs determined from Table 3. The decision tree correctly chose the appropriate purchasing method in three of the four winters evaluated. In the case of 1996/97, the climate forecast factors of

Integrating climate forecasts and natural gas supply information

Figure 3. Natural gas actual cost data per therm for 1995–99. Shaded areas show differences between the October fixed price and the rolling market price for the December−March period each year.

Table 2. Predictions of the three tools in the decision tree and overall pricing decision by winter for four years. Winter season Revised

ENSO model

CPC forecast

Natural gas supply

Decision tree purchase choice





1995/96 1996/97 1997/98 1998/99

Table 3. Monthly and seasonal natural gas usage and cost information for the December through March period for the 1995/96 through 1998/99 winters. Estimated total seasonal costs using (a) a fixed 1 October price each year and (b) a rolling market price. Winter season 1995/96

5 1996/97







Lock total

usage $/therm cost($)

1,035,800 0.24 248,592

1,125,900 0.18 202,662

991,500 0.27 267,705

927,800 0.32 296,896


usage $/therm cost($) usage $/therm cost($) usage $/therm cost($)

1,011,000 0.42 424,620 868,620 0.26 225,841 830,992 0.19 157,888

1,145,900 0.45 515,655 917,741 0.23 211,080 1,053,570 0.18 189,643

869,100 0.33 286,803 711,289 0.20 142,258 784,794 0.21 164,807

778,200 0.19 147,858 729,800 0.23 167,854 781,012 0.22 171,823


the decision tree indicated that the decision maker should ride the market, while the actual natural gas supply, which was much (8%) below the average supply (Figure 1), suggested going with a fixed price (the correct decision). This example demonstrates how important the non-climate forecast tools can be in developing a purchasing decision. Based on discussions with the heating plant manager, it appears that in years when the natural gas supply anomaly is >6% either above or below the average then greater weight should

Ride total


$1,374,936 $1,000,510 $747,033 $828,088 $684,161

be given to the natural gas supply tool. Likewise, in years when there is greater certainty with the climate forecast information, such as was the case during the 1997/98 El Niño (Changnon, 1999), then more emphasis should be placed on using those tools. Both the ENSO model and the natural gas supply tool correctly predicted the price outcome in three of four years, while the CPC forecast only provided a skill forecast (other than CL) in two years, with one correct


D Changnon, M Ritsche, K Elyea, S Shelton and K Schramm Table 4. Decision tree purchasing decision compared to lowest total cost option determined from actual natural gas usage and cost information (Table 3).

with continual interactions between applied climatologists and weather-sensitive decision-makers.

Winter season


1995/96 1996/97 1997/98 1998/99

Decision tree purchase choice

Lowest total cost option



and one wrong. The performance of the decision tree during the period of study indicated that it can be used in the future.

3. Conclusions In an ongoing applied climatology project with the NIU heating plant, relationships between natural gas usage/supply and two climate forecast tools were investigated. From this, a natural gas purchasing decision tree was created that incorporated past climate in northern Illinois and ENSO events, CPC long-lead climate outlooks, and natural gas supply information. In future years, near 1 October, NIU’s heating plant manager and staff meteorologist will assess the outcomes from the three parts of the decision tree and develop a natural gas purchasing decision. The decision tree made the correct purchasing decision in three out of four winters. Due to the small data samples associated with each of the three tools, the decision tree may be of limited use to decision-makers in some years. However, in years when (a) the ENSO model emphasizes a particular temperature anomaly (e.g. all six El Niños following a non-ENSO event produced above average temperatures in DeKalb), or (b) the CPC temperature forecast anomaly is large (e.g. >10% in the 1997/98 El Niño), or (c) the natural gas supply values are far from average (e.g. >6% below average in 1996/97), the heating plant manager should place greater weight on those factors in the natural gas purchasing decision. While 1 October was arbitrarily chosen as the decision time, it is important to realize that when choosing different times of the year the decision period may affect the performance of the decision tree. This applied climate research is clearly a ‘work-inprogress’. Continued testing of the tree is suggested in order to assess its ability to make the right purchasing decision prior to a heating season. In addition, continual modification of the ENSO model and CPC’s outlooks, as new information becomes available, will further enhance their usefulness in the tree. Overall, results indicate that there is economic value associated


The project team wishes to express appreciation to Gilbert Sebenste and Mike Saari of NIU for their input in this project and the natural gas information they provided. Meetings with Gilbert Sebenste, NIU’s staff meteorologist, gave the group the necessary information to tailor the study to the needs of the NIU heating plant. The team would like to thank those at the Midwestern Climate Center for providing data, Leonard Walther from NIU’s cartography lab for creating the figures, and Stanley Changnon and the anonymous reviewers for their comments. This research was supported by NSF CAREER grant #ATM-9508038.

References Barnston, A. G., van der Dool, H. M., Zebiak, S. E., Barnett, T. P., Ji, M., Rodenhuis, D.R., Cane, M. A., Leetmaa, A., Graham, N. E., Ropelewski, C. R., Kousky, V. E., O’Lenic, E. A. & Livezey, R. E. (1994). Long-lead seasonal forecasts – Where do we stand? Bull. Am. Meteorol. Soc., 75: 2097−2114. Barnston, A. G., Leetmaa, A., Kousky, V. E., Livezey, R. E., O’Lenic, E. A., Van den Dool, H., Wagner, A. J. & Unger, D. A. (1999). NCEP forecasts of the El Niño of 1997−98 and its U.S. Impacts. Bull. Am. Meteorol. Soc., 80: 1829− 1852. Brown, B. G. & Murphy, A. H. (1987). The potential value of climate forecasts to the natural gas industry in the United States. SCIL Report 87-2, Oregon State University, Corvallis, OR, 68 pp. Changnon, D. (1998). Design and test of a ‘hands-on’ applied climate course in an undergraduate meteorology program. Bull. Am. Meteorol. Soc., 79: 79−84. Changnon, D., Creech, T., Marsili, N., Murrell, W. & Saxinger, M. (1999). Interactions with a weather-sensitive decision maker: a case study incorporating ENSO information into a strategy for purchasing natural gas. Bull. Am. Meteorol. Soc., 80: 1117−1125. Changnon, S. A. (1999). Impacts of 1997-98 El Niño-generated weather in the United States. Bull. Am. Meteorol. Soc., 80: 1819−1827. Changnon, S. A., Changnon, J. M. & Changnon, D. (1995). Uses and applications of climate forecasts for power utilities. Bull. Am. Meteorol. Soc., 76: 711−720. Energy Information Administration (1999). Natutral Gas Weekly Market Update. Office of Oil and Gas, Department of Energy, Washington, D.C. Knox, J.B., Moses, H. & MacCraken, M. C. (1985). Summary report on the workshop on the interactions of climate and energy. Bull. Am. Meteorol. Soc., 66: 174−183. Schoonmaker, D. (1997). El Niño and underemployment. Am. Sci., 85: 319−321. Weiss, E. B. (1982). The value of seasonal climate forecasts in managing energy resources. J. Appl. Meteorol., 21: 510− 517.