26-28 April 2017 15th Itron Energy Forecasting Conference Presenter: Kristin Larson, Ph. D. Data Scientist Kristin.Larson@stormgeo.com Weather Trends & Climate Normals
Our Resources Decision Support for Weather Sensitive Operations 2
Our Experience 3
YOUR 24/7 Wx Department FORESIGHT: Proprietary long range weather forecasting solutions for energy markets THREAT ID: Early identification of potential severe weather grid disruption for load shedding PRECISION FORECAST DATA: Increase base load certainty with precise hourly point forecasts for LFEs NENA ANALYTICS: Coal & dry freight fundamentals influencing supply, demand, freight cost and arbitrage opportunities DEEP STORM: Predictive & Prescriptive machine learning algorithms for utilities 4
Machine Learning DeepStorm Uncover unknown patterns in your data through machine learning. Create predictive algorithms to drive improved business decisions for competitive differentiation. 5
Weather forecasting petabytes of data Advances in algorithms and GPU hardware has enabled deep neural networks to beat competing algorithms by large margins in fields like speech recognition and object detection in images. These same deep neural networks are used in StormGeo to discover complex structures in vast weather datasets and and relate them to business-impacting phenomena. A traditional statistical approach (like MOS) will use 10 to 20 variables to train the model and is restricted to linear relantionships. A deep neural network can include 1,000,000 variables in 3 dimensions and is not restricted to linear relationships. 6
ForeSight: Long Range Forecasting ForeSight 30 30 day outlook updated 3x per week Service includes: Forecast of temperature and precipitation anomalies across the globe On-demand video updates by our senior scientists 3x per week Live access to expert meteorologists via phone, email, and web conference Access to information through StormGeo's unique customer portal ForeSight 90 This quarterly outlook provides a long-range forecast in a 90-day format. StormGeo uses thorough global teleconnections and in-house developed forecast indices to provide market intelligence for expected weather conditions for the next 90 days. ForeSight Seasonal Whether its hurricanes, snow storms, severe weather, or other seasonally focused sensitivities - our meteorologists are able to provide long-range weather forecasting to help you prepare for and make decisions now that will affect the future. 7
ForeSight: Long Range Forecasting Tropical Cyclone (TC) Risk Assessment A detailed examination and discussion of near-term and long-range weather parameters/models with respect to the potential tropical cyclone risk in a particular basin. Analog seasons (seasons with a similar weather pattern and ocean temperature setup) are compared to the currently-assessed risk to a particular basin. The assessment has 3 main parts: Short-Term: 7-14 days Medium-Range: 30-45 days Long-Range: Seasonal outlook 8
ForeSight: Long Range Forecasting Hosted long range forecast weather portal at StormGeo.com 9
Nena Analysis A leading analysis house delivering energy market insights Coal Market Analysis We cover the global coal market. Web-based platform with daily coal market comment from our analysts, a directional price forecast for API2 and data access for your own use Weekly reports Monthly reports Request a free trial Get full access to StormGeo s analysis for one week free of charge. Follow the market closely with Nena s price forecasts and market updates throughout the day via Nena s website and in reports issued by email. A web meeting with an analyst is included in the free trial. Ken.Carrier@StormGeo.com Nena Coal Weekly report offers short term directional coal price forecasts for API2, API4 and API8, while Nena Coal Monthly presents scenario-based price forecasts for API2 (CIF ARA) front month and next four quarters. We calculate the coal power output for major countries and the coal-to-gas switching price. We assess the Global Arbitrage Coal Matrix in order to depict trade flow opportunities and trends. https://www.nena.no/global-coal/ 10
Climate Normals Supplemental Climate Normals available from: https://www.ncdc.noaa.gov/normalspdfaccess/ Download before the talk with internet access for your locations 11
Load Changes Predicting the Response of Electricity Load to Climate Change Patrick Sullivan, Jesse Colman, and Eric Kalendra National Renewable Energy Laboratory #64297 http://www.nrel.gov/docs/fy15osti/64297.pdf 12
Load Changes Temperature sensitivity to load using 2005-2006 (Platts, FERC form 714) Baseline Electricity demand based on the U.S. EIA Annual Energy Outlook (degree days at 1990-2010 averages) Temperature changes from MIT s Integrated Earth System Model (scenario RCP4.56, radiative forcing stabilizes at 4.5 W/m 2 in 2100) 13
Load Changes Annual Load change in 2050 compared to baseline Summer increases and winter decreases 14
Load Changes Summer Load Increase in 2050 compared to baseline 15
Weather Trends Change in Heating Degree Days 16
Weather Trends Change in Cooling Degree Days 17
Climate Normals Defined by World Meteorological Agency, a "normal" of a particular variable (e.g., temperature) is defined as the 30-year average. For example, the minimum temperature normal in January for a station in Chicago, Illinois, would be computed by taking the average of the 30 January values of monthly averaged minimum temperatures from 1981 to 2010. Each of the 30 monthly values was in turn derived from averaging the daily observations of minimum temperature for the station. In practice, however, much more goes into NCEI's Climate Normals product than simple 30-year averages. Procedures are put in place to deal with missing and suspect data values. In addition, Climate Normals include quantities other than averages such as degree days, probabilities, standard deviations, etc. Climate Normals are a large suite of data products that provide users with many tools to understand typical climate conditions for thousands of locations across the United States. 18
Climate Normals 30 year average updated every 10 years by NCEI Compared to the previous Climate Normals, the new Climate Normals includes the decade of the 2000s and loses the decade of the 1970s. As the 2000s were warmer than the 1970s, this has had a warming influence on the Climate Normals. Comparing these decades using our best dataset for climate change analysis, the USHCN, we find that the decade of the 2000s was about 1.5 F warmer than the 1970s. For maximum, minimum, and mean temperature the difference, respectively, was 1.37 F, 1.55 F, and 1.46 F. As the Climate Normals are an average of three decades, this warmed the new Climate Normals by approximately 0.5 F. The difference between these values and the actual difference between the reported 1971 2000 Normals and the new Normals are caused by station moves, changes in observing practices or instruments, etc. 19
Climate Normals Supplemental Climate Normals available from: https://www.ncdc.noaa.gov/normalspdfaccess/ 20
NOAA s 1981-2010 Climate Normals: Monthly Temperature Normals various definitions of normal climate Denver-Stapleton, CO GHCN-ID: USW00023062 Latitude: 39.7633 N Longitude: 104.8694 W Elevation: 1611.2m Maximum Temperature ( F) Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Ann 1981-2010 Normal 44.5 46.1 53.6 60.9 70.5 81.2 88.2 85.7 77.2 64.9 52.3 43.3 64.1 1991-2010 Normal 44.9 47.1 54.6 61.0 71.3 81.4 88.8 85.8 77.8 65.3 52.7 44.7 64.6 1996-2010 Normal 45.5 46.7 54.5 61.2 71.5 81.6 89.5 85.9 77.9 65.4 54.0 44.7 64.9 2001-2010 Normal 45.7 45.3 54.2 62.1 71.0 81.6 89.8 86.1 78.0 65.3 54.7 44.6 64.9 2006-2010 Normal 43.9 45.6 54.7 61.0 70.9 82.1 88.5 85.9 77.4 64.4 56.0 41.9 64.4 Optimal Climate Normal 45.5 45.9 54.2 61.0 71.2 81.7 89.7 86.2 77.7 65.3 54.2 43.8 64.7 Hinge Fit Normal 46.2 46.7 56.2 61.8 71.6 82.0 89.6 86.3 78.3 65.1 54.3 44.0 65.2 Minimum Temperature ( F) Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Ann 1981-2010 Normal 17.4 19.6 26.4 34.1 43.8 52.8 59.0 57.4 47.5 35.7 24.9 16.7 36.4 1991-2010 Normal 17.4 19.9 26.8 33.9 44.0 52.7 59.1 57.3 47.4 35.7 25.0 17.7 36.4 1996-2010 Normal 17.8 19.4 26.5 33.7 43.7 52.6 59.8 57.3 47.4 35.9 25.8 17.7 36.5 2001-2010 Normal 17.9 18.4 27.2 34.3 43.5 53.0 59.6 57.0 47.2 35.8 26.3 17.5 36.5 2006-2010 Normal 17.1 18.6 27.3 33.1 43.3 53.1 58.8 57.4 46.8 35.7 27.0 16.0 36.2 Optimal Climate Normal 17.7 19.4 27.4 33.9 43.4 52.8 58.9 57.3 47.7 35.7 25.2 17.1 36.4 Hinge Fit Normal 18.4 19.7 28.2 34.3 43.8 53.1 59.2 57.2 47.5 35.4 25.9 16.9 36.6 For Further Reading: Arguez, A., R. S. Vose, and J. Dissen, 2013: Alternative Climate Normals: Impacts to the Energy Industry. Bulletin of the American Meteorological Society, 94, 915-917. Arguez, A., and R. S. Vose, 2011: The Definition of the Standard WMO Climate Normal: The Key to Deriving Alternative Climate Normals. Bulletin of the American Meteorological Society, 92, 699-704. 21
2006-2010 Normal 2006-2010, the most recent 5 years in the period of record. 1996-2010 Normal 1996-2010, the most recent 15 years in 1981-2010 Normal 30-yr 1981-2010, the most recent 30 years in This is the traditional way that NOAA has computed Climate Normals. 5-yr 15-yr Optimal Climate Normal An average of the most recent N years in the period of record. A formula based on time series features, such as long-term trends, determines the value of N. A smaller (larger) value of N typically implies a stronger (weaker) trend in the time series. Included it in the average for that pre-determined number of years. The value was a candidate (it could have been used), but it wasn't ultimately used. Not sure about this description 2001-2010 Normal 2001-2010, the most recent 10 years in 1991-2010 Normal 1991-2010, the most recent 20 years in N Hinge Fit Normal 10-yr 20-yr A normal calculated using a statistical fit (black line segments) through the data points. Prior to the hinge point, the fit must be flat. Thereafter, the fit can be increasing, decreasing, or flat. The hinge fit normal (black dot) is defined as the value of the fit through the most recent year in the period of record (i.e., 2010). Frequently asked questions 1.What are climate normals? Traditionally, NOAA defines a climate "normal" as a 30-year average. For example, we compute the January temperature normal for a station by averaging the 30 January values of monthly temperatures from 1981 to 2010. Climate Normals are used to determine the rates a power company can charge its customers, where and when to schedule an outdoor wedding, and countless other applications. 2. Why do you provide seven different normals instead of one? Is it because of global warming? Many users of NOAA s Climate Normals products have expressed concerns about using a 30-year average in an era of observed climate change (see the For Further Reading section on the previous page). In fact, some of our users have begun calculating their own 10-year averages, for example. NOAA provides these additional computations to help users make better-informed decision. NOAA also recognizes that alternative ways of defining normal may work better than the 30-year average given observed global warming. Note: The values plotted in the schematics, as well as the depicted size of N in the Optimal Climate Normals schematic, are for illustrative purposes only, and do not reflect actual climate data. Page 2 of 2
2006-2010 Normal 2006-2010, the most recent 5 years in the period of record. 2001-2010 Normal 1981-2010 Normal 2001-2010, the most recent 10 years in 1996-2010 Normal 1996-2010, the most recent 15 years in 1991-2010 Normal 1991-2010, the most recent 20 years in 30-yr 1981-2010, the most An average of the most recent N in the period of record. A formula based recent 30on time years series features, in the period of such as long-term trends, record. This is the traditional way that determines the value of N. A smaller (larger) value of N typically implies a stronger (weaker) trend in the time series. NOAA has computed Climate Normals.
1991-2010 Normal 1991-2010, the most recent 20 years in the period of record. 20-yr
1996-2010 Normal 1996-2010, the most recent 15 years in the period of record. 15-yr
2001-2010 Normal 2001-2010, the most recent 10 years in the period of record. 10-yr
2006-2010 Normal 2006-2010, the most recent 5 years in the period of record. 5-yr
2006-2010 Normal 2001-2010 Normal 2006-2010, the most recent 5 years in the period of record. 5-yr 2001-2010, the most recent 10 years in 10-yr 1996-2010 Normal 1996-2010, the most recent 15 years in 15-yr 1991-2010 Normal 1991-2010, the most recent 20 years in 20-yr
Weather Trends Estimation and Extrapolation of Climate Normals and Climatic Trends Robert E. Livezey, Konstantin Y. Vinnikov, Marina M. Timofeyeva, Richard Tinker and Huug M. van den Dool Climate Prediction Center JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY, Nov. 2007 1759-1776 http://journals.ametsoc.org/doi/abs/10.1175/2007jamc1666.1 29
2006-2010 Normal 2006-2010, the most recent 5 years in the period of record. 2001-2010 Normal Optimal An average Climate over Normal 2001-2010, the most recent 10 years in 1996-2010 Normal 1996-2010, the most recent 15 years in 1991-2010 Normal 1991-2010, the most recent 20 years in N An average of the most recent N years in A formula based on time series An average of the most features, such as long-term recent N years in the period of record. A formula based on time series features, trends, such as long-term determines trends, the value of determines the value of N. A smaller (larger) value of N N. typically A smaller implies a stronger (larger) value of N (weaker) trend in the time typically implies a stronger series. (weaker) trend in the time series.
2006-2010 Normal 2006-2010, the most recent 5 years in the period of record. 2001-2010 Normal Hinge Fit Normal 2001-2010, the most recent 10 years in 1996-2010 Normal 1996-2010, the most recent 15 years in 1991-2010 Normal 1991-2010, the most recent 20 years in A normal calculated using a statistical fit (black line segments) through the data points. Prior to the hinge point, the fit must be flat. Thereafter, the fit can be increasing, decreasing, or flat. The hinge fit normal (black dot) is defined as the value of the fit through the most recent year in the period of record (i.e., 2010).
2006-2010 Normal 2006-2010, the most recent 5 years in the period of record. 2001-2010 Normal 2001-2010, the most recent 10 years in 1996-2010 Normal 1996-2010, the most recent 15 years in 1991-2010 Normal 1991-2010, the most recent 20 years in An average of the most recent N years in the period of record. A formula based on time series features, such as long-term trends, determines the value of N. A smaller (larger) value of N typically implies a stronger (weaker) trend in the time series.
Climate Normals NCEI engaging users for additional products. Working on ENSO (El Nino Southern Oscillation) normals. Send suggestions to: Anthony Arguez, Ph.D. Physical Scientist NOAA's National Centers for Environmental Information (NCEI) anthony.arguez@noaa.gov 828-271-4338 33
Kristin Larson, Ph. D. Data Scientist Kristin.Larson@stormgeo.com Ken Carrier Industry Manager 12650 N. Featherwood, Suite 140 Houston, TX 77034 Ken Mobile (832) 258-8086 ken.carrier@stormgeo.com www.stormgeo.com Thank you! 34 34