METEOROLOGY AND STATISTICAL SCIENCE
AT IOWA STATE UNIVERSITY

A group of statisticians and metorologists at Iowa State University have formed an informal working group we call Meteorology and Statistical Science (MASS). The group meets regularly to discuss possible directions and collaborative projects between these two disciplines. Based on interests and current research thrusts of the individuals in their own disciplines, we define and pursue projects of mutual benefit.

Current Projects of MASS:

Since early 2010, the cooperating members of MASS have focused primarily on two projects.

1. Improved Forecasts for Wind Speed

The development of renewable sources of energy is being pursued from many angles by researchers across the world. In the United States, wind energy is produced primarily by moderate to large "wind farms"-- collecitons of wind turbines at a general location. The state of Iowa is a major contributor to the production of wind energy, ranking near the top of states in both absolute wind energy production and per-capita use.

Companies involved in the production of wind energy must buy and sell power from power grids, and initial bids are made up to 48 hours in advance. While adjustments are made to these bids at various intervals (including real-time decisions), buying energy due to a shortfall in what was anticipated from forecasts becomes more expensive over time. Similarly, selling energy produced that exceeds what was anticipated from forecasts becomes less profitable. Dispsosal of energy that has been produced but for which there are no consumers is also expensive. Quality forecasts of wind speed at wind farms and the amount of potential power generation are necessary for the efficient use of wind power and, as a consequence, maximizing the decrease in power that must be obtained from non-renewable sources such as coal-fired power plants.

Wind speed forecasts have generally been produced by Numerical Weather Models (NWM), which are deterministic or mathematical process models. A unique aspect of forecasting wind for power production is that forecasts need to be made for altitudes at which turbines are placed (typically about 80 meters) rather than ground-level forecasts such as are publicly reported at many weather stations. This is not a problem for NWM which are based on three-dimensional grids in space, but it can complicate the process of obtaining observed wind speeds for comparison to the forecasts.

(image obtained from www.midamericanenergy.com)

Our current work involves examination of idividual "cases" of wind speed forecasts. A case consists of a 54 hour period at a given wind farm for which we have available

• Hourly wind speed forecasts at turbine height from about 12 NWMs.
• Hourly observed wind speeds from a weather tower at turbine height.

For each case, we use the first 30 hours of forecasts and observed wind speeds to model the differences between forecast and observed values for each NWM. A number of different statistical model structures have been examined, most of which produce either summary values of how well the NWM forecasts are matching observed values for the first 30 hours, or estimates of NWM bias as a function of additional covariates (such as forecasted temperature or wind sheer). Models fit to this data record are then used to produce pure forecasts of wind speed for the final 24 hours of the case. The manner in which these forecasts are produced depends on the type of model used in the 30 hour "training period", but may take the form of weighted averages of NWM forecasts or averages of "bias-adjusted" NWM forecasts.

2. Comparison and Assessment of Regional climate Models

Meteorologists at Iowa State University have been participating in the North American Regional Climate Change Assessment Program (NARCCAP), which is an international effort among agencies and academic institutions in the United States, Canada, and northen Mexico. A central objective of this project is to develop improved methods for comparing regional climate models to each other and to observed values. In general, statisticians have approached this objective along two avenues we think of as (1) overlaying stochastic structure on deterministic climate model output and (2) quantifying uncertainty with climate model ensembles. This division seems useful to help organize the types of activities that statisticians are involved in relative to climate model comparison and assessment, although they are not mutually exclusive approaches.

1. Model Output Statistics
   The first approach, overlaying stochastic structure on the output of deterministic climate models, is often called Model Output Statistics in the literature. In this approach, statistical models are fit to climate model output as if that output was observed or measured data. The logical basis for this approach is that although climate model output is actually deterministic in nature, climate models contain such complex mathematical dynamics that treating model output as if it were a realization of some spatial and temporal stochastic process can help us understand the overall structure of climate models.
2. Climate Model Ensembles
   Another approach that has been taken is to construct probabilistic structures for outputs from collections of climate models known as ensembles. Ensembles of climate models arise from different model structures, use of different "boundary conditions", and use of different future scenarios for model inputs such as anthropomorphic sources of carbon dioxide. For regional climate models, boundary conditions provide sets of constraints on, for example, the total amount of energy available in a given region over a given time span. Boundary conditions for regional models often are quantified in terms of model output from global climate models. There exist a variety of particular statistical models that have been used with climate model ensembles in the literature. The basic objective of these models, however, is to quantify the amount of uncertainty we have in climate model projections based on variability of those projections among members of the ensemble.

(image obtained from www.ucar.edu)

Our current work on this project involves developing methods that can be used to compare the output from several regional climate models to each other and compare regional climate mode output to historical observations. We have chosen to focus initially on a region in the upper midwest of the United States, and to consider the variables of temperature, atmospheric pressure, and precipitation. We are developing an approach that differs from both the Model Output Statistics and use of Climate Model Ensembles described previously. The methodology we are attempting to develop combines aspects of randomization probability, spatial and spatio-temporal subsampling, and the use of empirical distribution functions to quantify variability in regional climate models.

MASS Participants

The primary contributors at Iowa State University to the efforts described on this web page are as follows:

MASS Meetings