Renewable power generation and electricity demand forecast reply on weather forecast. The report ‘Forecasting wind and solar generation: improving system operations’ introduces what methods and weather forecasting data are used to generate wind and solar forecasts that inform system operators. Below are some characteristics of operational weather forecasting people should know.
Weather forecasting has a predictability limit around 10-14 days. The weather forecast people learn from radio, TV and online is short-range forecast. Mid-range forecast is limited to about 15 days. Although monthly and seasonal forecasting cannot reveal detailed weather conditions, they address some information on the general trend a few weeks and seasons ahead in terms of anomalies relative to climatology.
Ideally, weather forecast informs the state of future atmosphere in every location, both horizontally and vertically. Due to limited computing power, the atmosphere is divided into three-dimension grid boxes and the evolution of atmospheric variables is modeled within boxes. The atmospheric motion is govern by dynamical equations that describe the conservation of momentum, mass, energy, and water, and the physics pertains to their sources and sinks. With dynamical equations, pressure, temperature and water fields in the next time step can be derived from the previous time step.
Initial states are important to output accurate forecasting, known as initial value problems. Due to simplifications in physical parametrizations and numerical errors to solve equations, forecast can deviate from reality and therefore the forecast at this time step does not serve well as initial conditions for the forecast at the next time step. Data assimilation is the method used by NWP models to generate initial conditions. Data assimilation blends observations with the short-range forecast from the previous run of the model, and therefore it includes the laws of physics and real information about the atmosphere at specific times and locations. Assimilation process takes place in a simplified and lower resolution version of a NWP model in a window that is typically 3-6 hours. Through iterations of backward and forward runs, the model trajectory after correction gets closer to observations.
Monthly and seasonal predictability depends on boundary conditions of the system – that is the Earth’s surface (ocean and land) and top of atmosphere. To consider the impact of boundary conditions on atmosphere, one approach is to use empirical (statistical) methods to describe the relationship between the predictand and predictor variables. The other approach is to assimilate boundary conditions from observations in NWP models. Longer range predictability is largely dependent upon general atmospheric patterns over large areas, which are driven by internal variability of the atmosphere. Hence, NWP models for season predictions run at lower resolution and incorporate information on ocean variables that change with time, which are different from models for short-to-medium range equivalents which use fixed SSTs. To isolate the predictability that comes from boundary conditions, forecasters run an ensemble of forecasts and then average them to see if a signal is left.
Thanks to advancing computing power and more observations, weather forecast becomes more accurate these days but is not perfect yet. Therefore, human forecasters still play important roles in the NWP age. For example, they would assess NWP models by using radar and satellite imagery. Senior forecasters in forecast centers sometimes need to modify forecast based on their knowledge about systematic errors and local atmosphere.