Evaluation of weather datasets for building energy simulation

In recent years, calibrated energy modeling of residential and commercial buildings has gained importance in a retrofit-dominated market. Accurate weather data play an important role in this calibration process and projected energy savings. It would be ideal to measure weather data at the building location to capture relevant microclimate variation but this is generally considered cost-prohibitive. There are data sources publicly available with high temporal sampling rates but at relatively poor geospatial sampling locations. To overcome this limitation, there are a growing number of service providers that claim to provide real time and historical weather data necessary for building modeling at 15–40 km2 grid across the globe; common variables such as temperature and precipitation have been constructed on ∼1 km2 grids [1]. Unfortunately, there is limited documentation from 3rd-party sources attesting to the accuracy of this data. This paper compares provided weather characteristics with data collected from a weather station inaccessible to the service providers. Monthly average dry bulb temperature; relative humidity; direct normal, diffuse and global solar radiation; wind speed and wind direction are statistically compared. Moreover, we ascertain the relative contribution of each weather variable and its impact on building loads. Annual simulations are performed for three different building types, including a closely monitored and automated energy efficient research building. The comparison shows that the difference for an individual variable can be as high as 90%. In addition, annual building energy consumption can vary by ±7% while monthly building loads can vary by ±40% as a function of the provided location's weather data.

► Survey of weather vendor data statistically compared to weather station data.
► Peak difference of temperature was 8 °C monthly, 11 °C daily, and 17 °C hourly.
► Annual energy consumption varied by ±7% and monthly conditioning loads by ±40%.
► Weighting factor of each weather variable's impact on energy use is calculated.