Bias Correction Method for Solar Radiation Based on Quantile Mapping to Provide Weather Data for Building Energy Simulations

Abstract

The weather and climate model output has systematical errors called the bias. Bias corrections are necessary in order to use the model output for an application field, such as in building energy simulation (BES). In general, climate models can predict the daily maximum amount of solar radiation on clear days with sufficient accuracy. However, it is difficult to accurately model cloud physics processes, with model results sometimes predicting less cloudy days compared with the actual observations. When we correct solar radiation bias using a conventional bias correction method (BCM), which uses only the average solar radiation, the daily maximum value deviates significantly from the observed results, even when the daily average is accurate. In this paper, we present a BCM called quantile mapping (QM), which considers both the daily integrated and the maximum amount of solar radiation to provide the bias corrected weather data for the BES. In addition, we conducted BESs using the corrected weather data to evaluate the efficiency of the QM. When using the weather data corrected only by the monthly average, the BES could predict the average energy consumption, but the maximum cooling load was underestimated by 12%. Conversely, when using the data corrected by QM using either the daily cumulative or the maximum amount of solar radiation, the BES predicted the maximum cooling load with only 6% and 2% respectively. model results sometimes predicting less cloudy days compared with the actual observations. When we correct solar radiation bias using a conventional bias correction method (BCM), which uses only the average solar radiation, the daily maximum value deviates significantly from the observed results, even when the daily average is accurate. In this paper, we present a BCM called quantile mapping (QM), which considers both the daily integrated and the maximum amount of solar radiation to provide the bias corrected weather data for the BES. In addition, we conducted BESs using the corrected weather data to evaluate the efficiency of the QM. When using the weather data corrected only by the monthly average, the BES could predict the average energy consumption, but the maximum cooling load was underestimated by 12%. Conversely, when using the data corrected by QM using either the daily cumulative or the maximum amount of solar radiation, the BES predicted the maximum cooling load with only 6% and 2% respectively. model results sometimes predicting less cloudy days compared with the actual observations. When we correct solar radiation bias using a conventional bias correction method (BCM), which uses only the average solar radiation, the daily maximum value deviates significantly from the observed results, even when the daily average is accurate. In this paper, we present a BCM called quantile mapping (QM), which considers both the daily integrated and the maximum amount of solar radiation to provide the bias corrected weather data for the BES. In addition, we conducted BESs using the corrected weather data to evalua...

High-Resolution Solar Climate Atlas for Greece under Climate Change Using the Weather Research and Forecasting (WRF) Model

“…However, as Table 1 demonstrates, during the statistical comparison between WRF climatology and ECA dataset, the error analysis results yielded that the WRF model (black line) tended to overestimate constantly the GHI values of the ECA dataset (grey line), presenting a systematic bias error against observed values, with average values of the order of 57 W/m 2 for the whole dataset, resulting in the need for bias adjustment. Hence, the use of bias correction was applied in order to adjust the WRF model output according to the existing climate regime, which is necessary for the production of valuable outcomes in climate applications [29,30]. Thus, the bias in mean monthly model values is corrected by subtracting the value of 57 W/m 2 , which is the difference between modeled and observed mean monthly GHI values of the time period; see Table 1.…”