Search

This study evaluates the performance of different analog-based downscaling models for probabilistic quantitative precipitation forecasts over the metropolitan area of Bilbao, Spain. The analog-based statistical downscaling models used are a probability of the exceedance models in which the probability is derived from the probabilities given by a set of analogs and three Bayesian models. Results show that the differences in the performance of the models are subtle. The simplest model, which makes no use of Bayesian methods, performs better than the other models in the forecast of very low precipitation events, very likely due to the quality of the found analogs, since these categories are very populated in the phase space and forecast is very easy. However, as precipitation rates increase, Bayesian models perform better than the simple one based on probability of exceedance from individual analogs. The Bayesian models involving precipitation in the computation of the likelihood show in general the smallest biases in the forecast versus observed probabilities. [ABSTRACT FROM AUTHOR]

Abstract: In Bilbao (Spain), an air quality network measures sulphur dioxide levels at 4 locations. The objective of this paper is to develop a practical methodology to identify redundant sensors and evaluate a network''s capability to correctly follow and represent SO2 fields in Bilbao, in the frame of a continuous network optimization process. The methodology is developed and tested at this particular location, but it is general enough to be useable at other places as well, since it is not tied neither to the particular geographical characteristics of the place nor to the phenomenology of the air quality over the area. To assess the spatial variability of SO2 measured at 4 locations in the area, three different techniques have been used: Self-Organizing Maps (SOMs), cluster analysis (CA) and Principal Component Analysis (PCA). The results show that the three techniques yield the same results, but the information obtained via PCA can be helpful not only for that purpose but also to throw light on the major mechanisms involved. This might be used in future network optimization stages. The main advantage of cluster analysis and SOMs is that they provide readily interpretable results. All the calculations have been carried out using the freely available software R. [Copyright &y& Elsevier]

This work focuses on the prediction of hourly levels up to 8h ahead for five pollutants (SO2, CO, NO2, NO and O3) and six locations in the area of Bilbao (Spain). To that end, 216 models based on neural networks (NNs) were built. The database used to fit the NNs were historical records of the traffic, meteorological and air pollution networks existing in the area corresponding to year 2000. Then, the models were tested on data from the same networks but corresponding to year 2001. At a first stage, for each of the 216 cases, 100 models based on different types of neural networks were built using data corresponding to year 2000. The final identification of the best model was made under the criteria of simultaneously having at a 95% confidence level the best values of R 2, d 1, FA2 and RMSE when applied to data of year 2001. The number of hourly cases in which due to gaps in data predictions were possible range from 11% to 38% depending on the sensor. Depending on the pollutant, location and number of hours ahead the prediction is made, different types of models were selected. The use of these models based on NNs can provide Bilbao''s air pollution network originally designed for diagnosis purposes, with short-term, real time forecasting capabilities. The performance of these models at the different sensors in the area range from a maximum value of R 2 =0.88 for the prediction of NO2 1h ahead, to a minimum value of R 2 =0.15 for the prediction of ozone 8h ahead. These boundaries and the limitation in the number of cases that predictions are possible represent the maximum forecasting capability that Bilbao''s network can provide in real-life operating conditions. [Copyright &y& Elsevier]

This study evaluates the performance of statistical models applied to the output of numerical models for short-term (1–24 h) hourly wind forecasts at three locations in the Basque Country. The target variables are horizontal wind components and the maximum wind gust at 3 h intervals. Statistical approaches such as persistence, analogues, linear regression, and random forest (RF) are used. The verification statistics used are coefficient of determination (R2) and root mean square error (RMSE). Statistical models use three inputs: (1) Local wind observations; (2) extended EOFs (empirical orthogonal functions) derived from past local observations and ERA-Interim variables in a previous 24-h period covering a domain around the area of study; and (3) wind forecasts provided by ERA-Interim. Results indicate that, for horizons less than 1–4 h, persistence is the best model. For longer predictions, RF provides the best forecasts. For horizontal components at 4–24 h horizons, RF slightly outperformed ERA-Interim wind forecasts. For gust, RF performs better than ERA-Interim for all the horizons. Persistence is the most influential factor for 2–5 h. Beyond this horizon, predictors from the ERA-Interim wind forecasts led the contribution. Hybrid numerical–statistical methods can be used to improve short-term wind forecasts. [ABSTRACT FROM AUTHOR]