SO2 concentrations forecasting for different hours in advance for the city of Salamanca, Gto., Mexico

U.S. Mendoza-Camarena, F. Ambriz-Colin, D.M. Arteaga-Jauregui, A. Vega-Corona, J.M. Barron-Adame.

University of Guanajuato, Mexico.

OVERVIEW

Objectives

Predict SO2 concentration levels with different hours in Advance.
Convert the predicted levels to an Air Quality Index (AQI).
Use predictions to prevent environmental contingencies.
Analyze results to verify the potentiality to apply the method for other Pollutants.

Introduction

Salamanca, Gto., is one of the most polluted cities in Mexico (and probably in the world).
There have been a lot of Environmental Contingencies over the last 5 years.
Currently, there are 3 monitoring stations located in strategic places.
Nobody was seriously analysing data from those stations.
An Environmental Contingency Plan has recently been released.

SO₂ and PM₁₀ concentration levels activate the plan.
This Plan only considers corrective and not preventive actions.

What is being done and what to do?

Why prevention is not applied?

People that work in the Monitoring Stations are not trained.
A lot of information, but no capacity to analyse it.
Pollution has become a political issue.

What can be done to prevent instead of correct?

Analyse information (look for factors that have influence in pollution levels).
Search methods that can be easily implemented in Software to model the pollutants behaviour.
Automate the process of showing information.

Methodology

The methodology presented in the next slides, is only an overview of what was done to get the results. A different and deeper view of the methodology that was used is presented in the article.

Proposed Model

Prediction Phase

Results

Table shows only the best results obtained for different number of atmospheric variables.

The table shows 1-hr ahead predictions.

Used Variables                        RMSE (PPB) MAE(PPB)

SO₂ 28.41 17.28

SO₂, WS

SO₂, WD

28.40

29.06

17.23

16.97

SO₂, WS, WD 29.10 16.97

SO₂, WS, WD, T, RH 28.62 17.13

1-hr ahead predictions

Results for 1-hr Ahead SO2 Concentration Predictions:

Root Mean Squared Error (RMSE):  28.41 (PPB)

Mean Absolute Error (MAE): 17.98 (PPB)

Results for 1-hr ahead AQI Prediction:

Effectiveness Percentage = 86.26%.

Graphical results are shown in the attached figures.

Results for 12-hrs ahead predictions

Results for SO2 Concentration in Parts Per Billion:

Root Mean Squared Error (RMSE): 31.96 (PPB)

Mean Absolute Error (MAE):  16.95 (PPB)

Results for Air Quality Index of SO2:

Effectiveness Percentage: 89.43%

Graphical Results can be observed in the attached figures.

Results for 24-ahead Prediction

Results for SO2 Concentration Levels:

Root Mean Squared Error (RMSE): 37.27 (PPB)

Mean Absolute Error (MAE): 22.26 (PPB)

Results for Air Quality Index:

Effectiveness Percentage (EP): 86.64%.

Graphical Results can be observed in the attached figures.

Conclusions

Results are very Reliable for a 1-hr and 12-hr ahead Prediction.

For this experiment, Atmospheric Variables didn't show too much influence.

More data is needed in order to train the GRNN to perform Predictions along all the year.

GRNN can be used to warn authorities over possible Environmental Contingencies.

Need to use Techniques to improve the Prediction Performance.

Further Work

Use all the available data from the Monitoring Stations.

Process data in a different way.

Evaluate different Artificial Neural Networks (ANNs) topologies.

Use the ANNs to predict other pollutants (CO, PM₁₀, NO, etc.)

Implement algorithms in a Web-based information service.