Portugal using an ensemble of model results

Transcription

1 Forest fires impact on air quality over Portugal using an ensemble of model results A.I. Miranda, A. Monteiro, V. Martins, C. Borrego, M. Schaap, P. Builtjes CESAM, University of Aveiro, Portugal TNO, Environmental Quality Dep. Apeldoorn, The Netherlands Workshop GLOREAM/ACCENT 27 November 27-3, Berlin, Germany

2 scope estimate the air pollution effects of 23 forest fires through the application of two air quality modelling systems - CHIMERE and LOTOS-EUROS 1.Air quality modelling 2.Ensemble results

3 forest fires: the year Area burned (ha) August 23 86% of the area burned in August

4 forest fire emissions: 23 emission factor - EF Method E i = EFi β B A global burning efficiency - fuel load - B area burned - A β 5 Total emissions 23 (t) FOREST FIRES 4 ROAD TRANSPORT INDUSTRY AND SERVICES Results CO CH 4 NMVOC PM 2.5 PM 1 NO x

5 forest fire emissions A t 23 August 23 PM (t) CO (t) Emissões CO (ton) -5 PM 1 (t) CO (t) Emissões PM1 (ton) ( )

6 air quality modelling: August 23 CHIMERE LOTOS-EUROS Boundary conditions MOZART and GOCART climatological models Emissions EMEP and Portuguese inventory Vertical structure 8 layers (up to 3 m) Chemical mechanism MELCHIOR Boundary conditions Logan climatological datasets Emissions TNO inventory Vertical structure 3 layers (up to 35 m) Chemical mechanism CBM-IV Horizontal grid resolution Horizontal grid resolution 1 x 1 km x 12.5 km 2

7 air quality modeling: domains European domain CHIM-EUR 58ºN Portugal domain CHIM-PT 35ºN 58 km 14ºW 5x5 km 2 25ºE 29 km

8 air quality modeling: PM 1 CHIMERE LOTOS-EUROS 6 6 no forest fires forest fires no forest fires forest fires 5 5 RMSE (µg m -3 ) RMSE (µg m -3 ) Aveiro Coimbra Leiria Lisboa Porto Santarém Setúbal no forest fires forest oestfires 1 Aveiro Coimbra Leiria Lisboa Porto Santarém Setúbal no forest fires forest fires.8.8 r RMSE decreases 9% for CHIMERE and 8% for LOTOS-EUROS CHIMERE and LOTOS-EUROS underestimate PM 1 concentrations Correlation increases with forest fire emissions r Aveiro Coimbra Leiria Lisboa Porto Santarém Setúbal Aveiro Coimbra Leiria Lisboa Porto Santarém Setúbal

9 air quality modeling: O 3 CHIMERE LOTOS-EUROS 7 6 no forest fires forest fires 7 6 no forest fires forest fires (µg m -3 ) RMSE (µg m -3 ) RMSE Aveiro Castelo Branco Coimbra Lisboa Porto Santarém Setúbal Aveiro Castelo Branco Coimbra Lisboa Porto Santarém Setúbal r no forest fires forest fires RMSE decreases 17% for CHIMERE r 1 no forest fires forest fires O 3 : CHIMERE underestimates and LOTOS-EUROS overestimates For some districts the inclusion of forest fire emissions did not improve the LOTOS-EUROS performance Aveiro Castelo Branco Coimbra Lisboa Porto Santarém Setúbal Aveiro Castelo Branco Coimbra Lisboa Porto Santarém Setúbal

10 air quality modeling with - without forest fires CHIMERE LOTOS-EUROS ΔPM 1 (µg m -3 ) ΔPM 1 (µg m -3 ) PM 1 avg August, 3 rd

11 air quality modeling with - without forest fires CHIMERE LOTOS-EUROS ΔPM 1 (µg m -3 ) ΔPM 1 (µg m -3 ) PM 1 avg August, 3 rd

12 air quality modeling with -without t forest fires CHIMERE LOTOS-EUROS O 3 max ΔO diff 3 O3 (µg (µg.m-3) m ) ΔO 3 (µg m -3 ) August, 3 rd

13 air quality modelling: August 23 CHIMERE LOTOS-EUROS Run 1 Forest fire emissions distributed in the first 3 Layers + MELCHIOR2 chemical mechanism Run 2 3 Layers + MELCHIOR1 Run 3 8 Layers (3 meters) + MELCHIOR2 Run 1 Resolution (17.5 x 12.5 km 2 ) + no SOA chemistry included Run 2 Resolution (17.5 x 12.5 km 2 ) + SOA Run 3 Resolution (8.75 x 6.25 km 2 ) + no SOA Run 4 Run 4 8 Layers (3 meters) + MELCHIOR1 Resolution (8.75 x 6.25 km 2 ) + SOA

14 Model validation Background air quality stations PM 1 O 3 LEC VNT COI COI FUN CHA REB REB

15 air quality modeling: PM RMSE (μg.m -3 ) BIAS (μg.m -3 ) LEC C OI FUN REB LEC C OI FUN REB ,9 ient corr relation coeffici,8 7,7,6,5,4,3,2,1 Ensemble CHIMERE Ensemble LOTOS Ensemble LEC C OI FUN REB

16 air quality modeling: O ) RMSE (μg.m VNT COI CHA REB -3 ) BIAS (μg.m VNT COI CHA REB 1.9 c orrelation coeff ficient Ensemble CHIMERE Ensemble LOTOS Ensemble.1 VNT COI CHA REB

17 Some comments Forest fire emissions can represent an important source of air pollutants to the atmosphere Models performance increase when forest fire emissions are considered. There is no clear trend for best model performance between CHIMERE and LOTOS. Model results increase when ensemble techniques are applied.

18 ensemble =?

19 questions Ensemble modelling or selecting the best model? Many could be better than one. Instead of losing potential sources of information, models could be combined to yield better performance. How to select models/runs for final ensemble? Cross validation can be used for model training/selection i ti and the estimation of the expected classification error. The models are initialized with different model parameters and cross validation helps us to find proper values for these parameters and to select the best performing models for the final ensemble. How to choose model/run weights? How to perform spatial model ensemble?

20 Obrigada!!!