China testbedFMI-ENFUSER in Langfang

Res. Manager, Adj.Prof. Ari KarppinenModel developer, Res. Scientist Lasse Johansson

• What is the FMI-ENFUSER model?• A brief description

• Setting up the system in China• Main objective and the selected test region• Implementation of GIS-datasets for environment profiling• Gaining access to AQ measurements and meteorological data

• Status and preliminary results• What works and what needs more work• First fusion results in Langfang

2

Outline

02.05.2023 3

What is FMI-ENFUSER? (1/3)

ENFUSERHistorical

concentration time series in the region

Land use mappings

Population density

mappings

Emission source mappings

Modelled input

Observed input

Traffic HouseholdsIndustrial

Based on ALL available input, estimate pollutant spatial and temperal variation

of concentrations

Understand and describe the environment

Understand the historical behavior of pollutants (in various environments)

Understand the conditions and pollutant concentrations at hand

02.05.2023 4

What is FMI-ENFUSER? (2/3)FMI-ENFUSER = (The Finnish Meteorological Institute’s ENvironmental information FUsion SERvice)

The fusion of information (a separate task for the model) has been described in(Johansson et al, 2014)

• Combines land-use regression (LUR) and dispersion modelling into a novel approach named as ”dynamic land-use regression”

• Essentially, this is 3D land-use regression taking into account the meteorological conditions, especially the evolution of the wind direction.

• There are several different layers of ”land-use” for which the method is applied simultaneously.

• Strengths:• High resolution, especially suitable for urban areas

(if supporting information available !)• Information on emission sources not ABSOLUTELY necessary

• Automatic calibration : learning a continuous process • Information on emission sources, if known, can be included (e.g. shipping,small

scale wood combustion)• Fusion algorithm => latest sensor measurements & modelled data can be

included in the pool of information• Weather forecast + regional background forecast => ENFUSER• forecasting possible IF forecast model information available

• Challenges:• Statistical relationships between pollutant concentrations and extreme

meteorological conditions is difficult to define and utilize (rare situations always hard for statistical models!)

• Calibration is difficult with incomplete/low quality GIS-dataset

02.05.2023 5

What is FMI-ENFUSER? (3/3)

Main issue in China at the moment

Objective: Fuse PM2.5 measurements in Langfang

1. Describe the environment in the surronding region as accurately as possible

• Source and the nature/quality of GIS data unknown2. Gain access to AQ measurements in the surrounding

region• Pegasor + other unclassified sources of information• For calibration and operational use

• Decent calibration: 20+ stations, (minmimum of) full annual time series3. Gain access to weather data

• For calibration and operational use• Forecasts?

02.05.2023 6

China Testbed setup

Optimal calibration: weather data for the same period as

AQ measurements

02.05.2023 7

Testbed region

To obtain realistic behaviourin the model it is not enough toconcentrate only on Langfang

The surrounding area is equallyimportant for the calibration ofthe model.

The selected testbed region also includes Beijing, Tianjin, Tangshan, Baoding and several other cities.

For all of these other cities the environment has been mapped with the same detail as in Langfang.

=> When calibrated and operational ENFUSER should work all across the selected region.

02.05.2023 8

Open source Land-useFinland/Europe Langfang/China

Forests, plains, parks, lakes, sea, roads (5), residential, industrial, buildings

Lakes, sea, roads (5)

02.05.2023 9

Enhancing Open source Land-use with satellite images

• New approach: analyse rudimentary land use from satellite images => Fill in the gaps in OSM mappping

• Vegetation, Urban, Suburban• Simple image processing technique

• Deduction based on • Dominant color• Brigtness• Saturation

• Approach seems to work well in Hebei province when the ”eye altitude” of satellite is approx. 100km

• 100 x 100m resolution acheived • Better resolution would require more

sophisticated image processing

• Important for ENFUSER• Is used as a proxy for PM2.5 emissions• Desired resolution: 250 x 250m

• Best ”dataset” found for this purpose was an image describing the population in a 5 x 5km resolution

• This was converted into Googe Earth layer file (kmz) and fitted to the area => coordinates for the data

• Gives only indicative information on the population

02.05.2023 10

Population density mapping (1/3)

02.05.2023 11

Population density mapping (2/3)

Average aerosol optical depth, indicating the relative amount of particles that absorb sunlight. Based on satellite remote sensing during 2007-2011. (available near-real-time)Modis Terra (NASA), aerosol optical depth at 550nm 2007-01 to 2011-12 average.Data source:http://daac.gsfc.nasa.gov/giovanni

02.05.2023 12

Indicative population mapping 250 x 250m resolution

Indicative population mapping 5x5km resolution

Satellite data

Population density mapping (3/3)

Original population data redistributed emphasizing urban and suburban areas

Langfang

Langfang

Satellite data enhances both land-use and population mapping

02.05.2023 13

OpenStreetMap & street canyonsOpenSreetMap (OSM) is an open access map service provider that offers high resolution maps world wide.

FMI-ENFUSER uses OSM-maps with 5 x 5m resolution, covering all main cities in Finland

Street canyons and buildings can be analyzed from the image.

This is how FMI-ENFUSER ”sees” the crossing of Lönrotinkatu and Fredrikinkatu after image processing. The vicinity of buildings can be taken into account when the concentration is being estimated in urban areas.

02.05.2023 14

OpenStreetMap & street canyonsAn example of NO2 fusion at the center of Helsinki based on local measurements.

Colorscale: [10 -120] µg/m3.

With street canyon detection ENFUSER understands the input data (measurements) better and associates correctly higher concentrations to all (trafficed) street canyons.

OSM-data in China doesn’t contain buildings!

02.05.2023 15

PM2.5 from householdsIndividual buildings can be detected and their size can be evaluated using OpenStreetMap maps.

It makes sense then, to teach the model to associate PM2.5 area emissions to these small households.

=> gain coverage outside of HMA

Test variable setup:• building land-use from OSM-dataset (strictly required)• Smaller house => higher contribution per m2.

• Must be between 50m2 and 500m2.• Secondary land use information to help classification

• ‘Suburban’ => higher contribution per m2.• ‘Urban’, ‘Industrial or Economic’ => significantly smaller contribution per m2.

16

Working days, Winter Sundays, Spring

Seasonal PM2.5 averages (1/2)

Based on measurements from 45 stations.

Visualization: simple kriging extrapolation (with ENFUSER visualization toolbox).

No fusion of information, this kind of raw data is used for the calibration of the model together with meteorological data.

• Besides seasonal variation there’s also a clear diurnal variation to be seen in average PM2.5 concentrations (not shown here)• Highest concentrations during Winter (Monday-Friday)

• Current GIS-datasets cannot yet explain why the highest concentrations are observed near Baoding and Tangshan• More explaining factors (layers

of information, are needed• Demographics?• Wealth?• Industry?• Other?• Long-range transport?

02.05.2023 17

Seasonal PM2.5 averages (2/2)

Working days, Winter

02.05.2023 18

Tang

shan

: Xiao

shan

Baod

ing: H

uand

ian II

Baod

ing: S

WC

Tang

shan

: Sta

tion

Baod

ing: S

wH

Baod

ing: C

MS

Tang

shan

1Ba

oding

: FF

Tang

shan

2 T

angs

han

3Ba

oding

: RC

Tang

shan

: CC

Lang

fang

BDA

Beijin

g: D

D

Beijin

g: C

AEH

Beijin

g: S

huny

i

Beijin

g: O

lympic

Lang

fang

Aer

ospa

ce

Beijin

g: W

WP

Beijin

g: S

hijing

shan

Tian

jin: N

anjin

g Ro

ad

Lang

fang

TCM

Beijin

g: W

est P

ark

Beijin

g: H

aidian

Lang

fang

EMSC

Beijin

g: H

uairo

uTi

anjin

: JVF

Tian

jin: 3

Cang

zhou

: 1Ti

anjin

: 5

Beijin

g: C

hang

ping

Cang

zhou

: TV

Br.

Tian

jin: W

est S

tatio

n

Tian

jin: T

aifen

g IP

Tian

jin: 6

Tian

jin: T

iansh

an R

oad

Beijin

g: 7

Tian

jin:E

ast S

tatio

n

Tian

jin: Y

ongm

ing R

oad

Tian

jin: R

elay

Fact

ory

Tian

jin: D

ongli

Sch

ool

Tian

jin: H

angt

ian R

oad

Qinhua

ngda

o: 1

Cang

zhou

: Mun

iciple

EPB

Tian

jin: H

anbe

i Roa

d

40

50

60

70

80

90

100

110

120

130

Winter

Spring

Aver

age

PM2.

5 co

ncen

tratio

n [µ

gm-3

]

= in Langfang

The measurement height for the included sensors is unknown presumably 5-20m, and may cause additional bias in the calibration process.

Winter season includes February only (no data for Decemer and January)Spring: March ->

19

• Utilizing Pegasor’s PUAQ sensors

LANGFANG demonstration

Figure: A closer look at the hourly PM2.5 concentration in Langfang, given by FMI-ENFUSER

An example of estimated PM2.5 concentration time series based on the sensor data in Langfang. The

selected example location is just outside of Silver City Hotel, Langfang

Describe the environment as accurately as possible• New approach: OSM layer implemented

• Information content low in China• No buildings => no street canyon detection

• New approach: Satellite data implementation with image processing• Enhances the OSM-data

• A population density mapping implemented• Quality and reliability still poor for Chinese data• Enhancement based on satellite data

• Road specific traffic flow mapping for Langfang • Implementation ongoing, should prove to be useful

• Road traffic is not expected to be the key driver for observed PM2.5 concentrations

02.05.2023 20

Summary (1/3)

Gain access to AQ measurements in the surrounding region• 5 (+ more now!) Pegasor sensors installed in Langfang• Hourly data ”available” from 900+ stations in China

• 45 stations were identified and utilized in study area • Several pollutant species• Data available since Feb 2015 => calibration for Winter/Spring/Summer

can be done

Gain access to weather data• CMA agreed to provide weather data for the calibration period

(pending)• Backup solution: open access weatherdata (with forecasts) since

Apr 2015

02.05.2023 21

Summary (2/3)

• Despite the difficulties in obtaining GIS-data a preliminary collection of information has been implemented for environment profiling

• A satisfactory amount of pollutant and weather data is available in China

• Quality will further improve after the addition of PEGASOR and CMA data• Denser sensor network will reveal better the ”micro structure” of PM2.5

concentrations • ENFUSER can provide useful information already now:

• The full value of ENFUSER is revealed only when the quality and availability of input data improves and the training is based on sufficiently long statistics (like it already is in Finland)

02.05.2023 22

Summary (3/3)

www.fmi.fi

Johansson, L., Epitropou, V., Karatzas, K., Karppinen, K., Wanner, L., Vrochidis, S., Bassoukos, A., Kukkonen, J. and Kompatsiaris I. Fusion of meteorological and air quality data extracted from the web for personalized environmental information services. Environmental Modelling & Software, Elsevier, Volume 64, February 2015, Pages 143–155, 2014.