Table 1.

Candidate predictor variables available for model selection.

Variables and categories	Unit	Buffer size (m)	Source
Traffic variables			OpenStreetMap (2019) [27]
Total length of major roads	m	50, 100, 200, 500
Total length of secondary roads	m	50, 100, 200, 500
Distance to the nearest major road	m	—
Distance to the nearest secondaryroad	m	—
Land use variables			World Bank [28] 20 m × 20 m
Commercial/business/industrial	m2	50, 100, 200, 500
High-density residential	m2	50, 100, 200, 500
Low/medium-density residential	m2	50, 100, 200, 500
Peri-urban areas	m2	50, 100, 200, 500
Normalized difference vegetationindex (NDVI)	—	50, 100, 200, 500	United States Geological Survey [29]—Landsat 8 imagery (30 m × 30 m)
Waterways (total length)	m	50, 100, 200, 500	OpenStreetMap [27, 2019]
Counts of building	N	50, 100, 200, 500	Maxar/Ecopia.ai [30, 2020]
Population			Ghana census (2010) data [31]
Biomass use	%	50, 100, 200, 500
Population density	pop km−2	50, 100, 200, 500
Human activities			Google Places (retrieved in 2020)
Number of restaurants	N	50, 100, 200, 500
Number of schools	N	50, 100, 200, 500
Presence of bars	N	50, 100, 200, 500
Presence of shops	N	50, 100, 200, 500
Meteorological variables
Temperature	˚C	—	Kestrel weather meters
Relative humidity	%	—	Kestrel weather meters
Wind speed	m s−1	—	Kestrel weather meters
Mixing layer depth	m	—	HYSPLITE model [32]
Solar radiation	W km−2	—	HYSPLITE model [32]
Water vapor mixing ratio	kg kg−1	—	HYSPLITE model [32]

N: number.