Remote Sensing Specialist (Project: R-Smog)

Bhatti, S. (Consultant)

Activity: Consultancy


This study explored a variety of satellite-based remote sensing data and methods to map and analyse active fires (biomass burning), air pollutants and smog in the province of Punjab, Pakistan during the period from October 2016 to February 2017. The data of active fires, Aerosol Optical Depth (AOD), particulate matter 2.5 (PM2.5), particulate matter 10 (PM10), carbon monoxide (CO), sulphur dioxide (SO2), nitrogen dioxide (NO2) and surface reflectance was acquired from different data sources, and was processed using a variety of GIS- and remote sensing- based methods to meet the requirements of this study. Based on the temporal availability, some of the data was prepared and analysed at weekly time scale, while others were examined at monthly intervals. The selection of data and methods was based on an extensive review of relevant literature.

With the focus on mapping and analysing the spatial distribution and temporal variations in fire occurrences/thermal anomalies, air pollutants and smog/fog, a two-part methodology was developed where the first one explored the processed data in both space and time, while the second examined the spatiotemporal relationships among the different data variables. Three frameworks were established for conducting correlation analyses using generalized linear regression: (1) weekly correlation analysis of smog/fog with predictor variables (active fire, AOD, CO, SO2 and NO2); (2) weekly correlation analysis of active fire with predictor variables (AOD, CO, SO2 and NO2); and (3) monthly correlation analysis of active fire with predictor variables (PM2.5 and PM10). Animated time-series images illustrating the spatiotemporal variations in each data variable, and group animation images for each correlation framework were also created which provided an intuitive visual representation of how the variables corresponding to each framework varied in both space and time.

The spatiotemporal variations in each data variable were examined individually. A fluctuating trend was observed in weekly data of active fires, whereas the monthly data indicated high spatial clustering of fire occurrences towards east and north-east of Punjab during October 2016, which increased and extended towards central parts of the province in November 2016. The fire clusters were more prominent in central and southern Punjab during December 2016, while the southern cluster became more noticeable in January 2017. Highest number of fire occurrences were recorded during November 2016 (7323 occurrences), whereas the lowest during January 2017 (1628 occurrences). The AOD data indicated high spatiotemporal variability in weekly measurements during the study period. PM2.5 data exhibited some spatial clustering where concentration of high values was found towards north-west during October 2016, which shifted and slightly expanded spatially towards central and eastern parts of the province during November 2016. This cluster of high PM2.5 values started dispersing and shifting towards south during December 2016, and later formed a denser cluster in this area during January and February 2017. The temporal trend indicated that the highest PM2.5 values were observed during October 2016, which gradually reduced till January 2017, and again increased slightly in February 2017. PM10 exhibited a spatiotemporal pattern almost similar to that of PM2.5 (with very few differences) during the study period. CO has been somewhat persistent in terms of spatial distribution throughout the study period: low values were mostly observed towards north and west, whereas high values occurred generally towards south and east of the province. The only anomaly was observed during the first week of February 2017 when high values were observed towards north. Highest CO values were recorded during December 2016, while October 2016 saw some of the lowest CO values. Contrary to CO, SO2 has been fluctuating spatially and temporally during the study period. Although very little variation was observed in the low values of SO2, the high values showed some erratic behaviour. The temporal trend indicated that high SO2 values (weekly) were quite variable during October 2016 and February 2017, which was in line with its erratic spatial distribution during these months. Highest SO2 value was observed during November 2016, while the lowest in February 2017. NO2 exhibited somewhat persistent spatial pattern where high levels were recorded towards east and in some areas towards north of the province during majority of the study period, while low to moderate values were generally observed in majority of the study area. The temporal trend indicated very little variation in low NO2 values, whereas high values indicated somewhat increasing trend during October and December 2016, which however started decreasing from week 3 of December 2016 till week 3 of January 2017. A sharp increase in high NO2 value was observed during week 4 of January 2017, which again decreased during February 2017. The spike in the data of January 2017 week 4 might be because of some data error as rest of the trend did not show any irregular behaviour. Speaking of smog / fog, some noticeable patches were identified as being covered by these during the first week of November 2016, weeks 2-4 of December 2016 and first week of January 2017. Animated images produced using the time-series maps also assisted in understanding the spatiotemporal variations in each data variable.

The first framework of correlation analysis, which examined the relationships between smog and predictor variables (active fire, AOD, CO, SO2 and NO2) at weekly time scale, indicated a variable degree of association. A positive correlation between the response and predictor variables was observed during a few weeks while the relationship between these was either negative, or could not be established during other time periods. Interestingly, during most of the weeks where high smog /fog (large covered areas) was identified, the relationship of smog / fog with SO2, CO, NO2 and AOD turned out positive (with a few exceptions where this relationship was not clear but more or less, the trend was observed). However, the relationship of active fire was often found negative, which indicated that active fire during a given week had less or no contribution towards the formation of smog. This behaviour affirms that the impact of biomass burning or active fire during a given week could be observed in the coming week/s since the pollutants accumulate in the atmosphere to a certain level before mixing with the fog to form smog. Moreover, biomass burning occurring outside the study area (e.g. in the neighbouring regions of Indian Punjab) could also have contributed heavily to the formation of smog in the study area.

The second framework of correlation analysis examined the relationship between active fire and its predictor variables (AOD, CO, SO2 and NO2) at weekly timescale. Mostly, a positive relationship was observed between active fire and the predictor variables during October 2016. It is important to note that October and November are crop harvest months in the study area where farmers often burn the crop remains to prepare the fields for sowing of wheat crop. The increased crop burning during this period resulted in formation of heavy smog during the first week of November, which gradually thinned out during the next two weeks. No smog was identified during the last week of November and first week of December; however, heavy smog came in again and remained for the next three weeks. Biomass burning is practiced for space heating and cooking during severe cold weather during the months of December and January not only in the rural parts of the study area which are not well connected to Natural Gas Grid, but also at places in the urban areas because of shortage in the supply of natural gas. The relationship between active fire and predictor variables was not significant during the first half of November 2016; however, the trend picked up again and resulted in a strong relationship in the second half of November and first half of December, thus conforming to our earlier conclusion that active fires in a given time might have contributed to the formation of smog in the proceeding weeks.

The third framework of correlation analysis examined the relationship between active fire and its predictor variables (PM2.5 and PM10) at monthly timescale. A positive relationship was observed between active fire and PM2.5, whereas PM10 exhibited a negative relationship during most of the time period. This trend needs further scientific evidence at a better temporal resolution, e.g. weekly data (data of PM2.5 and PM10 was available only at monthly time scale).

Nevertheless, this research study provided very useful insight into the spatiotemporal relationships between smog, active fire, and different pollutant variables in Punjab, Pakistan, and elaborated the potential of satellite-based remote sensing data and geospatial methods for conducting such studies. Only freely available data sources were exploited in this research. Mapping of smog has been challenging in this study, and further research is recommended for improving the mapping process by integrating data on precipitation, water vapour content, cloud cover, etc. with the current methods. In addition, modelling of pollutant transport trajectories (taking into account the areas adjoining Punjab), in conjunction with the methods and results presented in this study, could provide a deeper understanding of the spatiotemporal relationships between smog and biomass burning.
Period1 Jun 2017 - 31 Oct 2017
Work forFood and Agriculture Organization of the United Nations, Italy
Degree of RecognitionNational


  • Remote sensing
  • GIS
  • Smog
  • Crop burning
  • Thermal remote sensing
  • Spatio-temporal analysis
  • Punjab, Pakistan