1 Introduction

Among the techniques and tools capable of assisting continuous monitoring of different environmental phenomena, remote sensing and GIS have proved to be the most practical to study and analyse earth’s atmospheric and surface processes. Fog in northern India is one such phenomena occurring every year over the Indo-Gangetic Plains affecting normal human activities, especially driving on the roads, agriculture, aviation and also health. Radiational fog and smog (smoke + fog) occur several times during the winter months (December–February) every year. The Indo-Gangetic Plains is a vast stretch of land in South Asia, spread over parts of Pakistan, India and Bangladesh and houses a large portion of the population of South Asia. Meteorologically, radiational fog is very low stratus cloud, which even touches the ground (Anthis and Cracknell 1999). The basic requirements for radiational fog to form are sufficient moisture in the air with high relative humidity, low air temperature and very low wind speed. There must also be condensation nuclei viz. dust particles, aerosols or pollutants onto which water can condense. Fog often forms after sunset as the air and ground surface begins to cool and condensation replaces evaporation. Air cools best with clear skies with little moisture in the mid and upper troposphere. High concentration of aerosol content in the lower atmosphere over the study area also enhances the formation and permanence of fog by additional condensation nuclei (Ramanathan and Ramana 2005; Prasad et al. 2006). Smog that is a toxic cocktail of fog, ash, acids, aerosols and other particles can bring damaging effects on agriculture etc.

During the past few decades, several studies have been done (e.g., Eyre et al. 1984; Dybbroe 1993; Park et al. 1997; Anthis and Cracknell 1999; Bendix 2002; Ahn et al. 2003; Bendix et al. 2004, 2005; Ellrod and Lindstrom 2006; Yoo et al. 2006) related with fog using remote sensing data. Few have also focused on fog dissipation and forecasting (e.g. Gurka 1974, 1978; Gustafson and Wasserman 1976; Anthis and Cracknell 1999; Choudhury et al. 2007; Ward and Croft 2008). Whereas the present study mainly focuses on the identification of the fog prone zones, mapping of fog, its dissipation and migration patterns and forecasting particularly over northern India.

2 Data used and methodology

2.1 Data used

NOAA-AVHRR multispectral images provide an excellent database for mapping and analysis of radiational fog in Northern India. Both daytime and nighttime coverages provide more means for choosing these data sets for the present study. Clouds and fog can be easily differentiated in NOAA daytime false colour images. In the nighttime scenes, infrared channels 3 and 4 are very useful for identifying the same. However, METEOSAT data are also useful for fog studies. In the present study, NOAA-AVHRR data are mainly used in order to carry out the analysis and mapping of the fog-affected areas. METEOSAT-VISSR data were utilized only for verifying the affected area in the scenes where the limited coverage of NOAA was available. MODIS (Terra and Aqua) could not be used in this case owing to the poor coverage of northern India in a single scene in spite of getting high spatial resolution (250 m) images. In order to study the dissipation and migration patterns of fog, large coverage of the study area is necessary. From all these aspects, NOAA-AVHRR images were found to be more suitable. The orbits of NOAA are set up in such a manner that an active pair of satellites provides the coverage of almost the Earth’s entire surface twice daily during morning, afternoon, early evening and night passes. The cloud-free NOAA-AVHRR images used in the present study were acquired and archived regularly by the Indian Institute of Technology Roorkee—Satellite Earth Station (IITR-SES) since October 2002.

Detection of fog can be done using all channels of NOAA-AVHRR. But it is best detected in channels 1 and 2 because of high reflectivity during daytime. False colour composite (geometrically corrected) images were prepared from Level 1b data. Fog detection is also possible using the two infrared channels (3 and 4) of NOAA-AVHRR, and using the difference of these the characteristics of various clouds and nighttime fog were studied (Park et al. 1997).

The analysis was done on a few selected NOAA images that were found to be fog affected. While selecting the images, proper coverage of the study area in the images and the maximum fog cover were taken as the main basis. In the present study, additional data sets of 2005–06, 2006–07 and 2007–08 have been included and integrated with the results and interpretations made on the basis of the earlier studied fog periods (2002–03, 2003–04 and 2004–05) by Choudhury et al. 2007. The details of the selected NOAA-AVHRR scenes are listed in Table 1.

Table 1 NOAA-AVHRR data used in the present study for analysis of different fog seasons between 2005 and 2008
Full size table

The daily data (different times for a single day) of air temperature, relative humidity and wind speed were obtained from Russia’s Weather Server (2008) and organized (for the same dates as that of satellite images as shown in Table 1) for meteorological stations (Fig. 1) within and nearby the fog-affected belt. Since fog formation is maximum between midnight to dawn given the cumulative radiative cooling effects of the atmosphere and surface, the early morning data (between 4 and 5 a.m. local time) were collected for the analysis. For the elevation data, geocoded Global USGS DEM (US Geological Survey-Digital Elevation Model) (1 km resolution) are used (US Geological Survey 2008).

Fig. 1
figure 1

Shows locations of meteorological stations of which the air temperature, relative humidity and wind speed data were used in the study

Full size image

2.2 Methodology

The prime objective of the present work was to develop a GIS-based fog forecasting model. But before doing any forecasting, it is necessary to identify the influencing parameters in the process of fog formation. Hence, to find out the optimum meteorological parameters for fog formation, different data sets (air temperature, relative humidity, wind speed and elevation) were integrated and analysed in the GIS platform (Arc GIS 9.2). For integrated analysis of the satellite and meteorological data necessary preprocessing of the satellite images is performed in ERDAS IMAGINE 9.0.

2.2.1 Favourable topographic and meteorological conditions for fog

It is necessary to ascertain the favourable topographic and meteorological conditions for performing the prediction of fog. According to Choudhury et al. (2007), air temperature between 3 and 13°C, relative humidity >84%, wind speed <2 m/s and elevation <350 m are the favourable conditions for the formation of fog. On the basis of the analysis of meteorological data for the years 2005–06, 2006–07 and 2007–08, the aforementioned conditions have been applied for verification. Surfaces were created for the temperature, relative humidity and wind speed data using Spatial Analyst (of Arc GIS 9.2) for all specific dates for which satellite data were available. The values of these parameters sustained within the fog-affected area were acquired by extracting the surfaces with the digitized polygons for the fog area. The same procedure was applied with the USGS DEM for finding out the favourable elevation value.

2.2.2 Preprocessing

The most important steps in the preprocessing of images are their georeferencing. The main aim of the georeferencing is to describe the correct location and shape of features by removing the geometric distortions. The sampled NOAA-AVHRR satellite images of the fog-affected dates were georeferenced using an already georeferenced boundary of India in the geographic projection system.

2.2.3 Mapping of fog in northern India for the winter months of 2005–06, 2006–07 and 2007–08

For mapping the fog-affected area, standard Spatial Statistics Extension of Arc GIS 9.2 was used. Before measuring the area, the fog cover boundaries were digitized from the satellite images (which show fog areas very distinctly) for the months from December to February for years 2005–06, 2006–07 and 2007–08. ‘An overlay was then performed for the digitized polygons representing the foggy areas. The cumulative fog-affected area (which also includes repeat fog occurrences) for each year was found by taking the union of the polygons representing fog areas for a single day. The area for each year was then measured using the above-mentioned tool. The variation in fog-affected areas during 2002–2008 was also compared with the variation in the meteorological parameters. The methodology applied for the mapping of fog is shown in Fig. 2.

Fig. 2
figure 2

Flow chart depicts the main data processing steps which have been taken in the present study towards fog mapping

Full size image

2.2.4 Classification of fog-affected areas on the basis of frequency of occurrence of fog using the satellite data from 2002 to 2008

A classification of the fog-affected area was performed on the basis of frequency of occurrence of fog in the winter months for the years 2002–2008. To perform fog classification, 28 evenly distributed places (meteorological stations for which data sets were available) over Indo-Gangetic Plains were taken with geographic locations. Available fog-affected 72 scenes for 6 winter seasons (2002–03, 2003–04, 2004–05, 2005–06, 2006–07 and 2007–08) of NOAA-AVHRR were considered for performing this classification. When the shapefile for the selected stations was overlaid on the georeferenced satellite images, the fog-affected locations were assigned 1 and fog-free locations were assigned 0. After performing the same exercise for all 72 scenes, the frequency of fog against each station was determined. Then by interpolating (Krigging) the frequency values, a classified map of the fog-affected zone was prepared with seven classes at equal intervals.

2.2.5 Study of the dissipation pattern of the North Indian fog

NOAA-AVHRR images were used for understanding the dissipation pattern of fog in the study area. For this, available satellite images (after georeferencing) were arranged in a time series to observe the manner in which an extensive area of fog or stratus dissipates. The same methodology was also applied for studying the changes in fog area in consecutive days within a fog episode. Careful analyses of the images were done to find out the patterns (if any) of fog dissipation or movement, so that it could be introduced in the model for fog forecasting.

2.2.6 Developing the GIS-based fog forecasting model for Northern India

Forecasting of the fog was attempted by analysis of the meteorological data for the studied years. Raster Calculator (Spatial analysis of ArcGIS) tool was used for modelling which allowed overlay analysis of different data layers. The surfaces prepared for the meteorological parameters and the USGS DEM were used as the input for prediction (Fig. 3). The surfaces prepared by both Krigging and Spline methods of interpolation were tried (to know which interpolation technique provides more realistic surface) and results were compared. The conditions used for the same have been identified by studying several fog-affected satellite images (between 2002 and 08) and then compared with corresponding meteorological data sets. The conditions that were found to be best suited for forecasting are given in Fig. 3.

Fig. 3
figure 3

Flow chart depicts the key steps of methodology developed towards fog prediction process

Full size image

3 Results and discussions

3.1 Mapping of fog-affected area

Mapping of fog-affected area was performed for the winter months of the years 2005–06, 2006–07 and 2007–08, and the corresponding measured areas are shown in Figs. 4 and 5. From the calculated area, an overall decreasing trend of the total fog-affected area was observed. The year 2007–08 is found to be the least fog-affected year.

Fig. 4
figure 4

Shows variation observed in the total fog-affected area during 2002–2008 winters

Full size image
Fig. 5
figure 5

Maps showing the total fog-affected area (derived from satellite images) in the winter seasons 2002–03 to 2007–08 and also depicting decreasing trend in the fog-affected area during study period (2002–08)

Full size image

3.2 Classification based on frequency of fog occurrence

The main aim of this classification was to delineate the zones, which remained most fog-affected in the recent years and were more prone to the same in successive years. The variations in the frequency of fog occurrence for different areas are shown in the Fig. 6. The area adjacent to Himalayan foothills between 80°E and 90°E longitude found to be the most frequently fog-covered area (marked as red zones in Fig. 6). The frequency shows a decreasing trend towards south direction. Some of the places like Gwalior, Kanpur, Lucknow, Varanashi, Allahabad (for locations, see Fig. 1) show a more prominent decreasing trend of fog occurrence from 2002 to 2008. These places remained almost fog free in the year 2007–08 (Fig. 7). In case of fog frequency also, the similar type of decreasing trend is observed as in the case of fog-affected area. The frequency was found to be much higher in the year 2002–03, while after a gradual decrease the fog occurrence frequency became lowest in the year 2007–08 (Figs. 7 and 8). In this estimation, only those days were included for which satellite images (complete scenes) were available.

Fig. 6
figure 6

Map shows the frequency classification of the fog-affected belt of northern India during 2002–2008 winters

Full size image
Fig. 7
figure 7

Bar diagrams showing the variations in the frequency of occurrence of fog over some selected places (as shown in Fig. 1) in the Indo-Gangetic Plains during the fog period 2002–2008

Full size image
Fig. 8
figure 8

Bar diagrams and graphs showing the variations in the total frequency and a comparison of occurrence of fog over some selected places in the Indo-Gangetic Plains during 2002–2008

Full size image

3.3 Variations in air temperature, relative humidity and wind speed for the years 2002–2008

In order to analyse the decreasing trend of fog occurrences both in terms of area and in terms of frequency of occurrence, the variation in controlling meteorological parameters for the same time period was also studied. The early morning metrological recordings between 4 and 5 a.m. (this is based on the day-to-day observations of fog formation) were taken for this analysis as it was found to be the closest available data to the time of fog formation process. The datasheets for air temperature, relative humidity and wind speed were prepared at an interval of three days for the winter months December to February for the years 2002–08, and different graphs were produced.

From the air temperature variation graphs, it can be clearly observed that in the years 2002–03 and 2003–04, the temperature for almost all meteorological stations first decreased from December to the first two weeks of January and then starts increasing until February. In these years, no rapid fluctuation in temperatures was observed and hence obtained relatively smooth graphs (Fig. 9). The troughs in the graphs showing the lowering of temperature also coincide with the more frequent and widespread fog events in the years 2002–04. The local air temperature fluctuations are observed to be very high during 2007–08. In this year, these fluctuations are observed to be the highest as depicted by the sharp hinges of the air temperature variation graphs. For long-lasting fog occurrences, such rapid variations are not favourable. So the air temperature variation can be considered as one of the influencing factors responsible for the decreasing fog occurrences in Northern India.

Fig. 9
figure 9

Graphs showing the air temperature variations for the selected meteorological stations during the years 2002–2008

Full size image

In case of relative humidity except for some local variations, it remains almost constant (Fig. 10); however, rapid fluctuations were observed for the year 2007–08.

Fig. 10
figure 10

Graphs showing the relative humidity variations for the selected meteorological stations during the years 2002–2008

Full size image

From the wind speed variation graphs (Fig. 11) no particular pattern can be observed; however, it was found to be generally less than 2 m/s during the winter months (December to February) for the years 2002–08. In the year 2007–08, stations like Patna, Lucknow showing a clear decreasing trend of fog also show some wind disturbances.

Fig. 11
figure 11

Graphs showing the wind speed variations for the selected meteorological stations during the years 2002–2008

Full size image

After the analysis of the variations in fog formation and the above meteorological parameters, it may be concluded that air temperature, relative humidity and wind speed variation patterns had been fluctuating since last few years. Rapid fluctuations do not favour more widespread and frequent fog formations as sustainability in meteorological parameters is also required for fog formation.

3.4 Dissipation and migration pattern of fog in northern India

Generally, the dissipation process of fog does not maintain any particular pattern, since it always maintains a non-uniform thickness and density, which is difficult to detect with the satellite data. Fog cover usually dissipates inward from its outer edges (Gurka 1974). In case of north Indian fog, almost the same pattern had been observed. But in more than 70% cases, it dissipated from the southern edge of the belt, i.e. southern portion of the foggy area dissipates quicker than the northern parts (Fig. 12) because of nearby drier air area. Nearer to the Himalayan foothills, fog sustains for relatively longer time during a single fog event.

Fig. 12
figure 12

NOAA-AVHRR images showing diurnal northward dissipation of fog. Black arrows show the northward dissipation pattern

Full size image

In the present study, it has been observed that generally fog often occurs initially for few days followed by non-foggy days. During such period, an interesting pattern of migration of the same was observed in Northern India. It has been observed that the migration was actually not due to movement by the same patch of fog, but a simple shift in the location of occurrence. This shifting is from the northwestern side towards northeast. At the beginning of such a phase, fog was observed to occur in the western side and then slowly proceeds towards east. This phenomenon is observed for last four years of study period but the pattern was difficult to identify in case of the years 2002–03 and 2003–04 when a single-day event covers the entire Gangetic Plains. The migration pattern is more observable when fog occurs in small patches over the study area (Fig. 13). The probable reason behind such movement pattern was the prevailing wind flow directions as shown in Fig. 14a.

Fig. 13
figure 13

NOAA-AVHRR images showing eastward migration or shifting in the location of occurrence of fog during 2006–07

Full size image
Fig. 14
figure 14

Maps showing (a) wind patterns in the month of January, (b) Regional distribution of natural and anthropogenic aerosol optical depth (AOD) over India at 0.55 μm derived from Moderate Resolution Imaging Spectroradiometer (MODIS) instrument onboard the Terra satellite (December 2002) (Ramanathan and Ramana 2005)

Full size image

3.5 Natural and artificial factors making northern India favourable for fog

Geology and Geography of any area greatly influences the climate. In Indo-Gangetic Plains, such factors also play a great role favouring the formation of fog in the winter season. Among them, the prevailing topographic conditions is one of the most important factors. Presence of the Himalaya in the north contributes a lot in the process of fog occurrence, and during the fog season, winds are considerably calm. Westerly winds are dominant mostly from February to May. But it often disturbs the fog events before February also. The Himalaya marks and controls the northern extent of fog and acts as a barrier to the frigid katabatic winds flowing down from Central Asia. Thus, North India is kept warm or only mildly cold during winter. Extremely low temperature (<3°C) is also not favourable for fog. Therefore, topography and geographic location helps the region to maintain the meteorological optimum for the same.

Another important geological parameter that plays an important role is the soil type (Alluvial soil) of the region. Alluvial soil contains sand, silt and loam generally having high moisture-retaining capacity and good thermal conductivity in the presence of moisture. Due to the thermal gradient, this type of soil shows a vertical moisture flux (Rosenburg et al. 1983). Because of the heat storage and relative constant temperatures of the deeper soil layers, the moisture flow is normally upward in winter and downward in summer due to temperature gradients. This upward movement of moisture can add further moisture just above the ground allowing the radiational cooling process to dominate and induce fog formation. This process is generally dominant in the early winter when we can observe thin layer of radiation fog over any open ground.

The Indo-Gangetic Plains along the Himalaya behaves as a trough resulting in more collection of relatively colder air. Further, the high concentration of the river systems, canals, reservoirs and other local water bodies (Fig. 15) may also have good contributions in raising the relative humidity. Normally, clear sky condition allows faster cooling of the earth due to high radiation during the winter period. During this, the Indo-Gangetic Plains receives occasional rainfall.

Fig. 15
figure 15

Map showing high density of both natural and artificial water bodies along the Indo-Gangetic Plains which contributes in increasing the relative humidity and soil moisture (source: Oxford student atlas for India 2004)

Full size image

After the rainfall, the increased moisture content of the earth’s surface also gets condensed due to the radiative cooling of earth’s surface which may also contribute towards the formation of a thick fog over a wide region. Another type of fog that affects the study area is advection fog, which is caused by the induction of moist winds from southwest (Arabian Sea) over the northern region. Thus, relatively moist and warmer air overrides a substantially cooler surface, causing fog formation just above the ground. Such fog is often related to synoptic weather features and is more persistent than the radiative fog (Wallace and Hobbs 1977). This moist air intrusion is not a very regular event between December and February.

The highly fog-affected region of the northern India remains under extensive agriculture practices like sugar canes, wheat, rice and presence of several important industries like thermal power plants, sugar mills, cement, chemical (Fig. 16). So the haziness due to both pollution and aerosol concentration in the atmosphere is observed to be more during the same period (Figs. 14b and 17). Particulates and aerosol content is very necessary for the suspension of water particles in the atmosphere or simply for the formation of fog. So the main sources of pollution in this region can also be considered as important contributors for the process of fog in Northern India.

Fig. 16
figure 16

Maps showing the locations of major industries and plants causing pollution along Indo-Gangetic Plains in Northern India

Full size image
Fig. 17
figure 17

MODIS false colour composite image of 04 December 2001 showing the haziness over the Indo-Gangetic Plains

Full size image

3.6 Forecasting of fog

Forecasting of fog formation has been attempted for known fog periods in the study area on the basis of available predicted meteorological data using GIS-based analysis. The results were then verified with the NOAA-AVHRR satellite images. Forecasting was performed and achieved a comparable match with the observed fog occurrence as depicted in satellite images. Using interpolated surfaces of the meteorological parameters by both Spline and Krigging interpolation techniques produces almost same results however; in some cases, Krigging has been proved to be better. There are several meteorological agencies, which provide the forecasted temperature, relative humidity and wind speed data that could be used for the forecasting in forward mode. In the present study, forecasted meteograms available at the link http://www.monsoondata.org/five days in advance were used for performing the fog formation modelling. Predicted fog for the dates 6 January 2006, 30 December 2006 and 4 January 2007 are shown in the Figs. 18, 19, 20. It is important to note that the predicted fog area always shows some mismatch with that of satellite image in the southern part of the fog area. The reason for this might be the difference of time of acquisition of the two data sets. There is a time difference of about 4–5 h between the meteorological data and the satellite data. Due to the dissipation of the fog during this period, we may get the mismatch. Sometimes radiation fog may take less than an hour to disappear. In that case, the degree of mismatch will be the maximum. Other than this, the quality of the meteorological data may also affect the output of the model. If the input data are not very accurate (being forecasted meteograms are in analogue form), then the prediction may also be erroneous. The density of the meteorological stations also has an important effect on the results. Since the calculation is based on different parametric surfaces derived through interpolation of meteorological data, the areas where the density of stations is low may also involve some amount of error. From the prediction for certain dates, it was observed that in spite of sustaining the favourable meteorological conditions, sometimes fog did not occur. It often happened when the area was covered by cloud on that particular day or on consecutive previous days. The presence of cloud absorbs both heat and aerosols, thereby reducing the chances of fog formation. During the fog period, it was seen that the meteorological parameters remain favourable for several days. While if within a non-foggy period, the same condition prevails for a single day fog may not develop. In this study, it was observed that if fog favourable conditions prevail for few days at a stretch, then only fog developed. So forecasting made on the basis of analysis of a single day’s data was found to be inaccurate. For better results in the forecasting, the factors like cloud cover and trends in the meteorological data in previous days should also be taken care of.

Fig. 18
figure 18

Prediction of Fog for 6 January 2006 based on the analysis of meteorological data in GIS using Spline and Krigging interpolation techniques

Full size image
Fig. 19
figure 19

Prediction of Fog for 30 December 2006 based on the analysis of meteorological data in GIS using Spline and Krigging interpolation techniques

Full size image
Fig. 20
figure 20

Prediction of Fog for 4 January 2007 based on the analysis of meteorological data in GIS

Full size image

One of the main limitations in the present study (while predicting the fog) is the forecasted meteorological data. Most of the agencies provide these data sets either in the form of meteograms or in the form of thematic maps. Reading the values of the above-mentioned parameters accurately from meteograms and thematic maps might involve undesirable errors.

4 Conclusions

The processes of fog formation and dissipation are controlled by a complex combination of meteorological conditions and no single parameter dominates. Air temperature, relative humidity, wind speed and elevation are the parameters having critical importance in these processes. In the present study, all fog occurrences over Northern India during 2002–08 were studied using NOAA-AVHRR satellite images. After mapping the fog areas for these years, an interesting decreasing trend in the affected area was found. Noteworthy decrease in fog occurrence in the year 2007–08 is found to be attributed to high temperature fluctuation and non-uniform wind speed as well as the humidity. However, high fluctuations in meteorological parameters may be one of the reasons for the same. Detailed studies are required to understand and explain such trends. The forecasting model developed on the basis of analysis of satellite data as well as meteorological data gives sufficiently promising results for further investigation and development. The predicted fog was found to be in close match while comparing with the fog occurrence as depicted on satellite images. Based on the above study, it is now possible, to some extent, to make fog forecasting during winter periods in the Indo-Gangetic Plains, if all required meteorological parameters are available in advance, and fog formation can be verified using satellite data analysis in near real time.