environmental regulations and time space …nkumar/manuscripts/nk_est_final...environmental...

Environmental Regulations and Time Space Dynamics of

AOD and LULC in Delhi

Journal: Environmental Science & Technology

Manuscript ID: Draft

Manuscript Type: Article

Date Submitted by the Author:

n/a

Complete List of Authors: Kumar, Naresh; University of Iowa, Geography Linderman, Marc; University of Iowa, Department of Geography Foster, Andrew; Brown University Chu, Allen; NASA, Climate and Radiation Branch Buda, Travers; University of iowa Tripathi, Sachi; Indian Institute of Technology Kanpur, Civil Engineering

ACS Paragon Plus Environment

Submitted to Environmental Science & Technology

1

Environmental Interventions and Time Space Dynamics of AOD and LULC in

Delhi

Naresh Kumar

Marc Linderman

Allen D. Chu

Travers Buda

Sachchidanand Tripathi

Andrew D. Foster

Corresponding author:

Naresh Kumar

312 Jessup Hall

Department of Geography

The University of Iowa

Iowa City, IA 52242

[email protected]

Page 1 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

2

ABSTRACT

Utilizing spatial-temporal autoregressive models this paper examines the distribution of aerosol optical depth (AOD), an indirect measure of air quality and radiative forcing, with respect to different land-use and land-cover (LULC) types and changes in and around Delhi triggered by radical environmental interventions, such as conversion of vehicles to compressed natural gas (CNG) and displacement of more than 25,000 polluting industries. The daily AOD were retrieved at 2km spatial resolution using satellite data (from 2000 to 2004), and collocated with the LULC and meteorological data with the optimal spatial-temporal windows. Our analysis suggests that central areas of Delhi recorded a decrease in AOD and areas bordering Delhi recorded a significant increase in AOD (> 0.2, about 30% increase) after the environmental interventions. Change in AOD was inversely associated with the increase in distance from the city center. LULC coupled with meteorological conditions explained ~ 90% of the total spatial variability in AOD, and changes in AOD were significantly associated LULC changes. These findings have significant implications for health effects and local climate changes as a consequence of changes in AOD and air quality triggered by the recent environmental interventions.

Page 2 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

3

INTRODUCTION

While the economies of the two Asian giants, India and China, are thriving,1 an increase in the burden of environmental contaminants and recent environmental interventions in these countries are drawing researchers’ attention. Beijing and Delhi recently adopted radical measures to improve air quality. While most interventions in Beijing were temporary to host the 2008 Olympics games, policy interventions in Delhi were permanent. Delhi became an exemplary city by enforcing the compressed natural gas (CNG) regulations and closure of polluting industries towards the beginning of the 21st century.2 At present about 100,000 CNG based public and private vehicles are registered in Delhi and more than 25,000 industries previously permitted in residential (or non-conforming areas) were displaced to the peripheral areas. This paper examines the effects of these environmental interventions in Delhi on time-space dynamics of aerosol optical depth (AOD) and its association with land use and land cover (LULC) changes in and around the city. This paper is likely to advance the literature in many different areas. First, it will augment our understanding of the effects of policy interventions on the redistribution of air pollution burden, especially when the interventions are not ubiquitous;3, 4 for example Delhi was affected by the interventions, but areas outside Delhi remained unaffected. This is likely to have serious implications for air pollution redistribution and its associated health effects. Second, unlike many previous studies, the findings of this research will provide insight into local and regional trends of LULC and their effect on time-space dynamics of AOD. Third, the findings of this research are likely to shed light on the potential local and regional climate changes with respect to changes in AOD caused by LULC changes. Time-space varying AOD, i.e. extinction of beam power resulting from the absorption and scattering of radiation in the presence of aerosols in a column, can be used to quantify and characterize radiative forcing,5 and hence its associated effects on local and regional climate change.6 Finally, in this paper we evaluate how LULC changes trigger changes in emission sources and intensification of urban land use.7, 8 The remainder of this paper provides details on the study area, multi-resolution and multi-sources of data, methods used to analyze these data, and results of the analysis.

STUDY AREA, DATA AND METHODOLOGY

Delhi Metropolitan – Delhi, the third largest city in India with a population of about 12 million people and a geographic area of 1500 km2, has been in news for adopting radical measures to curb air pollution and provides an excellent case study of policy effects on air quality in emerging economies (Fig S1). At the turn of the 21th century it was declared as the world’s most polluted city.2 The Supreme Court of India directed the Delhi Government to convert all commercial vehicles to CNG (during 2000 to 2002)9 and the closure of more than 25,000 polluting industries in non-conforming areas. Since these interventions the city and its neighboring areas have witnessed unprecedented LULC changes and redistribution of air pollution.9, 10 Therefore, the chosen study area is important for examining LULC changes and its effect, in turn, on the time-space

Page 3 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

4

dynamics of AOD and will shed light on the direct and indirect effects on environmental interventions on air pollution distribution/redistribution. Data – Satellite based AOD has been useful to study time-space dynamics of aerosols and to identify and characterize their sources.11, 12 The daily AOD at 2km spatial resolution were computed using the data from MODerate resolution Imaging Spectroradiometer (MODIS), onboard Terra and Aqua satellites.13 Hourly meteorological data, such as relative humidity, atmospheric pressure and temperature, were acquired from the National Climatic Data Center.14 For extracting LULC, cloud-free Landsat 5 TM data were obtained from the USGS Global Visualization Viewer for February and March of 1998 (path 40-41 and row 146-7) as well as the corresponding anniversary imagery from 2003.15 METHODS

AOD Extraction – The spectral channels used in retrieving AOD over land include 250m, 500m, and 1km bands. The 250m (0.66 and 0.86µm) bands are used to detect water bodies (such as lakes, rivers, etc.). Combinations of 500m (0.47µm) and 1km bands (1.38, 4.7, and 15µm) are used for clouds detection. The aggregated 0.66µm and 0.86µm channels together with other 500m channels (0.47, 0.55, 2.13µm) were used to derive aerosol optical depths. The methodology for retrieving the standard 10km AOD retrieval is detailed elsewhere.16 The same algorithm was used for retrieving 2km AOD except for the pixel requirement threshold. Since the number of pixels available in a 2 x 2km grid was only 16, we set the minimum requirement to 2 pixels. It is arguable that a different set of values could be used. The decision we made for 2km is more restrictive than that for the standard 10km products. Therefore, the quality of 2km AOD is likely to be as good as (or even better than) that of the 10km AOD products.

LULC Extraction – LULC for the years 1998 and 2003 was classified for New Delhi and its surrounding region encompassing an area of approximately 118,000 km2. The six visible, near and mid-infrared bands of the annual four images were mosaicked. The 1998 and 2003 mosaics were then stacked to produce a 12 band image. Vegetation in 2003 was thresholded based on the normalized difference vegetation index. To improve spectral separability, principal components analysis (PCA) was conducted for the vegetated and non-vegetated scenes separately17, 18 with the first five bands of each PCA providing the majority of the variance (~90%). These bands were clustered using a K-means algorithm in ENVI software initialized with 30 clusters.19 The resulting clusters were labeled as six land cover categories and the corresponding 98-03 change classes based on Indian Remote Sensing satellite IRS-1C panchromatic 5.8 meter resolution data (resampled to 6.25 meter) and Digital Globe high resolution imagery provided by Google Earth.20 Independent validation data were derived from the Digital Globe imagery as well as the IRS-1C data. The validation data set centroids were obtained for 50 meter radius circles containing purely one land cover class in the thematic classification as Google Earth imagery is estimated to have a horizontal positional RMS error of approximately 39.7m.21

Page 4 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

5

The accuracy of the classification was found to be 93% and 87% for the overall land cover classification and the land cover change classes, respectively (Table S5). Ancillary data consisting of polylines and polygons of roads and industrial parks were acquired from the Survey of India and combined with the thematic classification. Road polylines were encoded at a 30m resolution irrespective of thematic class. Industrial areas were overlaid on the thematic classification and recoded to a separate class where the polygons intersected the built thematic class. The resulting industrial class is comprised entirely of the intersection of these layers. Data integration – The data used in this research come from three different sources and were available at different spatial-temporal scales. For example, the spatial resolution of the daily AOD was 2km (i.e. multiple locations every day) and hourly meteorological data were available at several static stations. LULC data were extracted just for the springs of 1998 and 2003. These datasets were collocated with the AOD data using optimal spatial-temporal windows.22 Let Lik denote a 30m +pixel at ith location (represented by a pair of coordinates) with kth LULC type. Since AOD locations were distributed sporadically and did not correspond with the location and time of LULC data, AOD locations are distinguished from LULC locations; τatp denotes AOD from MODIS at locations a =1,…,A; days t = 1,…,T; and satellite overpass time (or hour of AOD) p=1,…,P. On a given day τat was observed at multiple locations (a) and there were many 30m LULC pixels, which remained the same for several years, around the ath site; likewise the overpass (or recording) time (p) of AOD does not correspond with the duration and time (h) of meteorological data (Mh) on the corresponding day. The proportion of area under kth LULC type around ath location (Lak) was computed by dividing the number of pixels under kth LULC type by the total number of pixels within 1km radius of ath site. Meteorological estimates at ath location and pth overpass time (Map) were derived by averaging the meteorological data at all stations within 90 minutes time interval of the pth overpass time of AOD data on tth day. The final disaggregated dataset consisted of 703,756 AOD values (from 2000 to 2004) collocated with % of area under different LULC and meteorological conditions. These data were grouped into two time periods – 2000-01 (pre-intervention) and 2003-04 (post-intervention). The year 2002 was skipped, because it was the year of transition when air quality interventions were implemented. Yearly data were further aggregated by 1km grid (~6,450 pixels) overlaid onto the study area (Fig S1). Analysis was performed on both disaggregated and aggregated datasets. While disaggregated dataset helped us evaluate the temporal trend of AOD, the aggregated data helped us understand spatial variability in AOD, largely associated with the anthropogenic sources of air pollution, in and around Delhi. The methodology used to analyze these data is detailed below. In the aggregated dataset, let τa be the average AOD corresponding to ath pixel, a = 1, …, N. Likewise, % area under each LULC type and meteorological conditions were averaged for ath pixel. Assuming averaging these for the entire duration accounts for the temporal

Page 5 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

6

noise, τa can be written as a function of LULC type (La) and average meteorological conditions (Ma) as τa = α + βL'a + φM'a + ua + εa (1) where β and φ are unknown regression coefficients, ua is the spatial random effect and can be modeled using conditionally autoregressive model as suggested by Besag 23

ua | uj ~ ∑∑ =

=

k

j

ajjk

j

aj

wu

w 1

1

ρ (2)

where ρ is the spatial autocorrelation parameter; waj = 1 if j is a neighbor, 0 otherwise. The decision about neighbors can be guided by a variogram which graphs the average semivariance on x-axis and distance interval on x-axis. The sill value that levels off when the spatial autocorrelation become insignificant can help us choose the distance range within which spatial autocorrelation exists. The spatial random effect has the following joint distribution u ~ Nn (0, σ2 (I – ρW)-1) (3) where Nn is the n-dimensional normal distribution, I is an n x n identity matrix and W is the weight (or adjacency) matrix, which can be replaced by inverse distance weighted matrix, in which jth neighbors close to ath location are given higher weight and vice-versa. In the disaggregated analysis (1) was extended to temporal domain, and τat observed at ath location and on tth day can be expressed as a function of LULC type (La) within one km radius around ath location and meteorological conditions (M'at) within 90 minutes time interval of pth overpass time on tth day

τat = α + βL'a + φM'at + uat + δat (4) where uat is spatial-temporal (non-separable) random effect, and can be defined by extending (2) to the temporal domain as

uat | uj|t±l| ~ ∑∑∑ =

±±

= =

±

k

j

ltjltjk

j

L

l

ltj

wu

w 1||||

1 0||

θ (5)

where θ is spatial-temporal autocorrelation parameter and wj|t±l| = 1 if it is (a neighbor) within D distance of ath location and L days of tth day, 0 otherwise. Since LULC types changes by geographic location (not by time) and meteorological conditions change by time (not by geographic location on a given day), the δat has three components

δat = υa + ωt + εat (6)

Page 6 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

7

υa is the time-specific random effect that accounts for intra-day meteorological conditions across all locations (ath); ωt is the random effect for constant LULC types that remain the same across time, and εatp ~ iid N(0, σ2).

RESULTS

AOD Distribution – The daily average AOD (for the entire duration from 2000 to 2004) in the study area was 0.669±0.003, and the value for Delhi was 0.785±0.001, significantly higher than that recorded outside Delhi (0.637±0.003) (Table 1; Fig S2a and S2b). From the descriptive analysis three important findings emerge – (a) the average AOD in the study area increased from 0.65 in 2000-01 to 0.68 in 2003-04 (~4.5% increase) despite the environmental interventions in place since 2002, (b) AOD concentrations gradually declines with the increase in distance from the city center, and (c) areas outside Delhi witnessed a significantly higher increase in AOD as compared to that inside the city (Fig 1; Table S2). For example, increase in AOD within 10km distance of the city center was less than 2.5%; increase in the areas more than 50km away from the city center was more than 6.5%. Unchanged LULC Area 1998 and 2003 – Despite the fact that Delhi metropolitan area has a population of more than 12 million, 66% of the study area was under vegetation canopy in both years 1998 and 2003 (Table 1). The percentage of vegetation canopy cover in Delhi was significantly smaller (~ 45.5%) as compared to areas outside Delhi (~72.8%). The second largest LULC category in Delhi was builtup areas that occupied ~29% of the total area, of which 12.8% was densely populated. Outside Delhi, however, 6.2% of the total area was occupied by builtup area, of which only 2.4% was under dense residential category. Bare ground was the second largest LULC category outside Delhi (Fig S3a & Fig S3b). Change in AOD and LULC from 1998-2003 – 7.3% of the total area underwent LULC changes from 1998 to 2003 (Fig 2). LULC changes from 1998 to 2003 were grouped into five categories, namely from bare to builtup, vegetation to bare ground, bare ground to vegetation, water to bare, bare to water. The largest LULC change across the study area was the conversion of vegetated areas to bare ground consisting of approximately 5% of the total area that underwent LULC changes. Shifts from bare ground to vegetation were approximately 1.5% across the study area with slightly higher rates outside of the city center. With 1.5% gain in vegetation canopy cover, the net loss of vegetation canopy cover was 3.6% in Delhi and 3.1% in areas outside Delhi. It is interesting to note that there was a significant increase in the builtup area in both Delhi and outside Delhi (~ 1% and 0.6%, respectively). Increase in builtup area and loss in vegetation canopy cover account for more than 80% of the total changes in LULC. The rates of LULC and AOD changes register significant spatial heterogeneity (Fig 2), for example southern and southeastern parts of the city recorded very high increase in both AOD and builtup areas. The central parts of the city, however, recorded a decline in the average AOD value. A visual analysis suggests that areas with a significant increase in AOD, especially southwestern parts (inside and outside Delhi), correspond closely

Page 7 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

8

with the areas of significant increase in builtup areas (Fig 2). It is interesting to note some central parts of the city registered a decline in the average AOD from 2000-01 to 2003-04, but the areas bordering Delhi (inside and outside) recorded a significant increase in AOD. As evident from Fig 2 and Table S2, increase in the builtup area was only 0.68% and the net increase in AOD values in southeastern parts (closer to Delhi border) was more than 0.2 (~25% increase). Linkages between AOD and LULC – The spatial resolution and temporal scale of LULC, AOD and meteorological data used in this research varied significantly. Therefore, these data were arranged and analyzed in both aggregated and disaggregated forms. In the remainder of this section, the results of both disaggregated and aggregated analyses are presented sequentially and then change in AOD is examined with respect to city context and change in LULC types. Disaggregated analysis – The results from linear and spatio-temporal autoregressive models are presented in Tables S3a and S3b. From this analysis three major findings emerge. First, individual LULC types do not account for a significant proportion of variability in the daily AOD. Given these eight different LULC types remained the same for the entire duration, differences in LULC can account for only spatial variability of AOD. Second, LULC associated with urbanization, such as industries, roads, builtup and residential areas, are positively associated with AOD. Vegetated areas and bare ground are inversely associated with AOD. Third, among the eight LULC types percent industrial and major road areas emerge as the most significant predictors of AOD.

As expected, among all meteorological conditions dew point (that indirectly indicates the presence of water vapors) contributed substantially more to explain AOD variance as compared to LULC variables alone. Individual LULC classes explained less than 5% of the AOD daily trends, while individual LULC type in association with dew point and other contextual information accounted for ~ 20% of the total variance in daily AOD levels. Dew point as well as city and year contexts were positively correlated with AOD and distance to the city center was significantly negatively correlated. Dew point’s marginal effects were consistent across all LULC classes while year and city explanatory variable effects varied with roads and vegetated areas.

Aggregated Analysis – Aggregated analysis is organized into two parts. In the first part, yearly average of AOD (aggregated by 1km spatial resolution) was examined with respect to individual LULC type with and without meteorological conditions and contextual factors (Table S4a and S4b). In the second part, AOD was examined with respect to five uncorrelated factors of different LULC types and changes (Table 2). These five factors together captured more than 97% of the total variability across all LULC types. (Table S5) The first factor alone, which registered high positive factor loadings (> 0.9) for residential, industrial, roads, builtup and residential LULC categories and moderately high negative loadings for area under vegetation, explained more than half of the total variability. This factor can be considered as an indicator of urban and industrial development; high scores on this factor indicate high population concentration and road density, high percentage of builtup and residential areas, and vice-versa. The second

Page 8 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

9

factor captured 20% of the total variability and represented bare ground and change from bare ground to builtup area. Factor 3, 4 and 5 together captured additional 23% of the variability in the LULC dataset and represent water, loss of area under vegetation canopy cover and light residential areas, respectively (Table S5). In the first aggregated analysis, industrial, road, bare ground, residential and builtup area were positively associated with AOD. The explanatory power of each LULC type was ~30% (Table S3a), which increased to more than 50% when dew point, distance city center, city dummy and year dummy were added in the model (Table S4b). Although the explanatory power of the model improved significantly once controlled for these covariates, the strength of regression coefficient of LULC type dropped significantly, except for bare ground. For vegetated areas, however, this association became positive.

In the second aggregated analysis, AOD was modeled with respect to five factors using ordinary least square regression (OLS) and spatial autoregressive models. From this analysis three important findings emerge. First, urbanness, loss of vegetation canopy cover and sub-urbanization are positively associated with AOD. Bare ground that lack anthropogenic emission sources is negatively associated with AOD. Second, the pooled analysis suggests that all LULC types in association with dew point and city-context explain ~90% of the total spatial variability in AOD. The regression coefficient for the city context (Delhi versus outside Delhi) is negative; i.e. when controlled for all LULC types and dew point, AOD in Delhi is relatively less as compared to areas outside Delhi. Third, LULC type and AOD have a strong spatial structure (in the aggregated dataset) and OLS under-predicts the influence of urbanness, loss of vegetation canopy cover and bare ground. The performance of conditional autoregressive (CAR) and spatial autoregressive model (with the use of geostatistical model to control for spatial structure in the residuals), measured by AIC value, is better than that of OLS. The next section in which AOD and LULC data were aggregated before and after the interventions will provide insight into how the changes in AOD from 2000-01 to 2003-04 are associated with LULC changes.

Change in AOD and LULC Changes – Our major focus here is to examine spatial variability of changes in AOD (from 2000-01 to 2003-04) with respect to four different classes of LULC changes (between 1998 and 2003), namely bare ground to builtup, vegetation to bare ground, bare to vegetation and net loss in vegetation canopy cover (Table 3). From this analysis three findings emerge. First, among these four categories change in bare to builtup area was the only significant predictor of increase in AOD. Second, when other variables, particularly change in dew point that indirectly account for change in AOD through increase/decrease in liquid aerosols, are included in the model all four categories of LULC changes show a significant association with the change in AOD. Increase in builtup area and loss of vegetation canopy cover were positively associated with the increase in AOD and vice-versa. As expected, areas that recorded increase in vegetation canopy cover showed a significant decline in AOD. Third, increase in the distance to Delhi border (outside Delhi) shows a statistically significant and positive increase in AOD. This suggests that migration of pollution industries and vehicles to the areas outside Delhi could be responsible for this increase in AOD through the increase in the emission from anthropogenic sources. Another interesting thing to observe from this

Page 9 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

10

analysis is that the increase in AOD is positively associated with the increase in distance from the city center. This analysis (and also the descriptive analysis presented above) suggests that environmental interventions have been more effective, at least, in the central parts of the city.

CONCLUSION AND DISCUSSION

Using high resolution satellite data and state of the art spatial-temporal analytical methods, this research examined AOD distribution in Delhi metropolitan across different LULC types before and after the environmental interventions. Four important findings emerge from this research. First, overall aerosol load in the study area was three/four times greater than that observed in comparable locations in the U.S.24 Second, the average AOD gradually declines with the increase in distance from the city center, but the rate of increase in AOD (from 2000-01 to 2003-04) increases with the increasing distance from the city center. The central parts of the city recorded a decline in the average AOD. But areas closer to southern parts of Delhi boarder recorded an average increase of more than 0.2 (~25% increase). Third, locations of significant increase in AOD correspond with the locations of increase in the builtup areas and decline in the vegetation canopy cover. A significant part of this increase in AOD from 2000-01 to 2003-04 and LULC changes in areas bordering Delhi can be attributed to environmental interventions that enforced the closure of polluting industries and conversion of public transport vehicles to CNG. Fourth, most LULC types and changes in LULC type observed a significant association with AOD. In the full model, all LULCs type coupled with meteorological conditions and other contextual variables captured ~ 90% of the total spatial variability of AOD. This suggests that any changes in LULC are likely to trigger long-term changes in AOD, and hence changes in air quality, radiative forcing and local climate. The findings reported above are likely to have major implications for the future research studies. First, high resolution satellite data are critically important for evaluating time-space dynamics of LULC and AOD. In situ monitoring is typically limited in extent and frequency, and does not capture distributed processes, particularly those related to the spatial-temporal changes in LULC, such as industrial displacement or development of industries, urban intensification, as well as year-to-year variations to long-term changes in vegetated land cover.25 These spatial temporal patterns and changes of land use, industrial activities and transportation can dictate time-space dynamics of AOD and air quality at local and regional scales. The high resolution MODIS AOD that have daily global coverage is critically important to examine spatially and temporally disaggregated effects of environmental interventions on time-space dynamics air quality and spatial and temporal patterns of emissions.12 Second, this study has important implications in terms of climate change and variation in health impacts (with the shifting burden of air pollution load in response to changes in emission sources) across space and time. The study covers the city of Delhi and also the suburban and rural areas outside the urban fringe, densely occupied by various types of vegetation. The AOD value being higher for the urban locations imply that the major

Page 10 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

11

source of aerosols over this place is anthropogenic. Anthropogenic aerosols are mostly small in size (in the submicrometer range) and strongly absorb long wave radiation. Aerosol absorption is one of the key properties for determining the effect of aerosols on the climate. Strong aerosol absorption can lead to a large difference in aerosol forcing at the top of the atmosphere and at the Earth's surface, which results in a significant increase in solar radiation absorption in the atmosphere, accordingly raising some intriguing scientific questions that are concerned with the global and regional circulation, temperature change and the hydrological cycle.26, 27 Therefore, LULC changes are critically important to trace the sources of aerosols and to further study the various properties of aerosols with respect to their sources. There are many uncertainties in quantifying the variation of aerosol properties on a global scale. Thus, the analysis of high resolution LULC and AOD is a step towards a better understanding of variation in aerosols at local and regional scales. Third, socio-physical and bio-chemical environment of the megacities of developing countries, especially of two Asian economic giants – India and China, are undergoing changes at a rapid rate.1 Economic growth, urbanization and industrialization and recent environmental interventions are the main drivers of these changes. The high resolution satellite data used and methodological framework presented in this paper could be useful to quantify and characterize these changes. Since the rate of LULC and AOD changes is very different, the methods presented in this research to collocate and analyze data that come from many different sources and available at very different spatial-temporal scales could be useful for future studies that aim to utilize multi-resolution spatial-temporal datasets. Based on the results reported in this paper we can conclude that LULC and AOD (in the study area that includes Delhi and its surrounding areas) have recorded significant changes after the environmental interventions. This finding, however, should be interpreted with caution, because there is not a direct way to attribute these changes in LULC and its associated effects on AOD to environmental interventions due to data limitations. But our analysis clearly does demonstrate that the average AOD declined in the central parts of the city and peripheral areas showed relatively high increases in the average AOD after the interventions. The rate of change in both LULC and AOD was significantly higher in areas bordering Delhi (inside and outside). With the control for meteorological conditions and spatial structure, LULC types together explained ~ 90% of the total spatial variability in AOD. Thus it is safe to assume that changes in AOD from 2000-01 and 2003-04 may be attributed to LULC changes, which are directly influenced by the displacement of polluting industries in the peripheral areas, urban intensification and unprecedented increase in the foreign direct investment (FDI). Our major focus in this paper has been on LULC and its associated effects on AOD. The future research should be geared towards teasing out the effects of environmental interventions from the effects of urban intensification and FDI on LULC and AOD changes. These two keys agents are important for the study areas because of a phenomenal increase in the FDI and its heterogeneous distribution in the study area. For example, Gurgoan located to the south of Delhi and in its very close proximity has been successful in attracting a vast majority of the total FDI.28 Therefore, it is important to develop an understanding of the

Page 11 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

12

direct and indirect effects of these two important agents (heterogeneous distribution of urban intensification and FDI, also linked with the globalization of economy) on LULC and AOD, which are likely to play an important role in the regional climate change.

ACKNOWLEDGMENTS

We gratefully acknowledge data provided by the National Aeronautic and Space

Administration, the National Climatic Data Center and the United States Geological

Survey. Funding and support for this work was provided by NICHD/NIH (R21

HD046571-01A1) and NIH/NIEHS (1 R21 ES014004-01A2).

Page 12 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

13

References

1. Schuman, M., India vs. China: Whose Economy Is Better? TIME Magazine Thursday, Jan. 28, 2010 2010.

2. Bell, R. G.; Mathur, K.; Narain, U.; Simpson, D., Clearing the Air: How Delhi Broke the Logjam on Air Quality Reforms. Environment Magazine 2004, 46, (3), 22-39.

3. Jerrett, M., Global Geographies of Injustice in Traffic-Related Air Pollution Exposure. Epidemiology 2009, 20, (2), 231-233 10.1097/EDE.0b013e31819776a1.

4. Ranft, U.; Sugiri, D.; Gladtke, D.; Eberwein, G.; KrÃ¤mer, U.; Wilhelm, M., Temporal-Spatial Trends of Health Indicators in Children in a Highly-Exposed Industrial City in Germany. Epidemiology 2007, 18, (5), S25 10.1097/01.ede.0000276490.76293.57.

5. Hansen, J.; Sato, M.; Ruedy, R., Radiative forcing and climate response. Journal of

Geophysical Research-Atmospheres 1997, 102, (D6), 6831-6864. 6. Yu, H.; Kaufman, Y. J.; Chin, M.; Feingold, G.; Remer, L. A.; Anderson, T. L.;

Balkanski, Y.; Bellouin, N.; Boucher, O.; Christopher, S.; DeCola, P.; Kahn, R.; Koch, D.; Loeb, N.; Reddy, M. S.; Schulz, M.; Takemura, T.; Zhou, M., A review of measurement-based assessments of the aerosol direct radiative effect and forcing. Atmospheric Chemistry and Physics 2006, 6, 613-666.

7. Hixson, M.; Mahmud, A.; Hu, J. L.; Bai, S.; Niemeier, D. A.; Handy, S. L.; Gao, S. Y.; Lund, J. R.; Sullivan, D. C.; Kleeman, M. J., Influence of regional development policies and clean technology adoption on future air pollution exposure. Atmospheric

Environment 2010, 44, (4), 552-562. 8. Weng, Q. H.; Yang, S. H., Urban air pollution patterns, land use, and thermal landscape:

An examination of the linkage using GIS. Environmental Monitoring and Assessment

2006, 117, (1-3), 463-489. 9. Narain, U., Policy Monitor: Urban Air Pollution in India. Rev Environ Econ Policy 2008,

2, (2), 276-291. 10. Kumar, N.; Foster, D. A., Have CNG regulations in Delhi done their job? Economic and

Political Weekly 2007, 42, (51), 48-58. 11. Gupta, P.; Christopher, S. A., Particulate matter air quality assessment using integrated

surface, satellite, and meteorological products: 2. A neural network approach. Journal of

Geophysical Research-Atmospheres 2009, 114, -. 12. Kumar, N.; Chu, A.; Foster, A., Remote sensing of ambient particles in Delhi and its

environs: estimation and validation. International Journal of Remote Sensing 2008, 29, (12), 3383-3405.

13. NASA Level 1 and Atmospheric Archive and Distribution System. http://ladsweb.nascom.nasa.gov/

14. NCDC NOAA National Operational Model Archive & Distribution System http://nomads.ncdc.noaa.gov/

15. USGS USGS Global Visualization Viewer http://glovis.usgs.gov/ 16. Remer, L. A.; Tanré, D.; Kaufman, Y. J., Algorithm for remote sensing of tropospheric

Aerosol from MODIS. In NASA, Ed. Greenbelt, MD: Goddard Space Flight Center, NASA.: 2006.

17. Deng, J. S.; Wang, K.; Deng, Y. H.; Qi, G. J., PCA-based land-use change detection and analysis using multitemporal and multisensor satellite data. Int. J. Remote Sens. 2008, 29, (16), 4823-4838.

18. Henry, M. C., Comparison of single- and multi-date landsat data for mapping wildfire scars in Ocala National Forest, Florida. Photogrammetric Engineering and Remote

Sensing 2008, 74, (7), 881-891. 19. ITT Visual Information Solutions ENVI.

Page 13 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

http://ladsweb.nascom.nasa.gov/

http://nomads.ncdc.noaa.gov/

http://glovis.usgs.gov/

14

20. Monkkonen, P., Using Online Satellite Imagery as a Research Tool Mapping Changing Patterns of Urbanization in Mexico. Journal of Planning Education and Research 2008, 28, (2), 225-236.

21. Potere, D., Horizontal Positional Accuracy of Google Earth's High-Resolution Imagery Archive. Sensors 2008, 8, (12), 7973-7981.

22. Kumar, N., What Can Affect AOD–PM2.5 Association? Environmental Health

Perspectives 2010, 118, (3), A2-3. 23. Besag, J., Spatial Interaction and Statistical-Analysis of Lattice Systems. Journal of the

Royal Statistical Society Series B-Methodological 1974, 36, (2), 192-236. 24. Chu, A.; Kumar, N. Validation and Spatial-Temporal Analysis of Multi-Resolution

MODIS AOD. http://jh302-nk-01.iowa.uiowa.edu/papers/NK_Chu_MRes_MODIS_AOD.pdf

25. Liu, Y.; Paciorek, C. J.; Koutrakis, P., Estimating Regional Spatial and Temporal Variability of PM<sub>2.5</sub> Concentrations Using Satellite Data, Meteorology, and Land Use Information. Environ Health Perspect 2009, 117, (6).

26. Ramanathan, V.; Crutzen, P. J.; Lelieveld, J.; Mitra, A. P.; Althausen, D.; Anderson, J.; Andreae, M. O.; Cantrell, W.; Cass, G. R.; Chung, C. E.; Clarke, A. D.; Coakley, J. A.; Collins, W. D.; Conant, W. C.; Dulac, F.; Heintzenberg, J.; Heymsfield, A. J.; Holben, B.; Howell, S.; Hudson, J.; Jayaraman, A.; Kiehl, J. T.; Krishnamurti, T. N.; Lubin, D.; McFarquhar, G.; Novakov, T.; Ogren, J. A.; Podgorny, I. A.; Prather, K.; Priestley, K.; Prospero, J. M.; Quinn, P. K.; Rajeev, K.; Rasch, P.; Rupert, S.; Sadourny, R.; Satheesh, S. K.; Shaw, G. E.; Sheridan, P.; Valero, F. P. J., Indian Ocean Experiment: An integrated analysis of the climate forcing and effects of the great Indo-Asian haze. Journal of Geophysical Research-Atmospheres 2001, 106, (D22), 28371-28398.

27. Satheesh, S. K.; Ramanathan, V., Large differences in tropical aerosol forcing at the top of the atmosphere and Earth's surface. Nature 2000, 405, (6782), 60-63.

28. Government of India FDI in India: Statistics. http://dipp.nic.in/fdi_statistics/india_fdi_index.htm (Feb 1, 2009),

Page 14 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

http://jh302-nk-01.iowa.uiowa.edu/papers/NK_Chu_MRes_MODIS_AOD.pdf

http://jh302-nk-01.iowa.uiowa.edu/papers/NK_Chu_MRes_MODIS_AOD.pdf

http://dipp.nic.in/fdi_statistics/india_fdi_index.htm

15

Table 1: AOD and LULC Summary Statistics

Delhi Outside Delhi Both

Aerosol optical depth (AOD) 0.785±0.001 0.637±0.003 0.669±0.003

Dew point (degree C) 14.593±0.007 13.912±0.033 14.057±0.030

Change in AOD (2003-04 – 2000-01) 0.036±0.005 0.028±0.002 0.029±0.000

% Change in AOD (2003-04 – 2000-01) 4.817 4.431 4.46

Unchanged LULC Type (% area under)

Industrial clusters 0.366±0.001 0.095±0.005 0.153±0.006

Major roads 0.577±0.002 0.069±0.004 0.178±0.008

Minor roads 1.543±0.004 0.511±0.016 0.731±0.019

Bare ground 15.252±0.045 12.468±0.238 13.062±0.212

Vegetation 45.469±0.090 72.758±0.365 66.936±0.440

Water 1.028±0.003 0.508±0.011 0.619±0.013

Residential 12.877±0.039 2.356±0.096 4.601±0.154

Light residential 7.936±0.011 2.704±0.057 3.820±0.073

Built-up 7.548±0.028 1.205±0.042 2.558±0.093

Total - Unchanged LULC type 92.596 92.676 92.659

LULC Change from 1998 to 2003

Changed from bare to built-up 0.990±0.003 0.597±0.012 0.681±0.012

Changed from vegetation to bare ground 4.885±0.006 4.769±0.042 4.794±0.036

Changed from bare ground to vegetation 1.214±0.003 1.740±0.014 1.627±0.013

Changed from water to bare 0.176±0.001 0.102±0.003 0.118±0.003

Net loss of vegetation canopy cover 3.674±0.005 3.107±0.004 3.214±0.004

Changed from bare to water 0.139±0.000 0.116±0.003 0.121±0.002

Total – LULC Change from 1998-2003 7.404 7.324 7.341

GRAND TOTAL 100 100 100

Number of observations 1,376 5,074 6,450

Page 15 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

16

Table 2: Average AOD (from 2000 to 2004) and LULC factors of Urbanness, Bare ground, Water, Deforested, and

Suburban LULC, Dew point, City Context.

Spatial autoregressive models Variables OLS

IDW Kriging CAR

0.0569*** 0.0644*** 0.0642*** 0.067*** Urbanness

-0.0012 -0.00096 -0.0009 0.002

-0.0113*** -0.00864*** -0.0104*** -0.0105*** Bare ground

-0.00075 -0.00059 -0.00054 0.0015

0.00571*** 0.00874*** 0.00901*** 0.0146*** Water

-0.00071 -0.00059 -0.00059 0.0016

0.00628*** 0.0106*** 0.0112*** 0.0171*** Deforested

-0.00081 -0.00065 -0.00063 0.00155

0.0393*** 0.0340*** 0.0390*** 0.0363** Suburban

-0.00087 -0.0007 -0.00067 0.0016

0.0468*** 0.0330*** 0.0369*** 0.0277*** Dew point (°C)

-0.00084 -0.00073 -0.00064 0.0007

-0.00589** -0.00536** -0.0093*** -0.0025*** City Context (1=Delhi, 0 otherwise) -0.00283 -0.00217 -0.00205 0.0037

1.798*** 0.803*** 0.904*** Autoregressive term

-0.027 -0.012 NA

Constant 0.0123 0.207*** 0.152*** 0.2798

-0.0119 -0.0102 -0.00908 0.0106

Observations 6450 6450 6450 6450

R-squared 0.789 0.874 0.878 NA

AIC -19159 -22204 -22427 -22611

Robust standard errors in the second row of each variable; *** p<0.01, ** p<0.05, * p<0.1

Page 16 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

17

Table 3: Change in AOD with respect to change in LULC Only LULC change and AOD LULC change and other variables

Variables bare to built

vegetation to bare

bare to vegetation

net loss in vegetation

bare to built

vegetation to bare

bare to vegetation

net loss in vegetation

0.0263** 0.0439 0.0015 -0.0003 0.0293** 0.0019** -0.006** 0.00316** Change in LULC 0.0018 0.0038 0.0020 0.0007 0.0039 0.0007 0.0021 0.0007

0.0217** 0.0226** 0.0227** 0.0226** Change in dew point (°C) 0.0006** 0.0006 0.0006 0.0006

0.0950 0.0822** 0.0841** 0.0671** Distance to border (km) 0.0148 0.0168 0.0163 0.0017

-0.0525 -0.0927 -0.088 -0.069** Distance to City Center (degree) 0.0131 0.0151 0.0150 0.0168

0.025** 0.0439** 0.0406** 0.0441** 0.0283** 0.053** 0.0718** 0.0468* Intercept

0.0022 0.0038 0.0035 0.002 0.0039 0.006 0.0034 0.0058

Auto-regressive term

.148** 0.03 0.031 0.031 .378** 0.239** .2438** 0.247**

AIC -12784 -12649 -12650 -12649 -14000 -13824 -13828 -13833

Standard errors in the second row, *** p<0.01, **p<0.05

Page 17 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

18

.4.5

.6.7

.8.9

AOD

0 20 40 60distance to city center (km)

2000-01 2003-04

Fig 1: Change in AOD with respect to distance to the city center (Connaught Place, i.e. central business district of Delhi)

Page 18 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

19

Fig 2: Change in AOD from 2000-01 to 2003-04

Page 19 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

environmental regulations and time space …nkumar/manuscripts/nk_est_final...environmental...

Documents