Monitoring Built-up areas using DMSP-OLS nighttime lights data: A - - PowerPoint PPT Presentation

Monitoring Built-up areas using DMSP-OLS nighttime lights data: A study from Indo Gangetic Plain Region lights data: A study from Indo-Gangetic Plain Region, India

Pranab Kanti Roy Chowdhury

S i ti t Scientist Government of India

DMSP Nighttime Lights of India Workshop

APAN 32nd New Delhi Meeting

26th August, 2011, New Delhi g , ,

Need of monitoring built-up areas

In developing countries like India, the urban built-up areas are expanding in an unplanned manner on their peripheries leading to an irreversible transformation of contiguous agricultural and forest lands into built-up areas. g p This affects hydrological and ecological cycles at regional and global scale and has become a matter of concern for climate change, natural resource utilization, biological and become a matter of concern for climate change, natural resource utilization, biological and ecological sustainability etc. Thus monitoring the growth of built up areas at regional and global scale has become an Thus, monitoring the growth of built-up areas at regional and global scale has become an urgent task for taking preventive measures in order to reduce the negative effects associated with these rapid expansion.

Monitoring urban areas at regional scale using Remote Sensing data

Remote Sensing techniques provide an ideal means of monitoring built-up areas. But high and medium resolution datasets are often less popular for regional level studies and medium resolution datasets are often less popular for regional level studies. Firstly, using high and medium resolution satellite datasets for regional level studies i l hi h t f i i th d t t ti d l b i d f i d involves high cost for acquiring the datasets, time and labour required for processing and interpreting images, huge database size and these act as prohibitive factors. Secondly, as the swath of the high and medium resolution images are not wide, frequent cloud conditions make it difficult to collect a large number of good quality datasets within a specific time frame for the entire study area. p y These problems may be overcome using coarse resolution datasets, having wide swath, less data volume and no or very less cost. less data volume and no or very less cost.

Monitoring urban areas at regional scale using Remote Sensing data

However, the main problem in monitoring built-up areas using coarse resolution datasets is that of mixed pixels which makes the estimation of built up areas less accurate that of mixed pixels which makes the estimation of built up areas less accurate. Techniques such as spectral mixture analysis, artificial neural network, support vector hi t b d f thi b t th it l d h th i machine etc may be used for this purpose but they are quite complex and have their own constraints. Thus, there is a need to develop new methods which will help in monitoring the growth of built-up areas within limited time along with minimal labour and cost. In this study an attempt has been made to implement a new method for monitoring built-up areas using DMSP-OLS datasets. Indo-Gangetic Plain region of India has been chosen as the study area for this purpose. the study area for this purpose.

Defense Meteorological Satellite Program

Defense Meteorological Satellite Program run by the United States Air Force operates Defense Meteorological Satellite Program, run by the United States Air Force, operates mainly for meteorological, oceanographic and solar-physics environments monitoring since 1970s.

Operational Linescan System

OLS, onboard the DMSP satellites is an oscillating scan radiometer capable of detecting even very faint VNIR emission using a Photo Multiplier Tube (PMT) in two resolutions of 0 55 km and 2 7 km 0.55 km and 2.7 km. This unique capability results in detection of lights from human settlements and ephemeral events on cloud less nights under low lunar illuminance, which has been used in urban g areas and population related studies by researchers globally.

DMSP-OLS nighttime lights data

• NOAA NGDC has developed yearly composites of DMSP OLS datasets captured in
• NOAA-NGDC has developed yearly composites of DMSP-OLS datasets captured in

cloudless nights under very low or no lunar illuminance.

• Ephemeral events have been removed from datasets using local threshold levels leaving

Ephemeral events have been removed from datasets using local threshold levels, leaving lights from only from urban areas, which may be used as a surrogate of urban areas.

• The digital datasets, having 6 bit radiometric resolution (1992 onwards), are available at

g g ( ) 1km spatial resolution and may freely be downloaded from the NOAA-NGDC website http://www.ngdc.noaa.gov/dmsp/global_composites_v4.html In this study the DMSP-OLS nighttime stable lights datasets have been used in monitoring y g g g the built-up areas and their growth.

Indo-Gangetic Plain, India

• The study area comprises of 9 Indian states and Union Territories and accounts for about
• The study area comprises of 9 Indian states and Union Territories and accounts for about

19.87% (653,211 sq. km.) of the total land area of India (Census of India, 2001) Th i b t d ith t i f i t t d ll i i h il

• The area is bestowed with extensive expanse of uninterrupted alluvium, rich soil,

sufficient ground water sources and mild climate, resulting in it being one of the world's most intensely cultivated and populated regions.

• The area is inhabited by about 41.32% of total Indian population (Census of India, 2001).
• In order to accommodate this huge and ever increasing population, the urban areas are

undergoing a rapid expansion and as a consequence most of the rich agricultural land is being converted into built-up areas. being converted into built up areas.

• This not only affects the agricultural productivity but also affects the ecological and

hydrological cycles and sustainability hydrological cycles and sustainability.

Map not to scale

DMSP/OLS nighttime lights data in built-up area monitoring

DMSP OLS nighttime lights datasets have been extensively used for urban areas and DMSP-OLS nighttime lights datasets have been extensively used for urban areas and population related studies, however direct identification of urban areas made by OLS is not free from errors

• Low radiometric resolution of 6 bits results in data saturation over brightly lit built-up areas.

Due to this, the pixels having significantly different fraction of built-up areas may have similar DN values in DMSP-OLS datasets. similar DN values in DMSP OLS datasets.

• Data saturation also results in over-estimation of urban areas. To address this many

researchers have used value thresholding in such datasets. But using a single threshold in a large region may result in smaller urban areas being lost in the process.

• Factors along with the impacts of backgrounds such as sporadic ephemeral lights, glint of

light into adjacent water bodies etc lead to uncertainty in motoring built-up areas light into adjacent water bodies etc. lead to uncertainty in motoring built-up areas. In this study, an effort has been made to integrate data from multiple sensors for accurately mapping the growth of urban areas (in terms of built-up area) on a per pixel basis. pp g g ( p ) p p

DMSP-OLS and NDVI datasets

Vegetation indices like NDVI are negatively correlated with impervious surfaces and may Vegetation indices like NDVI are negatively correlated with impervious surfaces and may be used for estimation of built-up areas as demonstrated by past researchers. OLS d NDVI d t t l t i th th t hi h f ti f b ilt OLS and NDVI datasets are complementary in the sense that higher fraction of built-up areas in a pixel will be associated with less vegetation cover, resulting in higher DN values in OLS data and lower DN value in corresponding NDVI data and the vice versa. Combined use of both datasets may provide new insights in estimating the fraction of built- up areas on a per pixel basis through bringing out more information in the resultant p p p g g g datasets than contained by the constituent datasets individually. This approach also helps in minimizing the errors resulting from external sources in DMSP- This approach also helps in minimizing the errors resulting from external sources in DMSP OLS data.

Human Settlement Index

Th t ti l ti hi b t th NDVI d th b ilt f h b The strong negative relationship between the NDVI and the built-up surface has been utilized by Lu et al, 2008 for developing Human Settlement Index (HSI) that represents fraction of built-up area of a per pixel basis. A higher value in the HSI is related to higher proportion of built-up area in a pixel. (1-NDVImax) + OLSnor (

max) nor

Human Settlement Index = (1-OLSnor) + NDVImax + (OLSnor * NDVImax) (After Lu et al., 2008) Where, OLS = (OLS - OLS ) / (OLS

• OLS

) OLSnor = (OLS - OLSmin) / (OLSmax - OLSmin) and NDVI = MAX (NDVI NDVI NDVI NDVI ) NDVImax = MAX (NDVI1, NDVI2, NDVI3,..., NDVI12)

Human Settlement Index

OLS d t t (DN l 63) li d t fit th d t ithi th OLS datasets (DN value range 0 – 63) are normalized so as to fit the data range within the theoretical range of NDVI datastes (-1 to +1). OLSmin is the minimum value in the DMSP-OLS datasets OLSmax is the maximum value in the DMSP-OLS datasets Yearly NDVImax are calculated in order to better separate built-up areas from bare soils, remove cloud cover and account for seasonal variation of vegetation cover. NDVImax datasetshas been calculated using all corresponding datasets available in a year datasetshas been calculated using all corresponding datasets available in a year

H S ttl t I d 2001 Human Settlement Index, 2001

HSI V l

9.01475

HSI Values

0.00001

Landsat ETM+ FCC, 2001 (4,3,2)

Human Settlement Index

• HSI datasets show proportion of built up area in each cell

HSI datasets show proportion of built up area in each cell.

• Higher proportion of built up in a cell results in higher OLS and lower NDVI value

resulting in higher index value and the vice versa.

• Direct extraction of absolute urban pixels at coarse spatial resolution will be erroneous

because of the mixed pixel problems. Through using HSI such errors have been avoided.

• Two HSI datasets were prepared for years 2001 and 2007 using DMSP-OLS nighttime

stable lights and MODIS NDVI (MOD13A3) datasets of corresponding years.

• Both datasets are available at 1km Spatial resolution and may freely be downloaded from
• Both datasets are available at 1km. Spatial resolution and may freely be downloaded from

their respective websites.

Monitoring built-up areas using HSI

A l t i ti f 500 i t d l l t d f th HSI l f

• A sample set consisting of 500 points was randomly selected from the HSI layer of year

2001.The values at the corresponding locations in the OLS data and NDVImax image of year 2001 were also extracted. Similar analysis was done for 2007 data.

• Regression 1: HSI values and the values at the corresponding locations in the OLS data
• Regression 2: HSI value and the values at the corresponding locations in the NDVI max

data

• For year 2001 the values for regression 1 and 2 were 0.77 and -0.79 respectively
• For year 2007 the values for regression 1 and 2 was 0.6 and -0.89 respectively

These results thus validated the assumptions that the HSI values were positively related to p p y the OLS values and negatively related to the NDVI values. Hence, an increase in OLS value will increase the HSI values, while an increase in NDVI value will decrease the HSI value

Monitoring built-up areas using HSI

• A map of HSI value change was generated for 2007-2001 period

A map of HSI value change was generated for 2007 2001 period

• The areas of HSI change were divided into the following three classes:

1 Class 1: which includes cells with Low change in HSI value 1. Class 1: which includes cells with Low change in HSI value 2. Class 2: which includes cells with Medium change in HSI value 3. Class 3: which includes cells with High change in HSI value From all the change areas, the following analysis was done

• 46% of the growth cells belonged to Class 1
• 48% of the growth cells belonged to Class 2
• 48% of the growth cells belonged to Class 2
• 6% of the growth cells belonged to Class 3

Monitoring built-up areas using HSI

Delhi Lucknow Kolkata Kanpur Kanpur

Monitoring built-up areas using HSI

An analysis of the spatial distribution of the three classes revealed, that the Class 1 comprised mainly of the existing built-up areas, while the Class 2 and Class 3 were located in the city’s fringe areas. y g These results on further analysis indicate:

• These results indicate that the built-up areas in the study area are experiencing an
• utwards spatial growth as the Class 2 and Class 3 cells are located in the fringe areas
• While a densification of the existing built-up areas is also taking place at a slow rate as

While a densification of the existing built up areas is also taking place at a slow rate as the growth cells with a Class 1 value are located in the existing built-up areas

State wise analysis of HSI

State wise analysis of HSI

A statewise analysis of growth cells belonging to three classes reveal that A statewise analysis of growth cells belonging to three classes reveal that

• For all states but Delhi and West Bengal, the growth cells mainly belong to class 2.

Chandigarh having the maximum while Bihar has least number of class 2 cells

• Delhi and West Bengal have maximum number of Class 1 cells while Chandigarh has the

least

• Bihar and Punjab have maximum number of class 3 cells while Punjab has the least

j j number. These results indicate that: These results indicate that:

• State of Delhi and West Bengal are experiencing predominantly densification of existing

urban areas

• States like Uttarakhand, Jharkhand and Chandigarh are experiencing outward urban

sprawl into surrounding contiguous land.

Conclusion

• This study demonstrates a simple cost and time efficient method for monitoring built-up

This study demonstrates a simple, cost and time efficient method for monitoring built up areas at a regional level in Indian context.

• Usefulness of DMSP-OLS datasets in Indian context may be observed from this study.
• Through this study, the efficiency of free, downloadable datasets has been established.
• Internet networks have been revolutionizing scientific researches around the world and

has also enables this study.

• The method implemented here may also be implemented in other areas of world for

mapping built up areas mapping built-up areas.

• An urgent requirement of a new sensor capable of producing nighttime lights at better

spatial and radiometric resolution was strongly felt during the course of this study. Such a p g y g y sensor may advance the application of DMSP-OLS data in even smaller urban areas.