Temporal changes in land cover types and the incidence of malaria in Mangalore, India

BACKGROUND
Malaria contributes to 881000 deaths worldwide annually and India is a major contributor in the region. This study aimed at detecting land cover changes and assesses their relationship with the burden of malaria in Mangalore taluk of southern India.


METHODOLOGY
Landsat TM images were obtained from the U.S. Geological Survey data repository. The statistics for the malaria incidences in the region were obtained from the National Vector Borne Diseases Control Program division of the State of Karnataka. The images were preprocessed, classified and change detection statistics were employed for major land cover types.


RESULTS AND CONCLUSIONS
An increase in the urban land cover by 20% with a reduction in the mountainous terrain by 34.7% and vegetation by 38.7% was noted between the years 2003 and 2005. The annual incidence of malaria increased five-fold from 203 to 1035/100000 population during the period. This study demonstrates the application of publicly available remote sensed data as a cost effective approach to study the agent, host and environment relationships in resource scarce settings which would provide valuable information planning and policy making at regional levels.


Introduction
Malaria is a complex and finely evolved vector-borne infectious human disease. In spite of global initiatives from various angles to curb this disease, an estimated 247 million people are infected and about 881,000 deaths occur annually 1 . Poorer tropical and subtropical regions of the world in Africa, South Asia, parts of Central and South America, the Caribbean, Southeast Asia, the Middle East, and Oceania are the regions with hi gh malaria transmission 2 .India greatly contributes to the global malaria incidence. The plethora of various species of malaria vec tors, a hospitable environment for growth and proliferation of the parasites and vectors and a malaria-susceptible human lineage have established India as a hot-spot for malaria infection 3,4 and India alone continues to contribute about 60% of Southeast Asian malaria incidences. The launch of the National Malaria Control Program (NMCP) resulted in a drastic drop of malaria cases from 75 million cases annually in 1947 to less than 50,000 in mid-sixties. However, after its near eradication malaria staged a dramatic comeback with nearly 2 million recorded cases every year since 1995 5 and a clear change in malaria epidemiology: an increase in the insecticide resistance, in proportions of malaria due to P. falciparum throughout the country, and rise in urban malaria [6][7][8][9] .
While the role of environmental factors, such as temperature, rainfall, relative humidity in the epidemiology of vector -borne diseases is well known, urbanization is emerging as an important factor associated with an increase in malaria [10][11][12] . A study done in Brazil showed that for the years 1997 and 2006, a 4.3% increase in deforestation per year was associated with a 48% increase in the risk in malaria after adjusting for the area and access to health care in the districts 13 .
While the environmental requirements for the maintenance of lifecycle vary between different vectors 14 , all vectors, including female Anopheles mosquitoes need water to complete their life cycle. Studies on the characteristics of the preferred habitats have shown that deforestation and land clearing contribute to the dynamic malaria patterns in a region. The mosquito biting rates and larval counts have been shown to increa se with the degree of deforestation and demonstrate seasonality with significantly lower bite rates during rainy reasons which could be a result of the a wash-out effect of breeding sites and reduced larval survival with changes in the ambient temperature 12 .
Over the last few decades, newer technologies have been developed which can better describe and characte rize changes in land cover types. The advent of remote sensing (RS) technology since 1970, a tool for the surveillance of habitats, densities of vector species and even prediction of the incidence of diseases, has opened up new vistas in the epidemiology of malaria and other vector-borne diseases. Number of studies has demonstrated successful utilization of RS data for better understanding the relationship between environmental factors and malaria transmission on the local and global scales. It has been shown that changes in the incidence of malaria transmission in Thailand are associated with changes in vegetation covers detected by Landsat satellite imagery using the Normalized Difference Vegetation Index (NDVI) 15 . In the study conducted in coastal Kenya changes in mosquito larva habitats associated with wetland areas and flooded vegetation were detected withRADARSAT-1 wetland mapping to help in assessing potential risk of vector-borne disease environment 16 .Using GIS technology to detect effects of landscape structure on Anopheles mosquito population it has been shown that in Northern Thailand Anopheles diversity was negatively correlated with landscape diversity, percentage cover of agriculture and fruit orchards and positive ly correlated with forest cover 17 . A study looking at the incidence of malaria in different regions of Jalpaiguri district in West Bengal, India has shown the malaria related morbidity and mortality to be highest in areas with higher tea gardens 18 .A feasibility study using multi-date IRS 1A and 1B satellite data was undertaken in collaboration with the Indian Space Research Organization. It was found that false color composite images can help in the dev elopment of base maps of the study area and macro stratification of mosquito conditions was possible. Correlation of changes in the area of land use features namely water bodies and vegetation with mosquito density was found to be significant in few of the six selected sites in and around Delhi. The limita tion of satellite resolution (36.5 meters) was felt as the smaller habitats of Anopheles mosquitoes were not detectable 19 . www.ssjournals.com The objective of this study was to examine the changes in land use and land cover types, which favor mosquito breeding, and t he trend in the incidence of malaria in Mangalore taluk of the State of Karnataka in South India using remote sensing imagery (with the 30m resolution) for the years 2000 and 2003. The selected study region has one of the highest malaria incidence in India and the incidence has been increasing 5 . The availability of favorable natural habitats and increasing artificial water collections due to construction related activities has rendered this region more sus ceptible to malaria and has been projected that the urban sprawl in this coastal belt of the state would contribute to 30% of the total available area by the year 2050 5,20 .

Study area
Mangalore taluk, on the western coast of Karnataka in South India spans an area of around 854 km 2 , situated between 13 0 8'0.11"N and 76 0 46'24.23"E coordinates ( Figure 1). This geographic region bordered by the Arabian Sea on its West, by the thick forested area on its east is rich with natural resources. This coastal region has experienced a dramatic increase in the extent of urbanization over the last three decades: between 1972 and 1999 the amount of developed land increased by 146%. During this period, the population in the region grew by about 54%. As per the 2001 census, the total population of the taluk was 882,856 with 68.08 % living in urban areas; males accounting for 49.24% and females accounting for 50.76% of the total population.. The population density of the taluk is 1048 persons/km 2 , which was 2.5 times higher than the district average of 416 persons/km 220,21 . Figure 1.Location of Mangalore taluk in South India.

Health Data
The annual counts for the malaria cases in the region for the period 1990 to 2007 were obtained from the National Vector Borne Diseases Control Program (NVBDCP) division of the State Public Health Department of Karnataka 5 . The population in the study area was computed based on the annual growth rate and the State level projections provided by the Census bureau, Ministry of home Affairs, Government of India 21,22 . The population densities over time were calculated using the projected population and the overall area of the region as estimated by the RS images. Annual malaria incidence rates (per 100,000 population) were calculated for the region. The trend in rates was assessed using a linear regression model.

Remote Sensing and GIS Data
Remote sensing (RS) data for the region were obtained as Landsat TM images for the years 2000 (dated 14th April and 20th December 2000) and 2003 (dated 23rd April and 27th January 2003) from the U.S. Geological Survey data repository. The Geographic Information System (GIS) shape files for India and the study region were obtained from the Tufts University GIS repository. The GIS vector files were processed and mapped using ArcGIS 9.1 (Environmental Systems Research Institute Inc., Redlands, CA, USA).
The RS data was processed and analyzed using the software ENVI 4.6 (ITT Visual Information Solutions, Boulder, CO, USA). Initially, the RS images were mosaicked after basic layer stacking, and were clipped using the shape file for the study area. The mosaicked and clipped RS images for the study area are shown in  IJBR (2014) 05 (08) www.ssjournals.com Both the unsupervised and the supervised classifications for the year 2000 were compared with the supervised maximum likelihood classification for four major classes of land cover types (Urban, Water, Vegetation, Mountain) with a minimum of 15 ROI polygons for each class and a minimum of 300 ROI points for each class for accuracy assessment. Based on the accuracy reports and commission and omission errors between t he two methods, supervised maximum likelihood classification was performed on all further images. Change detection statistics was performed on the classified maps using the post classification comparison method 23 .

NDVI transformation
NDVI transformation was performed on the raw, mosaicked images using bands 4(Near Infra Red) and 3(Red) for the Landsat TM images. The range of NDVI varied from -1 to +1, with values of 0 to 0.2 indicating bare soils (with scanty vegetation), and 0.2 to 0.7 reflecting different categories of green vegetation. Using image exploration and also based on earlier studies from this region, density slicing was performed on the NDVI transformations and vegetation with NDVI more than 0.2 (representing dense thick vegetation) [24][25][26][27] were highlighted and exported as separate classes for each study year.

Results
Mangalore is endemic for malaria and nearly 40% of cases of malaria are caused by P. falciparum infection. The absolute numbers of malaria cases in Mangalore taluk has been on the rise since the mid-1990s and the trend in the incidence rates are shown in Figure 3. Between the study years, absolute numbers of all malaria cases reported in the region increased five-fold from 1798 to 9138. Annual malaria incidence rates increased from 203 cases to 1035 cases per 100000 population in three years. The annual increase in malaria incidence rate, population and population density were 277/100000, 12182 and 18.33 persons/km 2 respectively.  The increase in urbanization was noted more in the regions adjoining the rivers and the coast in the study area. Figure 5 sho ws the changes over the 3 year period in a small section of Mangalore city.

Discussion
This pilot study illustrates dramatic changes in the land use, land cover, population density and malaria incidence in the co astal area of Mangalore. Considering that with even very modest deforestation of 4% per year, the risk of malaria can potentially increase by 50% in areas with conditions favorable for vector breeding and high malaria incidence 13 , we have observed changes that might have serious consequences for malaria transmission in the region. We demonstrated that four land cover types that affect life cycles of malaria vectors were changed between the years 2000 and 2003: with the annual increase in the urban land cover by 6.7%, a reduction in the mountainous terrain by 11.6% and vegetatio n by 12.9%. Over 3 years, a 4.5% reduction in the level of thick vegetation (NDVI>0.2) was noted in the region, which could be because of increased urbanization. The fresh water and stagnant water collections had increased by ~10% annually. Some of these artificial water collections, favoring mosquito breeding and transmission of the disease, are due to construction related activities and collections after rainfall in places like disused vehicle tires, tree holes, used coconut shells, open water tanks, open wells etc. which are associated with human activities 5 .
With its unique geographic position (8 0 4' to 37 0 6' N latitude and 68 0 7' to 97 0 25' E longitude), diverse topography and climatic variations, India contributes greatly to the global malaria burden 28 . It has been shown that in India the water bodies, coconut/arecanut plantations, marshy areas, moist soil, rocks with vegetation are the landscapes critical for mosquito breeding 4,29,30 . Recent increases in malaria related morbidity and mortality has been noted in the coastal regions of India which is mainly due to urban malaria. Mangalore taluk, located on the western coastal region of South India has experienced an increase in the urbanization over the last two decades and a corresponding increase in malaria related morbidity and morbidity. Recent studies show that malaria was on the rise among construction workers in this region 5 .
The data on annual malaria incidence in the region indicate a complex pattern with a clear decline between 1999 and 2003 and a sharp peak in 2004. One of the main reasons for this decline in the number of cases of malaria in this region after 2004 is the formation o f a special "malaria cell" in the area, which focused on active surveillance with special emphasis on migrant workers, construction workers, hotel workers and inmates of orphanages. The active surveillance for sources combined with source reduction activities, anti-larval and anti-adult insecticide spraying operations were intensified in the region 5 . However, the overall trend of malaria in this region is rising along with the increase in the population, population density and urbanization in t he area. The population density in this region has increased by over 100 people per km 2 in the last five years, which approximates to an addition of roughly around 25 new families per km 2 .We believe that the increase in the population density with increasing urbanization, deforestation is associated with an increase in the breeding sites of mosquitoes due to construction related activities in this region. This could lead to increased vector breeding, increased vectorhuman contact, resulting in an increase in the bite rates and consequently an increase in number of mosquito borne diseases i ncluding malaria in this region.
This study along with other investigations conducted in India has demonstrated a strong potential for using satellite imagery to del ineate the breeding habitats of the major malaria vectors and to forecast the risk of malaria transmission due to human activity 4,18,19,[29][30][31] . The use of remote sensing techniques to map disease risk and vector distribution has evolved and range from using simple correlations between spectral signatures from different land use-land cover types and species abundance to complex techniques that link satellite-derived seasonal environmental variables to vector biology [32][33][34][35] . In a study looking at comparisons of satellite-derived precipitation, temperature, humidity, vegetation and elevation measures in different districts of Kenya has demonstrated the differences in mosquito larval breeding habitats in relatively smaller geographical regions ranging from open sha llow sunlit puddles closer to homes to permanent vegetated aquatic habitats such as stream pools of rivers varying between different species of vectors 36 . A study looking at the dynamics of mosquito population in rice fields in California revealed that that rice fields with rapid early season vegetatio n canopy development, located near livestock pastures had greater mosquito larval populations than fields with more slowly developing vegetation canopies located further from pastures and the sensitivity of these predictions were up to 90% accurate 37 . The ability to foresee flooding of mosquito habitats by remote sensing was found to have importance bearing on developing strategies for mosquito control & disease prevention 18 .
The difficulties in obtaining good quality remotely sensed data for the region beyond the year 2003 due to high cost for purc hasing data for independent investigators and the non-availability of the information on malaria burden at sub taluk (at a refined spatial resolution) in India, limited information on local meteorological data, which are often obtained from different locations, incomplete and are not uniform, were some of the important limitations faced during the study. In India, the estimated economic loss due to malaria are huge and ranged between $506.82 million to $630.82 million between 1990-1993 38 and up to $1 billion in 1995-96 39 . The total Disability Adjusted Life Years (DALYs) lost due to malaria was 1.86 million years in 1997 based on the WHO reported incidence and still would be an underestimation in India, higher among (1.074 million) and the maxi mum DALYs lost were among the economically productive age groups from 15 to 44 years of age (53.25%), followed by children < 14 years of age (27.68%) 40 . India has spent up to 25% of its health budget on malaria control from 1977-1997, and starting in 1997, India planned to spend $40 million on malaria control, a 60% increase from the previous year 41 . From this point of view, availability of high quality data for research and prevention purposes can offer a low cost soluti on.
This study serves as a useful demonstration of how public access to remotely sensed data serves as a cost effective approach to study relationships between the environment, agent and the host. The presented approach can easily be adapted to other vector borne diseases, which contribute to a significant amount of morbidity and mortality in the developing world. In conclusion, with access to remotely sensed data of good spatial, spectral and IJBR (2014) 05 (08) www.ssjournals.com temporal resolution and effective disease surveillance systems it is possible to study the interactions between the environme nt and the diseases even in resource scarce settings using novel methods and this would prove invaluable to health planners and policy makers at regional levels.

Ethical approval
This study has been approved by the Institutional Review Board of the Christian Medical College for publication (Ref: IRB-EXP 4-30-01-2012).