Agricultural Land Cover Change in Gazipur, Bangladesh, in Relation to Local Economy Studied Using Landsat Images

Land classification is conducted in Gazipur district, located in the northern neighborhood of Dhaka, the capital of Bangladesh. Images of bands 1 - 5 and 7 of Landsat 4 - 5 TM and Landsat 7 ETM+ imagery recorded in years 2001, 2005 and 2009 are classified using unsupervised classification with the technique of image segmentation. It is found that during the eight year period, paddy area increased from 30% to 37%, followed by the increase in the homestead (55% to 57%) and urban area (1% to 3%). These changes occurred at the expense of the decrease in forest land cover (14% to 3%). In the category of homestead, the presence of different kinds of vegetation often makes it difficult to separate the category from paddy field, though paddy exhibits accuracy of 93.70% -99.95%, which is better than the values for other categories. In addition, the analysis based on digital elevation model reveals that paddy cultivation is implemented in lowland rather than highland. Homestead areas have spread from south (low elevation) to north (high elevation), in association with the decrease in forest-covered areas.


Study Area
The present research area is Gazipur (Figure 1), which is in the northern outskirts of Dhaka, the capital city of Bangladesh. The area is located between latitude 23.88˚ -24.34˚ and longitude 90.15˚ -90.70˚. Gazipur is situated in Old Brahmaputra (floodplains) and Madhupur tract (terraces). The main rivers in this district are Old Brahmaputra, Shitalakshya, Turag, Bangshi, Balu, and Banar. Annual average temperature varies from 13 to 36˚C. Annual rainfall is around 2400 mm. There are five sub-districts (Upazilla) in Gazipur, namely, Gazipur Sadar, Kaliakair, Kaliganj, Kapasia and Sreepur, as indicated in Figure 1. According to the BBS statistics in 2011 [8], the total population and population density of Gazipur district is 34,03,912 and 1884 km −2 , respectively, with annual growth rate of 5.2%. The literacy rate of Gazipur district is 62.60% [8]. Agro-ecologically, the study area belongs to AEZ-9 and 28 categories of FAO of the United Nations [16]. Rice of local and high yielding varieties (HYV), jute, vegetables, etc., are important agricultural products with relatively high economic impact, in addition to the forest products such as wood, wood fuel, wood pulp, forage etc. In terms of agricultural research institutes in Bangladesh, major organizations such as BRRI, BARI and university that has faculty in relation to agriculture are located in Gazipur. Also, the area has experienced rapid advancement in the industrial sector such as machine tool factories, a diesel plant, packaging industry, brick field manufacturers, etc., all of which can contribute to the regional economic development in the foreseeable future.

Satellite and Other Data Sources
The analysis of this study is implemented for the following two types of satellite images: Landsat 4 -5 Thematic Mapper (TM) (16 January 2005 with 0% cloud cover and 11 January 2009 with 1% cloud cover) and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) (29 January 2001 with 0% cloud cover). The present choice of January images is based on the fact that such nearly cloud-free data are available only in the dry winter season. Because of the choice, the paddy condition is for monitoring the growing phase of Boro rice. The analysis of satellite data in the same season is beneficial for attaining better comparison in different time periods, enabling the extraction of meaningful changes in LULC. Data are analyzed using the information from bands 1 -5 and 7 ( Table 1). The ETM+ sensor has a spatial resolution of 30 m for visible (bands 1 -3), near-infrared (NIR) (band 4 and 5) and shortwave infrared (SWIR) (band 7) wavelengths. The resolution of TM is the same for the blue to mid-IR reflectance bands. Landsat images were acquired from the USGS Global Visualization Viewer (http://glovis.usgs.gov/).
For the purpose of ground-reference evaluation of the classification results, we extracted 38 areas (4535 pixels in total) from the Google Earth imagery, which is originally based on the Digital Globe imagery observed in 2007 and 2010. Each area has 120 pixels, on average, and consequently, the pixel size unit of each area is around 11 × 11 (330 × 330 m 2 ). Also, we employed 262 images taken with a GPS camera occasionally in 2013 and 2014. The topographic feature of the area is obtained from the digital elevation model (DEM) data based on   Figure 2 shows the flow of the data analysis in the present study. Three satellite data were subjected to the segmentation analysis, which is the foundation for an object-based classification. It is the process of subdividing an image composed of a number of different image objects. For controlling segmentation procedure, Liu and Yamazaki [17] adopted the following four parameters: scale parameter (allowed level of in homogeneity in an object), layer weight (weight for each band), and shape factor (weight of spectral and shape features), and compactness (compactness of each object). In order to reduce the noise, the object-based segmentation is employed rather than the pixel-based segmentation. The values actually utilized in the present research are 5, 0.1, and 0.5 for scale parameter, shape factor, and compactness, respectively. The same value of unity is assumed for the layer weight of all bands. The choice of these parameters has contributed to the separation between forest and paddy classes.

Segmentation and Classification
In the present study, we use band 5 (1.65 μm) and band 7 (2.22 μm) of Landsat TM and ETM+ in addition to bands 1 -4, since these two bands with longer wavelengths are less susceptible to the atmospheric influence than shorter wavelength bands (bands 1 -4). By using the data for three different acquisition times (2001, 2005, and 2009), change detection has been achieved in the Gazipur area.
After the segmentation, unsupervised classification algorithm (ISODATA) is applied to classify the segmented images into four different classes, i.e., homestead, paddy, forest and urban. Prior to this procedure, the water bodies in association with major rivers are removed by applying masks (with the help of visual interpretation), since the discrimination between paddy and water body is often difficult. The major characteristics of these four categories can be described as follows. Paddy field is the arable land used for paddy cultivation. Forest indicates lands covered with deciduous forest, mixed forest and others. Homestead is the land used for the production of multipurpose trees, various fruits, vegetables, spices, fodder, forage, fuel wood and ornamental herbs/shrubs: recently such small-scale agriculture has increased considerably in the rural areas of Bangladesh [18]. The category of urban area stands for residential, commercial and services, industrial, transportation, roads, mixed urban and other urban structures.

Topography and Land Cover
The topographic feature of Gazipur was derived from the SRTM DEM data. Here the combination of the DEM information and the classification result of Landsat 2009 image (as described below) is exploited for the examination of altitude distribution of each land coverage type (homestead, paddy, forest and urban) (Figure 3). In Bangladesh, the great plain lies almost at sea level along the southern coast and the land elevation rises gradually toward the north. Since Gazipur area is in the mid of the great plain, it is included in medium-high to highland, i.e., 4 -24 m above sea level (ASL). In highland (20 -30 m ASL), the elevation is above the normal flood  level, while the medium-high area (below 20 m ASL) is normally flooded to the maximum depth of around 90 cm during the monsoon season from June to September. In Gazipur, homestead and forest areas are located in mid to high elevation areas, while paddy areas are found normally in lowland areas. From SRTM observation shown in Figure 3(a), it can be seen that western part of Gazipur is occupied by high land (>20 m ASL), whereas eastern part has middle to low range of elevation. Among the five sub-districts, land elevation is relatively high in Sreepur (northern area), while it is relatively low in Kaliganj (south-eastern area). As indicated in Figure 3(b), low-land areas (southern part) have many paddy areas owing to low topographical conditions. Other than paddy, the low land area is covered with floodplains, depressions, abandoned channels, swamps, and marshes. Also, the height of the land gradually increases from east to west: in the western part the highland is often called the Dhaka terrace, which is dissected by the tributaries of the Turag, the Buriganga and the Balur rivers. The elevation of Dhaka terrace ranges from 6.1 to 11.3 m ASL.  Urban area occupied 1% in 2001, which increased to 1.3% in 2005 and 3% in 2009. This change was observed especially in Gazipur Sadar, followed by Kaliakair. From Figure 4(a), it is evident that the rapid urbanization has occurred in the Gazipur-Sadar sub-district, in the vicinity of the capital Dhaka area. An important implication of growing urbanization is that some of the fertile agricultural land had to be converted to other uses. In the whole Bangladesh, close to 1% agricultural land is being lost every year to other uses including settlements for growing population [3], which is a potential threat to food security.

Land Cover Change
In contrast to the increase in the three categories mentioned above, the decrease in forest covered areas (14%, 10%, and 3% in 2001, 2005, and 2009, respectively) is noteworthy. Such decrease in forest cover is found especially in Sreepur (northern part). It is estimated that 25% to 26% of wood production in Bangladesh is used for burning bricks every year, causing deforestation. This is an alarming trend for environment because of the frequent incidence of drought-spells with the visually identifiable trend of desertification [19]. Serious environmental impacts of deforestation would generally include loss of habitats and bio-diversity, undesirable changes in local ecology, changes in microclimate, loss of carbon sink, soil erosion, less water retention, increased flooding, etc. The central Sal forest is distributed in Dhaka, Gazipur and some other districts of Bangladesh, corresponding to 86% of the Sal forest in the whole country. Anthropogenic threats such as over exploitation, illegal cutting, expanding agriculture, urbanization, pollution, as well as natural threats are the major causes of deforestation of Sal forest. According to the satellite data in 2009, forest coverage fraction was 28% in Sreepur, 11% in Kaliganj and Kaliakair, 19% in Kapasia, and 31% in Gazipur Sadar. According to the BBS report 2011 of Gazipur [8], on the other hand, 33% area in Sreepur was forest, whereas in Kaliakair, Kapasia, Gazipur Sadar, and Kaliganj, the forest fractions were 27%, 20%, 20% and 0%, respectively. The lack of forest area in Kaliganj, the south-eastern sub-district, is ascribed to a large-scale development of industrial complex that started in 2009. This destroyed nearly 1600 acres of crop land, forests, orchards and water bodies in Kaliganj for creating Purbachal New Town at about 16 km from the capital [20].

Validation of Classification Results
We conducted an extensive field survey using the global positioning system (GPS) from September 2013 to January 2014 to verify the classification results. Also with the help of the Google Earth imagery, the accuracy tables for the classification results in 2001, 2005, and 2009 are summarized in Table 2. Here the total number of validation pixels is 4535, corresponding to 38 points (areas) selected for the ground reference (GR). In the pixel-based accuracy table in 2001, correctly classified pixels are 972 (homestead), 1740 (paddy), 897 (forest), and 141 (urban area), whereas some pixels were misclassified in every class. Among the 1857 pixels that were classified as paddy for all of three different time periods, 117 pixels in 2001, 1 pixel in 2005 and 8 pixels in 2009 were misclassified as homestead. Regarding the percentage of accuracy, it is observed that paddy exhibits good results in different years (93.70% -99.95%), which is better than other categories. Since the κ values are 0.745, 0.852 and 0.820 for 2001, 2005, and 2009, respectively, it is understood that the accuracy of the present classification result is reasonable as a whole.
In the category of homestead, the presence of different kinds of vegetation often makes it difficult to separate the category from paddy field. Also, it is sometimes difficult to discriminate between paddy and water, since paddy rice plants grow well in low lying areas. This difficulty has been alleviated by using the river mask as derived from the three-year data using the very low reflectance of water surface in shortwave infrared and nearinfrared as compared with other objects.
Similarly, among the 213 pixels that belonged to urban areas, 141, 201 and 64 pixels were purely classified as urban, whereas 3 and 69 pixels, 5 and 7 pixels, and 146 and 3 pixels were misclassified as homestead and paddy in 2001, 2005, and 2009, respectively. As typically illustrated in Figure 5, the misclassification with different pixels tends to occur in association with the steady increase in the fraction of homestead during the eight year period. Since the Landsat images were taken in January in the dry season, soil contains less moisture. Thus, its reflectance becomes more or less similar to that of road or concrete in the urban category, and this can be the cause of the misclassification that occurs between the homestead and urban areas. Figure 5(b) shows the conversion from forest to homestead, exemplifying the expansion of homestead areas.

Relation to Regional Economy and Food Security
According to the present analysis of Landsat data, paddy area (in km 2 [8]. On average, the paddy areas account for 27.1% of the total area, which is slightly smaller than the satellite result of 34.8% (588 km 2 ). Nonetheless, the present analysis of satellite data has revealed that paddy area increased by around 7% during the period from 2001 to 2009. This recovery in the paddy area can be associated with the adoption of modern rice varieties. Although regional statistical numbers are unavailable, rice production increased by 6.4% from 2001 to 2008 in accordance with the social statistics in the whole country [8]. The change of three rice types in 2008-2009 was 6% to 6% (Aus, no change), 41% to 37% (Aman), and 53% to 57% (Boro  [8]. These changes indicate the recent improvement in the food security condition, especially in availability aspect. Records say that in the 1970s, 70% of the population were under the food consumption of poverty line in Bangladesh. In 2005, however, this went down to under half of the population. As such, nutrition situation in Bangladesh has improved, as manifested in per capita energy supply from 1800 kcal in 1970s to 3055 kcal in 2009 [8]. The present analysis with the SRTM DEM data has revealed that homestead areas are mainly situated at a height of ~16 m ASL, which is higher than other categories. The economic value of home gardens is the summation of the direct use values (e.g., fruit, fuel, and timber), indirect use values (e.g., soil conservation), optional values (e.g., biodiversity), existence values (e.g., endangered species), and bequest values (e.g., habitat) [13] [14].
From Table 2, it can be found that homestead accuracy was 70.03% in 2001 and 95.97% in 2009. In 2001, misclassification occurred with the category of forest, which is one of the components of homestead having similar spectral characteristics. One reason for the improvement in the accuracy is that gradually the forest fraction decreased, so that the accuracy in the homestead class improved in 2009. Homestead agroforestry practice and diversified food pattern can reduce the food gap [21]. A wide variety of trees, shrubs and thickets of bamboos form the groves that surround the village homesteads [8]. It is noticeable that the size of homestead forestland increases with the increase of landholding size class, but the percentage of homestead forestland in relation to total land decreases because of the devotion of higher proportion of the land to agricultural crops [22].

Conclusion
Detailed examination of land cover change has indicated the change in agroforestry conditions in Gazipur district in Bangladesh. The district is in the northern outskirts of Dhaka, and exemplifies the agricultural conditions in the midland of the country under the influence of population increase and resulting urbanization. In the present study period of 2001, 2005 and 2009, homestead areas increased by 2% and paddy area increased by 7%. In the same time span, forest land decreased by 11% due to urban expansion which has turned out to be significant in Gazipur Sadar and Kaliakair sub-districts. Such forest coverage reduction is alarming for maintaining ecological equilibrium, especially when the ever increasing population of the country is considered. The changes in land cover thus detected in Gazipur are in line with those anticipated from various observations previously made on the basis of ground-based, socio-economical surveys. The approach demonstrated in the present paper will be useful for monitoring the detailed changes on the regional level, in view of the capability of satellite remote sensing for providing stable datasets over an extended time periods. Further work is in progress for better monitoring of seasonal changes in rice phenology through the interpretation with long-term satellite data.