Land Use Change Detection and Impact Assessment on an Agricultural Area

This study focuses on detecting, mapping and analyzing the conversion of forests into agricultural uses and agricultural uses into urban/peri-urban uses in Nzega district, Tabora Region Tanzania over a period of 28 years since 1978. Land use classes were from multitemporal and multi-sensor satellite images and aerial photographs. Topographical maps at a scale of 1:50000 and onsite information gathered in the field were used for interpretation and ground truthing purposes. Land use changes were detected using land use change matrices and land use change maps. Four land use maps were compiled from Aerial photographs dated 1978, Landsat TM satellite imagery dated 1986, Landsat ETM+ satellite imagery dated 2000 and IRS satellite imagery dated 2006. The maps show agricultural areas are concentrated along road networks generally expanding towards neighboring grasslands and unprotected forested areas. Land use changes were detected from land use maps for three change periods: 1978 1986, 1986 2000 and 2000 2006, with emphasis on agricultural and forest land uses. The change maps and change matrices show that despite abandoned farms agricultural land increases with time. Change in the opposite direction was also common where abandoned agricultural land gradually regenerated back to grasslands and forests. An average of 16% of agricultural land has been maintained during the 1978 2006 period compared to 67% of forests that have been maintained in the same period of time.


Introduction
The dynamics in land use changes are brought about by a number of interacting factors including socio-economic, political, cultural and environmental factors (Lambin, Geist, & Lepers, 2003) and have prominent effects on the balance of ecosystems and on human productivity and welfare (Mertens, & Lambin, 1999;Srivastava, Han, Rico-Ramirez, Bray, & Islam, 2012).The control, management and prediction of future trends in environmental change, natural resources and food security are among the issues that require detailed information on land use and its changes over different spatial extents.
Tanzania has 80% of its population living in rural areas (Mbwambo, 2004) their main occupation being agricultural activities.Tanzania's fast growing population has led to higher and varied consumption demands, that in turn cause major changes in the use of land and natural resources to meet these demands (Mbwambo, 2004;PPU, 2007).The changes are reflected in the growing number and size of settlements (PPU, 2005), deforestation (FOSA, 2000), decline in agricultural productivity (Amani, 2005) and environmental degradation (Hubacek & Sun, 2001).To be able to control and manage land use change, systematic monitoring and mapping of change over spatial and temporal spaces is required to provide an understanding of the factors contributing to change and the role and significance of each of these factors.
The nature, amount and distribution of land use changes in Tanzania are not well known.There are several reports at National level (Kalenzi, 2006) documenting the occurrence of land use patterns and changes in the use of land resources but these reports mostly consist of non spatial information.Moreover, many existing reports are a result of studies conducted in the attempt to characterize global/regional environmental changes (Tunner, Moss, & Skole, 1993), thus lacking detailed information of what is actually happening on extents smaller than the national level.

Data Validation
Validation for topographical maps and aerial photographs consisted of examination of features present in the paper maps and the paper prints using distance measurements and by visual inspection.Both the topographical maps and aerial photographs were found to be in good order for the purpose of this study; the quality was good enough to allow interpretation of features both visually and digitally.The information present in topographical maps tallied with the corresponding information in aerial photographs; assuring consistency.Prominent features present on the ground were confirmed by visual observation to be in place in the topographical maps and aerial photographs implying that topographical maps and aerial photographs used in this study still form a reliable source of information despite the fact that they are outdated.
The quality of satellite images from both IRS and Landsat was checked by examining consistency with topographical maps and aerial photographs and found to conform to the requirements of this study.Both satellite image data were 100% cloud free within the study area although IRS and Landsat ETM+ imagery were acquired during the rainy season.

Data Processing
Data sets acquired were pre-processed to fit the requirements of this study.Satellite data sets used were preprocessed at Level -2 preprocessing, where radiometric correction and systematic geometric correction were performed.However some deviations from the mapping system used in Tanzania, especially in the position of features (coordinates) were identified therefore the position of features in the satellite image were rectified to conform to those of the mapping system of Tanzania.Aerial photographs covering the area of study were scanned and used to create a controlled mosaic.Due to lack of ground photo control points the aerial photographs could not be processed photogrammetrically.Instead, the photographs were georeferenced using coordinates scaled from topographical maps and then land use patterns were extracted from the photographs through visual interpretation and digitization

Image Classification
Supervised classification was selected to make use of ancillary data in the classification process in order to achieve high accuracy.Training sites were developed using knowledge obtained from field data, topographical maps and information from people familiar with the area of study.
The method used for supervised classification was maximum likelihood classification which is based on both the distances towards class means and the variance-covariance matrix of each class.The method groups together features in specified classes based on the likelihood of each feature to the training set representing a specified class.
Interpreting land use from satellite images require additional information that can link land cover classes captured by satellite images to human related activities that represent land use classes (Kandrika & Roy, 2008).Land use classes were inferred from land cover classes based on ground observations, visual patterns on both aerial photos and satellite images and ancillary information including local residents and expert experiences.Table 2 defines classes mapped.The main goal of image classification was to extract agricultural areas and the interaction between abandoned and newly acquired farms with other classes.Land cover classes were therefore categorized and grouped based on how they facilitate or limit changes in agricultural areas.Among the classes mapped agricultural land and peri-urban centers are referred to as land use classes while forests, woodlands, grasslands and marshlands are referred to as land cover classes.

Class Editing
The results of classification show a high spectral overlap between some of the classes resulting into mixed classes.Some settlements, bare soil and roads were classified as one category as they all possessed similar spectral values.In areas where settlements are intermixed with agricultural land i.e. mixed farming, majority of the houses (Matembe) are thatched with dry grass or has packed soil as their roofing.Roads are not paved and spectrally resemble bare soil and/or unplanted farms.This made it difficult to distinguish between settlements, bare soil, farms and grasslands using the conventional multispectral image classification algorithms.Mixed classes were rectified through class editing, where each pixel in the mixed classes is visited and labeled by its true class.The true class of each pixel is identified by visually interpreting the satellite image data using available ground truth information.The approach is based on generating binary masks or bitmaps over areas where mixed classes are identified.The masks are then used to either restrict the classification algorithm to a spectral range of pixels representing respective classes in which case the image is reclassified using generated bitmap masks, or transfer pixels from one category to another, or merge classes that are to form one class or delete undesirable classes falling under the mask.

Classification Accuracy Assessment
The results of classification were assessed using error matrices.An error matrix compares classification results to additional ground truth information as a standard.The strength of an error matrix lies in the fact that it identifies the nature of classification errors, as well as their quantities (Lillesand, Kiefer, & Chipman, 2004;Mather, 1999).Classification results were compared to ground truth data that was not used for training.Overall 80% and 77% was achieved for accuracy and reliability respectively.These results were influenced by peri-urban centers which had the least accuracy (53%) and reliability (56%); classes of interest namely forests, agricultural land and grasslands had better accuracy (86%) and reliability (89%).

Land Use Change Detection
This study made use of post classification comparison and GIS analysis of land use maps compiled from satellite images and aerial photographs to detect changes in land use categories.Classification results formed input to map calculation and overlay analysis functions where change maps and change matrices were generated.
www.ccsen     Forest cover had decreased by about 5% during this period with 79% of the forested land being maintained as forests by mid 1980s.About 1062 Ha of forests were cleared for agriculture during this period compared to 13066 Ha of grasslands cleared for the same thus the rate of conversion of forests to agricultural use in this period is low.4) and change maps emphasizing changes within agricultural and forested areas (Figures 7 and 8).Forested land acreage is observed to have increased during this period.This is partly because more forests were detected from the Landsat ETM+ satellite imagery (2000) which was acquired during the wet season as compared to those detected from Landsat TM imagery (1986) which was acquired during the dry season.It is also observed that 54% of the forested area was maintained during this period; decreasing from the 79% that was maintained during 1978-1986 period.Changes in the forested areas were observed in both directions, some of the forested land was converted to other uses while other uses were also converted to forests.Generally, most of land use changes during this period are in conjunction with agricultural land use which was identified to concentrate along road networks (Figure 3 5) and the change maps (Figure 8 and 9).Generally, most of the changes in land use during this period are concentrated along road networks thus proximity to road network is still one of the major factors facilitating the conversion from forest into agricultural and other uses and conversion form agricultural uses into peri-urban settlements.

Impacts of Land Use Changes on Agricultural Area
The major impact of land use changes on an agricultural area identified in this study is the rise of agricultural production.This rise is evident from the yields per farming season where 934 tonns of food and cash crops were produced as surplus during 2005-2006 in Puge South.The increased land acreage occupied by agricultural fields (Tables 3-5) further confirms the rise in agricultural production.
It is noted that the rate of conversion of all land use categories to agricultural land is generally highest among other land use categories in all the three periods of change detection.

Conclusion
This study was undertaken for the purpose of detecting, mapping and analyzing the conversion of forests into agricultural uses and agricultural uses into urban/peri-urban uses that have taken place in Tabora region for a period of 28 years since 1978 using IRS satellite images.A focus area selected within Tabora region comprised of part of Nzega district to allow a detailed study.To characterize land use and land use changes for the period when IRS satellite images were not available, Landsat TM and Landsat ETM+ satellite images dated 1986 and 2000 respectively, and aerial photographs dated 1978 were used.
The land use classes of particular interest to this study namely agricultural uses and forests were adequately compiled from all the data sets.However, peri-urban centers could only be mapped from IRS satellite images while individual houses were mapped from aerial photographs.The spatial resolution of both Landsat TM and Landsat ETM+ could not facilitate mapping of peri-urban centers or any other settlement patterns present in Nzega District.
Despite the differences in characteristics of data used in this study, land use change maps and land use change matrices were sufficiently and accurately generated from the compiled land use maps.GIS analysis functions (spatial overlay) and image processing functions (IHS/RGB transformation) provided capabilities of blending together data sets with different image resolution to achieve common resolution for comparison purposes.
The results show that an average of about 1042 Ha of forests were converted into agricultural uses during the study period i.e. 1978-2006 while an average of 3 Ha of agricultural uses were converted to peri-urban uses in the same period.The rate of conversion of forests into agricultural areas is low compared to that of grasslands, thus grasslands help to minimize and control forest clearance for agricultural uses.Moreover the rate of conversion of agricultural land to peri-urban centers was low compared to what was anticipated i.e. high conversion rate as a result of rapid urban/peri-urban growth to accommodate growing population.This was observed to be a result of the mixed farming practices of the area of study and poor infrastructure development retarding the growth from conventional to technologically advanced farming practices which could facilitate peri-urban/urban growth.Non agricultural land was converted to agricultural land at a relatively high rate compared to the conversion of agricultural land into non agricultural land uses.
The proximity of agricultural areas to roads, forests and grassland facilitated its increase in spatial coverage.Most of the agricultural land is concentrated along road networks where by people tend to expand their farms by acquiring adjacent unoccupied land.This implies that agricultural uses decrease with increased distance from road networks while forests on the other hand increase with increased distance from road networks.Similar results were found by Patarasuk and Binford (2012).Forests in most cases are adjacent to grassland which helps minimizing forest clearance for agricultural use.
It is important to acquire an in depth understanding of processes leading to land use changes so as to be able to monitor and manage changes occurring within small spatial extents and further to be able to manage natural resources including arable land for agricultural purposes.This study provides a starting point for improved mapping of land use changes in Tabora region and Tanzania at large.

Table 1 . Summary of data characteristics Dataset Date Spatial resolution/Scale Seasonality Land cover/use extraction
Two interviews were conducted at Puge division Local Government's office on 8 th and 9 th February, 2007.Each was done based on unstructured questionnaires prepared prior to the field visit as guidelines to the interviews.Two people namely the Puge Division's Secretary General and the Agricultural and Veterinary Officer, were interviewed.

Table 2 .
Description of lad use/cover classes mapped from aerial photographs were merged to forests.Agricultural land is increasing (Table3) as only 275.83 Ha were identified as agricultural land in 1978 compared to 14217.79Ha in 1986; a rise of about 65%.

Table 3 .
Land use change matrix giving total areas in Hectares of land use maintained/changed within land use classes during1978 -1986

Table 4 .
Land use change matrix giving total areas in Hectares of land use maintained/changed within land within land use classes during 1986 -2000 Agricultural land declined during this period.3554.18Ha were detected as agricultural land in 1986 compared to 1722.14 Ha in 2000 a drop of about 14%.Some of the agricultural land was abandoned during this period growing into grasslands.

Table 5 .
-6).Proximity to road networks is one the major factors facilitating the conversion of forest into agricultural and other uses and conversion of agricultural use into peri-urban settlements.Land use change matrix giving total areas in Hectares of land use maintained/changed within land use classes during 2000 -2006 Agricultural land has significantly increased during this period.About 5065 Ha were detected as agricultural land in 2006 compared to 1959 Ha in 2000 a rise of about 20% which is explained by policy implementation as discussed in section 4.1.4.Most of the previously cultivated areas were abandoned to gradually grow into grasslands while new areas were acquired for cultivation.Forests have generally maintained their spatial coverage during this period with about 7998 Ha detected in 2000 and about 7177 Ha in 2006.67% of the forests were maintained during this period; a rise of 13% compared to 1986-2000 period.Changes were observed from forest to other uses and from other uses to forest as shown by both the change matrix (Table