Land Use Land Cover Mapping with Change Detection: A Spatio- Temporal Analysis of Ambo Woreda, Ethiopia from 2000 to 2020

The main aim of this present study is to identify and detect the land use, land cover changes occurred in the Ambo Woreda of West Shewa Zone in Ethiopia and to integrate the Remote Sensing and GIS data for analyzing and evaluating the changes in land-use of study area. Based on remotely sensed data, the Land Use Land Cover (LULC) maps and field records have been considered for investigation. Landsat7 ETM+ image of 2000 and Sentinel 2A image of 2020 are the two remotely sensed images of study area used in this study. The supervised classification based on maximum likelihood classifier in ArcGIS 10.3 has been used to identify the five major categories of LULC. The observation on the period of twenty years reveals that the agricultural land and built-up areas have stretched rapidly to the adjacent fallow lands. Also, there is significant loss in Hilly Vegetation due to settlements and industrial expansion in the fastest growing region of the Oromia Zone.


I. INTRODUCTION
For a human being to survive on earth, land is one of the important natural resource next to water. Land use is explained as how man is utilizing the land whereas land cover is the vegetation spread over the land. LC data shows how much of an area is covered by natural phenomena but land use data shows how people utilize the landscape [1]. Drastic changes in land use and land cover (LULC) have been seen since the twentieth century and those traceable changes resulted in a complex land use structure [2]. Due to the increase in population and the flow of people from rural to urban areas, major towns and cities of the world have experienced rapid urban growth. Various driving factors like population pressure and development makes the LULC change dynamics non-uniform throughout the world [3]. "Like many other developing countries, Ethiopia has been experiencing environmental degradation problems including LULC conversion, soil erosion, loss of forest and other vegetation covers and water resource degradation" [4]. Studies on the Ethiopian highlands showed that the expansion of cultivated land increased through time at the expense of natural forest and also claimed that deforestation has been reduced in the recent times due to the plantation activities on degraded hillsides which led to improvement in vegetation cover [5]. Farmland, settlements and degraded lands were expanding considerably while grasslands and forest areas have been diminished driven by economic factors and policy issues [6]. Ethiopian highlands are prone to heavy degradation and frequently affected by drought and famine. More and more marginal lands are being used for agriculture due to the population growth in the highland areas [7]. Besides, the reliability on woody biomass for fuel, the expansion of agricultural activities at the expense of vegetation cover and demand for wood for construction activities are the two main factors led to the uncontrolled land cover change and deforestation in Ethiopia [8]. Analysis of LULC changes have significant roles in the understanding of earth-atmosphere, forest fragmentation and future management plans. Hence, LULC changes should be quantified and investigated. However, spatial pattern, extent and rates of such land cover changes have never been quantified and analyzed in the study area using the high resolution satellite imagery [9]. In recent times, multispectral and multi-temporal satellite data with medium to high resolutions, have emerged as an important tool for estimating various aspects like vegetation cover, forest degradation and urban expansion [5], [10]. Remote Sensing (RS) and Geographical Information Systems (GIS) has been used as a primary tool for identification, analysis and quantification of LULC changes [10]- [13]. High resolution satellite data like Sentinel -2A data can be used for preparing land cover maps and accurate results will be yielded in change detection analysis [14]. However, medium resolution data such as the Thematic Mapper (TM), Enhanced Thematic Mapper + (ETM+) and Operational Land Imager (OLI) have been used extensively in the studies for LULC change detection analysis due to the limited availability of Sentinel datasets [3], [15]- [18]. Despite, being the epicentre of Oromia, Ambo town lacks attention of scientific information on land use and land cover change. These kinds of analyses are very essential for the future planning and developments. The present research study was carried out with the following objectives (a) to evaluate land use and land cover change from 2000 to 2020 and (b) to perform change detection analysis to understand how various land use land cover types changed over this period.

 Study Area
The study was conducted in the Ambo Woreda, Ethiopia. The study area map is shown in Figure 1, which is a spa town located in the West Shewa Zone of the Oromia Region, at 125 kms west of Addis Ababa. The area lies within 8°45' -9°25'N latitude to 37°35' -38°00' E longitude at an altitude of 2101 metres above sea level. It covers an area of 95966.2 hectares. It experiences subtropical highland climate.

 Data Sources
Two period of satellite images were used to conduct this study as shown in Table 1. Digital map on shapefile of scale of 1:25000 from DIVA GIS were used as supporting spatial data for delineating the study area boundary. In addition, to that Global Positioning System (GPS) points marked during fieldwork, were used in the collection of Ground Control Points (GCP) as the support for the classification. Nearly 110 sample-training sites have been used in each year (2000 and 2020) from ancillary data like high-resolution Google Earth Imagery. The Landsat7 ETM+ data, which are acquired on the Worldwide Reference System-2 path/row system dated on 26. 11.2000 has been used in this study. It is a Level-1TP product which provides radiometric and geodetic accuracy. The Sentinel -2A data acquired on 12.11.2020 is another data used in the current study for extracting LULC information. It is a Level-1C product which is geometrically and radiometrically corrected. The Landsat satellite has the revisit period of 16 days where as the sentinel series has the revisit period of 5 days which makes it more useful for choosing from the cloud free day. The cloud cover of Landsat and Sentinel images used in this study are 0% and 0.12% respectively. Both of these images have been downloaded from the United States Geological Survey (USGS) Earth Explorer archives.

 Image Classification
The process of image classification aims to classify all the pixels in the processed satellite images to the desired land use and land cover categories. For the current study, Supervised Classification based on Maximum Likelihood Classifier has been selected for classifying pixels. To facilitate supervised classification, training samples were given using the signatures exhibited by each land use and land cover category. A total of five land use classes were established, by Anderson's Level 1 (1972) classification based on information from local communities, properties of Landsat and Sentinel images, data like Google Earth and field observation as shown in Table 2. LULC map is prepared with the help of ArcGIS 10.3 and ENVI 5.3.

 Accuracy Assessment
To determine the accuracy of the classification, the number of reference pixels is an important factor. An equalized stratified random sampling approach was used to assess the accuracy of each of the two land cover classifications. The overall accuracy and a KAPPA analysis were used to perform classification accuracy assessment based on error matrix analysis. Accuracy assessment was performed for 2000 and 2020 LULC maps. Based on the stratified random sampling method, a total of 110 pixels were randomly selected for each year. For the 2000 LULC map, the result shows an overall accuracy of 80% and a kappa index of agreement of 0.73 as shown in Table 3. In terms of User's accuracy, all classes were over 80% except fallow land and barren land which were 78% and 70% respectively. In terms of Producer's accuracy, all classes were over 85% except barren and agriculture lands. For the 2020 LULC map, the result shows an accuracy of 84.54% and a kappa index of agreement of 0.79 as shown in Table  4, In terms of User's accuracy, all the classes were over 83% except barren land which was 75%. In terms of Producer's accuracy, built-up class was 100%, whereas hilly vegetation and agricultural land were over 85%. And the fallow land showed over 77%. The confusion error matrix and Kappa statistics used for the classification accuracy of 2000 and 2020 LULC maps are presented in Tables 3 and 4. The kappa statistics showed a strong agreement for the years 2000 and 2020.

IV. CONCLUSION
The present work on the spatio-temporal analysis of LULC types in the Ambo Woreda in West Shewa Zone of Ethiopia has revealed that there were significant LULC changes in the area during 2000-2020 periods. The proportional changes in the area were found out for the total LULC types in two different periods. The multi-temporal analysis of satellite image data clearly showed that hilly vegetation and fallow lands have decreased; in the contrary agricultural land, barren land and built-