Study Area
Abaya-Chamo lake-wetland is located in 5°43'19''N - 6°38'51''N latitude and 37°21'55''E - 38°15’05''E longitude (Figure 1). In Figure 1, the large Northeastern water body is Abaya Lake and the smaller Southwestern one is Chamo lake. The area of Abaya-Chamo lake-wetland is 242,615 ha (Figure 1). Abaya and Chamo lakes, being Rift-Valley lakes in Southern Ethiopia, lie on a graven (depression) created by faulting due to divergent movement along the boundary of the Africa plate (westward) and the Somali plate (eastward). The Western part of the lake-wetland is largely plain, where some dome-shape and conical volcanic hills, and elevated spurs are observed.
Climatically, Abaya-Chamo wetland, based on data of 1987 – 2018 Mean annual temperature, was about 24 0C; and, the mean monthly temperature of the area is the highest in march (26 0C) and the lowest in July (230C), November (23.1 0C) and December (23.1 0C) (NMA, 2019). The wetland receives a somewhat low rainfall amount where the mean total annual (1982 - 2018) was 870.9 mm. The study area has two rainfall seasons: that is, spring (March, April and May) with total rainfall of 362.9 mm is the main rainy season. In spring, rainfall, averaged for 37 years (1982 - 2018), is the highest in April (153.7 mm). Autumn (September, October and November) with total rainfall amount of 265.6 mm is the second rainy season where it peaks in October (115 mm) (NMA, 2019).
Abaya-Chamo lake-wetland provides multiple ecological and economic benefits to people in the surrounding area. The wetland vegetation, being a vital nesting site and feeding source for hundreds of birds and hippopotamus, supports wildlife and serves as a spawning-area for crocodiles (Unbushe, 2013). Rich bird fauna, sport fishing for Tilapia, Nile Perch and Tiger Fish, the ‘Azo-gebeya’/‘Crocodile Market’ (where crocs are not exchanged rather crowds of crocs are visited), the ‘Forty-Springs’ (from which name of ‘Arba-Minch’ Town was coined) provide special attraction to tourists. Crocodile Ranching/Farming is important income source via tourism and the export of skin of crocs (Legesse, 2007). The lakes also harbor large population of common hippopotamus (Hippopotamus amphibius) and several rare bird species including migratory ones.
Lake Abaya-Chamo wetland revealed rapid change in land uses/land covers due to fast population growth-induced expansion of cultivated land and settlement at the cost forest and shrubland (Bekele, 2001). Crop farming and livestock rearing are important activities in the area surrounding the lakes. Extensive area to the West of Abaya lake was cleared in the 1960s and 1970s for expansion of large-scale farms for producing cotton, banana and other crops (Gelaw, 2019). State farms like Bilate, Arba-Minch and Sile (recently given for private investors) are examples of intensive farming in the plain area adjacent to the lakes.
Agroforestry is the main activity in the alluvial plain of the western shores of the lakes, where it is practiced using rain-fed and irrigation. Fruits (e.g. banana, mango, avocado, papaya, tomato,), cereals (e.g. maize), vegetables (e.g. cabbage, pepper), tuber and root crops (casava, onion, carrot) and cotton are cultivated on the fertile soils adjacent to the wetland (Gelaw, 2007; Gelaw, 2019). Wetlands, forest, woodland and bush-lands have changed to settlement and cropland (Kebede, 2012). These wetlands present a rich biodiversity in western shores of the lake Abaya-Chamo wetland even if it has been extremely impacted by anthropogenic pressure.
Research Design
This study, being viewed via the pragmatic lens, was conducted based on the mixed-methods approach. That is, data acquisition and analyses were carried out using a mixture of methods from both the quantitative and qualitative approaches (Creswell, 2009), were used for statistical based inferences about the LULC dynamics and the degradation of Lake Abaya-Chamo wetland. Methods of the qualitative approaches such as interview and observation were used to check, confirm and strengthen the findings of the quantitative approach. The concurrent embedded model was used to mix the quantitative and qualitative approaches (Creswell, 2009). Cross-sectional survey design was used to acquire and analyze data using both the methods of quantitative and qualitative approaches simultaneously (in parallel).
Acquisition and Processing of Satellite Imageries
Satellite data of Landsat TM of 1990, ETM of 2000, and OLI of 2010 and 2019 of Lake Abaya-Chamo wetland, having spatial resolution of 30 m were downloaded from the website (https://earthexplorer.usgs.gov/) of the US Geological Survey (USGS) (Table 1). Satellite data is the basic source of information which can be used for mapping and change detection in different land use/land cover categories of an area over the period of time. Landsat images captured during January and February were preferred since these dates enable to acquire satellite images free of the impact of cloud cover and to avoid the effect of seasonal variation on the classification of LULC classes. Ancillary data were also utilized during analysis. All data (images) were projected to the Universal Transverse Mercator (UTM) projection system, zone 37N and datum of World Geodetic System-84 (WGS84) to ensure consistency between datasets during analyses.
Table 1. Sensor Type, Resolution, Acquisition Date and Source of Satellite Images used for the Study
Sensor Type
|
Resolution
|
Path/Row
|
Acquisition Date
|
Source
|
Landsat-5 TM
|
30 m
|
169/56, 169/55
|
Jan 12, 1990
|
http://earthexplore.usgs.gov
|
Landsat-7 ETM+
|
30 m
|
169/56, 169/55
|
Jan 27, 2000
|
http://earthexplore.usgs.gov
|
Landsat-8 OLI
|
30 m
|
169/56, 169/55
|
Mar 05, 2010
|
http://earthexplore.usgs.gov
|
Landsat-8 OLI
|
30 m
|
169/56, 169/55
|
Mar 10, 2019
|
http://earthexplore.usgs.gov
|
Source: Own Summary, 2020
The imageries were checked against any defects such as striping. All image scenes were subjected to image processing using ENVI software (version 5.3), and each was clipped using the base-map of lake Abaya-Chamo wetland. Geometric and radiometric corrections were made for the images of the four periods (Table 1). The two scenes (i.e. the one that fall within path 169 and row 56, and the other that fall in path 169 and row 55) of each data set were mosaicked using linear contrast stretching and histogram equalization technique to create a single image covering the whole study area for each period.
Image Classification
Landsat TM of 1990, ETM+ of 2000, and OLI of 2010 and 2019 were also classified using supervised classification (maximum likelihood technique) separately to identify LULC classes of the study area. This method assumes the normal distribution of DN values, allowing the function to determine the probability of a pixel belonging to a specific feature class and assign each pixel to the highest probability class (Lillesand et al., 2004). The classifications were repeated numerous times by adding more training sites so as to come up with satisfactory results. Supervised classification was chosen to compare the outputs with results of the unsupervised classification; this was particularly vital for this study because it identifies and locates LULC types, which are known priori through a combination of interpretation of aerial photography, survey analysis and fieldwork.
In the accuracy assessment, confusion matrices and Kappa coefficient of agreement were calculated for each classification map. Estimation of Kappa coefficients yields statistics, which are measures of agreement or accuracy between the remote sensing–derived classification map and reference data (as shown by the major diagonal) and the chance agreement, which is indicated by the row and column totals (referred to as marginal) (Jensen, 2009). The classification results were compared with the ground truth (data) to confirm accuracy of the classification process. It is a way of assuring how many ground truth pixels were classified correctly, and how much errors were propagated during data acquisition, analysis and conversion (Edwards et al., 1998).
Accuracy Assessment
The accuracy of LULC maps produced was evaluated using overall accuracy (OA), producer’s accuracy (PA), user’s accuracy (UA) and Kappa statistics. PA quantifies the error of omission, while UA quantifies error of commission. Kappa is another method of expressing classification accuracy as it measures the chance agreement. Accuracy assessment was run in order to measure (statistically) the level of accuracy and degree of acceptance of analysis results of the GIS and remote sensing-based LULC classification and change detection of Abaya-Chamo lake-wetland, Southern Ethiopia (Table 3). In this study, reference data were collected during field work using Global Positioning System (GPS) and the reference points were independent of the ground truths that are used in the classification scheme. About 596 GCPs were collected from the field for accuracy assessment. Besides, Google Earth was also used to aid the validation process. Accordingly, the overall accuracy, Kappa coefficient, producer’s accuracy and user’s accuracy were computed from the confusion matrix. Kappa is expressing classification accuracy as it measures the chance agreement. It has been found to be stronger than the overall accuracy of images (Jensen 2005; Lillesand et al. 2014). The Ǩ ("KHAT') statistic is a measure of the difference between the actual agreement between reference data and an automated classifier and the chance agreement between the reference data and a random classifier (Jensen, 1996). Conceptually, Ḱ can be defined as:
………… (Eq.1)
Where K is Kappa coefficient, r is the number of rows in the matrix, xii is the number of observations in row i and column i (the diagonal elements), xi+ are the marginal totals of row i, x + i are the marginal totals column i, and N is the total number of observations (Bishop and Fienberg 2007). In reality, the value of Ǩ usually ranges between 0 and 1. Kappa coefficient of is calculated as follows:
…………… (Eq.2)
Where: 𝑁 is the total number of observations in the entire error matrix, 𝑘 is the total number of classes or categories, 𝑥𝑖𝑖 refers to the number of observations correctly classified for a particular category, and 𝑥𝑖+ and 𝑥+𝑖 refer to the marginal totals for row i and column i associated with the category. Assessing the overall levels of accuracy of the supervised classification for the years 1990, 2000, 2010, and 2019 was found as 88.9, 90.21, 91.20 and 97.92% respectively by adopting confusion matrix technique and 0.887, 089, 0.884 and 0.968 of kappa index value (Table3).
Collection of Field Data
Reference data were collected for training and validation of each LULC type of Abaya-Chamo lake wetland for each satellite image in each period. Geographic locations of ground truth LULC classes, used to calibrate the classification procedure, were identified using high spatial resolution imagery made freely available through Google Earth Pro. About 596 reference samples were derived from the LULC of the lake-wetland in 2019 and via the support of Spot map of 2019. Reference data for 2019 were collected directly from the field between September 2018 to February 2019 using handheld GPS. About 100, 100, 100, 100, 96 and 100 GCP samples were collected for 1990, 2000, 2010 and 2019, respectively, which were used for accuracy assessment.
Verification of the mapped lake-wetland was done through field visits, comparison with features on Google Earth and the DEM, and the use of prior knowledge of features present in the study area. The ground-truth points were randomly chosen. Ground-truthiness of the mapped features was necessary for the verification and accuracy assessment of the features.
Change detection
To analyze the patterns of LULC change, post classification comparison approach was performed. Tis technique provides the detailed “from-to” information and minimizes the possible effects of atmospheric variations and sensor differences (Lu et, al. 2004). Therefore, Landsat imageries of the three reference years were first independently classified. Now, the classified imageries were compared and change statistics were computed by comparing image values of one data set with the corresponding values of the second data set. The comparison values were summarized and presented in terms of area change in hectares, percentages and rate of change. The extent of area change (total change) for each LULC class was obtained by subtracting the area of initial year (oldest date) from the value of recent date (final year) of the study period.
Arc Map GIS 10.5 raster calculator was used to perform the change detection analysis. Change detection analysis was applied on results of the supervised classification about the six LULC classes of Abaya-Chamo lake-wetland for periods 1990 -2000, 2000 -2010, 2010 - 2019 and 1990 - 2019. The magnitude of area change of each LULC class in each period was calculated as follow:
M = (A2 – A1)/A1*100 ……………… (Eq.3)
Where: M is magnitude of area change of a LULC class in a period, A1 is area (ha) of the LULC class in the initial or earlier year, and A2 is area (ha) of the same LULC class in the recent year. The annual rate of change of each LULC class for each period was computed as:
R = (A2 – A1)/t*100 …………... (Eq.4)
Where: R is annual ‘Rate’ of area change (in ha and %), A1 is area (ha) of a LULC class in the initial year, A2 is area (ha) of the LULC class in the recent year, and ‘t’ is the time-interval between the initial and recent years.
Change matrix model (the raster calculator) was used to compute the area change from one LULC class to another type between the periods accounted in the study. The magnitude of change for each period was statistically tested using the Wilcoxon Signed-ranks test. The Wilcoxon Signed-ranks test is a non-parametric statistical test used to assess the difference between two conditions where the samples, in this case change of LULC class, are correlated. The data sets can be compared repeatedly over consistent periods (between initial and recent years).
Data Analysis
LULC changes of Lake Abaya-Chamo wetland were analyzed using GIS and remote sensing techniques. Different spectral signatures of similar pixel samples were selected from satellite imageries using the maximum likelihood method, which served as a separability measure for different land use/land cover classes which were later on grouped with spectrally identical signatures. Determination of appropriate classes was done based on level ‘I’ of the LULC classification and six classes were identified (Table 2). Computer aided interpretation of images was conducted using environmental resources data analysis system (ERDAS) Imagine 2014, ArcGIS 10.5, GPS (Garmin 5.1)-based data and environment for visualizing images (ENVI) 5.0 software, which were used for satellite image processing, classification of LULC, accuracy assessment and analysis of the wetland dynamics. Microsoft excel was also used for analysis
Table 2. Contextual Description of the LULC Classes of Abaya-Chamo Lake Wetland
LULC type
|
Description/Definition
|
Agriculture
|
Farmland used for growing cereals, tuber and root crops, agroforestry practice and horticulture including currently uncultivated (arable) land and fallowed plots.
|
Shrubland
|
Area covered with more of short, hard woody stem trees (bushes), limited herbaceous plants (shrubs) and isolated trees which often are mixed with undergrowth of grasses.
|
Forest
|
Area of dense, tall trees having interlocked canopies, including woodland and riverine forests
|
Swamp/
wetland
|
Is spongy, soft, wet/marshy land saturated with water, adjacent to Abaya and Chamo lakes, and on the banks of tributary-rivers (of the lakes) having permanent and seasonal grasses.
|
Water body
|
Area with temporary and permanent water cover which includes lakes, intermittent ponds and other areas with shallow water cover.
|
Settlement
|
Consists of homesteads of rural villages and buildings of urban areas (with commercial and residential purposes), camps, warehouses, roads and other infrastructures
|