Assessment of Land Cover and Land Use Change Dynamics in Kibwezi Watershed, Kenya

Land use and land cover (LULC) parameters influence the hydrological and ecological processes taking place in a watershed. Understanding the changes in LULC is essential in the planning and development of management strategies for water resources. The purpose of the study was to detect changes in LULC in the Kibwezi watershed in Kenya, using geospatial approaches. Supervised and unsupervised classification techniques using remote sensing (RS) and geographical information system (GIS) were used to process Landsat imagery for 1999, 2009, and 2019 while ERDAS IMAGINE™ 14 and MS Excel software were used to derive change detection, and the Soil and Water Assessment Tool (SWAT) model was used to delineate the watershed using an in-built watershed delineation tool. The watershed was classified into ten major LULC classes, namely cropland (rainfed), cropland (irrigated), cropland (perennial), crop and shrubs/trees, closed shrublands, open shrubland, shrub grasslands, wooded shrublands, riverine woodlands, and built-up land. The results showed that LULC under shrub grassland, urban areas, and crops and shrubs increased drastically by 552.5%, 366.2%, and 357.1% respectively between 1999 and 2019 with an annual increase of 35.55%, 35.38%, and 33.86% per annum. The area under open shrubland and closed shrubland declined by73.7%, and 30.4% annually. These LULC transformations pose a negative impact on the watershed resources. There is therefore a need for proper management of the watershed for sustainable socio-economic development of the Kibwezi area.


Introduction
Te watershed comprises both biotic and abiotic components, including human beings confned within a defned boundary. Te watershed provides essential services necessary for the well-being of society [1]. Despite the benefts of watersheds, the quantity and quality of services have been declining over time [2], which are attributed to land use/ cover, climate change, and the demand for these services from the rapidly increasing population.
Land use and land cover (LULC) and its resources support for social, cultural, material, and spiritual needs of human beings and have culminated in a signifcant transformation of the resource [3]. Te LULC change on the land surface is noticeable and takes place at a high rate, consequently limiting the potential of natural ecosystems to provide environmental services [4]. Te LULC change has been caused by the interaction between demographic, socioeconomic, and biophysical changes [5,6] and is believed to be a major force both locally and globally infuencing environmental change [7,8]. Te increasing alteration of the land surface due to anthropogenic activities have greatly afected the efectiveness of global environmental systems.
Rapid LULC change is occurring largely in the developing world [9,10] with natural vegetation converted to agricultural land [11,12] which has resulted in a decline in water, soil, and vegetation resources [13]. It afects hydrological processes in the watershed such as evapotranspiration, interception, and infltration, consequently altering surface and subsurface fows [14,15]. Te LULC through deforestation is believed to be the driver of climate change [7,8] that could result in loss of biodiversity and consequently have a negative impact on eco-tourism.
Te LULC changes originate from a local level, but due to their speed, extension, and intensity, they have various and critical global efects felt on the resources. Te everincreasing change is worrying and can have a signifcant impact on the local, regional, national, and global environment [16,17].
Te stakeholders, mainly scientists and water planners involved in decision making need the LULC information to determine the change that has occurred in the natural resources [18]. Tey consider LULC as a process that afects the natural environment and socio-economic situation signifcantly at local and global conditions. Te LULC information helps to understand and extend which human beings have impacted on the natural environment assessment, especially at the watershed level. Conventional methods for capturing and analyzing multidisciplinary spatial data are time-consuming and costly and have been replaced by digital, robust, and efective technologies such as remote sensing (RS) and geographical information systems (GIS). RS and GIS techniques have been used widely in diferent felds; agriculture [18] and environments [19] (T. Fung and E. Ledrew 1987) and integrated ecoenvironment assessment. However, many studies have concentrated on LULC due to its extensive impacts on ecology and vegetation [17] with limited studies at a watershed level which have a great impact on the hydrological processes. Soil and Water Assessment Tool (SWAT), a hydrological model, on the other hand, is a GIS interface that analyses raster data from RS to delineate the extent of the watershed based on the outlet details [20]. Te SWAT model has been used extensively globally due to its ability to capture the heterogeneity of the watershed by delineating the watershed using the river gauging station outlet.
Makueni County has undergone rapid land cover changes [21] attributed to population increase, urbanization, and agricultural development activities. Te changes have resulted from LULC modifcation over a period of time believed to have increased land degradation and impacted negatively on water resources in the area. Te LULC conversion may alter the hydrology of Kibwezi watershed through the intervention of precipitation, surface fow, infltration, evapotranspiration, and aquifer storage. Te study, therefore, aimed at delineating the Kibwezi watershed and comprehensively classifying and detecting the conversion matrix of the LULC using RS, GIS, and SWAT tools. Tis was achieved through classifying and analyzing LULC using 30 m resolution Lansat imagery for a period of 20 years (1999-2019) and using a 30 m resolution DEM to demarcate and characterize the watershed. It aimed at addressing the following questions: (i) what are the LULC categories present in Kibwezi watershed, (ii) what are the trends in LULC classes within the study period, and (iii) the major conversions in LULC that have taken place in Kibwezi watershed from 1999 to 2019. Te results can provide a guide to decision-makers and managers in the layout of structures, monitoring, and management of the watershed.

Location. Kibwezi watershed is located in Semi-Arid
Eastern Kenya, Makueni County, in the lower part of the River Athi basin as presented in Figure 1. Te watershed spans an area of approximately 700 km 2 in Kibwezi west and Kibwezi east subcounties and is described by the coordinates 37.73°E to 38.15°E and 2.32°N to 2.58°N at an altitude between 647 m and 1993 m above sea level.

Climate.
Kibwezi watershed typically semiarid climate characterized by varied and erratic rainfall. It receives a bimodal type of rainfall; March-May for long rains, and November-December for short rains with an annual average of 600 mm, while the minimum and maximum temperatures are 17°C and 27°C, respectively [22]. River Kibwezi is the main distributary in the Kibwezi watershed and is approximately 25 km long. Te watershed is also characterized by fat topography with a deposit of alluvial soil along the river valleys and sandy soils in the plain. Shallow to moderately deep soils are common and the landscape is bordered by the Yatta plateau on the eastern side.
Te prevalent vegetation of the Kibwezi watershed is savanna and shrub-woody vegetation with various species. It varies according to rainfall amount, soil type, and based on the species of grass or shrubs.

Land Use/Land Cover Maps
(1) Preparation of Field Base Maps. Landsat images covering the Kibwezi watershed were obtained from the USGS Global Visualization Viewer (https://glovis.usgs.gov). Te Landsat images were obtained in raster format at 30 m × 30 m resolution and were extracted for a ten-year interval from 1999 to 2019 for the study area. Te Landsat TM5 was available for the period 1999 and 2009 while Landsat 8 satellite imagery was available for 2019. Te images obtained were clear and free from cloud cover. Te individual spectral bands that were downloaded from the website were stacked together to form multispectral imagery scenes for the three periods, and thereafter subsets of the watershed (area of interest) were extracted from each of the multispectral Landsat scenes. Te digital image processing software ERDAS IMAGINE ™ 14 and ArcGIS 10.2 were used for the processing, analysis, and integration of spatial data to arrive at the objectives of the study. Te LULC processes development, included data acquisition preclassifcation to identify possible cover polygons, ground truthing to validate LULC classes, and development of fnal LULC maps ( Figure 2). Te imageries were terrain-corrected (ortho-rectifed) and hence were directly used for classifcation and GIS analysis. Te characteristics of satellite imagery are presented in Table 1.
Te ERDAS IMAGINE was used to perform ISODATA unsupervised classifcation on the 2019 imagery with 12 clusters/classes which were then classed using a minimumdistance classifer. Once the initial classifcation was completed, the 12 spectral classes were assigned to the 10 information classes in Table 2 based on the Kenya Soil Survey key to physiognomic classes presented in Figure 3. As with almost any automated classifcation technique, the initial raw land cover maps were characterized by speckling produced by the misclassifcation of a single pixel or small groups of pixels and hence the need for cartographic generalization. To eliminate much of the visual noise caused by these misclassifed pixels, the raw maps were converted from ERDAS raster format to ArcGIS shape fle and generalized using a series of both automated and manual procedures. Te eliminate tool in the ArcGIS ArcToolbox was used to remove the small sliver polygons by merging them with adjacent polygons. Te remaining coded polygons were the preliminary land cover polygons.
Te same procedure was also adopted to generate the 1999 and 2009 land use/cover maps. Te existing digital thematic data on roads, rivers, administrative boundaries, and settlements from the KSS GIS database were projected to WGS 84 UTM zone-37 south coordinate system, which was adopted for the study. Tese data together with preliminary land cover polygons and randomized observation points were overlaid on the Landsat 8 imagery of 2019 to produce a feld base map as shown in Figure 4. A preliminary legend was also constructed on the basis of the diferent land cover classes. Tis map together with GPS and digital Google Earth images assisted in the location of sites and was vital for the ground truthing and postclassifcation process.
(2) Field Data Collection Methods. Te unsupervised image classifcation method was carried out prior to the feld visit to determine the strata for ground truthing. Te aim of collecting data was to validate land use and land cover interpretation from the satellite images of 2019 and for a qualitative description of the characteristics of each land use/land cover. A random sampling procedure was applied to ensure representative samples in all preclassifed LULC polygons within the watershed. Te number of samples per LULC class was determined based on the size of the unit and they were equidistant from one another.
Te determination of position and general orientation was carried out using a Trimble SB hand held GPS receiver. Te fnal position data were recorded once the GPS's position dilution of precision (PDOP) values were less than three considering the land cover units were derived from Landsat imagery with a spatial resolution of 30 meters. Each time a randomized observation point was approached, a brief description of the land use and land cover observed (percent aerial cover of trees, shrubs, herbs, grasses, bare soil, and others) were recorded on the feld observation forms and geographical coordinates were saved into the GPSdatabank and photographs. Te dominant physiognomic classes (structure) of the vegetation were also recorded. Information on dominant plant species was also recorded. October and November mark the rainy season in lower eastern Kenya and therefore images were chosen during this season in order to best distinguish the spectral signatures of the land cover types since the vegetative growth is at its peak. Also, the near-anniversary dates were chosen for consistency between the two time points.
(3) Postclassifcation. Te image interpretation boundaries which needed adjustment were adjusted accordingly to the base map with reference to feld observations. Te gathered information on sampling points, tracking, and waypoints were used to fnalize the land cover codes used to describe each mapping unit depending on the land use/land cover class. Te land cover maps for 1999 and 2009 were converted from vector format to raster format using the ArcGIS software. Te land cover maps were then recorded into 9, 10, and 10 broad land cover classes respectively (i.e., built-up land, wooded shrublands, etc.), for GIS overlay analysis for change statistics and map production.
Te LULC maps for the Kibwezi watershed were prepared by clipping the mask area of the Kibwezi watershed using ArcGIS for the period 1999, 2009, and 2019.

Accuracy Assessment.
Accuracy assessment in LULC is crucial for the classifed classes before embarking on change detection (Owojori, A and Xie, H. 2005). It assists to assess the quality of data collected in the feld compared to the classifed satellite images to identify any error and its source.

Te Scientifc World Journal
Ground thruthing of LULC classes were compared to results from satellite images Landsat TM5 and Landsat 8. In this study, 547 pixel points were produced during LULC classifcation guided by a minimum sample size of 50 per LULC class [23]. Accuracy was assessed based on visual interpretation and the ground-truth data and the overall results were obtained by dividing the sum of the correctly classifed LULC pixels classes in each sampled unit and a total number of reference pixels which are then captured in the error matrix. Te Kappa test, a nonparametric statistic based on a discrete multivariate technique was used to determine the magnitude of accuracy between referenced and imagery data. Manomani and Suganya [24] have categorized Kappa statistics into six categories; less than 0 indicating no agreement, 0-0.
Where A i is the initial LULC class area (ha), A j is the fnal LULC class area (ha), and n is the number of years between the initial and fnal time period.   Figure 5. Using the DEM as input and the River Kibwezi discharge point to River Athi as an outlet, the SWAT model [20] was used to delineate the watershed presented in Figure 6.  [24]. Te LULC accuracy provides a basis for subsequent analysis of the changes in the watershed.  Table 4. Te percentage area of each LULC class in 1999, 2009, and 2019 indicated that wooded shrublands and cropland (rainfed) occupied the largest share of the watershed, representing on average 45% and 30% respectively across the period. Te classifcation also revealed that the area under built-up land, cropland (irrigated), shrub grasslands, and riverine woodlands each occupied less than 1% of the total watershed across the period. Tables 5 and 6 show the LULC transition matrix from one class category to another by hectares between 1999/2009 and 2009/2019. In  Te Scientifc World Journal       (perennial), and 32 ha of wooded shrubland were converted to built-up land or urban, while a total of 184 ha of cropland (irrigated and perennial) and crops and shrubs/trees were converted to cropland (rainfed). Furthermore, the area under rainfed crops (248 ha) and crops-shrubs/trees (34 ha), 1 ha under wooded shrublands, and 7 ha under riverine woodland changed to cropland (perennial) land. In addition, a total of 489 ha of cropland (rainfed, irrigated, and perennial) and 3437 ha of both open and wooded shrubland were transformed to crops-shrubs/trees land during this period. Te cropland (rainfed, perennial, and crops-shrubs/ trees) lost 770 ha, while open shrubland was reduced by 1,675 ha to wooded shrubland in 2009-2019 period. Table 7 summaries the dynamic changes of diferent land use/land cover from 1999 to 2019. Te results show a continuous increase of area under built-up land, shrub, grassland, and crops-shrubs/trees land, while the area under open and closed shrubs was on a decline. Te biggest increment between 1999 and 2019 was the area under crops-shrubs/ trees (3,606.5 ha), followed by cropland (rainfed) (1,206.5 ha), shrub grasslands (359.7 ha), riverine woodlands (349.2 ha), and built-up land (116.2 ha). Open shrublands area decreased by 4423 ha, followed by closed shrublands (963.3 ha) and wooded shrublands (177.2 ha), while cropland (perennial and irrigated) declined by 44.8 ha and 29.9 ha, respectively, as shown in the table. On the other hand, the area under cropland (perennial) and cropland (rainfed) increased between 1999 and 2009 but decreased between 2009 and 2019, while the area under wooded shrublands and cropland (irrigated) was reverse.

Land Cover Change Detection 1999-2019.
Moreover, from the results, the rate of expansion in shrub grasslands was the highest, followed by built-up land, and crops-shrubs/trees land, while the lowest expansion was in open and closed shrublands between the overall periods from 1999 to 2019.

Discussion
Global environment has undergone tremendous change that signifcantly infuences hydrological processes in the watershed resulting to hydrologic nonstationarity. Te LULC is believed to contribute signifcantly to change in global environment [25]. Remote sensing and GIS tools and integration with SWAT model have proved to be efcient, accurate, and cost-efective and give detailed information to analyze, detect, and monitor LULC changes at the watershed level. Te Kibwezi watershed has witnessed great LULC as shown in the transition matrix (Table 7). In the watershed,  Te results show that there is a great conversion of natural vegetation into cropland and built-up areas in the last 20 years in the watershed. A reduction in shrubs is associated with the expansion in urbanization within the study area. Te Kibwezi town and other trading centres are expanded due to commercial and business activities. Te urbanization is also attributed to increase in the population due to the growth rate in the area and sisal commercial farm within the watershed which provide employment. According to [26], population growth rate of Kibwezi stands at 1.4 per cent per year and a population density of 121 persons per square kilometer. Tis increase in population has led to a high demand for expansion of land through deforestation for the production of cereals, legumes, and horticultural crops for the population and also for commercial purposes. Tis growth has led to the clearing of shrubs for built-up area and also for agricultural activities. Te fndings are similar to [12,27], where population growth and GDP plays a big role in LULC conversion. Table 7 shows that open and closed shrublands have decreased by 5386 Ha while cropland increased by 3606.5 ha between 1999 and 2019 indicating a huge decline in the vegetation cover. Tis decline is attributed to encroachment of the rising population to the watershed despite the land use policy in this place. Te poor enforcement of land use policy may have resulted in decreased vegetation, consequently increased surface runof, and reduction in recharge of the watershed that has led to the drying of the distributaries in the watershed. In addition, riverine woodlands have increased signifcantly during the study period, which may have been as a result of deposition of sediments from the upstream. Te Kibwezi area has been experiencing occasional drought where the rainfalls have been erratic and temperatures increasing [28]. As a meaning of livelihood under this climate change, the population has been forced to clear the existing shrubs for timber, frewood, and charcoal for domestic and commercial use leading to land cover conversion.

Conclusions
Te use of RS and GIS has demonstrated the efectiveness of spatial digital techniques in detecting, assessing, and monitoring the LULC in the Kibwezi watershed. Te Kibwezi watershed was classifed into built-up areas, croplands, crops-shrubs/trees, shrublands, and riverine woodlands of LULC classes. Te diferent class categories underwent varied changes with some maintaining a constant increase or decreases in the area during the 1999/2009 and 2009/2019 period. Overall, the results showed that LULC under shrub grasslands, urban areas, crops-shrubs, and rainfed croplands increased by 552.5%, 366.2%, 357.1%, and 7.0%, respectively, between 1999 and 2019 while LULC under open shrubland, closed shrubland, cropland (irrigated), cropland (perennial), and wooded shrublands decreased by 73.7%, 30.4%, 10.4%, 0.8%, and 0.6%, respectively, during the same period. Te shrub grassland, urban areas, and crops and shrubs had an annual increase of 35.55%, 35.38%, and 33.86%, respectively, while open shrubland and closed shrubland declined by 73.7%, and 30.4% per annum. Te increasing rate of urban areas and cropland at the expense of vegetation will infuence land degradation, runof, and ground water recharge and results in decreased water and food productivity. Tese LULC transformations can assist the watershed managers and policy makers in developing management strategies for the watershed for sustainable socio-economic development

Data Availability
Te data that support the research fndings of the land use/ land cover (LULC) in the Kibwezi watershed are included within the article. However, a supplementary fle is attached, which indicates the land cover category codes and their land use description for the same. It also provides LULC for the period 1999, 2009, and 2019 with their respective areas/size in hectares for the study area.

Conflicts of Interest
Te authors declare that they have no conficts of interest.