Surface Water Quality Response to Land Use Land Cover Change in an Urbanizing Catchment: A Case of Upper Chongwe River Catchment, Zambia

The Upper Chongwe River Catchment has recently been overexploited for water resources with increased complaints by various water users about the deteriorating quality of surface water within the sub-catchment. This study was motivated by the need to investigate and understand the response of surface water quality to land use land cover (LULC) change due to urbanization. Water samples, collected at 9 sampling sites from 2006 to 2017, were analyzed for water quality using the weighted arithmetic water quality index and trend using the Mann-Kendall statistics. LULC change is detected and analyzed in ERDAS Imagine 2014 and ArcGIS 10.4 using 2006 Landsat 5 TM and 2017 Landsat 8 OLI imageries. The relationship between LULC change and water quality was performed with multiple regression analysis and Pearson correlation. The results reveal that Built-up area, Grassland and surface water increased by 5.48%, 13.34% and 0.03% respectively while Agricultural land and Forest Land decreased by −13.41% and −5.42% respectively. The water quality index ranged from 43.04 to 110.40 in 2006 and from 170 to 430 in 2017 indicating a deterioration in the quality of surface water from good to unsuitable for drinking at all the sampled sites. Built-up/bare lands exhibited a sigHow to cite this paper: Nguvulu, A., Shane, A., Mwale, C.S., Tena, T.M., Mwaanga, P., Siame, J., Chirambo, B., Lungu, M., Mudenda, F., Mwelwa, D., Chinyanta, S., Kawala, J., Bowa, V.M., Mutambo, L.S., Okello, N. and Musonda, C. (2021) Surface Water Quality Response to Land Use Land Cover Change in an Urbanizing Catchment: A Case of Upper Chongwe River Catchment, Zambia. Journal of Geographic Information System, 13, 578-602. https://doi.org/10.4236/jgis.2021.135032 Received: July 9, 2021 Accepted: October 24, 2021 Published: October 27, 2021


Introduction
River catchments play a significant role in the provision of water for domestic, agricultural and industrial purposes. Nevertheless, as observed by Reed et al. [1] and Avivor & Gordon [2], land use dynamics within river catchments have negative repercussions on river health. Usually triggered by population growth within a catchment area, urban land use dynamics often present themselves in form of large swaths of land being cleared to meet the increasing need for built-up area and agricultural use on one hand and high demand for water on the other. Such landscape dynamics tend to alter the natural ecosystem leading to loss of biodiversity, cause water fluxes and compromise water quality in lakes, rivers and streams within a catchment [3] [4] [5]. Therefore, revealing the relationship between land use change and water quality is of great significance to watershed protection.
Several studies have revealed that there is a significant correlation between land use land cover change and water quality. Hua [6] observed that land use dynamics in a watershed negatively impacts on water quality and indirectly affect the nature of a watershed ecosystem. This could arise from such factors as loss of wetlands, discharge from septic and sewer systems, airborne discharge from vehicles and wood-burning stoves, increased sediment and nutrient loading. Study by Huang et al. [7] has shown that the relationship between land use land cover (LULC) change and surface water quality parameters is generally Journal of Geographic Information System chment has been experiencing increased development activities that have triggered extensive LULC change in the recent past [8]. Extensive irrigation zones have mushroomed in the north, northeast, east, south and northwestern parts of the catchment while the central and western parts have seen increased urbanization [9]. Irrigated agriculture in the catchment depends largely on dammed water along the Chongwe River and its main tributaries. The catchment is under water stress and consequently, a great majority of the population lack access to good drinking water and good sanitation [10] [11] [12]. The challenges of water scarcity in the catchment seem to be compounded by severe biochemical and sewage pollution in the Chongwe River and its tributaries upstream.
In studying the relationship between LULC and water quality, several researchers have applied various statistical techniques including multivariate methods, linear models and redundancy approaches. For example, Li et al. [13] (2008) studied water quality in relation to land use and land cover in the Upper Han River Basin in China using multivariate analysis. From their work, it was concluded that agricultural and urban areas contribute to water quality degradation while forest cover plays an important role in keeping the water clean. Chen et al. [14] (2020) used Pearson correlation to determine the relationship between Land Use Change and Water Quality of the Mitidja Watershed in Algeria. From the results, urban settlement area was found to be a predictor for NH 4 -N, BOD 5 , COD, SS, PO 4 -P, DO and pH, while vegetation was a predictor for NO 3 -N. As observed by Tu [15], the relationship between land use and water quality varies significantly over space and geographic locations mainly due to the different catchment physiognomies and pollution sources. Other studies have, in fact, observed that land use on riparian-buffer scale influences water quality better than on a catchment scale [16] [17], while others hold that catchment scale better influences the water quality [18] [19].
Several water studies in the Chongwe River Catchment have been undertaken.
However, these studies have focused mainly on groundwater quality [20] and development of a groundwater information and management program for the Lusaka groundwater systems [21]. The outputs of these studies include 1) a land use map of Lusaka and surroundings [22], 2) discharge measurements and rating curves [23], 3) description of physiography, geology, climate, hydrology and The Chongwe River catchment can be divided into upper, middle and lower parts. The predominant land use in the upper and middle half is agriculture and livestock production. About 6500 ha of land is now cultivated under a variety of irrigation schemes and methods in both large-and small-scale farming. The main crops grown are wheat, maize, beans, groundnut, cotton, vegetables, flowers, and horticultural crops. The other middle half is predominantly a built-up area. The lower part is mainly forest and bushland providing valuable habitat for wild animals and birds. It is also one of the ecotourism sites in Zambia. Small scale river bank cultivation and fishing are common practices by the local community in the lower part providing a means of income and household food security.
The climate of Chongwe River catchment is described as humid subtropical,

Data
Two dataset types, i.e. spatial and nonspatial datasets, were used in this study. The spatial dataset comprised of 1) a Shuttle Radar Topography Mission database, water quality reports from the Zambia Environmental Management Agency (ZEMA) and authors' own water quality field and lab measurements for 2017.

Chongwe River Catchment Delineation
First, a field reconnaissance survey was conducted to have a visual understanding of the study area. Then the 30m spatial resolution SRTM DEM was prepro-

LULC Classification and Change Detection
Supervised image classification was performed on the Landsat 5 TM and 8 OLI multispectral images for the reference years 2006 and 2017 using the Maximum Likelihood Classifier (MLC) parametric decision rule. The MLC calculates a Bayesian probability function for each pixel from the inputs for classes established from training sites and then assigns the pixel to a class to which it most probably belongs [27] [28]. The classification was based on a predefined LULC classification scheme developed from the authors' field knowledge of the catchment. The scheme consisted of five classes namely built-up area, agriculture land, forest land, ranch/grassland and water bodies. After the class signatures were generated for each class, the images were then classified with appropriate colors and names for easy interpretation of classes. This process was repeated several times while comparing with a Google Earth images before settling for a suitable classified image on which to generate statistics for each class. The change detection analysis was done by comparing the 2006 and 2017 thematic images using confusion matrix operation. All the above operations were done in ERDAS Imagine 2014 and ArcGIS 10.4. The changes in LULC were expressed both in absolute and percentage proportions [29].

LULC Classification Accuracy Assessment
The image classification accuracy was assessed using the overall accuracy, producer's accuracy, and user's accuracy were determined in addition to the Kappa Coefficient (K) [30]. The Kappa coefficient is widely used because all elements in the classification error matrix, and not just the main diagonal, contribute to its calculation and because it compensates for change agreement. The Kappa coefficient lies typically on a scale between 0 (no reduction in error) and 1 (complete reduction of error). The latter indicates complete agreement, and is often multiplied by 100 to give a percentage measure of classification accuracy. In practice, the agreement is taken to be strong when K is greater than 0.80 (80%), moderate when K values fall between 0.40 (40%) and 0.80 (80%) and poor when K values are less than 0.40 (40%) [31].

Selection of Water Quality Parameters, Sampling Sites and Frequency of Sampling
The parameters selected for water quality analysis were pH, temperature, turbid-  were grouped into three zones namely 1) Upstream (Sites S1 -S4) 2) Midreach (Sites S5 -S7), and 3) Downstream (Sites 8 -9) Zones. The frequency of sampling was determined based on the reference data comprising of the WARMA quality monitoring data for 2006. were filled to the bream to avoid air contamination [35]. The samples were taken to the laboratory for analysis within 8 h.

Water Sampling and Analysis of Physical, Chemical and Biological Quality Parameters
On-site analysis equipment for pH, TDS, Temperature, and Turbidity were calibrated before analysis while their probes were rinsed with distilled water before taking the readings. Sampling time was conducted in the morning to avoid extreme temperature variations. After sampling, each of the triplet 500 ml were clearly labeled and stored in cooler boxes filled with ice cubes as a measure of maintaining a refrigerant temperature before laboratory analysis. Additionally, the analyses of physicochemical and biological parameters were carried out with the aid of local and international water quality compliance limits and guidelines [32] [36] [37]. Table 1 summarizes the methods, equipment and protocols used in both the in-situ and laboratory analyses.

Trend Analysis of Water Quality
The trend of the measured values of water quality variables is analyzed using the The expected value E(.) of S is and its variance, Var(.) is as in where t indicates the extent of any given time and t ∑ denotes the sum across all the ties in the water quality data. For n > 0, the standard normal variant is calculated as in Under the Kendall's t test, a positive value of S in Equation 4 indicates an increasing trend whereas a negative value indicates a decreasing trend. Under this (t) test, the following were the hypotheses: Null hypothesis, H0: There is no trend in the water quality. Alternative hypothesis, Ha: There is a trend in the water quality. The decision to reject the null hypothesis proceeds in the same fashion as standard tests of hypotheses of significance. That is, if the p-value of the test is less than the level of significance, the null hypothesis is rejected, but if the reverse is the case, then the null hypothesis is not rejected. Kendall's t test of significance was carried out in XL Stat 2014. All tests of significance were two-sided and considered significant at the 0.05 level.

Multivariate Analysis
In this study, the multivariate statistical techniques of Factor analysis (FA) and Hierarchical Cluster analysis (HCA) are used. First the raw data is standardized by subtracting the mean of the data set from each variable and dividing by the standard deviation to produce a normal distribution. The HCA approach using Ward Method is then applied to group sampling sites into clusters and homo- The Principal Component Analysis (PCA) is then used to evaluate selected parameters through a correlation matrix. The PCA is specifically used to extract the factors with correlated values while at the same time giving spatial and temporal changes in the water quality [41] [42]. The PCA is expressed mathematically as in where, z = component score, a = component loading, x = measured value of variable, i = component number, j = sample number, and m = the total number of variables. Factor analyses were used to determine the pollution factors affecting the water quality among the sampling sites where factor loading value > 0.75 is described as "strong" loading, 0.75 -0.50 is "moderate" and 0.50 -030 is described as "weak".

Weighted Arithmetic Water Quality Index
The weighted arithmetic water quality index (WQI A ) [43] [44] [45] is used to rate the water quality of our study area. The application of WQI A is done to determine the suitability of surface water for human consumption [46]. The WQI A can be expressed mathematically as in where n is the number of parameters, w i is the relative weight of the ith parameter and q i is the water quality rating of the ith parameter. The unit weights (w i ) of the various water quality parameters are inversely proportional to the recommended standards for the corresponding parameters [47]. The value of q i is calculated as in where V i is the observed value of the ith parameter, S i is the standard permissible value of the ith parameter and V id is the ideal value of the ith parameter in pure water.

Determining the Relationship between LULC Change and Water Quality
The relationship between land use and water quality was performed with multiple regression analysis and Pearson correlation. The correlation analysis developed a correlation matrix between the land use types and water quality in order to determine the type of interaction between them. Moreover, the multiple regression analysis explained the magnitude and influence of the land use (predictor variables) on water quality parameters (response variables). Stronger positive correlation showed values closer to 1 and those closer to 1 showed a stronger negative correlation between variables.

Land Use Land Cover Classification and Accuracy Assessment
The LULC of the study area for the years 2006 and 2017 and the LULC change are as in Table 2. Agriculture land, forest land and surface waterbodies decreased while built-up area and grassland increased during the study period.
Built-up and grassland increased by 5.35% and 13.31% while agricultural land More agricultural land is getting converted to urban built-up land while forest land is being cleared mainly for subsistence farming and charcoal burning [47] [48].
The above LULC results were adopted for further analysis based on the image classification accuracy assessment using the criteria shown in Table 3.

Water Quality Analysis
pH: pH measurements taken on site ranged between 5.7 and 8.3 during the sampling period at all the sites. The values shown in the month of July on the upstream sites such as Ngwerere at the Weir and at Kalimba farm were as low as 5.7.
These pH values were below the aquatic ecosystem limits according to guidelines prescribed in [36] [37]. The other sites were relatively within permissible limits for  The results fell below the WHO and ZABS drinking water limits which are set 1500 and 1000 respectively. However, the maximum values at the sites 2, 3 and 4 reflect the pollution burden to aquatic systems exerted by partially treated sewage discharged into the Ngwerere River at those sites since site 2 and 3 are almost at the discharge points.
DO: The values of dissolved oxygen ranged from 4.8 to 7.8 mg/l with the lowest values read at site 1 and 2. The WHO and ZABS maximum standard limit for DO is set at 5 mg/L. Therefore, site 1 and 2 readings were below the acceptable limit. The low levels of DO shown mainly in the Ngwerere River are an indication of high level organic matter discharged into the river by the commercial facilities located close to Site 1 and the Sewage treatment plant near Site 2 and Site 3 [12] [49]. This however, did not provide sufficient information to conclude heavy metal contamination is not present in the catchment. Perhaps, performing bottom sediments analysis for heavy metals would lead to a different inference. Table 4 shows the descriptive statistics of the physicochemical parameters and heavy metals that were analyzed in this study. It can be seen from the Table that the parameters, especially the chemical ones, were generally below the minimum permissible limits.

Trend Analysis of Water Quality
The results of the Mann-Kendall trend test on water quality using available data in the WARMA database for July 2006 to July 2016 combined with this study's monitoring exercise for July 2017 and 2018 for the Chongwe River Catchment are given in Table 5. The tests were all done at the α-value of 0.05 significant A. Nguvulu et al.

Hierarchical Cluster and Principal Component Analysis
The results of the Hierarchical cluster analysis yielded two clusters as shown in Figure 3. As can be seen from the figure and based on the Euclidean distance, cluster 1 consisting of 6 Sites (S5, S4, S1, S8, S7, S6, and S9) was characterized by low Euclidean distance while Cluster 2 consisting of 2 Sites (S2 and S3) was characterized by a high Euclidean distance.

A. Nguvulu et al. Journal of Geographic Information System
Principal Component Analysis in Table 6 shows extraction of five factors. Factor 1 accounted for 33.37% of the total variance in the water quality parameters, with strong factor loadings of electrical conductivity, TDS, Sodium, Potassium and Chloride and moderate loading of NO 3 and  of these ions could be due to the presence of organic matter and organic acids which indicates the influence of anthropogenic activities and geological formations on water quality. The concentration of NO 3 could be due to nitrification taking place in the rivers while the concentration of 2 4 SO − could be due to fertilizer from runoff which indicates human influence especially agricultural activities. Factor 2 accounted for 21.69% of the total variance with strong factor loading of Pb and moderate factor loadings of NH 4 , Turbidity, and Fe. This could be attributed to the sediment runoff from loose soils on agricultural lands into the rivers. Factor 3 accounted for 12.56% of the total variance in the water quality parameters and revealed moderate factor loadings of pH and DO indicating that despite domestic waste being discharged into the river, there is oxygen availability and sufficient pH levels to support aquatic life. Factor 4 accounted for 10.02% of the total variance with a moderate loading of temperature indicating the effects of shading of riparian forest which influences water temperature and aquatic productivity. Factor 5 accounted for 8.50% of the total variance with moderate loading of DO from domestic waste discharge. Similar findings were reported by Ayeni and Soneyanu [50] that land uses such as domestic and agricultural activities strongly influence the variation of the quality of surface water.
The variables from component loadings shown in Table 6 where then grouped according to their designated components and classified their factor loading as shown in Table 7. The factor loadings were classified "Strong", "Moderate" and  Table 8 shows the water quality index calculated and ranking of water quality at

Relationship between LULC and Water Quality
The relationship between land-use and water quality is presented in   land and TDS, conductivity, Cl, and turbidity are most likely to be ascribed to sediment runoff from construction sites, weathering of rocks and erosion from bare areas [56]. Road salts can be a great contributor to chlorides in receiving waters [57]. The positive correlation between agricultural lands and turbidity and Fe could likely be attributed to the sediment runoff from loose soils on agricultural lands into the river [55]. Forest and grassland depicted negative correlation with water quality parameters. This indicates that as forest land increases, degraded water quality decreases and vice versa. As observed by Sliva & Williams [52], vegetation can improve water quality deterioration by absorbing nutrients and blocking sediment runoff. Tu [15] revealed similar results, indicating forest and grasslands as indicators for good water quality.

Conclusion
The aim of this study was to analyse and understand the response of surface water quality to LULC change in the rapidly urbanizing Chongwe River Catch- The findings and methods used in this study could be useful to identify the sources of pollution and improve water management and land use planning within the catchment.

Recommendations
From the conclusions drawn, the following measures are recommended: • Buffering of the riparian zone should be implemented through Water User Association groups around the catchment. • Extending the enforcement of the Water Resources Management Act to monitoring and mitigation of surface water as its neglect has even more disastrous effects on the catchment considering its population growth rate against water demand among users for domestic, industrial and agricultural purposes.
• Investment in an integrated water quality management system which encompasses physical, chemical and biological indicators of water quality and prioritizes monitoring and mitigation according to the land use land cover distribution in the catchment, equipped with an up to date geospatial, physic-chemical and biological database.
• Establish an integrated sewer waste treatment within the catchment where the sludge could be scoped and properly managed while the effluent be properly monitored using real time monitoring at the point of discharge.