Subtle Impacts of Temperature and Rainfall Patterns on Land Cover Change Overtime and Future Projections in the Mara River Basin, Kenya

The interactive and cumulative effect of temperature and rainfall on land cover change is a priority at global, regional and local scale. This study examined changes in six land cover categories (forestland, grasslands, shrub land, bare land, built-up areas and agricultural lands) in four sub-catchments (Amala, Nyangores, Talek and Sand River), of the Mara River basin over a 30-year period (1987-2017) and made predictions of future land cover change patterns. Landsat Imageries of 90 m resolution were retrieved and analyzed using ArcGIS 10.0 software. Relationship between NDVI, temperature and precipitation was determined using Pearson’s correlation coefficient, while Markov chains analyses were performed on different land cover categories to project future trends. Results showed low to moderate (R = 0.002 to 0.6) trends of change in NDVI of different land cover categories across all sub-catchments. The greatest change (R 0.34 to 0.5) was recorded in bare land in three of the four sub-catchments studied. Precipitation showed a strong positive correlation with built-up areas, forestlands, croplands, bare land, grasslands and shrub lands, while temperature correlated strongly but negatively with the same land cover categories. The change detection matrix projected significant but varying changes in land cover categories across the four sub-catchments by 2027. This study underscores the impact of changing climatic factors on various land cover categories in the Mara River basin sub-catchments, with different land cover categories exhibiting strong positive sensitivity to high precipitation and low temperature and vice-versa. How to cite this paper: Mngube, F.M., Kapiyo, R., Aboum, P., Anyona, D. and Dida, G.O. (2020) Subtle Impacts of Temperature and Rainfall Patterns on Land Cover Change Overtime and Future Projections in the Mara River Basin, Kenya. Open Journal of Soil Science, 10, 327-358. https://doi.org/10.4236/ojss.2020.109018 Received: July 17, 2020 Accepted: September 7, 2020 Published: September 10, 2020 Copyright © 2020 by author(s) and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY 4.0). http://creativecommons.org/licenses/by/4.0/ Open Access


Introduction
Changes in land cover (LC) pattern globally reflect the interaction between human activities and the natural environment [1]. Climatic and land cover changes are increasingly becoming important components of sustainability especially for aquatic ecosystems [2] [3]. Due to anthropogenic activities, the Earth surface is under continuous alteration that impacts heavily on the natural environment [4]. Studies have linked changes in land cover to increased anthropogenic activities [5]. While the role of climate on land cover change has been extensively researched and discussed at the global and regional scales, knowledge of their impact at the local sub-catchment scale is limited, disjointed and anecdotal to draw any meaningful conclusions.
Studies show that the impacts of temperature and precipitation on land cover change are complex [6] and therefore require area specific studies to understand their correlation. This is because global analysis of the relationship between NDVI, precipitation and land surface temperature gives different views. While some researchers have not found any significant correlation [7], others have reported negative or positive relationships between climatic factors and land cover categories [8] [9]. A study in the northeast China by Luo et al. [10] established presence of a strong relationship between NDVI, precipitation and temperature for different vegetation types. The effect of temperature on NDVI was more obvious than that of precipitation in that particular study [10].
Zhang et al. [11] also reported the existence of a positive correlation between NDVI and temperature but pointed out that the effect of precipitation on NDVI was not as significant. Additionally, Zhang et al. [11] established that bushland NDVI correlated more strongly with precipitation than NDVI of other vegetation [11]. Based on these observations, it is apparent that global and regional responses to climate change show wide variation [12]. Therefore, there is a need to undertake studies that quantitatively measure the effect of changes in climatic factors on land cover change at the local level.
Given its many advantages, NDVI is best suited to monitor local or global vegetation changes resulting from a changing climate [13] [14] [15].
Normalized Difference Vegetation Index has been widely used for studying climatic effects on vegetation productivity since the 1980s, though results vary by complexity of vegetation characteristics and region [16] [17]. It is predicted that by 2050, temperature and precipitation are likely to show decreasing and increasing signals, respectively, across the East African region [18]. However, the magnitude of change is likely to vary by region and location. Predicting land and conservation practitioners in their attempt to manage and mitigate impacts [19]. Prediction of LC change has been used in different applications, such as urban planning [20]; selection of conservation priority areas and setting alternative conservation measures [21] studying dynamics of shifting cultivation [22] and in simulation of rangeland dynamics under different climate change scenarios [23]. A solid understanding of the trends in land cover change at different time scales (past, present and future) at the local scale is therefore critical for decision making and policy formulation.
A review of the most commonly used approaches to modeling and land use change prediction can be found in a study by Agrawal et al. [24]. Markov chain analysis has been extensively used to study dynamics of land use change at different scales [25]. It is a simple method for modeling land use change especially at large scales [26]. The stationary transitions assumed by the Markov chain models make it suitable for short-term projections [27]. However, given its' shortcomings, Markov chain analysis is often integrated with other empirical models [28]. The Markov-CA approach used in the current study is considered a spatial transition model as it combines the stochastic spatial Markov techniques with the stochastic spatial cellular automata method [29]. It has the advantage of predicting two-way transitions among the available LC classes, in contrast to the Geomod technique that only predicts one-way loss/gain from one class to another [30]. Lu et al. [20] noted that transition-based models that integrate spatial Markov model with spatial cellular automata model outperformed regression based models in predicting land use change.

Study Area
The Mara River Basin (Figure 1

Transition Probability Matrix
The transition probability matrix records the probability that each land cover category will change to another or remain in the same category. For the 6 by 6 matrix table, the rows represent land cover categories and the column represents corresponding NDVI values. Although this matrix can be used as a direct input for specification of the prior probabilities in maximum likelihood classification of the remotely sensed imagery, it was used in predicting land cover change by 2027.

Temperature and Precipitation Data Acquisition and Analysis
Temperature and rainfall data sets were obtained from Giovanni website. The data obtained were of high resolution (0.1˚ latitude × 0.1˚ longitude) daily grid-

Impacts of Temperature and Precipitation on Different Land Cover Categories
Kriging methods were employed in ArcGIS to produce monthly and annual  between monthly variables in MS Excel and P-values used to determine significance levels. Considering the lagged response of NDVI to temperature and precipitation, the correlation analyses were also carried out between each seasonal NDVI and the previous season's temperature and precipitation for the month of October.

Determination of the Accuracy of Land Cover Maps
The overall accuracy of the land cover maps for 1987, 1997 and 2017 was deter- is as shown below [32]. If p a = the proportion of observations in agreement and p ε = the proportion in agreement due to chance, then Cohen's kappa is:

Markov Chain Analysis
Markov chains were used to obtain the percentage and probability for each cat-Open Journal of Soil Science egory of land cover converted. Using the Markov model, the distribution of each land cover category was projected based on the transition probability p ij between two land cover categories (i and j). P ij was determined over a specific period, from time t to time t + 1, as follows: Let ij P P = denote the (possibly infinite) transition matrix of the one-step transition probabilities where: P = the Markov transition matrix P i, j = the land type of the first and second time period P ij = the probability from land type i to land type j t t + 1 = time The estimate of Markov chain is the relative frequency of transitions observed over the entire time period. The result of the estimation was used for prediction.
In practice, based on the map algebra principle, the class of land type utilizes the equation below to calculate the transfer map of land cover change under the ERDAS Modeler module. where:

Land Cover Changes between 1987 and 2017 across the Four Mara River Sub-Catchments
Overall, a considerable reduction in the spatial expansion of LC was observed between 1987 and 1997 compared to the period between 2007 and 2017 across all the four Mara River sub-catchments.

Nyangores Sub-Catchment
The findings presented in Table 2 (Table 3).  (Table 3, Table 4 and Figure 3). Generally, land cover categories appeared more sensitive to precipitation than temperature in Nyangores sub-catchments.

Sand River Sub-Catchment
Sand river sub-catchment showed varying trends in NDVI of different land cover   were observed. Greater NDVI changes of R 2 = 0.5 were observed in bare land within Sand river sub-catchment. Temperature also showed relative low change in mean R 2 = 0.22 while rainfall showed an even lower change in mean R 2 = 0.06 (Tables 8-10, Figure 5). Nevertheless, Pearson's correlation coefficient was very high between annual mean precipitation and different land cover categories i.e.
forest cover, bare land, shrub land and grassland whose NDVIs were 0.  Changes in maximum mean annual temperature (R 2 = 0.22) and mean total annual precipitation (R 2 = 0.06) have affected land cover categories differently in Sand River sub-catchment.

Talek Sub-Catchment
Varying trends in NDVI of land cover categories from weak (R 2 = 0.006 to 0.04) to moderate (0.33 to 0.5) levels were also observed in Talek (Tables 11-13; and Figure 6).

Transition Probability Matrix for Nyangores, Amala, Sand and Talek Sub-Catchments
On the overall, the transitional probability matrix of the four sub-basins showed an interesting pattern (Table 14). Row categories represent land cover classes in      (Tables 15-17 and Figure 7, Figure 8).

Projected Land Cover Change between 2017 and 2027
in Amala Sub-Catchment In the Amala sub-catchment, significant changes were observed in land cover classes over the 30-year period (1987-2017) (Figures 9-11). Some land cover categories increased over time at the expense of others while some decreased (Table 18). For instance, bare land decreased by −62.63%, shrub lands by −45.82%         (Table 20).      this resolution image to identify settlements and small crop lands.

Projected Land Cover Change between 2017 and 2027 in Talek Sub-Catchment
As was the case in the other three sub-catchments, Talek sub-catchment also exhibited some significant changes in land cover over the last 30 years (Tables   24-26

Discussion
Land cover change in the four sub-basins of the Mara River basin i.e. Amala, Nyangores, Sand River and Talek was evident though variable and apparently nonlinear. The dynamics of land cover change were however rather gradual, implying that different development trends in each decadal period depended on human activities and climatic factors; and area specific interventions. In fact, studies show that the pace, magnitude and spatial reach of human alterations of the Earth's land surface are unprecedented; with changes in land cover being among the most important [33]. In the current study, spatial analysis revealed that the observed land cover changes were distributed across the Mara River basin sub-catchments and included multiple change directions in LC. Similar land cover change trends have been reported in previous studies in other regions [34].
High rate of deforestation averaging 5.3% per annum was observed within the Mara River basin, but was more pronounced in the two upper Mara River sub-catchments (Amala and Nyangores). In addition, nonlinearity of land cover Rapid population growth and changes in climatic conditions have been singled out as some of the major drivers of deforestation in attempts to increase agricultural land and human settlements in the Mara River basin. Despite improvements in land-cover characterization made possible by earth observing satellites, global and regional land covers are still poorly enumerated [35]. However, scientists recognize that the magnitude of land cover change is massive. One earlier estimate, for example, holds that the global expansion of croplands since 1850 has converted some 6 million km 2 of forests/woodlands and 4.7 million km 2 of savannas/grasslands/steppes, while 0.6 million km 2 of cropland has been abandoned [36].
In the current study, the period between 1987 and 1997, exhibited a much higher reduction in spatial expansion in land cover, particularly in forest land and valuable forest of native trees that produce a number of benefits to the ecosystem and to the inhabitants [37]. In all the four sub-catchments, a significant difference in forest cover change was observed though to varying degrees. A similar trend was also observed in crop lands, grasslands and bare land.
Temperature and precipitation were identified as significant actors in the observed changes in land cover within the Mara River basin. When the mean total annual precipitation was high and maximum mean annual temperature was low, the NDVI of all land cover categories recorded highest values, whereas the opposite was true. Consistent with the findings of the present study, Herrmann et al. [38] reported a close relationship between rainfall and surface greenness in the Sahel region on a large scale, while variations in other climatic and environmental factors were considered of minimal effect. On the contrary, some researchers argue that the relationship between rainfall and vegetation greenness can be explained by changes in soil moisture conditions, which cause an instantaneous plant response [39]. Although variation of soil moisture in semiarid and arid regions is significantly controlled by rainfall amount, near-surface air tem-Open Journal of Soil Science perature may also be considered an additional climatic factor in soil moisture changes [40] [41] [42]. Therefore, the effect of air temperature on the relationship between rainfall and vegetation greenness should be examined in greater detail for tropical ecosystems, which most previous studies have not taken into account. projections in the present study. Nevertheless, for these projections to be realized, it is suggested here that deliberate attempts must be made by the government of Kenya and the local communities to deliberately protect the environment. This will in turn lead to improved economy for the community and the government as well as improve the overall health of the Mara River ecosystem.

Conclusion
This study highlights the impact of changing climatic factors on different land cover categories within the four Mara River basin sub-catchments. Generally, land cover categories exhibited positive sensitivity to high precipitation and low temperature. On the contrary, high temperature exhibited strong negative correlations with the different land cover categories including croplands. The results however suggest that, with proper interventions, forest land, grassland, shrub land and even built-up areas are likely to increase, while crop land and bare lands are likely to decrease by 2027.