Effects of Climate Variability on Normalized Difference Vegetation Index (NDVI) in the Gojeb River Catchment, Omo-Gibe Basin, Ethiopia

Vegetation dynamics have been visibly influenced by climate variability.+eNormalized Difference Vegetation Index (NDVI) has been themost commonly used index in vegetation dynamics.+e study was conducted to examine the effects of climatic variability (rainfall) on NDVI for the periods 1982–2015 in the Gojeb River Catchment (GRC), Omo-Gibe Basin, Ethiopia. +e spatiotemporal trend in NDVI and rainfall time series was assessed using a +eil–Sen (Sen) slope and Mann–Kendall (MK) statistical significance test at a 95% confidence interval. Moreover, the residual trend analysis (RESTREND) method was used to investigate the effect of rainfall and human induction on vegetation degradation. +e Sen’s slope trend analysis and MK significant test indicated that the magnitude of annual NDVI and rainfall showed significant decrement and/or increment in various portions of the GRC. +e concurrent decrement and/or increment of annual NDVI and rainfall distributions both spatially and temporarily could be attributed to the significant positive correlation of the monthly (RNDVI-RF � 0.189, P≤ 0.001) and annual (RNDVIRF � 0.637, P≤ 0.001) NDVI with rainfall in almost all portions of the catchment. In the GRC, a strongly negative decrement and strong positive increment of NDVI could be derived by human-induced and rainfall variability, respectively. Accordingly, the significant NDVI decrement in the downstream portion and significant increment in the northern portion of the catchment could be attributed to human-induced vegetation degradation and the variability of rainfall, respectively.+e dominance of a decreasing trend in the residuals at the pixel level for the NDVI from 1982, 1984, 2000, 2008 to 2012 indicates vegetation degradation. +e strong upward trend in the residuals evident from 1983, 1991, 1998 to 2007 was indicative of vegetation improvements. In the GRC, the residuals may be derived from climatic variations (mainly rainfall) and human activities. +e time lag between NDVI and climate factors (rainfall) varied mainly from two to three months. In the study catchment, since vegetation degradations are mainly caused by human induction and rainfall variability, integrated and sustainable landscape management and climate-smart agricultural practices could have paramount importance in reversing the degradation processes.


Introduction
Vegetation dynamics have been visibly influenced by climate variability due to biophysical responses of plant respiration and photosynthesis [1,2]. e remotely sensed Normalized Difference Vegetation Index (NDVI) derived from the Advanced Very-High-Resolution Radiometer (AVHRR) of National Oceanic and Atmospheric Administration (NOAA) satellite [3,4] has been the most commonly used index in vegetation dynamics [5,6]. e NOAA satellites contain AVHRR sensor [7], which collects spectral information in the visible and near-infrared regions, facilitating calculation of the NDVI [8]. e main reasons for using AVHRR-NDVI images in many environmental studies are its high temporal and adequate spatial resolution, good calibration, and low price [3]. NDVI is calculated as the difference between near-infrared and visible reflectance values normalized over the sum of the two and ranges from -1 to +1 for a given pixel [8]. Because of high reflectance in the NIR portion of the Electromagnetic Spectrum (EMS), healthy vegetation is represented by NDVI values between 0.1 and 1. On the contrary, nonvegetated surfaces such as water bodies yield negative values of NDVI. Bare soil areas represent NDVI values which are closest to 0 due to high reflectance in both visible and NIR portions of the EMS [9]. An NDVI value of zero means no vegetation and a value close to +1 (0.8-0.9) indicates the highest possible density of green leaves [8].
e NDVI is recognized as a reliable indicator of land cover conditions and variations in a wide range of studies spanning regional to global scales [10]. e growth and distribution of natural vegetation were significantly affected by climate variability at various spatial and temporal scales [11][12][13]. Rainfall is the most important climatic factors directly affecting vegetation dynamics and closely correlates with NDVI due to its irregularity and seasonal variability [14,15]. It is a particularly important predictor of vegetation distribution [15,16].
Many researchers have used NDVI to monitor the response of vegetation to rainfall fluctuations. e strong associations between rainfall and NDVI have been reported in various regions of the world [8,13,17]. Several studies have also investigated the relationship between rainfall and NDVI on global and regional scales to demonstrate a very close linkage between them [15,18]. Furthermore, the correlation results for NDVI and climatic variables from different parts of the world showed a significant relationship between NDVI and precipitation in the northern hemisphere semiarid regions [17]. Al-Bakri and Suleiman [3] obtained a significant positive correlation between NDVI and rainfall for both 10-day and monthly data in the Jordan region. Zhao et al. [19] found a direct association between vegetation growth and rainfall in relatively dry areas while an inverse association in relatively humid areas. Various studies [3,17,[20][21][22] have also found significant correlations between NDVI and rainfall in different regions. Hence, vegetation dynamics and their relationship with climate variability have become a hotspot in global change studies [21,23]. e most widely used global NDVI dataset is the Global Inventory Monitoring and Mapping Studies (GIMMS) database [24]. is dataset is famous for its high quality, having been corrected to remove some nonvegetation effects caused by solar angle and sensor errors. It has been widely used to monitor long-term trends in vegetation activity from regional to global scales [6]. In the Gojeb River Catchment, no studies have been carried out on the performance of the Global Inventory Monitoring and Modelling System (GIMMS3g) in monitoring vegetation dynamics in relation to monthly, seasonal, and annual climate variability in general and rainfall variability in particular. Hence, the aim of this study is to examine the spatiotemporal vegetation dynamics (Normalized Difference Vegetation Index (NDVI)) and its response to climate variability (rainfall) using GIMMS3g from 1982 to 2015 in Gojeb River Catchment, Omo-Gibe Basin, Ethiopia. e objectives of this study are to (1) investigate the monthly, seasonal, and annual NDVI trend in relation to climate variability in Gojeb River Catchment during the past 34 years; (2) analyze the correlation between NDVI and climatic factors in order to determine the impact of climatic variability in vegetation dynamics at temporal and spatial scales; and (3) distinguish the relative importance of the impacts of climate variability (rainfall) and human activities on vegetation dynamics using wavelet analysis in the Gojeb River Catchment of Omo-Gibe Basin, Ethiopia.

Description of the Study Area.
e study was conducted at the Gojeb River Catchment (GRC), a part of the Omo-Gibe basin in Ethiopia ( Figure 1). e Omo-Gibe basin, third largest perennial river in Ethiopia next to the Baro Akobo and Blue Nile rivers, lies between 5°31′ to 10°54′ N and 33°0′ to 36°17′ E and covers about 79,000 km 2 of land area in South and Southwest Ethiopia [25]. e GRC is located between 7°00′-7°50′ N latitude and 35°30′-37°20′ E and covers a total area of 6932.345 km 2 with altitudinal ranges from 817 to 2500 m.
In the Omo-Gibe River Basin, the climate varies from a hot arid climate in the southern part of the floodplain to a tropical humid in the highlands that include the extreme north and northwestern part of the basin [25]. e rainfall varies from over 1900 mm per annum in the north-central areas to less than 300 mm per annum in the south. Moreover, the rainfall regime is unimodal for the northern and central parts of the basin and bimodal for the south. e mean annual temperature in the Omo-Gibe basin varies from 16°C in the highlands of the north to over 29°C in the lowlands of the south [26]. In the GRC, the rainfall is unimodal and the amount increases with the increase in elevation. e mean annual rainfall of the catchment ranges from 1390.7 mm in the downstream portion to 1884.0 mm in the upstream portion. e mean monthly maximum temperature and minimum temperature are 25.9 and 14.4°C, respectively ( Figure 2). e land use pattern is dominated by five land use categories, namely, woodland, cropland/agricultural, swamp, forest, and shrubland [27]. e total population of the catchment is about 2,637,351 of whom 1,320,571 (50.07%) are men and 1,316,780 (49.93%) are women and while only 9.2% are urban inhabitants [28].

Data Used
e Normalized Difference Vegetation Index (NDVI) derived from satellite data is an important indicator that can be used to analyze green vegetation dynamics and reveal the response of vegetation dynamic to the climate variability [7]. NDVI is the most widely used index for monitoring the spatiotemporal dynamics of green vegetation and ecoclimatological studies [1,7,[29][30][31][32][33][34]. e index ranges between −1 (water bodies) and 1 (dense vegetation) [17] and computed as follows [35]: where NIR and RED denote spectral reflectance in the nearinfrared (800-1000 nm) and red (620-750 nm) portions of the electromagnetic spectrum, respectively. e most commonly used and unique Earth observation program, which enables the analysis of vegetation activity over the past three decades, is provided by the National Oceanic and Atmospheric Administration (NOAA) Polarorbiting satellites [4]. e NOAA satellites contain the Advanced Very-High-Resolution Radiometer (AVHRR) sensor [7,32], which collects spectral information in the visible and near-infrared regions, facilitating calculation of the NDVI [8]. e NDVI data from NOAA-AVHRR sensors have been widely used to analyze vegetation dynamics and trends in various regions of the world [5,6]. e most widely used global dataset of NDVI from NOAA-AVHRR series satellite data is the Global Inventory Monitoring and Mapping Studies (GIMMS) database [24], which covers a longer period than the other available datasets .
is dataset is famous for its high quality, having been corrected to remove some nonvegetation effects caused by solar angle and sensor errors, and has been widely used to monitor long-term trends in vegetation activity from regional to global scales [6].
In the Gojeb River Catchment, the NDVI dataset used is the third-generation version acquired from Global Inventory Modelling and Mapping Studies (GIMMS NDVI3g) available from January 1982 to December 2015. In this study catchment, the NDVI dataset used was the latest updated version of the third-generation Global Inventory Monitoring and Modelling System (GIMMS NDVI3g) [36,37] available at https://ecocast.arc.nasa.gov/data/pub/ gimms/3g.v1/as nc4 files. GIMMS3G NDVI data comprise a long-time series global vegetation index dataset based on NOAA meteorological satellites (NOAA 7,9,11,14,16,17,and 18) [32,36]. Among the long-term AVHRR-based datasets analyzed, the GIMMS3g NDVI is found to have the highest temporal consistency and is the most appropriate choice for NDVI trend analysis [30,[36][37][38][39]. e GIMMS NDVI3g is generated from NOAA's AVHRR data, and the spatial resolution is 1/12°. Its temporal resolution is 15-day intervals with 34 years' time span (1982 to 2015). e GIMMS3g NDVI dataset has already been shown to accurately represent the real responses of vegetation to climate variability [32,36,37,40]. e quality and consistency of the GIMMS3g data are assured by correction for (i) sensor degradation, (ii) sensor intercalibration differences, (iii) solar zenith and viewing angles, (iv) volcanic aerosols, (v) atmospheric water vapor, and (vi) cloud cover [4]. In the study catchment, a satellite orbital drift correction is performed using an empirical mode decomposition/reconstruction (EMD) method by minimizing the effects of orbital drift by removing common trends between the time series of Solar Zenith Angle (SZA) and NDVI [30,31,41].
In the study catchment, due to the ease of accessibility of the dataset, the GIMMS3G NDVI dataset, which spans from

Advances in Meteorology
January 1982 to December 2015 with a 15-day interval, derived from AVHRR sensors was used. e GIMMS3g NDVI dataset is considered the best dataset available for long-term NDVI trend analysis [36]. e GIMMS3g NDVI dataset has a spatial and temporal resolution of 0.05°× 0.05°a nd 15 days, respectively. Nowadays, it is considered as the best dataset available for long-term NDVI trend analysis [42]; it accurately represents the real responses of vegetation to climate variability [40]; it possesses the highest temporal consistency and is appropriate for trend analysis [42]. e biweekly NDVI data were converted into monthly values using the maximum value composite method to reduce cloud disturbance and increase the overall quality of the dataset [30]. A maximum value composite (MVC) method was applied to get the monthly NDVI data by reducing the atmospheric effects of clouds and aerosol [1,7,43,44]. An MVC technique is a process that minimizes cloud contamination, reduces directional reflectance and off-nadir viewing effects, minimizes sun-angle and shadow effects, and minimizes aerosol and water-vapor effects on NDVI signal [45]. e averaged NDVI for Bega (dry season) from December to February (DJF), Belg (small rainy season) from March to May (MAM), Kiremt (main rainy season) from June to August (JJA), and Meher (main harvesting period with moderate rainfall) from September to November (SON) was calculated for analysis. e GIMMS3g NDVI dataset was cut to the extent of Gojeb River Catchment and the bimonthly data are aggregated to monthly, seasonal, and annual composites by applying the arithmetic mean of each month. Finally, the monthly, seasonal, and annual NDVI for the entire study catchment from 1982 to 2015 has been made available.

Rainfall Data.
e monthly, seasonal, and annual distribution of rainfall is determinant for vegetation growth and productivity in various ecosystems and widely used to evaluate changes in vegetation dynamics [46]. In order to examine trends in the spatiotemporal variation of rainfall in the catchment, the product of Climate Hazards group Infrared Precipitation with Stations (CHIRPS) time series data from January 1982 to December 2015 was used. e CHIRPS data was created at the United States Geological Survey (USGS) and Earth Resources Observation and Science (EROS) Centre. e CHIRPS data used had a spatial and temporal resolution of 0.05°× 0.05°degree and one month, respectively. e CHIRPS (v2.0) project provides rainfall datasets based on conjunction geostationary infrared satellite rainfall estimates and rain gauge observations that are interpolated to produce robust precipitation grids [47]. e CHIRPS is updated frequently and provides data at higher spatial and temporal resolutions. It is high quality, temporally consistent, and near real-time precipitation datasets that can help identify environmental changes, quantify the important role played by warming air temperatures, and play an important role in seasonal drought prediction [47,48]. Monthly gridded precipitation data over the 1982-2015 study periods were downloaded (http://chg.geog. ucsb.edu/data/chirps/) and averaged to match the 0.05°s patial resolution of NDVI data. In the study catchment, NDVI and rainfall analysis was focused on the basis of four seasons, i.e., Bega (dry season) from December to February (DJF), Belg (small rainy season) from March to May (MAM), Kiremt (main rainy season) from June to August (JJA), and Meher (main harvesting period with moderate rainfall) from September to November (SON). e long-term NDVI climatology  was created by averaging data for all cloud-free pixels for DJF, MAM, JJA, and SON for the same period. en, four seasons were summed on an annual basis between 1982 and 2015 as the independent variable.

Image Processing.
All images were processed in MAT-LAB version 2016a and 2017b and ArcGIS Version 10.3 and then the NDVI and rainfall datasets were further processed in Microsoft Excel. e raw NDVI data was downloaded and processed following different steps. Firstly, the satellite data was analyzed using remote sensing tools. In ArcGIS, the coordinate system of all images was defined into GCS-WGS-1984 and projected into UTM-37N, which is a coordinate system used in the Gojeb River Catchment. e NDVI values were calculated using a raster calculator in ArcGIS 10.3 and MATLAB tools and divided into relevant classes. ese classes when converted into spatial data indicated the spatial and temporal changes in vegetation. Secondly, all the images that were refined using ArcGIS and MATLAB tools and clipped using the study catchment boundary shapefile and the mean values of NDVI were extracted for each land cover class. Finally, the mean NDVI values were exported into Microsoft Excel, and graphs were produced to show the monthly, seasonal, and annual patterns of NDVI and rainfall. Moreover, using ArcGIS and MATLAB, different maps were produced to show the spatial and temporal variation of NDVI with respect to rainfall.
In the study catchment, the rainfall data was downloaded in a gridded dataset form for the study periods from January 1982 to December 2015. Firstly, to get the total monthly rainfall for each year, the images were added in MATLAB and ArcGIS tools. Secondly, the coordinate system of all images was defined into GCS-WGS-1984 and projected into UTM-37. irdly, all the total monthly rainfall images were clipped using the catchment boundary shapefile and/or Tiff file. en, the mean value of rainfall was extracted for each land cover class.
e mean values were exported into Microsoft Excel for trend analyses and graphs were produced to show patterns of rainfall. As a result, maps were produced to show the spatial and temporal variation of rainfall with respect to NDVI.

Data Analysis: Impact of Climate Variable (RF) on Vegetation Dynamics (NDVI).
In the GRC, correlation analysis was carried out using monthly, seasonal, and annual NDVI and rainfall at each grid point to understand the relationship between them. Correlations between NDVI and rainfall were computed using the Pearson correlation method. e coefficient of Pearson enabled the magnitude of each correlation to be measured on a per-pixel basis for the period of 1982-2015. e statistical significance was tested using Pearson's correlation coefficients at a confidence level of 95% (P � 0.05). If the correlation coefficients were greater than 0.3 (P � 0.05), it was considered that NDVI had a strong correlation with rainfall [49]. e significance of the correlation coefficients was mapped to show the spatial distribution of the relationships.

Analysis of Trends of NDVI and Rainfall.
e trend in rainfall and NDVI time series was assessed at both spatial and temporal levels during monthly, seasonal (DJF, MAM, JJA, and SON), and annual basis. e trends of the NDVI and rainfall were analyzed using a eil-Sen (Sen) slope, which is a robust nonparametric statistical estimation of the trend magnitude which is insensitive to small outliers and missing noises values [49]. e magnitude of NDVI and rainfall changes was assessed using Sen's slope at a 95% confidence interval where positive and negative values indicate increased and decreased in rainfall and vegetation, respectively. Statistical significance of Sen's slope trend was tested using the Mann-Kendall (MK) significance test, which is widely applied to long-time series trend analysis with nonnormal data [49]. e MK test requires the data to be serially independent in order to avoid large uncertainty in estimating the trend of serially correlated data like rainfall and NDVI [34,50]. e MK test for assessing trends in both NDVI and rainfall using their detrended data component was used. NDVI time series encompass vegetation seasonal growth in relation to the amount and distribution of rainfall.
e MK test is one of the widely used nonparametric tests for assessing seasonal variations mainly in hydrological time series [51,52]. However, it is currently more commonly applied for testing the significance of changes in NDVI trends [34,53]. e MK trend test is less sensitive to missing data values, irregular data distribution, and outliers [12]. In the study area, the significant trend in rainfall and NDVI time series was assessed using the MK significance test at 95% confidence interval where black area showed a significant trend (H 1 ) while white area showed no significant trend (H 0 ).

Residual Trend Analysis (RESTREND) Method.
In Gojeb River Catchment, natural and human effects on vegetation degradation were assessed using residual trend analysis (RESTREND) method where negative values indicate degraded area while positive values shown improved vegetation performance. e method is simple and can be used to reveal the spatial heterogeneity of factors driving land degradation at cell resolution [54]. e RESTREND method has been used widely to explore the relationship between climatic and anthropogenic causes of land degradation in various regions of the World [55][56][57]. e RESTREND method can be used to quantitatively and effectively differentiate climate and human drivers of land degradation [54,56]. e residuals can be acquired from the difference between observed NDVI and predicted NDVI and were analyzed to detect the trend over time [58]. If there was a significant decreasing trend of residuals, it would presumably be derived by human activities. Otherwise, if the Advances in Meteorology residuals showed no trend over time, the declining vegetation production would be attributed to climate factors such as rainfall [49]. If the residual trend is insignificant, the NDVI variations can be explained with climatic variables; otherwise, NDVI variations might be affected by human activities [1,31,59].

Time-Lag Effects of Vegetation (NDVI) Response to
Climatic Factor (Rainfall). Since the vegetation is affected by both current and previous rainfall events, the relationship between NDVI and rainfall is characterized by a time-lag response of NDVI in relation to rainfall. Moreover, this relationship can differ significantly depending on the plant growth stage [42]. e rainfall from the CHIRPS is used as the independent variables, whereas the NDVI data from the GIMMS3g, which represents vegetation activities, are used as the dependent variables to evaluate their dependency at different time-lag scales. In this study, the Windowed Cross-Correlation (WCC) method [19] was used, which has been frequently used [60], to analyze the time-lag effects of vegetation greenness response to rainfall. e WCC methods aimed to identify the most significant correlation between NDVI and climatic factor (rainfall), and the corresponding lag time. e models showing the relationships between the NDVI and rainfall were as follows [19]: where k rf refers to the regression coefficient with a time lag of i months; i ranges from 0 to n (i.e., 0 represents no timelag effects, and 1 − n represents a one-to n-month lag, respectively); NDVI refers to the GIMMS3g NDVI time series ; and RF is the time series for rainfall with a time lag of i. For the climatic factor, the lag month (i) that has the highest determination coefficient (R 2 ) was the best time lag for the vegetation responses to this climatic factor. erefore, combined with the rainfall data, correlation coefficients were calculated to compare the monthly lag of NDVI from rainfall. e analysis was based on the pixel scale, where (i, j) refers to the pixel located in the row i and column j. e determination coefficient R 2 of the regression equation was used as an index to evaluate the correlations between NDVI and climatic factors. e highest R 2 indicated the strongest NDVI response to a climatic factor. e time lag corresponding to the highest R 2 was regarded as the optimum time lag.

Annual NDVI and Rainfall Trends in the Gojeb River
Catchment, Ethiopia. In the GRC, Sen's slope trend analysis indicated that the magnitude of annual NDVI showed significant decrement especially in the southeast and west of the catchment, while the annual NDVI trend magnitude was significantly increased in the northern and some other portions of the catchment (P < 0.05; Figure 3). e continent with the highest proportion of significant increasing trends is Europe (49.71%), followed by South America (40.73%).
Nevertheless, the majority of the study catchment (around 85% of the area) did not show significant changes in vegetation dynamics. e continent with the highest proportion of significant decreasing trends of NDVI is North America (7.63%), followed by South America (7.31%) [61]. Studies in Asia revealed that vegetation greenness has significantly decreased over the last 33 years [32]. e vegetation trends have shown significant decreasing trends in the southern Amazon forest [62] and in the southern Central African Republic and Uganda [59,60]. On the other hand, the annual rainfall trend magnitude increased significantly (P < 0.05) in the southern part of the catchment. Increasing trends in precipitation are observed in high latitudes in Asia, in middle Europe, and in parts of Africa, while the southwestern United States, northern Mongolia, and northeastern China show decreasing trends [61]. However, the major portion of the study catchment did not show a significant reduction in the annual rainfall trend (P > 0.05, Figure 3). Similar studies in Northwestern Ethiopia revealed a statistically nonsignificant increasing trend in annual mean rainfall time series during the study period between 1983 and 2014 [34].
In the GRC, the mean annual NDVI and rainfall were found to be 0.72 (±0.026) and 1652. e close resemblance in NDVI and rainfall patterns (in trend and amplitude) in GRC might reveal a close relationship between vegetation and rainfall. roughout, NDVI shows a clear response to the cycle of rainfall.

Monthly NDVI and Rainfall Trends in the Gojeb River
Catchment, Ethiopia. Spatially, the monthly distribution of NDVI was found to be higher in the northern and northwest portion of the catchment, while it is lower in the southern and southeastern portion. Temporally, the monthly minimum and maximum NDVI values were found to be 0.55 and 0.84, respectively, with an average value of 0.72 (±0.06). Sen's analysis of trend magnitude showed that the monthly NDVI decreased significantly (P < 0.05, black area, Figures 4(a) and 4(b)) in the southeast (downstream) and northern and other significant portions of GRC. Concurrently, the mean monthly rainfall trend magnitude showed a significant reduction, especially in the southern portion of the catchment (black area, Figures 4(c) and 4(d)) which indicates the strong and positive significant correlation of NDVI and rainfall (R � 0.189, P ≤ 0.001). 6 Advances in Meteorology During DJF, Sen's trend analysis and MK significant test indicated that the NDVI trend magnitude increased significantly (P < 0.05) in the northern upstream portion of the GRC, while vegetation cover decreased significantly in the downstream portion of GRC. However, the major portions of the study catchment have remained unaltered in vegetation cover during DJF (Figures 5(a) and 5(e)). Concurrently, though not statistically significant, the rainfall trend magnitude has shown decrement in almost all portions, particularly in the downstream portion, of the study catchment during DJF of the study period (Figures 6(a) and 6(e)). However, it remained unaltered in the western part of the upstream part of the catchment.

Seasonal NDVI and Rainfall Trends in the Gojeb
During MAM, Sen's trend analysis showed that the NDVI trend magnitude decreased in most portions of the study catchment, particularly in the downstream and west of the upstream portions ( Figure 5(a)). However, MK significant test indicated that the NDVI trend decreased significantly in the downstream portion of the GRC (P < 0.05, Figure 5(e)). During JJA, Sen's trend analysis and MK significant test showed that the NDVI trend magnitude increased significantly in the mid and northern upstream portions of the study catchment (Figures 5(c) and 5(g)).
On the other hand, during MAM, Sen's trend analysis and MK significant test showed that the rainfall trend magnitude did not increase significantly throughout the GRC (P > 0.05, Figures 6(b) and 6(f )). Concurrently, the rainfall trend magnitude did not show any significant changes across the catchment, except in some downstream portions of GRC (Figures 6(c) and 6(g)). During SON, the NDVI trend magnitude showed a significant reduction in some parts of the catchment, while some other portions of the catchment also showed some increment in vegetation cover not statistically significant (Figures 5(d) and 5(h)). Simultaneously, Sen's trend magnitude of rainfall did not show significant increment, except for some small portions of the catchment which showed a significant increase in rainfall trend magnitude (Figures 6(d) and 6(h)).
Annual NDVI trend magnitude Value Annual NDVI trend significance  Annual rainfall trend magnitude  Annual rainfall trend significance  Annual RF trend significance l High: 1 (significance)   Figure 5: Seasonal NDVI trend magnitude and significance during DJF, MAM, JJA, and SON between 1982 and 2015 in GRC.

NDVI RESTREND in the Gojeb River Catchment,
Ethiopia. In principle, RESTREND can identify the pixels that have experienced a reduction or an increase in biomass production per unit rainfall over a period of time [57]. It provides an effective tool to distinguish between the effects of climatic factors and human activities on vegetation changes when used properly where negative values indicate degraded area while positive values shown improved vegetation performance [56]. If there was a significant decrease and no trend of residuals (difference of observed and predicted NDVI [58] over time), the vegetation degradation would be derived by human activities and climate factors such as rainfall [49], respectively. If the residual trend is insignificant, the NDVI variations can be explained with climatic variables [1,31,59]. e NDVI linear slopes (Figure 7) examine the spatial patterns of the rate of NDVI changes and the minimum value of NDVI and how they relate to changes in rainfall from 1982 to 2015. In the GRC, the negative values (Figure 7) indicated the vegetation degradation, while the positive values showed improved vegetation performance. And, the zero values of NDVI linear slope indicated the unaltered vegetation cover in the study catchment.
In the GRC, a strongly negative decrease of NDVI, for instance, during 1984, 2000, 2008, and 2005 (Figure 8), could be derived by human-induced vegetation degradation. On the contrary, the strong positive increase of NDVI, during 1997 and 2007, could be attributed to the rainfall variability in the study catchment. e RESTREND analysis showed that the NDVI trend magnitude residuals decreased significantly in the downstream portion of the GRC (Figure 9) which might be derived by human activities.
e trend magnitude residuals also showed a significant increment in the northern portion of the upper catchment which could, in turn, be attributed to the variability of rainfall. In the study catchment, there are also places in which the residuals showed no trends over time which could be affected by climatic (rainfall) variability ( Figure 9). Similarly, studies, elsewhere, [56] reported that NDVI increases in response to higher rainfall in the areas with positive slopes. e RESTREND method assumes that areas that show a negative trend are degraded, while those with a positive trend are improved or at least not degraded. Figure 9 shows areas with positive and negative trends of vegetation productivity that have been derived by either human activities or rainfall. e RESTREND residual trend magnitude based on rainfall shows both areas with positive and negative trends (Figure 9). e RESTREND maps show areas that are likely subject to land degradation (negative trend of residuals) or not (positive residuals) (Figure 9). e dominance of a decreasing trend in the residuals at the pixel level for the  Figure 6: Seasonal rainfall trend magnitude and significance during DJF, MAM, JJA, and SON in GRC. NDVI from 1982NDVI from , 1984NDVI from , 2000NDVI from , 2008 to 2012 indicates vegetation degradation, whereas a strong upward trend in the residuals was evident from 1983, 1991, 1998 to 2007, indicative of vegetation improvements. e significant relationship between vegetation production and rainfall was the basic premise of the RESTREND method [49]. Evans and Geerken [63] first proposed the NDVI-based RESTREND method to distinguish between vegetation degradation due to climatic variations and that caused by human activities in Syria. If the significant climate effects could be removed from the long-term trends of NDVI, human-induced land degradation might be distinguished [58]. He et al. [54] differentiated human-induced drivers from climatic drivers of grassland degradation in the Liao River Basin using the RESTREND method. Studies [54] showed that the RESTREND method is an effective and reliable tool to help identify the driving factors of vegetation degradation. Accordingly, they reported that climate variability and human activities contributed equally to the vegetation degradation in the Otindag Sandy Land. Rainfall is a major driving force and the most important factor in promoting vegetation growth and natural grasslands and forests of the world [64].

Advances in Meteorology
In the study catchment, the residuals may result from a number of factors, including climatic variations (mainly rainfall), natural disturbances (e.g., flooding), and human activities (livestock grazing, population, rural settlement, agriculture, fuel wood and charcoal extraction, and timber production).

NDVI-Rainfall Correlation Analysis in the Gojeb River
Catchment, Ethiopia. In the GRC, Pearson's correlation was utilized to analyze the relationship between overall vegetation dynamics (NDVI) and climatic variations (rainfall) over the period 1982-2015 (Table 1). e annual NDVI had a significant positive correlation with rainfall (R NDVI-RF � 0.637, P ≤ 0.001). Similarly, the monthly NDVI had a significant positive correlation with rainfall (R NDVI-RF � 0.189, P ≤ 0.001). In consistent with this finding, Qiu et al. [65] observed significantly positive NDVI-climate relationship on the denoised time series in most vegetated regions. However, the seasonal NDVI did not show any significant (P > 0.05) correlation with rainfall. On the contrary, Chu et al. [1] reported that the NDVI had a quite significant seasonal correlation with precipitation. e annual and monthly spatial distributions of NDVI and rainfall correlation analysis have been shown in Figure 10. In the GRC, spatially, positive correlations between annual NDVI and rainfall in almost all portions were obtained, while very small portions of the western part of the catchment showed negative correlations between them during the study periods (Figure 10(a)). e negative NDVI-rainfall correlation coefficients could be mainly associated with the negative dynamic trends of either NDVI or climatic factors [65]. In line with our findings, the spatial relationship between NDVI and precipitation was quite different among vegetation types and seasons [1]. In Gojam of Ethiopia, Getahun and Shefine [20] reported a strong   positive linear correlation between NDVI and rainfall. ey also confirmed very high annual correlation coefficients between NDVI and precipitation. Similarly, Bao et al. [5] reported the strong positive correlation between NDVI and precipitation in most areas of Mongolian plateau for the periods between 1982 and 2010 suggesting that precipitation is the most direct influencing factor on vegetation dynamics. On the contrary, in Northwestern Ethiopia, the parallel assessment of the NDVI and precipitation time series showed the lower dependency of vegetation change on the long-term precipitation changes [34]. Moreover, the major portions of the catchment showed positive correlation  Advances in Meteorology though there have been some portions in which negative correlations were obtained between the monthly NDVI and rainfall relationships (Figure 10(b)). Regarding the seasonal NDVI and rainfall correlation, there were relatively higher strong positive correlations during December, January, and February, DJF, though other seasons also showed positive correlations. On the contrary, there were also portions of the catchment which showed negative correlations during all seasons which might be associated with the quality of GIMMS datasets and with their negative dynamic trends. e vegetation growth has the strongest correlation with the precipitation of last month in most of Africa (excluding central Africa) [61].

Lag Time Effects of NDVI Responses to Rainfall in the Gojeb
River Catchment, Ethiopia. An image showing the spatial distribution of time lag between NDVI and rainfall was provided in Figure 11. e time lag between NDVI and climate factors (rainfall) varied with different climatic regions and vegetation cover of the study catchment. Regarding the time lag between NDVI and rainfall, the majority of the pixels (>85%) were examined with 1.6-3.3 months' time lag. Other areas were observed with a time lag of less than 1.6 and a very small portion of the catchment with 3.3 to 4 months. In the study catchment, the minimum and maximum NDVI values were observed during March (0.66) and September (0.78) while the minimum rainfall and maximum rainfall have been observed during December (41 mm) and January (41 mm) and July (233 mm), respectively, which indicated that the NDVI is getting maximum three months after the rainfall is maximal. erefore, in the GRC, the majority of time lags are two to three months. In consistent with this finding, other authors [66,67]

Conclusion
is study analyzed the spatial and temporal NDVI variations in the annual, monthly, and four different seasons (DJF, MAM, JJA, and SON) in the Gojeb River Catchment and their response to climatic factors (rainfall) from 1982 to 2015 based on GIMMS datasets and CHIRPS time series. e NDVI-rainfall relationship evolves both temporally and spatially. In the GRC, Sen's slope trend analysis indicated that the magnitude of annual NDVI showed significant decrement especially in the southeast and west of the catchment and significantly increased in the northern and some other portions of the catchment. e annual rainfall trend magnitude increased significantly (P < 0.05) in the southern part of the catchment. Nevertheless, the majority of the study catchment (around 85% of the area) did not show significant changes in vegetation dynamics and a significant reduction of annual rainfall trends. In the GRC, concurrent decrement and increment of annual NDVI and rainfall distributions both spatially and temporarily were observed. e close resemblance in NDVI and rainfall patterns (in trend and amplitude) in GRC might reveal a close relationship between them. Sen's analysis of trend magnitude showed that the mean monthly NDVI and rainfall decreased significantly in the southeast (downstream) and northern and other significant portions of GRC which indicate the strong and positive significant correlation of NDVI and rainfall (R � 0.189, P ≤ 0.001). In the GRC, the lowest and highest seasonal NDVI values and rainfall distributions were observed during MAM and JJA, respectively, indicating similar patterns of seasonal rainfall distribution with NDVI values during the study periods. Consequently, the reduction of vegetation cover is observed during DJF and MAM which might be attributed to the significant positive correlation of seasonal NDVI and rainfall during the study periods.
In the GRC, the RESTREND analysis showed that a strongly negative decrease of NDVI, for instance, during 1984, 2000, 2008, and 2005, could be derived by humaninduced vegetation degradation. On the contrary, the strong positive increase of NDVI, during 1997 and 2007, could be attributed to the rainfall variability in the study catchment.
e RESTREND analysis showed that the NDVI trend magnitude residuals decreased significantly in the downstream portion of the GRC which might be derived by human activities. e trend magnitude residuals also showed a significant increment in the northern portion of the upper catchment which could, in turn, be attributed to the variability of rainfall. In the study catchment, there are also places in which the residuals showed no trends over time which could be affected by climatic (rainfall) variability. e dominance of a decreasing trend in the residuals at the pixel level for the NDVI from 1982NDVI from , 1984NDVI from , 2000NDVI from , 2008 Figure 11: Spatial distribution of lag time for NDVI and rainfall from 1982 to 2015 in the Gojeb River Catchment.

Advances in Meteorology 13
In the GRC, unlike the seasonal, Pearson's correlation analysis revealed a significant positive correlation of the monthly (R NDVI-RF � 0.189, P ≤ 0.001) and annual (R NDVI-RF � 0.637, P ≤ 0.001) NDVI with rainfall. In the study catchment, spatially, positive correlations between annual NDVI and rainfall in almost all portions of the catchment were obtained during the study periods. Regarding the seasonal NDVI and rainfall correlation, there were relatively higher strong positive correlations during December, January, and February, DJF, though other seasons also showed positive correlations. e time lag between NDVI and climate factors (rainfall) varied with different climatic regions and vegetation cover of the study catchment. Regarding the time lag between NDVI and rainfall, the majority of the pixels (>85%) were examined with 1.6-3.3 months' time lag. Other areas were observed with a time lag of less than 1.6 and a very small portion of the catchment with 3.3 to 4 months. In the study catchment, the minimum and maximum NDVI values were observed during March (0.66) and September (0.78) while the minimum rainfall and maximum rainfall have been observed during December (41 mm) and January (41 mm) and July (233 mm), respectively, which indicated that the NDVI is getting maximum three months after the rainfall is maximal. In the study catchment, the majority time lags are two to three months.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e authors declare that they have no conflicts of interest.