Accuracy of the Land Cover Classification
The error matrix for the land cover map of 2015 revealed that the overall accuracy is 86.57% with a kappa coefficient of 0.84 (Table 1). These accuracy values indicate that it is possible to undergo analysis since they fulfill the accuracy level demanded from land cover maps, which are derived from satellite imageries (Anderson et al. 1976 cited in Berakhi et al. 2015).
Table 1
Error Matrix Generated from the 2015 Land Cover Map of the Study Area and the Reference Data
Classified data | Reference data | Row total | User’s accuracy* |
Forestland | Shrubland | Grassland | Cropland | Settlement | Bare land |
Forestland | 39 | 4 | -- | -- | -- | -- | 43 | 91 |
Shrubland | 6 | 50 | 2 | -- | -- | 1 | 59 | 85 |
Grass land | 5 | 1 | 49 | -- | 1 | 3 | 59 | 83 |
Cropland | -- | 2 | 7 | 77 | 6 | 6 | 98 | 79 |
Settlement | -- | -- | -- | -- | 46 | -- | 46 | 100 |
Bareland | -- | -- | -- | 3 | -- | 42 | 45 | 93 |
Column total | 50 | 57 | 58 | 80 | 53 | 52 | 350 | -- |
Producer’s accuracy** | 78 | 88 | 84 | 96 | 87 | 81 | -- | -- |
Overall accuracy (summation of the diagonal land covers/ summation of column or row totals): 86.57%, and Kappa coefficient: 0.84 (see the methodology section for details) |
* Diagonal value of a land cover (bold)/row total of the same land cover |
** Diagonal value of a land cover (bold)/ column total of the same land cover |
Description Of Land Cover Types
Six major land cover types were identified from 1986, 2001, and 2015 satellite images of Hita-Borkena watershed. These include barelands, croplands, forestlands, grasslands, settlements and shrublands. This does mean that these land cover types are the only ones in the watershed. There are rather other types like gullies, Borkena River and smaller streams, and main asphalt road that connect Addis Ababa to Kombolcha-Dessie-Woldia-Mekelle, but their spatial coverage is insignificant compared to the major ones. The description of the major land cover types is given in Table 2.
Table 2
Description of Land Cover Types in Hita-Borkena Watershed
Land cover types | Descriptions |
Forestlands | Areas devoted for the growth of relatively taller trees that form closed or nearly closed canopy (70–100%) (Alemayehu et al. 2016), are mostly dominated by acacia species. |
Shrublands | Areas that contain shrubs and thorny bushes and are less in density than forestlands. They include a bush canopy (> 50%) which is mixed with some trees and that of grass cover (< 50%). They are non-herbaceous species whose branches begin from the base of their stem and are usually < 5 meters in height (Alemayehu et al. 2016; Belay 2002). The dominant plant species in this category in the study area is Euclea racemosa (locally called “Dedeho”) |
Grasslands | Non-woody areas dominated by grasses with no or few shrubs (Messay 2011) and are used for communal grazing by the residents of the watershed. |
Croplands | Land used for the growth of seasonal and perennial crops, which are mostly grown by rainfall and sometimes by traditional irrigation means. The source of irrigation water is mostly Borkena river. |
Settlements | These are referring to both rural and urban settlements. Those located in rural areas can be of clustered or scattered dwelling units. Urban settlements are those with better road and social service facilities than the rural counterparts. |
Barelands | Parcels of land mostly covered with no or little plant cover and contain exposed rocks. They are indicators of high rate of degradation in the given area. |
Trends Of Land Cover Dynamics
The study found that forestland and shrubland diminished during the analysis period. The available forest was mainly deforested between 2001 and 2015 that give rise to the clearing of forest that covers 110.3 ha, i.e. 50% of the forest cover in 1986. The decreasing trends of forestland and shrubland, and increasing trends of cropland, bareland and settlement in the study watershed happened due to population pressure.
Table 3
Land Cover Changes in Hita-Borkena Watershed in1986, 2001 and 2015
Land cover type | Land cover changes |
1986 | 2001 | 2015 |
Ha | % | ha | % | ha | % |
Forestland | 220.6 | 10.5 | 135.9 | 6.5 | 25.6 | 1.2 |
Shrubland | 774.0 | 37.0 | 441.8 | 21.1 | 285.4 | 13.6 |
Grassland | 254.4 | 12.2 | 234.8 | 11.2 | 298.9 | 14.3 |
Cropland | 766.5 | 36.6 | 1143.7 | 54.6 | 1315.2 | 62.8 |
Settlement | 35.7 | 1.7 | 59.4 | 2.8 | 78.6 | 3.8 |
Bareland | 42.6 | 2.0 | 78.1 | 3.7 | 90.0 | 4.3 |
Total | 2093.8 | 100 | 2093.8 | 100 | 2093.8 | 100 |
Grassland exhibited a unique trend over the 29-analysis period (Table 4). Its coverage increased by 17.5% from 1986 to 2015. This happened due to the fact that cut-and-carry system has begun to be implemented in the study watershed after 2001. The result of this study is different from Gete & Hurni (2001), Messay (2011), Mohammed (2011) and Belay (2002). The first three authors found out that grassland show a decreasing trend over their respective analysis periods. On the other hand, Belay (2002) revealed that grassland followed an increasing trend in Derekolli catchment between 1957 and 2000 due to the shrinking and modification of shrubland.
Table 4
Trends of land cover changes in Hita-Borkena Watershed in different periods
Land cover type | Area dynamics of each land cover type |
1986–2001 | 2001–2015 | 1986–2015 |
ha | % | ha | % | ha | % |
Forestland | -84.7 | -38.4 | -110.3 | -81.2 | -195.0 | -88.4 |
Shrubland | -332.2 | -42.9 | -156.4 | -35.4 | -488.6 | -63.1 |
Grassland | -19.6 | -7.7 | 64.1 | 27.3 | 44.5 | 17.5 |
Cropland | 377.2 | 49.2 | 171.5 | 15.0 | 548.7 | 71.6 |
Settlement | 23.7 | 66.4 | 19.2 | 32.3 | 42.9 | 120.2 |
Bareland | 35.5 | 83.3 | 11.9 | 15.2 | 47.4 | 111.3 |
Total | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
Cropland showed an increasing trend between 1986 and 2015 (Table 4). This result is in agreement with most studies (for example, Aklilu 2006; Belay 2002; Gete & Hurni 2001; Kebrom & Hedlund 2000; Menale et al. 2016; Messay 2011 among others). Figure 4 also displays the expansion of cropland to the steepy slope in the study watershed.
Like the results given by Aklilu (2006) and Menale et al. (2006), the present study witnessed that settlement recorded an increasing tendency between 1986 and 2015 unlike forestland and shrubland. Bareland, which is mostly an overgrazed land containing rock outcrops (Woldeamlak 2002), exhibited an increasing trend alarmingly between 1986 and 2015 (Table 4). This implies that there has been a threat of land degradation (Gete & Hurni, 2001; Menale et al. 2016) in the study watershed. Though it recorded a rapid rate since 1986, the rate after the implementation of landscape restoration including area closure (i.e. after 2001) is relatively slower. However, Mr. Ali, who has been working as an agricultural expert in the study area for a long period of time, witnessed that the interest of the community to conserve the watershed relatively declined immediately after MERET project had stopped in 2012 (personal communication, 2015). The recent railway construction that connects Wollo to Afar passing through the watershed is disturbing the watershed and is contributing to the increase in area coverage of bareland. The researcher also observed that more care is being given for hillsides where area closures are found, but some farmers are letting their animals to graze freely at the foot of the hill (Fig. 5)
Transition Between Land Cover Types
As mentioned above, forestland declined massively in Hita-Borkena watershed between 1986 and 2015. It converted significantly into shrubland, followed by cropland, between 1986 and 2001 and into shrubland, followed by grassland, between 2001 and 2015. Considering the conversion of other land cover types into forestland, the conversion from shrubland into forestland was the most important one during the two periods: 1986–2001 and 2001–2015 (Tables 5 and 6).
In a similar fashion with forestland, shrubland coverage in the watershed decreased extensively. It changed to cropland, followed by grassland in 1986–2015 and 2001–2015. There was of course conversion of other land cover types into shrubland. The main conversions took place between forestland to shrubland and grassland to shrubland during the aforementioned periods (Tables 3, 4 and 5).
In the first year of the analysis period, grassland covered 12.2% of the total area of the watershed. Due to the prohibition of free grazing, though some violations still exist, its coverage increased to 14.3% in 2015 unlike other land cover types. It mainly transformed into cropland, followed by shrubland during the periods between 1986–2001 and 2001–2015. On the other hand, it gained 114.6 and 106.6 ha from shrubland and between 1986–2001 and 2001–2015, respectively. Larger area of bareland was relatively transformed into grassland during the second period than the first one (Tables 5 and 6).
Cropland has been occupying huge portion of the watershed compared to other land cover types. It changed to shrubland, followed by grassland, between 1986 and 2001 and to grassland, followed by shrubland, between 2001 and 2015. It gained larger area from shrubland and grassland in both periods compared to the area it lost (Tables 5 and 6). This is confirmed by Aklilu (2006). Besides, Mohammed (2011) found out that conversion of grassland, forestland and shrubland gave rise to the expansion of cultivated land.
Bareland gained more area through 1986–2015. It expanded at the expense of mainly cropland, grassland and shrubland between 1986 and 2015. It completely lost to cropland, followed by shrubland and grassland, in both periods: 1986–2001 and 2001–2015 (Tables 5 and 6). As indicated by Aklilu (2006), bareland transformed into cropland, plantations and grazing land during the period between 1957 and 2000.
In response to the increased population in Hita-Borkena watershed, rural and urban settlement greatly expanded through the analysis period. It converted into cropland, followed by bareland, between 1986 and 2001 and into cropland, followed by shrubland, between 2001 and 2015. The area it gained was greater than the corresponding area it lost in the two periods. It gained land from grassland, followed by cropland and from cropland, followed by shrubland, in 1986–2001 and 2001–2015, respectively (Tables 5 and 6). Similarly, Kebrom & Hedlund (2000) revealed that rural settlements transformed into cultivated areas, remaining open areas, shrublands and other land cover classes in 1958–1986. The same authors noted that conversion of other land cover categories into rural settlements constituted more than 50%.
Table 5
Land cover change matrix in Hita-Borkena watershed between 1986 and 2001
| From land cover type in 1986 |
To land cover type in 2001 | Land cover type | Forestland | Shrubland | Grassland | Cropland | Bareland | Settlement | Total |
ha | ha | ha | ha | Ha | Ha | ha |
Forestland | 81.1 | 52.6 | 0.0 | 2.2 | 0.0 | 0.0 | 135.9 |
Shrubland | 86.1 | 254.7 | 19.3 | 73.6 | 4.8 | 3.2 | 441.8 |
Grassland | 5.4 | 114.6 | 45.3 | 63.0 | 3.6 | 2.9 | 234.8 |
Cropland | 46.6 | 328.6 | 154.1 | 573.2 | 34.2 | 7.0 | 1143.7 |
Settlement | 1.1 | 10.6 | 17.8 | 13.6 | 0.0 | 16.4 | 59.4 |
Bareland | 0.2 | 12.8 | 18.0 | 41.0 | 0.0 | 6.2 | 78.1 |
Total | 220.6 | 774.0 | 254.4 | 766.5 | 42.6 | 35.7 | 2093.8 |
Table 6
Land cover change matrix in Hita-Borkena watershed between 2001 and 2015
| From land cover type in 2001 |
To land cover type in 2015 | Land cover type | Forestland | Shrubland | Grassland | Cropland | Bareland | Settlement | Total |
ha | ha | ha | ha | Ha | ha | ha |
Forestland | 11.6 | 6.5 | 0.1 | 5.6 | 0.0 | 1.8 | 25.6 |
Shrubland | 72.5 | 94.0 | 32.6 | 61.9 | 13.6 | 10.7 | 285.4 |
Grassland | 30.6 | 106.6 | 58.3 | 84.2 | 10.3 | 8.9 | 298.9 |
Cropland | 20.1 | 204.0 | 126.7 | 920.2 | 22.7 | 21.5 | 1315.2 |
Settlement | 1.1 | 19.1 | 12.5 | 29.7 | 0.0 | 16.2 | 78.6 |
Bareland | 0.0 | 11.6 | 4.6 | 42.0 | 31.5 | 0.3 | 90.0 |
Total | 135.9 | 441.8 | 234.8 | 1143.7 | 78.1 | 59.4 | 2093.8 |
Link Among Land Cover Dynamics, Slope And Area Closure
Attempts were made to see the association among land cover dynamics, slope and area closure in Hita-Borkena watershed, northeastern Ethiopia. The study watershed was classified into 3 slope classes, i.e. gentle slope (0–12%), mid slope (12–36%) and steep slope (> 36%), based on Aklilu (2006).
As can be seen in Fig. 6a, cropland was the major land cover type on the gentle slopes in the watershed. On the other hand, forestland, grassland and shrubland shrank down on the same slopes through 1986–2015.
Though the mid slopes in the watershed were dominated by cropland, grassland showed an increasing trend in the analysis period (Fig. 6b) mainly due to area closure. However, due emphasis should also have been given to forestland and shrub-land. This is because the watershed could be protected not only by grassland but also by forestland, shrubland and different SWC measures. In relation to grazing land, Aklilu (2006) identified that it is mostly located on gentle and mid slopes of Beressa watershed, Ethiopia.
In 1986 and 2001, unlike gentle and mid slopes, steep slopes in the watershed predominantly contained shrubland. In 2015, three land cover types: cropland, grassland and shrubland, were dominating the watershed (Fig. 6c). Grassland’s coverage on the steep slopes significantly increased as a result of the presence of area closure in the watershed under consideration, but forestland on the same slope class drastically decreased. This implies that there was negligence of tree planting and high rate of deforestation in the study area. In line with this, Gete & Hurni (2001) noted that 90% of the forestland on the steep slopes in Dembecha area, northwestern highlands of Ethiopia had deforested between 1957 and 1982.
Comparisons were made in order to show how the amount of vegetation including forests, grasses and shrubs in combination and cropland on gentle, mid and steep slopes changed in the watershed between 1986 and 2015.
As revealed in Fig. 7a, cropland showed a considerable increase in the gentle slope throughout the analysis period for the reason that such slope is favorable for crop production. On the other hand, forestland, grassland and shrubland (F-G-S) in combination exhibited opposite trend.
In 1986, F-G-S on mid slopes covered almost 30% of the area of the watershed while cropland constituted only 11%. F-G-S on the same slopes declined to 18 and 14% in 2001 and 2015, respectively, whereas there was the corresponding rise in the size of cropland in the indicated years (Fig. 7b). Though not possible to resume the vegetation cover in 1986, the present F-G-S has been conserved as a result of the restoration project that incorporates area closure in the study watershed.