Object‐based image analysis (OBIA)‐based gully erosion dynamics, sediment loading rate and sediment yield study in Lake Hawassa Sub‐basin, Ethiopia

Gully erosion is a serious environmental issue in Lake Hawassa Sub‐basin. It has affected a large portion of the catchment, and a large amount of sediment has been transported into the lake. This condition needs to be addressed. However, before conservation techniques are implemented, the gully systems should be mapped, their extent and changes over time need to be measured, and their capacity to produce sediment from their catchments should be determined. The objective of this study were, therefore, to map and analyse the change in gully erosion and quantify the sediment loading rate and sediment yield from the active and connected gullies in Lake Hawassa Sub‐basin using object‐based image analysis (OBIA) with high‐resolution SPOT 5 satellite image of the years 2011 and 2020. This method was integrated with field observation for mapping gully features of the study area, in comparison with automatic digitization carried out with the help of eCogenition Developer Version 9.1 and ArcGIS tools. Overall accuracy and Kappa coefficient were determined and were found to be 85.2% and 0.81, respectively, for the image of year 2020 and 81.1% and 0.76, respectively, for the image of year 2011. Based on the OBIA method, the extent of gullies (in area) were found to be 63.5 km2 in 2011 and 79.9 km2 showing a rapid increase between 2011 and 2020 (an increase of 16.4 km2 (24.4%) in the 10 years considered). The later result shows that 5.53% of the area of the Lake Hawassa Sub‐basin is affected by gully erosion. The maximum gully density in the study area was found to be 589 km/km2 in 2011 and this increased to 884 km/km2 in 2020. The sediment loading rate from the Lake's catchment was found to be in the range of 12.62 to 38.59 ton per hectare per year. The sediment yield from the Lake's catchment was 8.83 to 27.02 t/ha/year. The total annual volume‐based sediment yield at the Lake generated from the gully was 2.39 million cubic meter considering sediment delivery ratio of 70% for fully connected gullies. This result shows that 0.21% of the storage capacity of the Lake was being lost due to sedimentation from the gully system every year. From the result by dividing the total volume of the sediment by the surface area of the lake, one can see that a silt thickness of 2.51 cm was being deposited in the Lake every year.

determined and were found to be 85.2% and 0.81, respectively, for the image of year 2020 and 81.1% and 0.76, respectively, for the image of year 2011. Based on the OBIA method, the extent of gullies (in area) were found to be 63.5 km 2 in 2011 and 79.9 km 2 showing a rapid increase between 2011 and 2020 (an increase of 16.4 km 2 (24.4%) in the 10 years considered). The later result shows that 5.53% of the area of the Lake Hawassa Sub-basin is affected by gully erosion. The maximum gully density in the study area was found to be 589 km/km 2 in 2011 and this increased to 884 km/km 2 in 2020. The sediment loading rate from the Lake's catchment was found to be in the range of 12.62 to 38.59 ton per hectare per year. The sediment yield from the Lake's catchment was 8.83 to 27.02 t/ha/year. The total annual volumebased sediment yield at the Lake generated from the gully was 2.39 million cubic meter considering sediment delivery ratio of 70% for fully connected gullies. This result shows that 0.21% of the storage capacity of the Lake was being lost due to sedimentation from the gully system every year. From the result by dividing the total volume of the sediment by the surface area of the lake, one can see that a silt thickness of 2.51 cm was being deposited in the Lake every year. a large source of silt in a variety of settings (Nachtergaele et al., 2002) and comprehensive monitoring and improved gully placement prediction are crucial (Valentin et al., 2005).
Lake Hawassa lost about 4% of its storage capacity in between 1999 and 2011 (Abebe et al., 2018) due to sedimentation. In addition to this, a considerable portion of the Lake's catchments are having under extensive gully erosion (Belete, 2013;Belete et al., 2021;Hoogenbooh, 2013;Moges & Holden, 2008). However, the exact figure of sediment yield from gully erosion into the lake has not yet been studied. Moreover, it is vital to understand how much gully erosion contributes to the overall sediment output to make better decisions for sediment reduction and soil protection (Nyssen et al., 2004;Valentin et al., 2005). Therefore, this study aimed to examine gully dynamics in 10 years period (2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020) of the study area and estimate the sediment yield from the connected gullies over the indicated period. Therefore, mapping existing gullies and their dynamic activity over time is critical for monitoring land degradation and its environmental and socioeconomic impacts, that is, quantifying gully erosion dynamics can help understand gully formation and is used for spatial and temporal evolution of the study watershed and used as sustainable planning strategies designed for the stabilization of gullies and prevention of gully development in sensitive locations (Jetten et al., 2003;Martha et al., 2011).
There are several studies on gully dynamics by using different approaches. Some applied visual stereoscopic analysis of aerial photographs to measure temporal changes in length, area, and density for various gullies (Mararakanye & Nethengwe, 2012a;Shruthi et al., 2011Shruthi et al., , 2015. To measure and monitor changes in gullies, visual stereoscopic analysis of a series of aerial photos, pixel-based satellite image analysis, object-based analysis of data from airborne LiDAR data, and low-altitude unmanned imaging platforms are among some the effective methods (Nachtergaele et al., 2002).
Object-based image analysis (OBIA) principles and methodologies have found widespread use in image classifications and object annotation (Gonçalves et al., 2019;Ma et al., 2018). It is based on image segmentation, edge detection, and feature extraction methods that have been around for decades (Blaschke, 2010). OBIA uses properties such as size, shape, texture, color, and contextual information to identify or detect desirable features Csillik et al., 2014;Kadir & Brady, 2001;Ma et al., 2018). OBIA develops "extremely homogeneous" zones (Gonçalves et al., 2019) and it is supposed to be a knowledge-driven strategy that is aimed to generate a set of homogeneous image objects that disclose current real-world qualities (Abdi, 2019;Bangalore Venkata Chalapathi et al., 2015;D'oleire-Oltmanns et al., 2014;Francipane et al., 2018) which will produce an object in his or her mind that mirrors the real-world object. A programming tool called eCognition Developer was developed for this type of image-based object analysis. It has been widely used in the development of rule sets for the automatic analysis and classification of remote-sensing data (Trimble, 2014). It also has functions for feature extraction, change detection, and object identification (Nachtergaele et al., 2002;Poesen et al., 2011;Trimble, 2014). eCognition has the advantage of avoiding errors made in manual digitization and taking much less time.
After all the mapping of the gully erosion, the prediction of sediment loading rate and sediment yield from the gully erosion in the catchments is very crucial. Because of, high sediment loading rate and sediment yield (Belete et al., 2021;Ghimire et al., 2013) generated from gully erosion are about 80% in the magnitude of sediment generated by sheet and rill erosion.
For the determination of sediment loading rate and sediment yield, activeness and connections of the gully network and their development history identification are very crucial one. Thus, the criteria stated by (Moges & Holden, 2008;Nyssen et al., 2006); if the gully is active, it has at least one of the following are major criteria: namely (1) a cave-in-causing undercut or plunge pool; (2) a vertical or almost vertical cliff; (3) the wall of the gully lacks vegetation; (4) stress fractures and finally; and (5) a collapsed side wall and debris on the gully floor.
The connectivity of the gully system gives the information on the sediment delivery ratio (SDR) used by characterizing specific gully situation at field level investigation. Because not all gully material ends up in the river or reservoir at the catchment exit since some sediment is deposited in sinks (Ndomba & Griensven, 2011;Roux et al., 2020) and their development history determine their ages for sediment loading rate and sediment yield. According to Hughes and Prosser (2012), Ndomba andGriensven (2011), andRoux et al. (2020), the information of the SDR ranged from 0% for disconnected gullies to 40% for possibly connected gullies, 50% for partially connected gullies, and 100% for fully connected gullies in the study. But most of the time 75% SDR adopted in different studies.

| Description of the study area
This study was conducted in Lake Hawassa Sub-basin. The study Lake Basin is a closed system with a total area of 1445.24 km 2 and situated in the northernmost central part of the Rift Valley Lakes basin of Ethiopia. The Sub-basin is divided into eight watersheds: Hawassa Loke, Tikur Wuha, Lalima, Shamana Hurufa, Shashemene-Toga Jara_Henisa, Boricha Dore and Deneba Watersheds (Belete, 2013). Geographically it lies between latitudes 6°48′45" and 7°14′49″ N and longitudes 38°16′34″ and 38°43′26″ E (Figure 1). The highest and the lowest altitudes of 1624 and 2987 m above sea level, respectively. The majority of the land (56%) is in the flat to gentle slope class (0%-8%), followed by the moderate slope class (8%-30%), the high to extremely steep slope class (>30%) (Belete, 2013).
There are three distinct seasons in the watershed (Belete, 2013;Legesse & Valletcoulomb, 2003). The long rainy season in the summer (June-September) accounts 50%-70% of the annual total rainfall. The second season is the dry season that lasts from October to February (Belete, 2013;Belete et al., 2021;Legesse & Vallet-coulomb, 2003). The "smallest rainy" season occurs between March and May in which around 20%-30% of the annual rainfall falls. According to Thornthwaite's system of identifying climate or moisture areas, the climate in the area ranges from dry to subhumid (Belete et al., 2021). Based on the rainfall data obtained from meteorological stations , the mean annual rainfall amount is 1050 mm.

| Data collection and analyses
Due to their typically excellent spatial resolution and capacity to discern wavelengths in a range of bands, the SPOT 5 satellite image of 2011 and 2020 with multispectral images of four spectral bands (green, red, near infrared, and shortwave infrared) and the spatial resolution of (the pixel size with 2.5 and 1.5 m) were used for this study. The image was collected from Geospatial Information Institute of Ethiopia. The SPOT 5 image was chosen since the study objective focused on the gully dynamic over a 10-year period, and this image was accessible from the SPOT 5 launched era. The highest resolution image was selected because gullies less than 2.5 m will become embedded inside the pixels; image analysis was not unable to detect it (Mararakanye & Nethengwe, 2012;Mararakanye et al., 2015).

| Gully dynamics analyses
For the analysis of gully change, two SPOT 5 images recorded of 10 years apart were compared to determine for the active gully systems and, how quickly the gullies changed the area and the density in the watershed. In the years between 2011 and 2020, any gully that expanded laterally in size, length, and density was classified as an active gully. This came in convenient for computing the normalized difference vegetation index (NDVI) and the modified normalized difference water index (MNDWI). The image analysis was done with F I G U R E 1 Location of the study area. eCognition developer Version 9.1 and the postprocessing of the image objects was done by using ArcMap. In eCognition, the OBIA ruleset was used to extract gullies from the SPOT 5 image of 2011 and SPOT 5 image of 2020. Following the gully extraction, the two vector data sets were overlaid in ArcMap, and the surface area was computed using a simple raster computation.

Classification approaches
To detect gully-affected areas, the classification approach combining the general concept of OBIA with a top-down approach was developed. The ruleset in eCognition based on "topdown" approach, with the smallest level being pixel-based and the largest level being the "entire scene," creating three levels of varying segment sizes from large to small was adopted. Using SPOT 5 image in eCognition, a ruleset was created that took into account futures brightness, texture, and connection to neighboring objects.
After overlaying the two vector data sets in ArcMap, the surface area difference, length of the gully, and their density was determined using a simple raster calculation. This calculation only accounted for lateral growth, length, and density, but not account for an increase in gully depth (Roux et al., 2020).
By top-down approach, the largest homogeneous regions, such as vegetation cover and settlement/residential areas were first masked out. The vegetation's features were identified based on high NDVI values. The addition of the NDVI as an additional image layer to the firstlevel segmentation method allowed for a more accurate delineation of objects for vegetation cover identification. NDVI was calculated at the image object level, rather than per pixel, by using eCognition's bespoke algorithm function, with the help of red and near-infrared bands of the SPOT 5 image.
The normalized difference water index (NDWI) provided for better differentiation between water bodies; structures and was used to remove the rivers and Lake body from the classification process, according to Xu (2006). Thus, the water body was also masked out by computing the MNDWI using the image's green and short-wave infrared bands, similar to the methods used by Ji et al. (2009) andXu (2006). Water reflects the most in the SPOT 5 image's green band, while absorbing the most in the NIR and SWIR bands. The MNDWI is a normalized index comparable to the NDVI, with positive values for water bodies. These two indices were used in the first step of the classification procedure to remove water and vegetation cover. This allowed a significant portion of the image to be removed from the rest of the classification process, resulting in more simplified results.

| Accuracy assessments
An accuracy assessment reflects the difference between the classified image and the reference data. Two maps were compared to determine the classification accuracy: the manually digitized map and the classification-generated map (OBIA map) in this study (Lillesand, 2008).
The results of the four independent accuracy evaluations must be completed and compared (Lillesand, 2008). This made it possible to compare different accuracy assessment methods against one another. Therefore, in this study, the error matrix findings are examined using the producer's accuracy, user's accuracy, total classification accuracy, and kappa coefficient.
The total number of test samples is multiplied by the number of samples that were successfully classified to determine the overall classification accuracy, which combines the user and producer accuracy. Tracking commission errors was done using the user's accuracy, whereas tracking omission errors is done using the producer's accuracy. The Kappa coefficient of accuracy measures the discrepancy between the actual and chance agreement in the error matrix (Persello et al., 2010).

| Total area of overlap
The total area of overlap approach works by calculating the total area of gullies that overlapped between the manually digitized data set and the OBIA data. This was done by using a simple raster calculation to compare the digitized gullies to the gullies retrieved using OBIA. Both data sets (digitized data and data extracted by OBIA) were converted to raster files in ArcMap as follows: no data values were assigned a code (i.e., areas where gullies were not found); while areas, where gullies were identified, were assigned another code for the OBIA extracted data set and other codes for the digitized data set. The first code for digitized data set denoted a nongullied area, while the second code for those digitized reflected a gullied area. The two data sets were added, resulting in new code for four classes. Between the two data sets, these classes reflected the various combinations of gullies and no gullies. This was done using the data sets from 2011 to 2020.

| Gully density expansion
The spatial distribution of gully density was also assessed and mapped based on the data obtained from OBIA 2011 and 2020 gully area and length. Hence, the total gully length (skeleton exported from OBIA) per area of the subwatershed was used to compute gully density within the subwatershed. The gully density for the entire area was calculated using ArcGIS's line density tool. It was calculated by multiplying the total length of the gully within the circular kernel (50 m search radius) by the total area of the circular kernel. The gully density difference between the 2 years (2011 and 2020) was used to compute the total change in gully density over the 10-year period.

| Determinations of volume and sediment loading rate
Quantification of volume is based on direct measurements in the field, while determinations of sediment loading rate are based on the required volume and age of the gully system. Therefore, input parameters used to quantify sediment loading rate include the historical formation time of the gully, and the activity, and connectivity of the gully network system. In this study; field-based data were collected by direct measurement on the depth, width and length for active and connected gully networks.
Therefore, it is important to determine the activities and connectivity of the gully system in the Lake Sub-basin under study which is shown below in Figures 2 and 3. This can be confirmed by field investigations. Therefore, the gully activity and connectivity of the Lake were evaluated using the criteria of Moges & Holden (2008) during field investigation period.
After identification of the active and connected gullies to the Lake system shown in Figures 2 and 3, their locational, depth as well as width measurements were carried during the field investigation for only active and connected gullies because of the significant (50%-100%) of the sediments loading rate are obtained at the reach of the watershed (Hughes & Prosser, 2012;Nnomba, 2010;Roux et al., 2020).

A. Historical period gully initiation
The various gully systems that historically initiation period from Landsat imagery are important information for calculating sediment loading rate and sediment yield from the gully network. For example, Landsat and stereo aerial imagery (Hoogenbooh, 2013), eventbased interviews and measurements (Moges & Holden, 2008;Nyssen et al., 2006) are among the methods previously used for studies of gully formation. For this study, a Landsat image from 1970 to 2016 of (LS1-5MSS, LS4-5 TM, LS7-ETM, and LS8 OLI TIRS) were used to determine the development age of the gully. B. Volume of gully erosion and sediment loading rate computations There are numerous ways for determine of sediment loading rate due to gully erosion (Gyssels & Poesen, 2003;Poesen et al., 2003). To do so, computation of the volume (m 3 ) of the soil loss by gullies, field level measurement length, width, and depth at various locations for the selected active and connected gullies were carried by using 50 m standard tape and GPS.

F I G U R E 2 Active gullies on spot image.
Finally the volume was computed by using the following equation (Nyssen et al., 2006), once these data are averaged: where L i is the gully segment's length in meters and A i is its cross-sectional area in m 2 . With the computed volume, the area-specific long-term gully erosion sediment loading rate (RL) in t/ha/year was derived.
where SLR the area-specific long-term gully erosion/sediment loading/rate (t/ha/y), V is the gully volume at the time span under consideration (m 3 ), Bd is the average bulk density of the soils in the contributing area (t/m 3 ) found from by laboratory tested the soil sample collected from the study area which is an average value of 1.86 t/m 3 , T is the time span under consideration (years), and Ca is the watershed area (ha) (Nyssen et al., 2006). The ground truth-based data as well as airborne approaches for the determination of gully volume and an erosion rate was adopted for this study.

C. Sediment yield computations
The amount of sediment deposition rate at the reach for the study Lake was determined based on the theory of SDR which was characterizing the gully situation at field level i.e. the sediment delivery ratio used by characterizing specific gully situation at field level investigation (Hughes & Prosser, 2012;Nnomba, 2010;Roux et al., 2020). Therefore the 70% of SDR was adopted for this study area based on the justification given by Roux et al. (2020) for active and connected gullies systems. Finally, 70% of the sediment loading rate is considered as sediment yield at the reach of the watershed.

| Quantifying gully changes
The dynamics of the gully system were investigated in the study Lake Hawassa Sub-basin in eight watersheds based on their drainage characteristics and their alterations measured (Watershed 1-8).

| Accuracy assessment
The basic accuracy assessment was carried for 10 span years. All four independent accuracy assessments shown in Figure 4, results indicated that they are excellent and higher in a range which was presented in Table 1. The accuracy assessment is based on the correctness of the reference data, which in this case was the manually digitized gullies, as stated by Yale's Centre for Earth Observation (2003).
The accuracy of the OBIA-derived gullies map versus the manually digitized gullies map is shown in Figure 4.
In comparison to manually digitized gullies, Table 1 summarizes the accuracy data of the OBIA-generated gullies map. The OBIA gullies map's overall classification accuracy is 85.2%, with a total kappa coefficient of 0.81 was found for 2020 whereas 81.1%, and 0.76 was for the year 2011. But the results shown for the two periods were different due to the SPOT 5 satellite images used, their resolution are different. SPOT 5 images of 2020 was 1.5 m × 1.5 m resolution, in which the gully future was easily identified during manual digitization and reduced man-induced error. In the case of the SPOT 5 satellite image for 2011 its resolution 2.5 m × 2.5 m, which was less than the resolution 2020 SPOT 5 satellite image. Thus, some of the gully futures are missed during manually digitization, in the case of automatic digitization with the help of eCognition developer software, gullies had a more defined border surrounding them, whereas manually digitized gullies' borders did not hug the gully's edges as tightly as those generated using OBIA, resulting in a larger region around the gullies, increasing the area accessible for successful overlap. In addition to this, the manually digitized gullies were rougher and included vegetated regions.

| Gully dynamics in the study area
From the viewpoint of the gully dynamics conceptualization, it is necessary to estimate both the change in gully area expansion as well as the density of gully systems within the watersheds (Shruthi et al., 2015). The extent of gully erosion in the watersheds was assessed for 10 year period, that is, from 2011 to 2020. From the image analysis results and field level investigation, it was possible to identify that the bulk of the gullies is concentrated in the South East and North East regions of Lake Hawassa Sub-basin; shallower gullies are on higher and steeper slopes, but very denser at higher and steeper slopes. The gully dynamics in the study area in terms of areal expansion and change in density are presented as follows.

| Changes in gully area extent
The results of analyses revealed that the areal extent of the gully system in the study area considerably increased in 2020 from the corresponding value in 2011 which was shown below in Table 2 and Figure 5. The extent of the gully-affected area in 2011 was 63.5 km 2 as shown in Figure 5, whereas it was 79.9 km 2 in 2020 as determined by using OBIA with the help of eCogintion Developer software (automatic digitization). This shows that the areal extent of the gullies increased by 16.4 km 2 (24.4%) in the last 10 years. However, based on manual digitization, the gully area extent was found to be 68.1 km 2 in 2020 and 51.5 km 2 in 2011, showing an increase of 16.6 km 2 (32.24%) over the same study period. These show that the results of manual digitization of gully areal extent is very close to the result obtained by OBIA. The slight difference was due to image resolution involved in the OBIA and manual digitization. In addition to this, the diversity of the gullies is another cause of the different result as justified by (D'oleire-Oltmanns et al., 2014;Shruthi et al., 2015) for similar studies.
The expansion of gullies was further investigated by dividing the study area into eight watersheds, and it was conducted using the OBIA and manual digitization. Table 2 shows the variations in area extent of the gully system in the eight watersheds for 2011 and 2020.
The result reveals that the Lalima and Boricha-Dore Watersheds are significantly affected by the gully whereas the Watershed Denba was low compared to the other subwatersheds in which the gully affected. In generally a total 5.53% (79.9 of 1445.4 km 2 ) of area was affected by the gully erosion developed in the Lake Hawassa Sub-basin for the last 53 years (1970-2020).

| Gully density dynamics
Gully density is another indicator of gully system changes. The overall change in gully density during the 10 year period is depicted in Figure 6a,b.  Figure 6a,b show that gully density in the study area ranged from zero to 589 km/km 2 in 2011 and from zero to 884 km/km 2 in 2020. The gully density during the 10 years of the study area increased by 33.37%. From field investigation, it was confirmed that the most dense and deepest gullies are found at the lower and flatter slopes. These findings are similar to the findings in the same study area by Belete (2020) and Moges and Holden (2008) and several other studies (Shruthi et al., 2015).

| Quantity of sediment from the gully system
To quantify the sediment yield, the historical gully initiation period is an important parameter, which was obtained from Land Sat image to distinguish the initiation period of the gully erosion. For this study, initiation of the gully period was assessed for selected active gullies from Land Sat image based on the criteria stated by Oostwoud et al. (2000).

| The historical period of the gully erosion formation
As detected from Landsat image, the oldest and the youngest gullies were appeared in 1970 and 2016, respectively. Further analysed on the images showed that active and connected gullies were found dominantly in five subwatersheds found in the Western part the Lake Sub-basin. Three active and connected gullies were identified in Deneba Watershed and they emerged in 1970; three gullies were identified in Jara-Henesa Watershed and they emerged 1978; nine gullies were identified in Lalima Watershed and they emerged between 1985 and 2003; one gully system was identified in Shamana Hurufa Watershed with its tributaries and it emerged between 1983 and 2002; one more gully system was found in Borecha-Dore Watershed that emerged in 1995. Considering the 17 active and contentious gully systems listed above, the tributary gullies were initiated between 1978 and 2016, out of, which 25% the tributary gullies emerged between 1973 and 1980; 25% of the tributary gullies appeared between 1986 and 1990; 35% of the tributary gullies appeared between 1993 and 2000, and finally, 15% of the tributary gullies came to existence between 2003 and 2016.

| Sediment volume, loading rate, and yield from gully erosion
The results of sediment volume, sediment loading rate, and sediment yield were determined for five Watersheds out of eight Watersheds in the study areas are presented in Table 3. The mean value of 1.86 t/m 3 of the dry density was used for mass base sediment yield computation which was obtained from collecting and analysing representative soil samples from all the watersheds at the filed level.
The sediment production from gully erosion for the OBIA gullies was determined with an average sediment loading rate of soil loss ranging from 12.97 t/ha/year and 38.59 t/ha/year. Thus, high sediment loading rate was recorded in Lalima Watershed whereas the lower value recorded 12.97 t/ha/year in Borecha-Dore Watershed.
Thus, the annual base soil loss rate by the gully erosion system are higher and increased by 29.33% when compared the result obtained by Moges & Holden (2008) in the Watershed for all the active and connected gully systems who find the sediment loading rate range from 6 to 30 t/ha/year.
The gully erosion contribution of the siltation to Lake Hawassa was also evaluated based on the result of sediment volume and annual sediment yield from the watersheds. The results were simplified by using a constant SDR of 70% (Hughes & Prosser, 2012;Nnomba, 2010;Roux et al., 2020). The annual sediment yield from the five subwatersheds was also determined with a total volume of 3.42 MCM of soil eroded by gully erosion per year. Therefore, by taking SDR of 70%, the Lake siltation by gully erosion system was estimated to be 2.39 MCM/year. Based on this siltation rate, the Lake's storage volume loss was evaluated based on the reconstructed Lake volume by bathymetry survey techniques (Abebe et al., 2018), which was found to be 1,174.61 MCM in 2011 and, thus, 0.21% of the Lake is being taken up every year by siltation from gully erosion. The annual thickness of the Lake sedimentation was also determined by taking the ratio of the sediment added to the Lake and the surface area of the Lake by considering constant Lake volume loss (2.39 MCM/year) and constant surface area of the Lake (95.56 MM 2 as determined from SPOT 5 Image 2020). Dividing the volume of the sediment that entered into the Lake by the surface area of the Lake, results in 2.51 cm thickness of silt, and this shows that this much silt had been deposited in the Lake from gully erosion in the 10 years period. Other studies (Abebe et al., 2018;Belete et al., 2021;Menberu et al., 2021) also showed that the Lake's volume was lost by 4% (46.67 MCM) from 1999 to 2011 as a result of siltation from the catchment with a rate of 0.33% (4.76 MCM) per year.
Considering the volume will decline by 0.21% or silt thickness of 2.51 cm annually, the estimated lifespan of Lake Hawassa is 476 years. Using the lake bathymetry survey data of T A B L E 3 The sediment yield rate from all the active gully systems 2020. 1999 and 2011, this conclusion was confirmed by the reconstructed reservoir volume by bathymetry techniques in 2011 (Abebe et al., 2018), where 46.67 MCM (4.63 cm average thickness) of silt got deposited in the lake during the 12-year (1999-2011) interval between bathymetric studies. The result difference in between (Abebe et al., 2018) and the finding of this study is about 8%. This is acceptable as the siltation in reservoirs from gully erosion is about 90% (Simon et al., 2011;Tibebe & Bewket, 2011;Zegeye et al., 2016) of the total siltation.

| CONCLUSION
Accurate and detailed spatial information on gully location and extent at an appropriate spatial scale is an essential part of evaluating the impacts of the gullies on sedimentation in catchments, lakes, and reservoirs. This is because, mapping of existing gullies and their dynamics over time is useful for monitoring land degradation as well as environmental and socioeconomic impacts. In addition, such information is vital for coming up with sustainable strategies in stabilization of gullies and prevention of gully development in sensitive locations. Accordingly, this study used OBIA method to identify the gully location, extent, and sediment yield at various times. A programming tool called eCognition Developer, developed for this type of analysis, was employed for this purpose in the study. Based on the results of field observation, measurements, and automatic digitization, eCognition Developer was found to be effective in identifying even finer gully-related edges within complicated gully networks. Based on the automatic digitization tool and further manipulation of the generated data, the areal extent of the gully network was found to be 63.5 km 2 in 2011 and this value was found to be 79.9 km 2 in 2020. These values show that the change of the areal extent of the gullies increased by 16.4 km 2 (25.83%) in the 10 years period. The result of gullies had developed and sediment yield in subcatchments indicated that gullies had developed rapidly down slope over the last 6 to 53 and 10 years, respectively.
Moreover, the analyses of gully density values, determined using the drainage density formula, showed that, it ranged from zero to 589 km/km 2 in 2011 and zero to 884 km/km 2 in 2020, with increase of 33.37%. These changes were found to be very fast when compared to findings in other studies. These values show that there have been considerable changes in the areal extent and density of gullies in the study area. Thus, if these gully dynamic continues in this trend, the various landuse/land cover types and the socioeconomic activities of the communities living in it will be affected considerably.
On top of this, sediment yield values were determined considering the gullies that were extracted by the OBIA method. In this regard, the sediment loading rate was found to range from 12.62 to 38.59 t/ha/year or the sediment yield 8.83 to 27.01 t/ha/year. Based on this value, it can be concluded that the Lake was silted and lost its capacity by the sediment from the gullies at a rate of 2.39 MCM/year. Furthermore, the majority of the gullies are very wide and deep in nature, and the runoff being channeled has made control measures and ameliorating interventions very difficult and expensive to implement. Thus, all concerned bodies should give the required level of attention and exert all possible efforts to manage the catchment and control gullies from further expansion so as to manage siltation in the Lake and protect it from losing its volume in the future.