Natural Pans as an Important Surface Water Resource in the Cuvelai Basin—Metrics for Storage Volume Calculations and Identiﬁcation of Potential Augmentation Sites

: Numerous ephemeral rivers and thousands of natural pans characterize the transboundary Iishana-System of the Cuvelai Basin between Namibia and Angola. After the rainy season, surface water stored in pans is often the only affordable water source for many people in rural areas. High inter- and intra-annual rainfall variations in this semiarid environment provoke years of extreme ﬂood events and long periods of droughts. Thus, the issue of water availability is playing an increasingly important role in one of the most densely populated and fastest growing regions in southwestern Africa. Currently, there is no transnational approach to quantifying the potential storage and supply functions of the Iishana-System. To bridge these knowledge gaps and to increase the resilience of the local people’s livelihood, suitable pans for expansion as intermediate storage were identiﬁed and their metrics determined. Therefore, a modiﬁed Blue Spot Analysis was performed, based on the high-resolution TanDEM-X digital elevation model. Further, surface area–volume ratio calculations were accomplished for ﬁnding suitable augmentation sites in a ﬁrst step. The potential water storage volume of more than 190,000 pans was calculated at 1.9 km 3 . Over 2200 pans were identiﬁed for potential expansion to facilitate increased water supply and ﬂood protection in the future.


Introduction
The Iishana-System is part of the western CB (Cuvelai Basin), which crosses the border between southern Angola and northern Namibia. The basin is one of the most densely populated areas in southern Africa; around 40% of all Namibian inhabitants live in this region. The high population density is expected to increase drastically at a growth rate of 1.4% in the next several years [1]. Projections for the year 2041 anticipate a population count 27% higher than what exists today [2]. One factor driving this growth is the presence of slightly more fertile soil than in the rest of the country [3]. The cultivation of rain-fed sorghum and millet and the practice of stock farming are the most important sources of rural income [4,5]. Most preceding research on floods and droughts in the Iishana-System focused on the estimation of the flood extents [13,17,18,30], flood modeling [16,27], the description of flood impacts on livelihoods [10], and drought hazard assessment [5]. Thus, most studies concentrated on selected parts of the Iishana-System, but never on the entire region. Additionally, research has been carried out by [31][32][33][34] to analyze the water supply function and quality of hand-dug wells. However, there is no quantitative information on the water storage and supply function of the natural depressions in the Iishana system. Nor does information exist about the types, distribution, connectivity, and storage volume of natural and artificial pans. Thus, this study aims to comprehend the hydrological complexity of the entire Iishana-System by identifying and classifying the pattern, volume, depth, and distribution of its small, natural water pans. These patterns were identified and characterized by using a Blue Spot Analysis, which was based on the high-resolution TanDEM-X DEM (digital elevation model). Further, surface area-volume ratio calculations were undertaken for determining suitable augmentation sites.

Study Area
The Iishana-System is located in the western part of the CB [35] (Figure 2). The size of the total CB is about 97,620 km 2 . The Iishana-System covers about 18,370 km 2 . Approximately 1.2 million people live in this region (34% in Angola and 66% in Namibia) [3,21,36]. The study area is bounded by the Kunene River and Okavango River (Angola) in the north and by the Etosha Pan (Namibia) in the south. The topography is characterized by very flat slope angles with average gradients of 1‰ in Angola to 0.5‰ in Namibia. With an elevation of about 1200 to 1100 m, the flat terrain plunges into the Etosha Pan [11]. The Iishana-System is defined as a diffuse network of shallow and ephemeral channels [3,5,10,11,13,23,[25][26][27]. Their sizes are highly dynamic with widths reaching from 1.750 to 100 m and depths ranging from 1 to 7 m [11].
Several shallow depressions (natural pans) have formed due to a lack of morphologically well-defined discharge channels. These isolated pans connect to broad, undefined channels once they are filled with local rainwater [3].
Some of these channels have been covered by windblown sand, especially towards the southeast where drier conditions are present. As a result, the southern Iishana-System is characterized by tens of thousands of small pans, which are usually never connected to each other [11] except during heavy flood events.
The elevated areas between the Iishana consist of Kalahari sand. These locations are feasible for crop production and subsistence rain-fed agriculture [3]. In the lower Iishana areas, aeolian and fluvial sediments are present. These are frequently characterized by fine-grained material (clay fraction), calcite and, according to own investigations, also by silicate crusts. This composition supports low infiltration capacities and high surface runoff [24,[38][39][40].
In terms of climate, the CB spans between a subtropical north and a semiarid south. The potential evaporation in the Iishana-System increases from west to east from 2600 to 3200 mm/a [7,11]. The annual precipitation increases from about 350 up to 950 mm along the same gradient. The rainy season runs from October to April and the dry season from May to September. The climate is characterized by a high interannual rainfall variability, causing floods and droughts with all their consequences, as mentioned in the preceding section [6,[8][9][10].

DEM Preprocessing
This study was based on the TanDEM-X DEM, which provides a horizontal spatial resolution of around 12 and a vertical resolution of less than 1 m (i.e., information on the surface elevation is stored in floating point precision). The relative vertical accuracy is less than 2 m for regions with subtle topography (slope < 10 • ) [41]. From a hydrological point of view, the DEM is a great improvement compared to former DEMs (e.g., the SRTM DEM), which provided spatial resolutions of up to 20 m and accuracies of 16 m horizontally and 10 m vertically [42,43].
In using the DEM, two challenges had to be addressed prior to the Blue Spot Analysis. First, the DEM comes with some local inconsistencies and errors, which are due to phase-unwrapping problems caused by water surfaces. This effect typically manifests in locally restricted, highly increased elevation values (up to tens of meters compared to the surrounding terrain). Second, the DEM is a surface model, and the provided information displays the height of the terrain plus the height of man-made objects, vegetation, etc. A removal of these objects (along with the above-mentioned errors) is preferred for the hydrographic modeling. The processing steps, outlined in (Figure 3a), were applied to account for these issues and the Water Indication Mask (Water Mask), which comes along with the DEM [44], was used to remove inconsistencies over water surfaces [45]. The resulting gaps were interpolated with the help of the Image Analyzer Function of ArcGIS, called 'Eliminate void and fill' [46]. In the following, the Invert Fill algorithm (Figure 3a) was applied. This method is based on the approach of [47] and the Fill-function algorithm implemented in ArcGIS. First, the previously edited DEM was inverted with the help of the raster calculator by multiplying the DEM raster values by minus one. Second, the ArcMap Fill-function algorithm was used to fill the inverted peaks. Next, the inverted and the filled DEM were retransformed with the raster calculator by multiplying them by minus one again. As a result, the former inconsistencies, which appear as peaks in the DEM, were removed, but natural pans remained.
In the next step, vegetation obstacles were reduced with a 3 × 3 Focal Mean Filter to produce the ADEM (adjusted digital elevation model). Finally, major road dams, culverts and bridges were implemented. Details on the origin of the data used and the specific procedure of implementation can be read in [46].

Modified Blue Spot Analysis
The following approach is based on the ADEM and aims to identify natural pans [48]. The Blue Spot identification took place in several steps ( Figure 3b): First, the ADEM was filled by the ArcGIS Fill function with a threshold of 10 cm to account for possible artifacts and noises [49]. This means all pans were filled to 10 cm for equalization. Second, the ADEM was entirely filled to the pour point of all pans (not using a limiting threshold). These two results were subtracted to create the Blue Spots in the new raster layer called DEM-BS (digital elevation model-Blue Spots) [50]. In further postprocessing, the results of the Blue Spot Analysis were cleaned and separated into unique groups through the use of the Boundary Clean, Majority Filter, and Region Group tools [51].
Finally, the data was transformed into a Feature Layer (i.e., vector geometry). From the resulting polygons, the attributes of area, depth, and sum were calculated amongst other relevant attributes. To calculate the volume values of each Blue Spot (Figure 4), the area was multiplied by the sum of the pixel depth then divided by the count of pixels (Equation (1)).

Validation
For validation, a mask was developed in cooperation with the Julius-Maximilians-University of Würzburg and the German Aerospace Center. This mask represents the maximum water coverage in the entire Iishana-System for a specified period. This MWE (Maximum Water Extent Mask) was developed using a multisensor approach [52]. It includes data from 12 different spaceborne remote sensors, which, in turn, are located on seven different satellites. The multispectral satellites used are Landsat-5 and Landsat-8, and the radar satellites are ALOS/PALSAR, ALOS-2/PALSAR-2, Envisat ASAR, Sentinel-1, and TerraSAR-X (Table A1).
All available images for three rainy seasons with floods were examined, covering the months of October to September during the years 2007/2008, 2008/2009, and 2016/2017. Single water masks were extracted for each available image within the three study periods. Detailed methods can be found for Landsat-5 and -8 data in [53], for Sentinel-1 radar data in [54], and for TerraSAR-X in [55]. The remaining data can be found in the RaMaFlood (Rapid Mapping of Flooding) process description of [56]. In addition, these water masks were used to calculate the total duration of annual floods [52] during the three study periods. If one pixel had two consecutive water detections, it was considered to be flooded and was assigned to the annual water mask for its respective study period. This subdivision fueled further investigation, which will be described in another study. Finally, the three individual annual water masks were merged into one mask, the MWE mask. Permanent water surfaces were excluded from the outset. Since one year with 365 days is not completely covered by any of the three study periods, water bodies are defined as permanent if the number of days with water presence within the total duration layer is higher than the maximum possible number of days minus 30 [57].
The validation process took place in two major steps and is outlined by an example in Figure 5.  Figure 7. Subfigure (a) shows the situation and water extend on site in 2017. The red shape in subfigure (b) determines the surface water area of (a) and corresponds to the water surface area detected via the satellite data in 2017 (b). Subfigure (c) shows the result of the calculated depths and extent of the pans. Additionally, the gray areas show the maximum flood water extent detected in the past. In subfigure (d) the appearance of the depression in a dried out state is illustrated. This condition was documented by photo in 2018 after a drought event. Subfigure (e) shows the same dry pan by a satellite image including the former red shape of the water extend of 2017. Here, the dark area does not represent water, but dry sediments.
In total, 11 photos of water-filled pans were taken in the countryside and in the city of Oshakati (Figures 6 and 7(FA 2)) for ground truth control. Ten of these pictures were taken between the 5th and 7th of March 2018, ranging from one to three days after a heavy rain event. This work was done at the same time as the road dam survey and culvert detection study, which were conducted during the field trip in March and April of 2018. Details of the survey can be read in [46]. One photo was taken during the study's first field trip on the 20th of September 2017 in Iipopo. This exception was included because it is the best-documented site in the study area.  Subsequently, the results of the Blue Spot Analysis (calculated pans) were compared to the onsite pictures. Next, the same calculated pans (Blue Spots) were compared to the MWE validation Mask. In both comparisons, overlapping analysis was performed by visual examination. For each approach, this process determined whether or not a Blue Spot matched both the photo and the MWE.

Results
The Iishana-System is typically considered and described as a unified and coherent hydrological system ( Figure 6). For better examination, references to individual focus areas are made (Figure 7). At the same time, the landscape is artificially divided based on geopolitical boundaries in order to highlight possible hydrological potentials in a country-specific manner.

Surface Area
The study location covers an area of 18,370 km 2 , of which 8726 km 2 belong to Angola and 9644 km 2 to Namibia. The total area of all pans is about 4021 km 2 . Thus, the proportion of the total surface area is approximately 22% for Angola and 27% for Namibia.
The total number of all pans is 190,623, making an average of 10.4 pans per square kilometer. There are fewer pans in Angola (84,755) than there are in Namibia (106,089 pans). Likewise, the ratio of pans per square kilometer differs between the two states from 9.7 to 11.0, respectively ( Table 1).
The size of the smallest pan in both states is 156.25 m 2 and is valid as the smallest value for the entire study due to the methodological approach (raster resolution 12.5 × 12.5 m). The largest pan, with an area of around 16.8 km 2 , lies in the Angolan territory. For Namibia, the largest pan is about 13 km 2 . This size imbalance can be further observed in how the mean values and median values differ considerably, pointing to a skewed distribution. When looking at the entire study area, the mean area of pans is 21,096 m 2 , while the median is 2500 m 2 . Similarly, the values of Angola show a mean of 23,978 m 2 and a median of 2343 m 2 while the Namibian territory shows mean and median values of 18,749 and 2656 m 2 (Table 1). These strong divergences between the mean and the median values result from a large number of small pans and a much smaller number of large pans, defined as outliers (Figure 8).
A comparison of the spatial distribution of the pans (Figures 6 and 7) shows that on the Angolan side of the Iishana-System there are fewer pans than there are on the Namibian side. For all three regions (the entire Iishana-System, the Angolan portion, and the Namibian portion), the statistics show unimodal right-skewed distributions (Table 1). These are characterized by strong outliers (e.g., particularly large areas and all areas to be assigned to the right-skewed area are smaller) ( Figure 8). Furthermore, there is a lower density of pans in Angola (Table 1). However, pans on the Angolan side are, on average, larger than those on the Namibian side. This is particularly evident along the more elevated areas in the north and west, as well as in the southwest (Figures 6 and 7(FA 1)) and in the northeast portion of the study area, which extends to the city of Oshakati (Figure 7(FA 2)). The southern part of the study area is characterized by numerous, small pans that are not connected to the drainage network (Figure 7(FA 3)).

Depth
The depths of the pans range from 0.1 to 27.12 m throughout the Iishana-System (Table 1). The lowest depth is limited to 0.1 m according to the methodological approach. It is also the most common value in both subareas, Angola and Namibia. Therefore, in Figures 6 and 7 a fixation to a limit of 0.21 m was made for visual reasons. All values up to a depth 0.20 m were disbanded in order to better identify relevant information. The deepest value of 27.12 m was identified on Namibian territory; the deepest value on the Angolan side is 19.26 m ( Table 1).
The median and average depths for the entire area are 0.24 and 0.29 m. In the countryspecific calculations, these values differ just slightly, although pans on the Namibian side are often shallower (Table 1). A similar picture emerges when looking at the quartiles: the vast majority of pans are very shallow, with depths of less than half a meter. When looking at the maximum depth values, it must be taken into account that borrow pits (Figure 1c) result in higher depth values (yellow areas in Figure 7(FA 2)).

Volume
The total volume of all pans was approximately 1.9 km 3 and thus, as an example, corresponds to less than 0.1% of the volume of Lake Victoria, which measures 2424 km 3 . It is striking that about 67% of the total calculated volume is allocated to the Angolan side and only 33% of the total volume is attributable to the Namibian territory. The volumes of individual pans range from 0.1 to 12,642,002 m 3 , with the largest pan volume being located in Angola. By comparison, the largest Namibian pan has a volume of about 6,055,794 m 3 ( Table 1).
The median and mean values of the pan volumes also differ from one another. For the entire study area, the median value for volume is 612 m 3 , whereas the mean value is 9848 m 3 . A comparison between the pan volumes in Angola and Namibia shows that the median and mean values for Angola (668 and 13,810 m 3 ) are considerably larger than those for Namibia (566 and 6662 m 3 ). From the 75% quartiles, it is obvious that the majority of pan volumes is lower than approximately 2100 m 3 . Pans with a larger volume span from around 2100 to 12,642,002 m 3 . The bottom line is that the volume of pans in Angola is about twice as large as the volume in Namibia.

Surface Area-Volume Ratio
The SA/V value (ratio between the area and the volume of the pans) is an important metric to assess the water storage function of pans (Figures 9 and 10). In deep pans with a small surface area, water loss by evaporation is much slower than in large, shallow depressions. For example, a reduction of 20% of the surface area results in a 20% reduction of evaporation [58].
The lowest SA/V ratio in the Iishana-System is 0.14 m −1 and the highest ratio is 538.95 m −1 , both extremes are found on the Namibian side (Table 1). In the Angolan region of the study area, the lowest SA/V ratio is 0.60 m −1 while the highest value is 102.40 m −1 .
The overall assessment of the SA/V ratios indicates that the rates for the Angolan side are lower than those for the Namibian side, and so the problem of water loss due to evaporation has a bigger impact in the Namibian region.
Presuming that an optimum SA/V ratio is ≤1 m −1 , [58] all values labeled blue in Figure 10 show the most qualified pans for potential artificial enlargement to increase the storage function. Under this condition, 2176 pans were identified as suitable for augmentation.

Example and Synopsis
As a specific example, the water filled pan of an Oshana next to the small village Iipopo is used as a natural flood water harvesting measure (Figure 5a,b). Water is stored during the rainy season and provides a resource throughout the dry season for irrigating cash crops like tomatoes [14]. The plantation is marked with a black rectangle and the remaining body of water is framed with a red polygon, both visible in Figure 5b,c,e. Picture (a) in Figure 5 was taken at the end of the dry season in September 2017 and shows the remaining water stored in the pan (Figure 5a,b). A few months later in March 2018, after the rainy season, the pan dried out completely (Figure 5d,e). This was the result of little rainfall during the wet season of 2017/2018, which was a continuation of a 3-to 4-year-long drought period. This drought did not end until the rainy season of 2019/20.
The processed drainage network (D8-algorithm) and the cumulative flooded area in Figure 5c shows that the floodwater typically fills all pans and the Iishana up to a certain point, depending on the amount of precipitation that has fallen in Angola and Namibia during the rainy season. This case clearly demonstrates that the priory calculated pans (Figure 5c) match when compared to the water surface area validated in September 2009 (Figure 5a,b). Additionally, the MWE (Maximum Water Extent Mask) verifies that the results are correct.
Regarding the metrics of the pans, about one-third of the total and both the largest and deepest pans are located within potential flood plains and in the ephemeral runoff trajectories of the Iishana-System. There are fewer pans, both by total number and per km 2 , in Angola than there are in Namibia. In contrast, the Angolan part of the region represents slightly more than two-thirds of the potential total volume of 1.9 km 3 . This difference is also reflected in lower SA/V rates and, on average, slightly greater depths of the pans on the Angolan side. However, both subregions and the entire study area are similar in having strong outliers in terms of area, size, volume, and depth, and thus they are all characterized by unimodal right-skewed distributions ( Table 1).

Error Analysis
As noted by [26], currently, no standardized validation procedures exist for the accuracy assessment of extracted discharge networks. Therefore, all methods applied must always consider the specific landscape characteristics of the individual study area, as well as those of the basic data used. This also applies to the current validation methods of pans.
Looking at the DEM data, possible errors can be found in the backscatter behavior of the radar signal. Rizzoli et al. (2017) [41] describes a weak backscatter behavior on sandy surfaces, which could be improved by increasing the angle of the emitting sensor. Thus, the predicted error rate of all pixels for the global model could be reduced from 3% to 0.1%. Furthermore, a vertical accuracy of 2 m with a confidence interval of 90% is given. In the range of 50%, an accuracy of 1 m is given. This accuracy is remarkable, but it still casts a critical light on the results of the volume calculation here; minimal deviations can cause a significant volume change. Currently, a quantification of this potential error is not possible.
Moreover, it must be determined whether very small elements (e.g., pan 156 m 2 ) are true pans or just errors resulting from water-induced interferences. In this study, the DEM preprocessing was based on an approach published in a preliminary work [46] in which algorithms for hydrographic modeling were well established and considered accurate. As such, we believe that an accurate result for the spatial characteristics of pans was derived, especially considering that the topographic setting in the Iishana-System allows for such small pans. Additionally, the ground truth control points, although few in number, confirmed these results. Mendelsohn et al. (2013) [11] emphasize that the northern Cuvelai Basin (in Angola) consists of broader channels while the southern part of the Iishana-System consists of many smaller pans measuring less than 100 m 2 , traditional hand-dug wells, surface dams, and smaller ponds. Furthermore, the authors of [39] describe the amount of pans as innumerable, and their size in the southern parts as minor due to the low topography. Additionally, in [23] the authors describe these small ponds as having diameters of 4-5 m and depths of 8-9 m and small SA/V ratios. These smaller ponds would not be detected as natural pans due to the raster resolution. Therefore, the results only show the natural pans described by [11] and [39] and not the small artificial pans. Here the advantages of DEM resolution can be seen.
Other inaccuracies may occur in the generalization of pixel groups. The grouping of pixels (clustering) helped to reduce the enormous number of individual small pans measuring 156 m 2 (i.e., area covered by one pixel in the DEM). According to [59], this approach has advantages and disadvantages, which have to be considered during the evaluation of the results. The advantage is the reduction of outliers as well as the connection of pixels that actually belong together, which were separated due to the mixed-pixel effect. On the other hand, pixels that do not belong together may also be erroneously connected. However, since the advantages of this approach outweigh the reduction of outliers, this method was used in preprocessing.
The calculated flow paths and the MWE mask fit well together and have a high level of agreement. Therefore, reliability is high for the flow path analysis (Figures 6 and 7).
The validation of the calculated pans, with the help of the identified flood areas, should also be viewed critically. This method increases the probability of the occurrence of a pan, but it does not show it in detail. Only validation by ground truth control, e.g., with photos, provides security. Nevertheless, the validation procedure shows a high degree of accuracy, as the result in Section 3.5 shows (Figure 7). Additionally, this method provides the best performance in rural areas.

Function as a Water Resource
The transboundary region suffers simultaneously from a poor socioeconomic performance and a high frequency of natural hazards, namely floods and droughts. Moreover, resilience in regards to climate change is insufficient [5,60]. Studies have shown that Namibia has more pans, in total, than Angola does, but those Namibian pans have less volume. Just one third of the total volume of pans belongs to Namibia. Additionally, individual pans with the largest volumes all appear in Angola. Considering the heightened ecosystem stress due to climate change with less precipitation, higher evapotranspiration and a growing population on the Namibian Iishana-System, efforts should be undertaken to artificially deepen these natural reservoirs to sustain the people's resilience. However, how to start this approach?
In the past, efforts have been made to push the concept and implementation of rainwater-harvesting techniques, especially rooftop harvesting in rainwater tanks [61,62]. However, these are expensive for a rural population living under subsistence economy conditions. A rainwater-harvesting tank costs anywhere from USD 460 to 870 for a 10 m 3 tank, depending on the material [62]. Meanwhile, many older, man-made depressions are no longer in use [23]. Therefore, the answer to the water resource question should be to reactivate the man-made depressions. This approach will be cheaper because no costly material is required for construction. To implement this plan, selected pans would have to be identified and deepened (excavated) as retention measures. The SA/V ratio could be an essential factor in the analysis for identifying suitable pans for enlargement. It is clear that these natural pans play a major role now and in the future in supplying the local population and livestock with water.
Additionally, the differences in the pan depths indicate that the process of linear surface runoff dominates on the Angolan side of the study area. As the distance to the Etosha Pan decreases, the ground slope also decreases, leading to a prevalence of extensive surface runoff.
It can be concluded that the deeper, linear forms in Angola are maintained by flood runoff while the shallow forms in Namibia are, perhaps, mainly filled during flood events. This is a question to be analyzed in future research. Consequently, the fluvial-morphology determines and affects the available water resource in the different parts of the study area.
Furthermore, one third of all pans are located in the inundation zone, making these spots major priority locations for enlargement. The use of the above-mentioned retaining measures on these pans would provide both flood protection and water storage capabilities. These priority locations would be filled by precipitation and floodwater, thus increasing the ability for water storage and enhancing the resilience of the rural population.
Accordingly, enlargement should be performed on pans that meet the following criteria: (1) Pan must be close to a location with a high demand for water/flood protection; (2) Pan must be primarily connected to the channel network within an inundation zone; (3) Pan needs a low SA/V-ratio to counter the problem of high evaporation rates.
For the pans in the southern part of the study area, the situation is different. Here the pans are filled mostly by precipitation. Surface runoff usually finds no entry into the depressions, since these are often geomorphologically encapsulated (Section 2.1). Therefore, it is proposed that these isolated pans be opened and connected via the calculated flow paths to enhance the probability that they will be filled during rainfall and surface runoff events. This can boost the water volume to better meet the increased water demand. Up until this point, this concept has only been applied for smaller, hand-dug pans in the northern regions [23].

Conclusions and Outlook
Ephemeral rivers and numerous natural and artificial depressions hydro-morphologically characterize the semiarid and transboundary Iishana-System between southern Angola and northern Namibia. A high inter-and intra-annual variation of precipitation regularly leads to floods and also to severe droughts.
Therefore, the aim of this work is to increase the resilience of the local people's livelihood by identifying suitable pans for expansion as intermediate storage. These pans will reduce the flood peak and provide water for drier periods.
The identification of the pans in this very topographically flat region was a great challenge. Therefore, a new strategy for processing the TanDEM-X elevation model was developed and applied in combination with a modified Blue Spot analysis.
The validation of the results was done with the help of calculated flood masks. Additionally, photos were taken during a field trip to verify the ground truth. Thanks to the double validation approach, reliable statements could be made despite less available data.
More than 190,000 pans were identified. The total potential volume of these pans was determined to be 1.9 km 3 , with the Angolan territory accounting for about 2/3 of the volume and the Namibian territory only accounting for about 1/3. The total surface area of all pans is 4021 km 2 . This means that about one fifth of the entire region is covered by pans. A distinction between artificial and natural pans could be made by calculating the mean depths. It was shown that the pans are very shallow, measuring barely 30 cm deep on average. At the same time, however, there are also numerous outliers with significantly greater depths (Figure 10), which will be the focus of investigation in the following work. Additionally, about one third of all pans are located inside the ephemeral flow paths.
Furthermore, the hydro-morphological differences of the pans between the northern area (Angola) and the southern area (Namibia) described in the literature can be confirmed [11,23,39]. In depth, width and volume, the pans in the northern region are larger than those in the southern part of the Iishana-System. In addition, the hydro-morphological characteristics of numerous small, isolated depressions in the very south can be confirmed.
In order to identify pans as potential water reservoirs, a further parameter was determined. The surface area-volume ratio takes into account the volume coupled with the surface of a pan and thus provides an important indication of the natural storage time of the water in the system with regard to the high evaporation rates in this region. With the help of these calculations, more than 2000 pans were identified and could be considered for potential further enlargement.
A prioritization procedure should be applied for the selection of potential sites. According to the criteria described in Section 4.2, potential sites should first be selected according to their distance from a settlement structure with increased water demand. In addition, the selection of the site should be based on its connection to the channel network in order to increase the probability of filling by flood or precipitation water. Nonconnected pans should at least be as close as possible to a drainage system in order to be artificially connected to it. To make the implementation worthwhile, the surface area of the pan should be large enough and, at the same time, should have a SA/V ratio smaller or equal to one.
Currently, a validation of the calculated volumes is missing. UAV flights were planned for April and September 2020. Using the Structure-from-Motion method, accurate 3D models of selected pans should be created, and the volumes calculated and compared with those derived from the DEM. Due to the COVID-19 pandemic and the resulting travel restrictions, this project has to be realized in the future.
For the first time in the literature to date, this investigation drew a detailed morphometric picture of the pans within the Iishana-System. The relevance of this work is further demonstrated by the key characteristics of distribution and shapes as well as sizes and volumes. Especially with regard to climate change, this work provides initial answers for future work on flood protection and water storage for years of drought, which should be a priority task to protect the local people's livelihood and enhance their resilience.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.

Acknowledgments:
We thank the German Aerospace Centre (© DLR 2017) for their delivery and support with the TanDEM-X data (Proposal ID: DEM_HYDR1285). ALOS/PALSAR and ALOS-2/PALSAR-2 data are kindly provided by JAXA (Proposal number MTH 1153, PI number 3043). We are grateful for the collaboration with the University of Namibia (Campus Ongwediva). The publication of this article was funded by Freie Universität Berlin.

Conflicts of Interest:
The authors declare no conflict of interest. The cofounders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.