*4.3. Results from UAVs*

The case study revealed many opportunities but also a number of challenges for UAV data capture as a technical solution to provide a spatial database for capturing land rights and cadastral boundaries in Kenya. In most countries, before commencing a UAV flight mission, regulatory clearance has to be in place to ensure the safety of airspace users, people, and property on the ground [56]. In that regard, Kenyan UAV legislation underwent changes during the case study. Before the official regulations were gazetted [57], the use of UAVs was heavily restricted, with a mandate to seek flight permission from Ministry of Defense and Kenya Civil Aviation Authority. At the time the regulations were passed, processes for flight authorizations seemed to be straight forward. However, a reality of a too costly and restrictive procedure largely impeded the rise of UAV technology in Kenya. Soon after release in June 2018, the regulations were nullified by the Government, leaving a regulatory vacuum in the country. Both data acquisition flights were carried out with a temporal flight authorization and awareness of the local government.

After an extensive sensitization of the local government and community, the UAV data, as well as GNSS measurements, were completed in March 2018 (Mailua) and September 2018 (Kajiado). The RGB pictures were processed with Pix4D to create an orthophoto (Figure 10). Flight specifications and information on geometric accuracy are summarized in Table 5.


**Table 5.** Flight characteristics and geometric accuracy of Kajiado and Mailua dataset.

However, our case study showed that UAV workflows are easy to transfer to different contexts: data acquisition always follows a standard procedure following the steps of flight planning, data collection and postprocessing. Prices of UAV equipment vary largely, offering technical platforms for almost every budget without compromising too much on data quality. Nevertheless, the purchasing costs might give an indication of the longevity and the reliability of the UAV components, which is beyond the results that the case study currently provides. Similar to the price for UAVs, the accuracy of the final orthomosaic can differ from several centimeters to meters, as it depends on the GNSS sensor of the UAV, the availability of a geodetic network, the visibility of satellites during data acquisition, and the strategy of ground control measurement. The insights from the workshop can be concluded in a SWOT analysis (Table 6).

**Table 6.** SWOT results on UAV data acquisition.


## *4.4. Results from ABE*

Delineating boundaries with indirect surveying from the remote sensing imagery requires knowledge about the boundaries. To recognize boundaries in an image, it helps to be familiar with their appearance on the ground. We, therefore, went to the area for which UAV data was captured and took images of example boundaries. A team of village elders and a local researcher joined us to communicate with land owners when passing and capturing their boundaries. The team explained which objects were typically used to demarcate boundaries and provided insights on local boundary demarcation challenges.

During fieldwork in Kajiado, we obtained an understanding of local boundary characteristics and demarcation challenges. The letters used in the following refer to Figure 11.

**Figure 11.** (**a**–**d**) Examples of visible boundaries in Kajiado. (**e**–**h**) Boundary demarcations challenging to identify correctly from remote sensing imagery collected during the field survey.

A majority of boundaries are demarcated by visible objects such as (a) stone walls, (b) corrugated metal fences, (c) vegetation, or (d) ditches. The following examples are extractable from remote sensing imagery though require local knowledge or context for a correct identification: (h) ditches can be confused with soil erosion when extracted from imagery alone. (d) Some fences demarcating boundaries are challenging to differentiate from its surrounding. High-resolution digital surface models (DSMs) can support the identification of such fences. (f) Beacons demarcate boundary corner points and (g) can be used in parallel with linear boundary demarcations, or as control points for hosting measurements.

The cadastral boundary has often remained on the connection of the beacons, instead of on the visible boundary. Based on the local knowledge obtained during fieldwork, and the large portion of cadastral boundaries in Kajiado being visible following the FFP principles, the boundary mapping approach could be applied to the captured UAV data (Figure 12).

**Figure 12.** (**a**,**b**) Cadastral boundaries delineated from UAV data.

Some challenges that we observed during delineation are shown on Figure 13 below.

**Figure 13.** Challenges observed during delineation: (**a**) undersegmentation, (**b**) oversegmentation, (**c**) fragmented segmentation, (**d**) redundancy of least-cost-path calculation, (**e**) visible boundary not demarcated by objects, but by context, and (**f**) identification of delineation areas through boundary mapping approach.

Existing reference maps for our area would mostly consist of Registry Index Maps (RIMs). RIMs show the outline of land parcels within a given jurisdiction using general boundaries along visible features. The boundaries' position is only indicative and not legally binding. RIMs and survey plans for urban areas have the highest accuracy specifications of 30 cm nominal positional accuracy [38]. Different types of RIMs exist that partly allow positional errors of up to 200 cm [58,59]. As the digital cadaster coverage in country is low the local experts shared that even meter level accuracy can be acceptable for certain areas. However, we observed how time-consuming and tiring this procedure is. The expert should zoom in and out continuously in searching for visible boundaries. Then, the accuracy for delineation of each parcel will be different depending on the skills and precision of the operator. The automatic approach that was proposed speeded this process. The automatically detected and suggested boundaries just have to be checked by the operator and with several clicks to be adjusted and approved. It was observed that for long curved objects manual delineations is much slower and requires continuous clicking and the automatic one requires to click only on the starting and ending point. For a small rectangular object it is required to click only inside of the object and the boundary will be automatically delineated. Using the new proposed method, we reduced the number of clicks with 80%, saved 38% of the time and achieved 71% accuracy compared to manual delineated boundaries [48].

The operational analysis showed that the approach is most suited for the delineation of visible cadastral boundaries demarcated through physical objects. In our study side area, walls and fences were partly covered by vegetation and not built consistently. From 211 parcels, 21 could be delineated without further editing, 24 required minor editing on <20% of the outline length, and the remaining parcels were digitized through snapping to the automatically generated lines and generating new ones. In general, the approach obtains the highest time savings for areas in which boundaries are visible, long and curved, whereas boundaries in our study side are often covered, short, and straight.

The feedback analysis investigated the strengths, weaknesses, opportunities, and threats (SWOT) of the approach. Feedback is derived from three one-day workshops for 57 land administration stakeholders from local government institutions, NGOs, private companies, and national government institutions. The SWOT feedback from the three workshops is shown on Table 7.


**Table 7.** SWOT results on automated boundary extraction approach.

The methodology was tested and works also for remote sensing data with different resolutions (0.02–0.25 m) acquired from other platforms such as satellite and aerial cameras on board of an airplane. Advantages are strongest when delineating in rural areas due to the continuous visibility of monotonic boundaries. Manual delineation remains superior in cases where the boundary is not fully visible, i.e., covered by shadow or vegetation. Although our methodology has been developed for cadastral mapping, it can also be used to delineate objects in other application fields, such as land use mapping, agricultural monitoring, topographical mapping, road tracking, or building extraction.

#### *4.5. Results from Analysis of Governance Aspects*

In the focus group discussions and individual semi-structured interviews held around each remote sensing application, several governance aspects were raised. As most of these apply to two or all three remote sensing methodologies, and are discussed here jointly along the lines of six aspects derived from the discussions: (1) legal versus informal rights, (2) government versus non-governmental actors, (3) the national versus regional/local government, (4) digital versus paper way of working, (5) use of open source software, and (6) lack of clear legislation for specific new tools and applications esp. UAVs.

Many different definitions of the term "governance" exist, but most important is that it stands for a broader concept than government, and also includes the influence of other actors on processes that affect all. Within the context of the research, a definition was developed where governance is "the process of interactively steering the land tenure society to sustain the use of the its4land tools" [60].

In Kenya the 2010 Constitution brought a number of changes that affect the governance aspects of our remote sensing methodologies. As mentioned earlier, customary tenures are now explicitly recognized in the Constitution, although the attention to them in specific laws and regulations is still lagging, and in peri-urban (and informal) areas, other forms of non-statutory tenure rights exist that are not specifically mentioned. The formal systems for land administration, that tend to only serve statutory rights, are embedded in laws and regulations, but also in the way the different formal land sector actors operate in practice; which tends not to focus on innovation or broadening of the beneficiary group. There is, currently, a lack of participatory mechanisms that can support the collaboration between the different governmental levels and the non-governmental actors. Political interests or corrupt practices were mentioned during the workshops and interviews. These practices happen due to both the lack of transparency in the decision making process and lack of an enforcing institutional environment. Further, there is no specific legislative framework that supports innovative approaches as the ones offered via our developed applications.

Allowing non-governmental organizations (such as private companies, NGOs and professional network associations) to take the lead in implementing the more participatory and innovative technical applications is also difficult. There is not really a tradition to do so, which is partly due to lack of resources: financial, human and technological. Further, the fear of losing jobs due to introduction of new ways of work make the street level bureaucrats wary, whereas the higher level workers fear of loss of the control of the currently used methods which involve political interests and corruption practices. As most of the national government and counties lack basic infrastructure, one way the national government could support the implementation of technical applications is by providing financial or legal incentives to non-governmental actors, as in many cases there are consultancies who have the expertise and could support the adoption of technical applications within a short time frame. However, neither governmental actors nor private companies are used to this type of participatory approaches. Until now, according to the different actors who participated in the workshops, there have not been real participatory approaches that could support their implementation. The capacity of the local levels to implement technical applications like ABE and SmartSkeMa face the challenge of variety in capacity among the counties, and some cases were reported where governmental employees need to use their personal computers to carry out their daily job activities.

The 2010 Constitution brought the devolution of powers to the 47 counties. There is still lack of clarity relating to the division of responsibilities between the county and national government level, and the different governmental levels currently often lack resources to implement, maintain or upgrade the use of innovative technical applications, especially when those require the specific IT knowledge that comes with geospatial techniques. The current governance structure favors a top-down implementation process where the national government is the main actor. While some counties have the capacity to support the implementation of the technical applications, others clearly lack infrastructure, financial resources or knowledge.

In addition to the limited capacity, it also became clear that only some governmental actors see the transition from paper-based data to digital based data as a priority. The transition from paper-based data to digital data is already set in some counties, but is not always perceived as a priority by all governmental actors. Due to the lack of political will, the implementation of our technical applications cannot be expected to occur in the short-term. Political interests or corruption practices around the possible implementation of the technical applications were also mentioned by the different interviewed actors. This situation is due to the lack of legislation for digital data and the current prioritization of paper-based data.
