IEEE Geoscience and Remote Sensing Magazine - June 2022 - 184

indicator 1.4.2). Secure access to land is essential for smallscale
agricultural producers to invest in their land and contribute
to the market (targets 2.3 and 2.4). It is also fundamental
for gender equality, ensuring women's rights to
land tenure (SDG 5). The authors in [166] show that secure
land tenure for women improves investments in agricultural
developments and enhances the chance of women's
involvement in family food and agricultural productivity. A
lack of tenure security also impacts the development of sustainable
cities (SDG 11), management of natural resources
(SDG 15), and synergy between land administration agencies,
courts, and legal support services (SDG 16).
Strategies to support these goals rely partly on the development
of land administration systems (LASs) to formalize
land rights and implement land-related policies [167]. There
is a clear need for innovation for the fast, accurate, and costeffective
cadastral mapping needed for LASs [168]. The traditional
surveying methods prove to be quite costly, slow,
and labor intensive. In response, fit-for-purpose (FFP) land
administration [169] advises and supports the development
of new technologies using remotely sensed data and taking
the country context into account. The FFP concept is also
included in the recently developed framework for effective
land administration developed by the UN Expert Group on
Land Administration and Management, which is acting as
a standard at the international level [170].
Spatial land rights recording, i.e., cadastral mapping, is
the most expensive part of an LAS [171]. The automation or
semiautomation of cadastral boundary delineation based
on satellite or UAV images has been investigated since physical
objects often coincide with visible cadastral boundaries
and can be detected through image analysis [172]-[176].
The extracted boundaries should be combined with legal
information, a procedure known as adjudication, and incorporate
local knowledge from human operators.
The initially explored methods for the automated extraction
of cadastral boundaries from EO images are based
on image segmentation and edge detection [173]. However,
the main disadvantages of these methods are sensitivity to
intraparcel variability and the dependence of the selected
parameters [177], [178]. Better results have been achieved
using learning-based contour detectors, such as the globalized
probability of boundary (gPb) [135], which combines
brightness, color, and texture cues into a globalization
framework using spectral clustering.
Recent studies explored DL methods [179], [180]. The
consortium of the its4land project (https://its4land.com/)
developed a suite of open source solutions for land tenure
recording using EO data [168], [181]-[183]. It developed
methods based on the gPb, simple linear iterative clustering
superpixels, and a CNN to extract cadastral boundaries
and a strategy to assign costs to each line incorporating local
user knowledge. This work resulted in an open source
plug-in for a Quantum Geographic Information System
providing user-guided delineation functions calculating the
least-cost paths along the extracted and weighted boundaries.
Experiments were conducted using aerial images
acquired in Ethiopia and UAV images from Rwanda and
Kenya (Figure 8). Overall, the obtained results based on
CNN-derived boundaries achieved a precision of 76%. The
use of this semiautomated interactive method leads users
to spend 38% less time and 80% fewer clicks compared to
manual delineation [179], [184].
Following this research line, Xia et al. [180] explored the
potential of FCNs to extract cadastral boundaries in urban
and semiurban areas in Rwanda using UAV data. The authors
adopted the FCN-DK architecture, which, compared
to gPb and multiresolution segmentation, resulted in overall
better performance. Nevertheless, the performance of
any automated method depends on the presence of visible
objects delimiting the boundary of the property (e.g., fences,
pathways, walls, roads, or land cover transitions).
In support of the full land recording process, Chipofya
et al. [185] developed an approach incorporating handdrawn
sketch maps with remotely sensed data. Their
method converts the raster sketch map into a vector automatically,
and the hand-drawn symbols are detected and
recognized using a CNN. The system performs a strokebased
image segmentation wherein the boundaries of
sketched objects are drawn and delineated. Finally, the
concepts corresponding to the detected symbols are applied
to the image segments based on the distance and
a fixed set of rules specifying the spatial constraints on
configurations of different types of features.
FIGURE 8. The cadastral mapping tool (its4land) applied to a rural
area in Ethiopia.
184
CLIMATE ACTION
The current scenario of climate change and projections
from climate models call for definite and urgent action,
as requested by SDG 13 [186], [187]. Extreme events are
more severe, frequent, and unexpected in space and time
[188]. The Earth system is actually changing globally but
also on the local and regional scales, with huge implications
for ecosystems, biodiversity, and agriculture, just to
name a few. In this scenario, humanity faces the challenges
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE JUNE 2022
https://www.its4land.com/
IEEE Geoscience and Remote Sensing Magazine - June 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2022
