THIS IS A POST
PRINT
Published in journal
Geocarto International
1
Land Use/Land Cover in view of Earth Observation: Data Sources, Input 1
Dimensions and Classifiers -a Review of the State of the Art 2
3
Prem Chandra Pandey
1,*
, Nikos Koutsias
2
, George P. Petropoulos
3,4
, Prashant K. 4
Srivastava
5
, Eyal Ben Dor
6
5
6
1
Center for Environmental Sciences and Engineering, School of Natural Sciences, Shiv Nadar 7
University, Greater Noida Dadri, Gautam Buddha Nagar Uttar Pradesh, India-201314; 8
prem26bit@gmail.com; prem.pandey@snu.edu.in 9
2
Department of Environmental and Natural Resources Management, University of Patras, G. Seferi 2, 10
GR-30100, Agrinio, Greece; nkoutsia@upatras.gr 11
3
School of Mineral & Resources Engineering, Technical University of Crete, Crete, Greece 12
4
Department of Soil & Water Resources, Institute of Industrial & Forage Crops, Hellenic Agricultural 13
Organization “Demeter”, Larisa, Greece 14
Email: petropoulos.george@gmail.com 15
5
Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, India-16
221005;, email: prashant.iesd@bhu.ac.in 17
6
Department of Geography, School of Geo-Sciences, Faculty of Exact Sciences, Tel Aviv University, 18
Israel- 6997801; bendor@post.tau.ac.il bendor@tauex.tau.ac.il 19
20
* Corresponding author: Email: prem26bit@gmail.com 21
22
ABSTRACT: Land use/Land cover (LULC) is a fundamental concept of the Earth's system 23
intimately connected to many phases of the human and physical environment. Earth Observation (EO) 24
technology provides an informative source of data covering the entire globe in a spatial and spectral 25
resolution appropriate to better and easier classify land cover than traditional or conventional 26
methods. The use of high spatial and spectral resolution imagery from EO sensors has increased 27
remarkably over the past decades, as more and more platforms are placed in orbit and new 28
applications emerge in different disciplines. 29
The aim of the present review work is to provide all-inclusive critical reflection on the state of 30
the art in the use of EO technology in LULC mapping and change detection. The emphasis is placed 31
on providing an overview of the different EO datasets, spatial-spectral-temporal characteristics of 32
satellite data and classification approaches employed in land cover classification. The review 33
concludes providing recommendations and remarks on what should be done in order to overcome 34
hurdle faced using above-mentioned problems in LULC mapping. This also provides information on 35
using classifier algorithms depending upon the data types and dependent on the regional ecosystems. 36
One of the main messages of our review is that in future, there will be a need to assemble 37
techniques specifically used in LULC with their merit and demerits that will enable more 38
comprehensive understanding at regional or global scale and improve understanding between different 39
land cover relationship and variability among them and these remains to be seen. 40
41
Keywords: LULC mapping; Landsat; Hyperspectral; spatial-spectral dimensions; 42
Multi-temporal; multi-source 43
44
2
1 Introduction 45
1.1 Introductory concepts 46
Land Use/Land Cover (LULC) and its changes has been considered as one of the factor of global 47
environmental change (Erdogan et al. 2015). Accurate identification and monitoring of LULC is 48
important for land resource management, since LULC mapping constitutes an important part of the 49
land management system (Chatziantoniou et al. 2017). Land cover demonstrates the terrain features 50
on the Earth surface whereas land use reflects the utilization of available land by the human beings i.e. 51
built environment/human use of terrains (Fisher et al. 2005; Hansen and Loveland 2012). Accurate 52
knowledge of LULC provides critical information for planning and management activities (Elatawneh 53
2015). This is attributed to the fact that land is one of the most important natural resource of the earth 54
system contributing to life and various other development activities (Whyte et al. 2018). 55
LULC information and its spatial distribution patterns are essential for a wide spectrum of research 56
themes especially urban studies characterized by heterogeneous classes and for maintenance and 57
developmental plans (Stefanov et al. 2001). LULC change has been perceived as a key driver of 58
worldwide environmental change by affecting the land surface (Petropoulos et al. 2013). Being in 59
steady change, urban perimeter, river basins, wetlands, agricultural areas are constantly subjected to 60
LULC changes, particularly by decreasing forest cover to give a path for agricultural extension, 61
urbanization, industrialization and so on (Stamou et al. 2016). Land cover in urban environments is 62
changing rapidly and conversion from agricultural/fallow to concrete forest resulting in urban sprawl 63
(Pandey et al. 2012), hence play a key role in environment changes (Vargo et al. 2013). 64
The assessment of LULC and of its change is important for understanding several environmental 65
issues related to urban as well as to surrounding landscapes. The primary impact on many other 66
processes need to be assessed, such as utilisation of land cover, surface temperature variation due to 67
concrete forest, (Rani et al. 2018), habitat fragmentation, biodiversity loss (Trisurat et al. 2010; 68
Theobald et al. 2011), soil and land degradation (Zucca et al. 2010; Bajocco et al. 2012; Pandey et al. 69
2013), decreased air quality, waste disposal problem (Pandey et al. 2012), decreased water seepage, 70
increased runoff along with subsequent flooding/flash flood, water quality deterioration (Tu 2011; 71
Uriarte et al. 2011), and decreased agricultural productivity. An improved understanding of historical 72
LULC change patterns provides a better means to understand the present and project future trends of 73
LULC change using different remote sensing (RS) datasets at multiple spatial, spectral and temporal 74
resolutions (Pocewicz et al. 2008). One of the key concerns about LULC and its impact has emerged 75
on a global stage due to the realisation that changes occurring on the land surface also influence 76
climate (Mahmood et al. 2014), ecosystem and its services and in return reduces biotic diversity 77
(Dezso et al. 2005). As a result, the requirements for mapping and monitoring LULC at multiple 78
scales are well-suited with demands associated to the EU habitats Directive (Petropoulos et al. 2013; 79
Singh Priyadarshini et al. 2017). 80
Nowadays remote sensing is the primary sources used extensively for LULC analysis in the recent 81
decades. Remote sensing often combined with Geographic Information System (GIS) has been used 82
extensively in mapping LULC in the analysis of their dynamics (Zucca et al. 2010). Several research 83
works were carried out by considering the importance of LULC changes at multiple scales, for spatio-84
temporal change patterns and identification of composition and its rate among different study sites 85
(Gessner et al. 2009; Chen G et al. 2012; Modica et al. 2012; Sharma et al. 2012; Grecchi et al. 2014). 86
The purpose of this review is to present LULC classification using multi-sensors, multi-source, multi-87
temporal datasets, input dimension and use of classifiers, and present the standard on improving the 88
3
change analysis, depending upon the user needs and requirements according to the landscape or data 89
availability. Figure 1 represents the user inputs, input dimension and classifier algorithms for LULC 90
mapping. The use of more than one data attributes helps in enhancing the results, such as high spatial 91
resolution, high spectral resolution, providing high temporal resolution to study change patterns at a 92
regular interval and may contribute a large coverage of the landscape. 93
94
95
Figure 1 An illustration showing types of datasets, spatio-spectral-temporal dimension and classifiers 96
algorithms for LULC and dynamic changes (author generated figure). 97
98
Thus, the objective of this review is two-fold: first to highlight various aspects of LULC classification 99
using multi-sensor, multi-source, multi-temporal datasets, input requirements and use of classifiers 100
and second to present the standard on improving the change analysis, depending upon the user needs 101
and requirements according to the landscape or data availability. In this background, the importance 102
of input dimension, remotely sensed datasets, as well as algorithms is discussed which is certainly 103
dependent upon how they are being utilised during LULC assessment. 104
105
1.2 EO datasets: Multispectral, Hyperspectral, LiDAR, SAR 106
Remote sensing has emerged as very powerful technology providing accurate spatial information and 107
LULC distribution in the temporal period (Bora and Goswami 2016; Gidey et al. 2017; Rani et al. 108
2018; Kabisch et al. 2019). The use of remotely sensed dataset depends upon the user's need, 109
4
requirement, and type of assessment of the landscape. While other factors such as regional coverage 110
(either large or small-MODIS, MISR), spatial and spectral (AVIRIS, ASTER, AVHRR), high spatial-111
spectral resolution (AISA -airborne hyperspectral images), temporal coverage (LANDSAT TM, MSS, 112
ETM+) and Synthetic Aperture Radar (SAR) data (to counter cloud effects) play an important role in 113
choosing the particular data for a specific type of study (See Figure 2). 114
LULC change patterns and dynamic changes have been presented with conventional methods, 115
individual remote sensing data, multi-sensor, multi-source, multi-sensor-temporal data are widely 116
used for assessment and evaluation of LULC change and patterns of the landscape (Figure 2). More 117
recently the synergy between different Earth Observation (EO) datasets in obtaining LULC mapping 118
has been examined. The motivation behind the synergy of different datasets is to harness the different 119
properties such as spatial, spectral, topographic, texture for improving the accuracy of land cover 120
mapping and temporal for improving the change dynamics. Therefore, user needs and requirements 121
play an important role in the selection of types of remotely sensed datasets, input dimension, and 122
implementing classifiers. With the advancement of EO technology, the broad spectral resolution 123
was replaced with high spectral resolution and filled the gap in limitation of multispectral imaging 124
(Heiden et al. 2007). EO datasets classification along with a range of classification approach varies 125
with the complexity of study site, the content and details of the classification scheme, spatial/spectral 126
resolution of datasets, and thus remains a challenge in the remote sensing community. 127
In overall, remotely sensed data are sharing the stages for LULC change and pattern analysis 128
according to need and availability. The implementation of different remotely sensed data is according 129
to the users' needs and the requirement for large area coverage, high spatial resolution, spectral 130
resolution, temporal resolution or combination of one or more together. 131
132
133
Figure 2: Conceptual model to demonstrate the user needs and requirements revolve around the basic 134
characteristics/properties of remotely sensed data and their combinations (author generated figure). 135
