Mapping Siberian Arctic Mountain Permafrost Landscapes by

Machine Learning Multi-sensors Remote Sensing: Example of

Adycha River Valley

Moisei Zakharov

1,2 a

, Sébastien Gadal

1,2 b

, Yuri Danilov

and Jūratė Kamičaitytė

Aix-Marseille Univ., CNRS, ESPACE UMR 7300, Univ., Nice Sophia Antipolis, Avignon Univ.,

13545 Aix-en-Provence, France

North-Eastern Federal University, 670000 Yakutsk, Republic of Sakha, Russian Federation

Kaunas University of Technology, Kaunas, Lithuania

Keywords: Permafrost Landscape, Remote Sensing Modeling, Landscape Mapping, Terrain, Landsat, ASTER GDEM,

Yakutia.

Abstract: The landscape taxonomy has a complex structure and hierarchical classification with indicators of their

recognition, which is based on a variety of heterogeneous geographic territorial and expert knowledge. This

inevitably leads to difficulties in the interpretation of remote sensing data and image analysis in landscape

research in the field of classification and mapping. This article examines an approach to the analysis of intra-

season Landsat 8 OLI images and modeling of ASTER GDEM data for mapping of mountain permafrost

landscapes of Northern Siberia at the scale of 1: 500,000 as well as its methods of classification and

geographical recognition. This approach suggests implementing the recognition of terrain types and

vegetation types of landscape types. The 8 types of the landscape have been identified by using the

classification of the relief applying Jenness's algorithm and the assessment of the geomorphological

parameters of the valley. The 6 vegetation types have been identified in mountain tundra, mountain woodlands,

and valley complexes of the Adycha river valley in the Verkhoyansk mountain range. The results of mapping

and the proposed method for the interpretation of remote sensing data used at regional and local levels of

studying the characteristics of the permafrost distribution. The work contributes to the understanding of the

landscape organization of remote mountainous permafrost areas and to the improvement of methods for

mapping the permafrost landscapes for territorial development and rational environmental management.

1 INTRODUCTION

The development of knowledge-based approaches to

object recognition is one of the most relevant research

areas in machine learning and artificial intelligence

algorithms for image processing and interpreting of

the Earth observation data (Arvor et al, 2019).

Landscape classification and mapping in geography

are traditionally represented by the classification of

the landscape types and categories according to the

characteristics of the vegetation cover, soil, relief,

geomorphology, lithology, etc. Permafrost

landscapes are a complex geographic object in the

https://orcid.org/0000-0002-8916-2166

https://orcid.org/0000-0002-6472-9955

https://orcid.org/0000-0003-4806-4938

https://orcid.org/0000-0002-8821-5764

zone of permafrost distribution and the development

of cryogenic processes. They have a complex

hierarchical classification structure (Fedorov, 2018).

Recognition and mapping of permafrost landscapes

objects are based on the multi-fusion data modeling

on the territorial and geographical features of

landscape components. It makes them a

multidimensional object for their recognition using

remote sensing data processing (Boike et al, 2015).

Given the lack of geospatial data of environmental

parameters, remote sensing modeling becomes one of

the main available tools for understanding the spatial

organization of mountain permafrost landscapes in

Zakharov, M., Gadal, S., Danilov, Y. and Kami

caityt

e, J.

Mapping Siberian Arctic Mountain Permafrost Landscapes by Machine Learning Multi-sensors Remote Sensing: Example of Adycha River Valley.

DOI: 10.5220/0010448801250133

In Proceedings of the 7th International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2021), pages 125-133

ISBN: 978-989-758-503-6

125

the Arctic region (Witharana et al, 2021). Accurate

mapping of landscapes is particularly important in

view of the richness of the territory in mineral

resources, as well as in assessing the possibilities of

territorial development and taking effective measures

for environmental management (Kalinicheva et al.

2019). In addition, permafrost landscape types are

used to account for agrobiological resources, such as

reindeer pastures. Landscape taxonomy has a

complex structure, being a heterogeneous knowledge,

so there are obvious difficulties in interpreting land

use/ land cover classes for landscapes and

geographical processes. The classification of

permafrost landscapes used for the territory of Siberia

and Central Asia is based on Milkov's theory of

landscape taxonomy and fractional hierarchical

classification of landscapes, represented on the

Permafrost-Landscape map of the Republic of Sakha

(Yakutia) in scale 1: 1 500,000 (Fedorov et al, 1989).

Classification and Geographic Information System

(GIS) mapping of permafrost landscapes of the

Republic of Sakha (Yakutia) implemented is based on

superimposed analysis of climate-geomorphological,

geological, biotic, and soil factors (Fedorov, 2018).

This methodology allows using the cryoindication

approach to apply remote sensing data and techniques

in the interpretation of vegetation cover. In addition,

remote sensing data are used as a tool for drawing

boundaries in the designation of permafrost

parameters (such as the type of distribution, depth of

occurrence, cryogenic processes), extracted from the

database of the geocryological observatory, and the

collection of field data. Data from multispectral

images are widely used in the analysis and modeling

of vegetation cover and their succession stages, as

well as the thermal regime of permafrost (Shestakova,

2011) from thermal images (Kalinicheva et al. 2019).

These examples allow us to see that remote sensing

data is a relevant and rapidly developing tool in the

study of the permafrost landscape. Machine learning

and artificial intelligence algorithms (including deep

learning), such as Support Vector Machine, (Pal, and

Mather, 2005) and Random Forest (Eisavi, 2015),

have shown significant performance in analyzing

large data sets when modeling mountain permafrost

landscapes on the example of Orulgan ridge in

Verkhoyansk Mountains system (Gadal et al, 2020).

The ability to perform complex hierarchical

classifications has become the main tool for analyzing

changes in the environment. At the same time, the

capabilities of remote sensing data in the paradigm of

geographic processes and complex geosystems

(landscapes), including a set of heterogeneous

knowledge, represent a significant gap in the

representation of geographic knowledge in image

analysis. Research on the development of a

methodology for mapping and recognizing

permafrost landscapes is increasingly combining

machine learning and artificial intelligence methods

in the analysis and the interpretation of remote

sensing data with geographic knowledge and

geographic classification (Huang, 2020). In this

study, we aim to develop a mapping methodology of

permafrost landscapes at an average scale of 1:

500,000 through modeling of intra-seasonal Landsat

8 OLI images and digital elevation model (DEM),

while building a knowledge-based approach to image

analysis and considering two main principles. The

first principle is a classification of permafrost

landscape types, made according to the approach of

permafrost-landscape classification and using the

criteria for their recognition for the possibilities of

correlation with another research. The second

principle is the application of multi fusion model for

integrating the results of image classification into a

spatial database that should be based on determining

the relationship between the ontological status of

image objects and objects of permafrost landscape.

2 METHODS AND MATERIALS

2.1 Study Area

The study area has a size of 60x80 km, and it is

located between 66°26' - 65°53' North latitude and

136°27' - 138°13' East longitude. This is the basin of

the Adycha river, which is the largest tributary of the

Yana River. Mountains belong to the Chersky range

(Adyche-Elginsky plateau) in North-Eastern Siberia.

According to the permafrost landscape map of the

Republic of Sakha (Yakutia) (Fedorov, 2018), this

Arctic region consists of mountain deserts, mountain

tundras, and mountain woodlands, as well as

intrazonal valley landscapes of mountain taiga and

mountain tundra. Medium-high mountains of the

study area are characterized by significant dissection.

The height above the sea level of the watersheds

ranges from 289 to 1715 m. Permafrost type is mainly

a continuous area of frozen strata from 80-100%. The

thickness of the permafrost ranges from 200-400

meters. In addition, according to the permafrost

landscape map, 7 types of landscape vegetation and

10 types of mountain-slope and valley areas are

distinguished in the study area (Figure 1).

GISTAM 2021 - 7th International Conference on Geographical Information Systems Theory, Applications and Management

126

Figure 1: The study area and fragment from the Permafrost Landscape Map of the Republic of Sakha (Yakutia) on Scale 1:1

500 000 (7 vegetation types and 10 terrain types).

2.2 Data and Methods

In this study, we used hybrid data fusion modeling for

landscape recognition based on the classification of

multispectral images based on differences in

photosynthetic activity of different vegetation types

during the growing season, classification of

landforms using TPI (Topographical Position Index)

and methods for mapping the permafrost landscape.

This allowed us to synthesize methods for classifying

objects (classes) of the Earth's surface, which are

closely related to the characteristics of data (mainly

spectral, spatial, radiometric, and temporal

resolution) with categories of permafrost landscapes.

The Landsat 8 OLI images and DEM data with a

spatial resolution of 30 m we used. This kind of

remote sensing is suitable for landscape mapping on

a scale of 1: 500,000 to 1: 100,000. These local scales

are intended to reveal in maps the spatial organization

of the landscape in scales of the types of landscapes,

and the types of terrain. At the same time, we follow

the criteria for selecting terrain types and landscape

types used in the permafrost-landscape mapping.

Terrain types are recognized by the correlation of

stratigraphic-genetic structure and geomorphological

structure of territory. In landscape types, the

recognition criteria are classes of vegetation

associations (vegetation unit). In previous studies

(Gadal et al, 2020) we have based analysis on the

reclassification of a series of multi-time land covers

for vegetation association recognition. In this study,

we conduct a combined classification for three

vegetation indices. This method has increased the

level of automation for selecting vegetation types in

permafrost landscapes (Figure 2). Landsat 8 OLI

images acquired on 15 June 2018, 31 July 2018, and

August 27, 2018, were used in this study. A

preprocessing procedure was performed with

multispectral channels (radiometric calibration,

atmospheric correction using the DOS method (Dark

Object Subtraction)).

Figure 2: Modeling workflow of the permafrost landscape

approach.

Relief data are collected by merging the ASTER

GDEM scenes into a mosaic. The ASTER GDEM

(Global Digital Elevation Model) product developed

by METI (Ministry of Economy, Trade, and Industry

of Japan) and NASA is based on data from the ASTER

sensor of the Terra satellite. ASTER GDEM is the most

improved DEM dataset that has been GDEM3,

Mapping Siberian Arctic Mountain Permafrost Landscapes by Machine Learning Multi-sensors Remote Sensing: Example of Adycha River

Valley

127

released in 2019 available at 30 meters’ resolution

(Abrams et al 2020). It covers an area up to 83 latitudes

and has high detail for mountainous areas.

3 RESULTS

3.1 Terrain Types by Landform

Classification using Topographic

Position Index

The main factor in determining the types of terrain is

the topography, geomorphological and lithological

features of the rocks. This means that the

stratigraphic-genetic complex, namely the nature of

surface deposits determines the type of terrain. There

are 10 types of the accumulative valley and mountain-

slope areas on the territory of the study (Fedorov,

2018). The boundaries of the slope types of terrain are

determined by its "upper" contact with the flat surface

of the watershed, and on the other side - by the

"lower" junction with the floodplain or above-flood-

terrace types of terrain. The transition of slopes to

accumulative valley areas is carried out using a well-

defined bend along the rear edge of the valley floor.

An exception to the recognition principle is the type

of inter-alas terrain, which is distinguished in flat-

plain territories with the development of thermokarst

formations (Savvinov, 2002).

TPI is often used for automatic calculation of

geomorphometric properties of the earth's surface

(Weiss, 2001, Jenness, 2006, Ratajczak et al, 2009).

Terrain types are determined according to their

comparison with landforms determined by comparing

TPI values. GRASS GIS (neighborhood analysis) and

QGIS software for TPI and slope position are

implemented for the processing with ASTER GDEM.

Positive TPI (>1) values represent locations that

are above the average for their surroundings, as

defined by the neighborhoods. Negative TPI (<-1)

values represent locations that are lower than their

surroundings. TPI values close to zero (1>TPI>-1) are

either flat areas or areas of constant slope (where the

slope of the point is significantly greater than zero).

By defining thresholds for continuous TPI values at a

given scale and checking the slope for values close to

zero, terrain types can be classified into discrete slope

position classes (Jenness, 2006). Through

neighborhood analysis, TPI's are generated in scales

300 m (Figure 3, c) and 1000 m (Figure 3, d).

Using the GIS-based Jenness landform

classification algorithm (Jenness, 2006), we were

able to identify 5 types of terrain: eluvial (rocky and

mountain top), colluvial (steep mountain slopes),

diluvial-colluvial (foothills and lower parts of slopes),

river valleys and glacial valleys (the bottom of the

trough valleys) (Figure 4). We had to combine inter-

alas and outwash and mid-terrace.

Figure 3: a) RGB (2-3-4 bands) Landsat 8, 27 august 2018;

b) DEM 30m, mosaic of ASTER GDEM scenes; c) 300m

Neighborhood TPI; d) 1000m Neighborhood TPI.

Determining the moraine type of terrain based on

slope analysis is difficult. When solving this issue, we

used the color composite of 2-3-4 bands of Landsat 8

of a summer image that can determine the side

moraines designed when the glacier melts into the

valley slopes in the form of ramparts or moraine

terraces.

Figure 4: Hisometric profile with terrain types of Adycha

river valley.

The low-terrace type of terrain is determined by

the height of the valley section with a threshold of 500

m. According to the criteria for identifying low

terraces, only the Adycha river valley is located

below 500 meters. The valley of Adycha River of a

large tributary belongs to well-drained low-terraced

terrain types.

The map of terrain types (Figure 5) shows a

significant difference in the spatial distribution of

terrain types, in comparison with the permafrost-

landscape map, while the general pattern remains.

The Adycha river basin in the study area is

characterized by a strongly dissected and well-

drained accumulative plain and by the presence of

many trough glacial valleys.

GISTAM 2021 - 7th International Conference on Geographical Information Systems Theory, Applications and Management

128



Figure 5: Permafrost landscape terrain type map by ASTER GDEM/ Scale 1:500 000.

3.2 Vegetation Unit of Permafrost

Landscape Recognition

For the interpretation of the vegetation types, we

applied the method of using the series of intra-season

multispectral satellite images. The processes of

accumulation and destruction of chlorophyll and

changes in the water content in them are associated

with phenological cycles and cause variations in the

spectral-reflective characteristics of vegetation

(Stytsenko, 2018). The seasonality of the behavior of

vegetation is the result of micro and macroclimatic

aspects, as well as the activities of other living

organisms (Dyah et al, 2012). While for permafrost

landscapes, a significant impact is made by cryogenic

processes and seasonal dynamics of the thawed

permafrost layer. Since the dependence of the spectral

brightness coefficients on the wavelength varies not

only for different objects but also for the same objects

depending on the chlorophyll state and humidity, first,

it depends on the vegetation phase (Stytsenko, 2018).

This method based on phenological patterns is actively

used to classify cropland and pastures by vegetation

indices of time-series images from Sentinel-2 (Belgiu

and Csillik, 2018) and MODIS. This method is

particularly applicable to woodlands and valley

complexes, where the sparsity of the tree layer allows

satellite images to capture the spectral reflections of

shrubs, bushes, and grass, underlying forest surface,

playing a leading role in the typification of classes of

vegetation associations. This feature and advantage

allow us to increase the quality of differentiation of

objects depending on the type of shrubs or herbage of

larch woodlands (Elovskaya, 1989).

Figure 6: a) NDVI on 15 June 2018; b) NDVI on 31 July

2018; c) NDVI on 27 August 2018.

Mapping Siberian Arctic Mountain Permafrost Landscapes by Machine Learning Multi-sensors Remote Sensing: Example of Adycha River

Valley

129

The classes of plant associations for training the

algorithm for the classification of vegetation

associations are based on geobotanical studies of the

Chersky ridge, as well as on the types of vegetation

identified on the agricultural map of the Yakut ASSR.

A detailed geobotanical description of the study area

is presented in the works of Nikolin E.G. (Nikolin,

2009) Kuvaev V.B. (Kuvaev, 1960), and others.

Within the Chersky ridge, 5 main landscape-

phytocenotic structures are distinguished, represented

by 4 altitudinal belts and a complex of valley

vegetation. In terms of floristic zoning, the study area

belongs to the Western Verkhoyansk. Woodland and

sparse forest represent the arboreal layer from Larix

cajanderi. The shrubs are dominated by Pinus pumila,

Betula divaricata, Betula exilis, while the layer of

dwarf shrubs is dominated by Ledum palustre,

Vaccinium uliginosum, and Vaccinium vitis-idaea.

The moss-lichen cover is represented by sphagnum

(Sphagnum warnstorfii, Sphagnum fuscum, etc.),

green mosses, and lichens (Cladonia stellaris,

Cladonia arbuscula, Cladonia rangiferina, Cetraria

islandica, Cetraria laevigata, Cetraria cucullata,

Cetraria nivalis, species genera Umbilicaria,

Parmelia, Hypogimnia, etc.) In addition, steppe

communities are formed on the slopes of the southern

exposure. In the valley landscapes, small Ivanchay

meadows are formed, adjacent to floodplain forb

meadows. The vegetation of the valley complexes is

dominated by dwarf birch-shrub and forest

communities, including poplar-chasonian forests

(Isaev et al, 2017).

The dataset compiled from the input images

generated by the Normalized Difference Vegetation

Index (NDVI) (Crippen, 1990) is a typical Vegetation

Index for Remote Sensing Vegetation Analysis. This

method is a local application of phenology-based

image classification (Son et al, 2014). The proposed

automated method of vegetation cover mapping,

based on the analysis of short time series, allows

circumventing the restrictions imposed by a single

classification date.

The maximum likelihood (ML) classification

algorithm based on calculating the probability

distribution for the classes, let us evaluate whether a

pixel belongs to the land cover class by Bayes'

theorem. This algorithm requires enough pixels for

each learning area to compute the covariance matrix

(Congedo, 2018). This algorithm is known for its high

efficiency and gives the greatest advantage to the

dominant classes of the study area. In addition,

among the class pairs that overlap in the spectrum,

ML favors the dominant class pair. Thus, ML causes

the retooling of most of the dominant classes in the

study area (Shivakumara et al, 2018). Training

samples for vegetation classes and water are

determined by the color composite (4-5-3), (2-3-4)

using the vegetation map of the Yakutian ASSR

(Elovskaya, 1989) to determine the spatial

distribution of vegetation communities and features

of their species by the analysis of NDVI during the

vegetation season.

Figure 7: Yandex color composite image (CNES 2018,

Distribution Airbus DS), the fragment of random point a),

g), Larch woodlands lichen; b), l) Complex of mountain-

tundra vegetation of trough valleys; c) Larch woodlands

lingonberry green moss-lichen; k) Larch sparse green moss-

sphagnum with bogs; d), e) epilithic-lichen stony deserts

with areas of mountain tundra and debris of the slopes of

valleys with areas of steppe vegetation; f) Larch woodlands

and sparse forests with green moss shrub birches.

Data from late June shows low NDVI (Figure 6,

a) responses in mountainous areas, in some areas

covered with snow from heights of more than 1600

meters. High NDVI values are observed in low-

terraced areas with open larch forests covered with

dwarf birches and green moss. In July (peak of the

green season) the spectral response of the valley

vegetation complexes is almost the same with a

resolution of 30 m, and the high NDVI values (Figure

6, b) are the reason for the classification for dark and

light wood cover. As expected, only areas with

epilithic-lichen vegetation and areas exposed to forest

fires remain with zero NVDI values. In August, it is

possible to separate the areas of valley larch

vegetation in sphagnum bogs and in humid areas by a

drop in NDVI values (Figure 6, c). In the valley areas,

it is possible to clearly distinguish the areas of larch

open spaces with lichens by the permanence of the

average NDVI values.

When there is a real lack of ground check data at

the appropriate scale, the only acceptable method for

assessing accuracy is the method of generating

random points and correlating the classification

results

with the available higher resolution data

GISTAM 2021 - 7th International Conference on Geographical Information Systems Theory, Applications and Management

130

Figure 8: Permafrost landscape vegetation unit map by Landsat 8 OLI image series 2018-2020. Scale 1:500 000.

Yandex Satellite, Google Earth (Figure 7). Overall

accuracy was 78% and Kappa coefficient 0,71 with

500 random points. Based on the classification

obtained, a vegetation map of permafrost landscapes

was created, showing 6 types of vegetation cover with

an acceptable level of classification accuracy. The

resulting map (Figure 8) reliably, at the present level

of exploration of the territory, conveys the spatial

organization of plant associations.

4 DISCUSSION

In the context of climate change and permafrost

degradation, qualitative modeling is of particular

importance (Fedorov, 2019). The quality of remote

sensing data modeling depends on basic landscape

and geographic knowledge, geobotanical descriptions

of the territory, and the availability of a variety of

cartographic materials in geology, geomorphology,

and soil distribution. The obtained maps and the

described method are intended to contribute to the

development of mapping of permafrost landscapes,

including by modeling remote sensing data. The

results obtained can be used to create maps on a local

scale that are suitable for considering the

agrobiological resources of areas, but also for

understanding the local cryogenic conditions of

mountain territories.

By comparing maps of vegetation and terrain

types, one can obtain the following information about

the mountainous permafrost landscapes of the

Adycha valley. The spatial distribution of classes of

plant associations is uneven (Figure 8). The most

widespread types are Larch woodlands lingonberry

green moss-lichen with areas of cedar elfin in

mountain sparse forests (47.41%), Larch sparse

forests and dwarf green moss sparse forests with

dwarf birch forests in mountain light forests (3%),

Green moss-sphagnum larch sparse forests with

marsh terraces on accumulative valleys (8.54 %),

Larch woodlands lichen (6%), and a complex of

mountain-tundra vegetation in trough valleys (4.5%).

In total, 4 plant types make up 67% of the total land

cover, 29.38% are epilithic-lichen stony deserts with

areas of mountain tundra and talus of valley slopes

Mapping Siberian Arctic Mountain Permafrost Landscapes by Machine Learning Multi-sensors Remote Sensing: Example of Adycha River

Valley

131

with areas of steppe vegetation. In Adycha river

valley, in low-terraced terrain types, because of the

warming effect of the river, three classes of plant

associations are formed, which are traced in the

dynamics of the green moss index - sphagnum larch

woodlands with marsh, larch woodlands, and dwarf

green moss woodlands with dwarf birch forests in

mountain woodlands and lichen larch forests.

Epilithic-lichen is distributed on the steep slopes of

the mountains of colluvial, near-watershed eluvial,

and rocky terrain types.

5 CONCLUSIONS

The proposed method for recognizing permafrost

landscapes formulates an approach to using

algorithms for processing remote sensing data in

landscape research. The criteria for combining the

results of remote sensing and the geographical

components of the permafrost landscape have been

established. The maps obtained using remote sensing

modeling are a compilation of geographical studies of

a given territory used in the interpretation of

processing results. Therefore, the quality of modeling

directly depends on the level of conceptualization of

geographical knowledge about permafrost landscapes

and the study area. This approach can be implemented

using spatial ontology in the future.

The method used is proposed for mapping at the

local level at scales from 1:500,000, 1: 200,000 to 1:

100,000, when mapping vegetation and mesorelief of

individual territories of mountain permafrost

landscapes that are still difficult to access and labor-

intensive for field research. The lack of opportunities

to interpret cryogenic parameters (such as freezing

depth, rock temperature) can be considered an

obvious shortage of this study. The data obtained on

the spatial distribution of vegetation and terrain types

can be considered a contribution to understanding the

landscape organization of mountain ranges in North-

Eastern Siberia. It can also be used to study the

cryogenic conditions of mountain regions.

The development of methods for mapping and

classification of the permafrost landscapes and other

geographic objects of the landscape is directly

dependent on the level of accumulated geographic

knowledge about the territory and the geographic

processes. Remote sensing can be used for

developing the knowledge-based approach for image

processing and image analysis. This study proposes

one of the possible approaches to remote sensing

modeling for mountain permafrost landscapes.

ACKNOWLEDGEMENTS

We express our gratitude to Fedorov A.N., for the

provided vector data of the Permafrost-landscape

map, consultation, and advice. We also thank the

Yakutsk Branch of the Russian Botanical Society for

their help in identifying classes of vegetation

associations.

This research was supported by the French

National Research Agency (ANR) through the PUR

project (ANR-14-CE22- 0015), the North-Eastern

Federal University, the CNRS PEPS RICOCHET

program, and the Vernadsky Grant of the French

Embassy in the Russian Federation

REFERENCES

Abrams, M.; Crippen, R.; Fujisada, H. 2020 ASTER Global

Digital Elevation Model (GDEM) and ASTER Global

Water Body Dataset (ASTWBD). Remote Sens., 12,

1156.

Arvor D., Belgiu M., Falomir Z., Mougenot I., Durieux, L.

2019, Ontologies to interpret remote sensing images:

why do we need them? GIScience & Remote Sensing,

56:6, 911-939, DOI: 10.1080/15481603.2019.1587890.

Boike J., Grau T., Heim B., Gunther F., Langer M., Muster

S, Gouttevin I., and Lange S. 2015 Satellite-derived

changes in the permafrost landscape of central Yakutia,

2000–2011: Wetting, drying, and fires, Global and

Planetary Change, vol. 139, pp. 116- 127.

Belgiu M., Csillik O. 2018 Sentinel-2 cropland mapping

using pixel-based and object-based time-weighted

dynamic time warping analysis Remote Sensing of

Enviroment 204, 509-523

Crippen, R.E. 1990 'Calculating the vegetation index faster,'

Remote Sensing of Environment, 34, 71-73.

Congedo L. 2018. Semi-Automatic Classiﬁcation Plugin

Documentation. Release 6.0.1.1 //, 189 c.

Dyah R. Panuju and Bambang H. 2012. Trisasongko

Seasonal Pattern of Vegetative Cover from NDVI

Time-Series,

Eisavi V., Homayouni S., Maleknezhad A., Yazdi A. 2015.

Land cover mapping based on random forest

classification of multitemporal spectral and thermal

images. Chemistry, Medicine Environmental

Monitoring and Assessment 187

Elovskaya, L.G.; Petrova, E.I.; Teterina, L.V.; Naumov,

E.M. 1989. Soil map. In Agricultiral Atlas of Yakut

ASSR; GUGK: Moscow, Russia; pp. 40–41. (In

Russian)

Fedorov, A.N. 1991. Permafrost Landscapes of Yakutia.

Classiﬁcation and Mapping Issues; Permafrost

Institute: Yakutsk, Russia. (In Russian)

Fedorov, A.N. 2019. Permafrost Landscapes: Classification

and Mapping. Geosciences, 9, 468

GISTAM 2021 - 7th International Conference on Geographical Information Systems Theory, Applications and Management

132

Fedorov, A.N.; Botulu, T.A.; Varlamov, S.P.; Vasiliev, I.S.;

Gribanov, S.P.; Dorofeev, I.V.; Klimovskii, I.V.;

Samsonova, V.V.; Soloviev, P.A. 1989. Permafrost

Landscapes of Yakutia (Explanatory Note to the

Permafrost Landscape Map of the Yakut ASSR (Scale

1: 2,500,000); GUGK: Novosibirsk, Russia,

Fedorov, A.N.; Vasilyev, N.F.; Torgovkin, Y.I.;

Shestakova, A.A.; Varlamov, S.P.; Zheleznyak,M.N.;

Shepelev, V.V.; Konstantinov, P.Y.; Kalinicheva, S.S.;

Basharin, N.I. 2018 Permafrost-Landscape Map of the

Republic of Sakha (Yakutia) on a Scale 1:1,500,000.

Geosciences, 8, 465.

Gadal S, Zakharov M, Danilov Y, Kamičaitytė J. 2020.

Remote Sensing of Mountain Permafrost Landscape by

Multi-Fusion Data Modeling. Example of Verkhoyansk

Ridge (Russia). IEEE International Geoscience and

Remote Sensing Symposium, Online, United States.

pp.3082-3085

Gadal S, Zakharov M, Kamičaitytė J, Danilov Y 2020Alas

Landscape Modeling by Remote Sensing Image

Analysis and Geographic Ontology. Study case of

Central Yakutia (Russia). 6th International Conference

on Geographic Information Sytems Theory,

Applications and Management, Polytechnic Institute of

Setúbal. pp.112-118

Huang W. 2020. Towards knowledge-based

geovisualisation using Semantic Web technologies: a

knowledge representation approach coupling

ontologies and rules. International Journal of Digital

Earth. Volume 13, - Issue 9

Isaev A.P., Mikhaleva L.G., Chikidov I.I. 2017. Plant

communities of the river basin Arkachan (Central part

of the Verkhoyansk ridge) // Successes of modern

natural science. No. 3. - P. 62-66;

Jeevalakshmi D; Narayana Reddy D; Manikiam D 2006.

Land cover classification based on NDVI using Landsat

8 time series: A case study Tirupati region

Jenness J. 2006. Topographic Position Index (tpi_jen.avx)

extensionfor ArcView 3.x, v. 1.3a. Jenness Enterprises.

Kalinicheva, S.V.; Fedorov, A.N.; Zhelezniak, M.N. 2019.

Mapping Mountain Permafrost Landscapes in Siberia

Using Landsat Thermal Imagery. Geosciences, 9, 4

Kuvaev V.B. 1960. Regularities of the distribution of the

vegetation cover of the Western Verkhoyansk region /

Problems of botany // Mater, on the study of flora and

vegetation of highlands. Vol.5.: M.-L., S. 72-84.

Nikolin, E.G., 2009. General regularities of the altitudinal-

latitudinal distribution of the flora of the Verkhoyansk

Range, in Soils and Vegetable World of Mountain

Territories. M.: KMK, S. 235-239.

Pal, M. and Mather, P. 2005. Support vector machines for

classification in remote sensing. In International

Journal of Remote Sensing Vol.5. pp. 1007-1011

Ratajczak, M., Jasiewicz, J. 2009. Application of free open-

source software tools to automatic and semiautomatic

classification of landforms in lowland areas.

Geomorphologia Slovaca et Bohemica, 1, p. 43-52

Savvinov D.D., Prokopyev N.P., Fedorov V.V., Gorokhov

A.N., Dmitriev A.I., Pesterev A.P., Gavrilyeva L.D.,

Poiseeva S.I., Arkhipov V.V., Averensky A.I.,

Larionov A.G., Bosikov N.P., Kilibeeva O.N.,

Gabysheva P.E. 2002. Alas ecosystems - Institute of

Applied Ecology of the North of the Academy of

Sciences of the Russian Federation (Y), - pp. 45-68

Shats, M.M.; Dorofeev, I.V.; Fedorov, A.N. 1987.

Experience the landscape and permafrost mapping

Southern Yakutia. In Natural Conditions Reclaimed

Regions of Siberia; Permafrost Institute: Yakutsk,

Russia; pp. 49–55.

Shestakova, A.A. 2011.Mapping of the Permafrost

Landscapes Taking into Account Vegetation

Succession (for Example of the Prilensky Plateau).

Ph.D. Thesis, Permafrost Institute, Yakutsk, Russia, (In

Russian)

Shivakumara B., Rajashekararadhya S. 2018 Investigation

on Land Cover Mapping Capability of Maximum

Likelihood Classifier: A Case Study on North Canara,

India. Procedia Computer Science Volume 143, Pages

579-586

Son N-T, Chen C., Chen R.,Duc N-N., Chang L-Y. 2014.

A Phenology-Based Classification of Time-Series

MODIS Data for Rice Crop Monitoring in Mekong

Delta, Vietnam Remote Sens., 6, pp. 135-156;

Stytsenko E.A. 2018. Development of a technique for

automated interpretation of a vegetation cover with the

integrated use of multi-season zonal space images, PhD

Memory of Engineering Sciences, Moscow

Weiss A. 2001. Topographic position and landforms

analysis. In: Poster presentation, ESRI user conference,

San Diego, CA.

Witharana, C.; Bhuiyan, M.A.E.; Liljedahl, A.K.;

Kanevskiy, M.; Jorgenson, T.; Jones, B.M.; Daanen, R.;

Epstein, H.E.; Griffin, C.G.; Kent, K.; Ward Jones,

M.K. An Object-Based Approach for Mapping Tundra

Ice-Wedge Polygon Troughs from Very High Spatial

Resolution Optical Satellite Imagery. Remote Sens.

2021, 13, 558.

Mapping Siberian Arctic Mountain Permafrost Landscapes by Machine Learning Multi-sensors Remote Sensing: Example of Adycha River

Valley

133