1
Topography EU-DEM and derivatives Elevation Slope Aspect General-, Plan-, Profile Curvature Vertical Distance to Channel Network SAGA Wetness Index Topographic Wetness Index Diurnal Anisotropic Heating Real Surface Area Channel Network Base Level MRVBF - Multiresolution Index of Valley Bottom Flatness MRRTF - Multiresolution Index of Ridge Top Flatness Mass Balance Index Stream Power Index LS Factor Topographic Position Index Distance to Actual Stream Network Land use and vegetation CORINE LandCover, CLC50 MODIS images Soil Legacy soil data Digital Kreybig Soil Information System (DKSIS) AGROTOPO 1:200.000 scale genetic soil map Climate average annual evapotranspiration average annual precipitation average annual temperature annual evaporation Lithology Geological Map of Hungary 1:100.000 – merged FAO categories Groundwater level Geological Atlas of Hungary SEQUENTIAL CLASSIFICATION Gábor ILLÉS 1 László PÁSZTOR 2 Katalin TAKÁCS 2 Acknowledgements: Our work has been supported by the Hungarian National Scientific Research Foundation (OTKA, Grant No. K105167). 2 Institute for Soil Sciences and Agricultural Chemistry, Centre for Agricultural Research, Hungarian Academy of Sciences 1 Forest Research Institute, National Agricultural and Innovation Centre Várkerület 30/A, 9600 Sárvár, Hungary Annamária LABORCZI 2 Segmentation and sequential classification of a synthetized image composed of spatial environmental data for the compilation of a soil type map 1:200.000 scale genetic soil map (A). Forest is a unique legend element (B); the map was synthetized from larger scale soil maps, which did not represent the country homogeneously, this fact is strongly reflected in its pattern (C). Hungary’s general geographical (A) and the related land use (B) conditions. Plains are dominated by arable lands; mountainous regions are mainly characterised by forests Genetic soil type layer of AGROTOPO (A). The representation of areas with different physiography/land use is rather inhomogeneous: hilly regions dominated by forests (B); plains characterised by croplands (C) SEGMENTATION A unified, national soil type map with spatially consistent predictive capabilities was compiled applying traditional and newly tested Digital Soil Mapping classification methods: segmentation of a synthesized image consisting of predictor variables and multi-phase, sequential classification by Classification and Regression Trees, Random Forests and Artificial Neural Networks. Object based classification using spatial-thematic segments was applied to define mapping objects. Classifications were carried out on two levels to achieve better results. Performance of classifiers was continuously assessed and applied for the identification of best performing predictions, which were combined for the production of the final map. Evaluation of the results showed that the object based, multi-level mapping approach performs significantly better than the simple classification techniques. The importance of the newly prepared map could actually be evaluated from the practical point of view. This is the first countrywide soil type map that unifies expert inputs and databases from both the agricultural farmlands and forested areas. As a consequence, this map can be equally used for agricultural or forestry oriented purposes providing interoperability between the sectors. Because of the robustness and huge data background, the map is suitable to be involved in nationwide spatial and land use management planning. Nationwide legacy soil type maps characterized forest dominated, hilly/mountainous regions either simply as forest, or with significantly lower thematic and spatial resolution. Soil survey on arable lands Hungarian Soil Information and Monitoring System (SIMS) Hungarian Detailed Soil Hydrophysical Database (MARTHA) ~ 1,200 points Recently sampled point database Laboratory measurements (pH, SOM, CaCO3, etc.) & soil types ~ 3,900 points hydrophysical and chemical information collected from various sources Soil survey in forests ~ 55,000 data points of forest compartments genetic soil type, texture class, rooting depth class Forestry database VALIDATION & 1st level: main soil type groups (MSTGs) multiple steps - overall classification: delineation only for 3 MSTGs sand, brown forest and lithomorphic soils - left out MSTGs were classified again to get the spatial distribution of the other six MSTGs skeletal, salt-affected, meadow, peat, alluvial soils and chernozems 2nd level: Soil types (ST) ENVIRONMENTAL CO - VARIABLES Classifiers on both levels: Classification and Regression Trees Random Forests Artificial Neural Network The final product: a unified, national, soil type map with spatially consistent predictive capabilities Nationwide main soil type group map mosaicked from partial results Harmonization of nomenclature Sporadically available (only for agricultural areas) digitized large scale soil type maps were used for external validation. Black lines are soil map delineations with type classification. In the background a part of the newly compiled soil type map. Both thematic and spatial representation of hilly/mountainous areas is much more detailed than on former national soil maps. The mosaic-like pattern of lowlands is retained and the large scale geographical structural elements are very well reflected. Agreement with the predicted soil type map Overall accuracy Overall kappa AGROTOPO lowlands 0,31 0,24 hilly areas 0,27 0,15 mountainous areas 0,29 0,18 1:200.000 genetic soil map lowlands 0,36 0,29 hilly areas 0,25 0,17 mountainous areas 0,17 0,09 This is the first countrywide soil type map that unifies expert inputs and databases from both agricultural farmlands and forested areas. It can be equally used for agricultural or forestry oriented purposes providing interoperability between the sectors. For forest area validation a collection of independent soil observations was set up from monographies, studies, and reports: accuracy of 65%. Comparison to the two earlier nationwide soil type maps on a pixel by pixel level 12 models Combination of best performing models None of the models overperformed 62% accuracy. However, by a proper combination we could finally produce 70% accuracy.

Segmentation and sequential classification of a ... · agricultural areas) digitized large scale soil type maps were used for external validation. Black lines are soil map delineations

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Segmentation and sequential classification of a ... · agricultural areas) digitized large scale soil type maps were used for external validation. Black lines are soil map delineations

TopographyEU-DEM and derivatives• Elevation• Slope• Aspect• General-, Plan-, Profile Curvature• Vertical Distance

to Channel Network• SAGA Wetness Index• Topographic Wetness Index• Diurnal Anisotropic Heating• Real Surface Area• Channel Network Base Level• MRVBF - Multiresolution Index

of Valley Bottom Flatness• MRRTF - Multiresolution Index

of Ridge Top Flatness• Mass Balance Index• Stream Power Index• LS Factor• Topographic Position Index• Distance to Actual Stream Network

Land use and vegetationCORINE LandCover, CLC50MODIS images

SoilLegacy soil dataDigital Kreybig Soil Information System (DKSIS)AGROTOPO1:200.000 scale genetic soil map

Climateaverage annual evapotranspirationaverage annual precipitationaverage annual temperatureannual evaporation

LithologyGeological Map of Hungary 1:100.000 – merged FAO categories

Groundwater levelGeological Atlas of Hungary

SEQUENTIAL CLASSIFICATION

GáborILLÉS1

LászlóPÁSZTOR2

KatalinTAKÁCS2

Acknowledgements: Our work has been supported by the Hungarian National Scientific Research Foundation (OTKA, Grant No. K105167).

2 Institute for Soil Sciences and Agricultural Chemistry,Centre for Agricultural Research,Hungarian Academy of Sciences

1 Forest Research Institute, National Agricultural and Innovation CentreVárkerület 30/A, 9600 Sárvár, Hungary

AnnamáriaLABORCZI2

S e g m e n t a t i o n a n d s e q u e n t i a l c l a s s i f i c a t i o n o f a s y n t h e t i z e d i m a g e c o m p o s e d o f s p a t i a l e n v i r o n m e n t a l d a t a f o r t h e c o m p i l a t i o n o f a s o i l t y p e m a p

1:200.000 scale genetic soil map (A). Forest is a unique legend element (B); the map was synthetized from larger scale soil maps, which did not represent the country homogeneously, this fact is strongly reflected in its pattern (C).

Hungary’s general geographical (A) and the related land use (B) conditions. Plains are dominated by arable lands; mountainous regions are mainly characterised by forests

Genetic soil type layer of AGROTOPO (A). The representation of areas with different physiography/land use is rather inhomogeneous: hilly regions dominated by forests (B); plains characterised by croplands (C)

S E G M E N T A T I O N

A unified, national soil type map with spatially consistent predictive capabilitieswas compiled applying traditional and newly tested Digital Soil Mappingclassification methods: segmentation of a synthesized image consisting ofpredictor variables and multi-phase, sequential classification by Classificationand Regression Trees, Random Forests and Artificial Neural Networks. Objectbased classification using spatial-thematic segments was applied to definemapping objects. Classifications were carried out on two levels to achievebetter results. Performance of classifiers was continuously assessed and appliedfor the identification of best performing predictions, which were combined forthe production of the final map.Evaluation of the results showed that the object based, multi-level mappingapproach performs significantly better than the simple classificationtechniques.The importance of the newly prepared map could actually be evaluated fromthe practical point of view. This is the first countrywide soil type map thatunifies expert inputs and databases from both the agricultural farmlands andforested areas. As a consequence, this map can be equally used for agriculturalor forestry oriented purposes providing interoperability between the sectors.Because of the robustness and huge data background, the map is suitable to beinvolved in nationwide spatial and land use management planning.

Nationwide legacy soil type maps characterized forest dominated, hilly/mountainous regions either simply asforest, or with significantly lower thematic and spatial resolution.

Soil survey on arable lands

Hungarian Soil Information and Monitoring System (SIMS)

Hungarian Detailed Soil Hydrophysical Database(MARTHA)

~ 1,200 pointsRecently sampled point databaseLaboratory measurements (pH, SOM, CaCO3, etc.)& soil types

~ 3,900 pointshydrophysical and chemical information collected from various sources

Soil survey in forests

~ 55,000 data points of forest compartmentsgenetic soil type, texture class, rooting depth class

Forestry database

VALIDATION

&

1st level: main soil type groups (MSTGs)multiple steps- overall classification: delineation only for 3 MSTGs sand, brown forest and lithomorphic soils

- left out MSTGs were classified again to get the spatial distribution of the other six MSTGs skeletal, salt-affected, meadow, peat, alluvial soils and chernozems

2nd level: Soil types (ST)

ENVIRONMENTAL CO-VARIABLES

Classifiers on both levels:

Classification and Regression TreesRandom ForestsArtificial Neural Network

The final product: a unified, national, soil type map with spatially consistent predictive capabilities

Nationwide main soil type groupmap mosaicked from partial results

Harmonizationof nomenclature

Sporadically available (only foragricultural areas) digitized large scale soil type maps were used for external validation. Black lines are soil map delineations with type classification. In the background a part of the newly compiled soil type map.

Both thematic and spatialrepresentation of hilly/mountainousareas is much more detailed than onformer national soil maps.

The mosaic-like pattern of lowlands is retained and the large scalegeographical structural elements arevery well reflected.

Agreement with the predicted soil type mapOverall

accuracy

Overall

kappa

AGROTOPO

lowlands 0,31 0,24hilly areas 0,27 0,15mountainous

areas 0,29 0,18

1:200.000 genetic soil map

lowlands 0,36 0,29hilly areas 0,25 0,17mountainous

areas 0,17 0,09

This is the first countrywide soil type map that unifiesexpert inputs and databases from both agriculturalfarmlands and forested areas.It can be equally used for agricultural or forestry orientedpurposes providing interoperability between the sectors.

For forest area validation a collection of independent soil observations was set upfrom monographies, studies, and reports: accuracy of 65%.

Comparison to the two earlier nationwide soil type maps on a pixel by pixel level

12 modelsCombination of best performing models

None of the models overperformed62% accuracy. However, by a propercombination we could finally produce70% accuracy.