Research On Uncertainty Of Geomorphic Classification Based On Digital Elevation Models

Geomorphic classification is one of the important theoretical issues and basic problems in geomorphology study.It is also a quite complex research topic in the field of digital geomorphology.Uncertainty is an inherent property of the world,and it is no exception in the study of geomorphic classification.Geomorphic classification is mainly affected by classification principles,classification indicators,classification methods and data resolution.Changes in these factors will lead to uncertainty in the process and results of geomorphic classification.For example,the fuzzy or transitional of the landform type boundary is an embodiment of the uncertainty of geomorphic classification.The research on the uncertainty of geomorphic classification is an important scientific problem in the field of geomorphology.At present,most of the studies of classification uncertainty are based on remote sensing images,and they mainly focus on measurement of the uncertainty in remote sensing information,the uncertainty analysis of remote sensing image classification based on object-based image analysis(OBIA).These methods are rarely used in the study of the uncertainty of geomorphic classification.The development of digital elevation model(DEM)and digital terrain analysis(DTA)provides important data and theoretical basis for the study of geomorphic classification uncertainty.Currently,as far as the research on geomorphic classification uncertainty based on DEM is concerned,there is still a lack of systematic research ideas.Because the main factors affecting the uncertainty are classification indicators,classification methods,and data resolution,the research entry point should also be started here,but how to carry out systematic research on the uncertainty of geomorphic classification is an important issue facing now.In this dissertation,we combine DEM with existing geomorphological map,and comprehensively apply theories and methods such as OBIA,DTA,landscape pattern analysis,and rough set theory to analyze the main influencing factors of geomorphic classification uncertainty,and take Shaanxi Province as a case study.The dissertation focuses on the landscape metrics of geomorphological types,and explores the uncertainty of geomorphic classification from three aspects: classification index,data and classification method.The main contents and conclusions of this study are as follows:(1)For an area to be classified,the dominant geomorphological types have a strong ability to resist changes of classification indexes,resulting in less classification uncertainty;the rate of amplitude(ROA)and variation coefficient of non-dominant types landscape metrics are larger,which will lead to more classification uncertainty,even lead to incorrect classification results.When the number of indexes increases to a certain level,the landscape metrics of most landforms tends to be stable,and the classification uncertainty is relatively low.When indexes expressing elevation changes(such as relief amplitude,elevation variation coefficients)are used as geomorphic classification parameters,the landscape metrics values change significantly.Therefore,the control role of such indexes should be fully considered in geomorphic classification.(2)The landscape index of six geomorphological types in Shaanxi Province has different responses to changes in DEM resolution,and generally have random,fuzzy,periodic,and unknown changes.The dominant geomorphic types in the classification area are less disturbed by DEM resolution,and the changes are not sensitive.The nondominant geomorphic types are on the contrary,such as low mountains and high mountains.The grain effect of the landscape index can further analyze the uncertainty characteristics of landform classification results caused by changes in data resolution.The fitting function of the landscape metrics with the changes of the DEM resolution can better excavate the uncertainty characteristics.For example,the linear change trend of the aggregation index(AI)helps to predict the uncertainty,but it will not reduce or eliminate the uncertainty.The appropriate grain of the landscape index can reflect the scale or interval of the geomorphic type,and making the uncertainty of the geomorphic classification controllable.In addition,the grain effect of the landscape index has a multi-scale nesting characteristic,especially the landscape index that periodically changes with DEM resolution,such as the edge density(ED),this paper considers it a novel grain effect.(3)We discussed the impact of supervised and unsupervised classification methods on the uncertainty of geomorphic classification in this paper.For supervised classification,the changes of landscape metrics of geomorphological types can clearly indicate the hierarchy of landform types,which will lead to lower uncertainty of landform classification.Under unsupervised classification,the indication of landscape metrics changes is unclear,and the predominance and diversity of geomorphological types cannot be mined,which leads to increased uncertainty in geomorphic classification and lower accuracy of classification results.(4)The rough entropy,approximation classification accuracy and the approximation classification quality based on rough set theory are effective indicators in the geomorphic classification uncertainty assessment from multiple scales,and the assessment results are more accurate.Meanwhile,the three indicators can also measure the influence degree of different factors on the uncertainty of geomorphic classification and the results obtained in this paper as follows: the data resolution is higher than the classification index,and the classification method is higher than the data resolution.This conclusion can provide references for the control of uncertainty in other study areas or geomorphic classification for different goals in the future.This research systematically analyzes the main factors that affect the uncertainty of geomorphic classification,combines the object-oriented classification ideas and landscape pattern analysis to conduct a comprehensive study on the uncertainty of landform classification.A multi-scale geomorphic classification uncertainty evaluation method based on segmented objects is proposed,which effectively makes up existing ideas and methods of uncertainty evaluation.In short,the research in this paper has a good reference and application value for the classification of geomorphology and its uncertainty control,and it is a useful supplement to the geomorphology research based on digital terrain analysis.