Ταξινόμηση αναγλύφου με υδρολογικά κριτήρια: Εφαρμογή σε λεκάνες της Αττικής

Abstract

Στόχος της παρούσας μεταπτυχιακής εργασίας είναι η βέλτιστη ταξινόμηση των γήινων επιφανειών δύο λεκανών απορροής με τη χρήση υδρολογικών κριτηρίων. Τα κριτήρια της διαδικασίας ταξινόμησης είναι υδρολογικού χαρακτήρα και υπολογίζονται από υψομετρικά δεδομένα από Ψηφιακό Μοντέλο Εδάφους (ΨΜΕ). Στο πλαίσιο αυτό επιλέγονται τρία κριτήρια: ο δείκτης HAND (Height Above the Nearest Drainage), η κλίση του εδάφους και η «υδρολογική» απόσταση κάθε γήινης επιφάνειας από το πλησιέστερο κλάδο υδρογραφικού δικτύου. Εφαρμόζονται δύο ταξινομήσεις με βάση συνδυασμούς των κριτηρίων: η πρώτη με βάση το δείκτη HAND και την κλίση του εδάφους και η δεύτερη με το σύνολο των κριτηρίων. Χρησιμοποιούνται δύο μέθοδοι ταξινόμησης: η πρώτη με ασαφή λογική και η δεύτερη με βάση τα νευρωνικά δίκτυα. Παρουσιάζονται αναλυτικά τα αποτελέσματα για όλους τους συνδυασμούς και αυτά συγκρίνονται για την εξαγωγή συμπερασμάτων. Αποτέλεσμα της διαδικασίας είναι η κατάταξη του συνόλου της έκτασης των λεκανών απορροής σε κλάσεις και ο προσδιορισμός του εύρους των τιμών κάθε κριτηρίου ανάλογα με την κλάση. Για το σκοπό αυτό, ορίζονται πέντε κλάσεις και εξετάζεται η επιτυχία του αποτελέσματος κάθε ταξινόμησης. Αναλυτικότερα, η περιοχή μελέτη αποτελείται από δύο υδρολογικές λεκάνες του νομού Αττικής, τη λεκάνη του ποταμού Ερασίνου στην Ανατολική Αττική και τη λεκάνη του ποταμού Σαρανταπόταμου, στη Δυτική Αττική. Τα δεδομένα για τη μελέτη των λεκανών, είναι από ΨΜΕ υψηλής ανάλυσης και δείγμα σημείων από ορθοφωτομωσαϊκό χαρακτηρισμένων και καταχωρημένων στις πέντε κλάσεις. Ο δείκτης HAND υπολογίζεται ως η υψομετρική διαφορά που έχει κάθε στοιχείο της λεκάνης με το κοντινότερο σε αυτό στοιχείο που χαρακτηρίζεται ως υδρογραφικό δίκτυο, ακολουθώντας τη διεύθυνση ροής των υδάτων. Όμοια γίνεται και ο υπολογισμός της απόστασης, από το κοντινότερο κελί που χαρακτηρίζεται ως υδρογραφικό δίκτυο, ακολουθώντας τη διεύθυνση απορροής. Αρχικά, επιχειρείται η βέλτιστη ταξινόμηση με τη μέθοδο της ασαφούς λογικής, Fuzzy c-means (FCM) λαμβάνοντας υπόψη και τους δύο συνδυασμούς ως προς τα κριτήρια (δύο κριτηρίων και τριών κριτηρίων), που αναφέρθηκαν προηγούμενα. Στη συνέχεια, εξετάζεται η εφαρμογή των τεχνητών Νευρωνικών δικτύων για τους δύο αντίστοιχους συνδυασμούς κριτηρίων αντίστοιχα. Στη δεύτερη μέθοδο, το δείγμα των χαρακτηρισμένων σημείων που εξ αρχής κατατάχθηκε στις κλάσεις, αποτελεί τα δεδομένα εκπαίδευσης του συστήματος του Νευρωνικού Δικτύου Ακολούθως, επιτακτικής ανάγκης θέμα είναι ο ορισμός των κλάσεων στις οποίες ταξινομήθηκαν οι υδρολογικές λεκάνες. Στόχος του ορισμού των κλάσεων είναι αφενός η σαφής απόδοση της κατάστασης που επικρατεί στην πραγματικότητα σε κλίμακα λεκάνης απορροής και αφετέρου η μετέπειτα ερμηνευσιμότητα των αποτελεσμάτων (αριθμητικά και ποιοτικά), ως προς τα χαρακτηριστικά (δείκτης HAND, κλίση, «υδρολογική» απόσταση) της κάθε κλάσης. Οι κλάσεις είναι πέντε και χαρακτηρίζουν τις περιοχές που αντιστοιχούν στην πραγματικότητα στις παρόχθιες εκτάσεις, πεδινές εκτάσεις, πλαγιές ήπιας κλίσης, πλαγιές απότομης κλίσης και οροπέδια. Τα χαρακτηριστικά κάθε κλάσης, αποτελούν τα όρια και τα στατιστικά χαρακτηριστικά κάθε κριτηρίου για τον χαρακτηρισμό μιας γήινης επιφάνειας στην αντίστοιχη κλάση. Ως προς την ερμηνεία των αποτελεσμάτων, η βέλτιστη ταξινόμηση υλοποιείται με χρήση του συνδυασμού δύο κριτηρίων και εφαρμογή της μεθοδολογίας των νευρωνικών δικτύων. Η κατάταξη με βάση αυτή την ταξινόμηση αντικατοπτρίζει με ακρίβεια την πραγματικότητα. Αντίθετα, η ελευθερία ως προς τη διαδικασία της ταξινόμησης, των κριτηρίων που προσφέρει η ασαφής ταξινόμηση, δεν βοήθησε προς την κατεύθυνση της επίτευξη της βέλτιστης ταξινόμησης καθόσον οι αντίστοιχες κλάσεις παρουσιάζουν δυσκολία ως προς την ερμηνεία των αποτελεσμάτων.

The development of hydrological landscape units constitutes an interesting framework for scientists in the last decade. The concept of hydrological landscape units is based on the idea that the whole hydrological system interacts with a single and simple physiographic feature, as well as that this feature represents the basic building block of all hydrological landscape types. The drainage basin is the fundamental hydrologic unit for studying the flow of rivers, surface runoff, ground water flow, precipitation and, finally, predominant runoff regimes. Digital Terrain Models (DTMs) allow us to make calculations in order to describe, understand and predict water storage and movement on Earth. To study hydrological landscape units the DTMs should be represented as grids. The quantitative analysis of DTMs has led to the development of a number of hydrologically relevant numerical descriptors of landscapes. Every landscape unit contains a plethora of these descriptors, and fields such as elevation, surface slope, flow direction, flow accumulation and qualitative characteristics. A number of previous studies investigate relationships between topography and hydrological behavior in order to identify hydrologically different functional landscape units and better characterize the model structure, model parameter sets as well as metrics of catchment similarity. Based on this concept, this thesis presents a landscape classification based on hydrological criteria. Topography, land use and geology have also been used to directly infer classes according to predominant runoff processes. Regarding to the classification of the above hydrological units, three major criteria are taken into account: HAND (Height Above the Nearest Drainage), surface slope and «hydrological» distance. HAND is a new terrain model than normalizes DTMs according to distributed vertical distances relative to the drainage channels. HAND can be easily calculated from the DTM for each cell of the catchment, as the vertical difference from the nearest cell characterized as drainage network, following the flow direction path. Very similar to the determination of HAND, is the calculation of the «hydrological» distance. Study catchments The main objective of the present thesis is the optimum classification of the hydrological landscape units of two study catchments located in Attica, Greece. During the DTM processing of each catchment, the morphological characteristics are calculated and displayed. The first study area is the Erasinos Catchment located in eastern Attica. The drainage basin covers nearly 202 km2, the elevation ranges from 0 to more than 1017 m, with an average of 150 m. The slope varies from 0 to 86.9°, with an average of 8.4°. The predominant land uses are urban and cultivated areas; so, the catchment is composed of flat areas with low elevation. In spite of the flat areas, a variety of landscape units can be shown from lowlands to uplands. A 10 m cell size DTM was used to calculate the HAND index. Ground truth data were collected, in order to check the validity of the classification results. A number of 98 points were measured by using a Global Positioning System (GPS). Additionally, a total number of 6500 points were collected from an orthomosaic, which was produced through photogrammetric procedures. The point sample is used as the training sample and test sample in the artificial network analysis. Also, this sample provides check points for fuzzy classification. An orthomosaic is an image composed of many single orthophotos. An orthophoto is an aerial photograph geometrically corrected ("orthorectified") so that the scale is uniform: the photo has the same lack of distortion as a conventional map. It is thus ideally suited as a background image for overlaying other information of interest, e.g. GIS vector data. The second catchment is the Sarantapotamos Catchment located in western Attica. The drainage basin has an area of 328.5 km2, while its ground elevation varies from 0 to 1314 m, with an average of 415 m. The slope varies from 0 to 88.7 °. The predominant land uses are urban and cultivated areas; so the catchment is composed of flat areas with low elevation. In spite of the presence of hilly areas, a variety of landscape units can be shown from lowlands to uplands. A 5 m cell size DTM was used. A total number of 7000 points were measured from the orthomosaic. Classes of hydrological landscape units A typical but mandatory step for precisely defining the predominant regime is producing a finite number of classes as indicators for the characterizations of the study catchments. After studying the terrain morphology, it is noted that Greek catchments contain very mountainous areas, agricultural areas, hills, and steep slopes/ All these types of terrain are considered as the predominantly representative landscapes in Greece. Five landscape classes are distinguished which are described next. In the first category the studied regions are classified as riparian areas (riparian) and represent areas that are very close to the river network and have gentle slopes. The slopes are not related to the banks of rivers, but rather to flat areas designated as wetlands or delta of rivers. The indicator HAND, in these fields is too small as the altitude of the river almost coincides with the altitude of the above regions. The regions defined by the second category can be considered as the extensions of the regions described in the first category, where the terrain can be considered smooth. These areas can hereafter be termed as “lowlands”. Even though these regions are theoretically flat areas, their average slope value is slightly higher than that of the first category. Within Greek territory, the regions which surround mountains cover a very large area, due to the intensely mountainous terrain. As is natural, these areas form a separation barrier between the lowland regions and the mountainous regions. These areas are thus classified in the third category and labeled “gentle slope” regions. As would be expected, the HAND indicator is higher in this case than in the previous two categories. This is caused by the fact that a slope augmentation brings about an increase in the altitude of these regions, relative to the altitude of the hydrographical network, from which the indicator is calculated. The high slope regions are classified in the fourth category. These regions are the commonly identifiable element of the plethora of predominant mountainous areas in Greek territory. Their higher terrain slope creates conditions that are favorable to runoff formation. The flow which occurs in this category is responsible for the highest percentage of floods in Greece. The fifth and last category represents plateaus and is titled accordingly. The appearance of plateaus is not common within Greek hydrological basins. Fuzzy C-Means Classification One of the most widely used fuzzy clustering algorithms is the Fuzzy C-Means (FCM) algorithm (Bezdek, 1981). The FCM algorithm attempts to classify a finite collection of n elements , into a collection of c fuzzy clusters with respect to some given criterion. Given a finite set of data, the algorithm returns a list of c cluster centers and a partition matrix U, where each element uij indicates the degree of membership, which means that element xi belongs to cluster cj. Artificial Neural Network Classification An Artificial Neural Network (ANN) is an adaptive mathematical model or a computational structure that is designed to simulate a system of biological neurons to transfer information from its input to output in a desired way. ANN Classification is the process where the ANN learns how to separate samples into different classes by finding common features between samples of known classes. Because of clustering unknown samples, the point sample is often used as a means of testing the ANN classifier. It is widely used for hydrological landscape classification, as it easily recognizes landscape patterns. Conclusions With regard to data used for hydrological landscape classification the following general conclusions can be drawn: 1. The HAND indicator was applied to two selected Greek catchments, in order to find a reliable correlation between predominant conditions of water flow and terrain characteristics. The necessary condition is that the density of the hydrographic network is maintained, corresponding to real conditions in wet or dry periods. 2. The fact that the DTM grid was initially used as an input data for the calculation of the HAND indicator, does not permit an accurate determination of the classification. The results proved to be more adequate, after normalizing the DTM grid and iterating the calculation of the HAND indicator regarding the closest areas. As a result an improved classification was obtained. The normalized grid has a maximum altitude diminished by 10 m and the number of cells which contribute to the flow at the exit point of the catchment is slightly smaller. 3. Accurate in situ measurements need to be used for the validation of the input data so that the results of the classification be reliable. A coarse DTM grid can easily lead to limited information in terms of the exact values of slope in riparian areas. The selection of adequate data is highly related to the results, nevertheless taking into account the cost effectiveness factor. Creating DTM grid on the basis of maps is the usual method, although it is proved to be less accurate, owning to the errors in contours, the inadequate distribution of points and the errors in digitization. . 4. Contrary to the usual method, the use of in-situ surveying methods (with the aid of a total station or GPS) are highly accurate but the number of observation points is restricted. Photogrammetric procedures, whether semi-automatic or fully automatic, produce DTMs of accuracy that is analogous to that of the input data. Thewse overcome problems of previous methods that are mentioned earlier. 5. Of course, other methods to produce DTM could be use such as LiDAR and SAR especially when high resolution is sought. . 6. Geographical Information Systems can manage data from different sources and their main advantage seems to be the study of complicated relations between the data, displaying them though in an understandable way. With regard to the studied watersheds the following conclusions can be drawn: 1. As evidenced by the success rates, the optimal classification of the eastern study basin was obtained through the use of neural networks by combining two criteria, HAND indicator and slope. This adjustment has an average success rate of 65.8% containing all classes. 2. Based on the success rates, the optimal classification of the western study basin achieved also through using neural networks and a combination of two criteria, HAND indicator and slope. This adjustment has a success average up to 54% containing all classes. 3. Concerning fuzzy classification, although this has not shown high success rate, it is certain that it allowed to classify data without constraining factors. Moreover, even if the display seems to follow a logical spatial allocation in relation to the hydrographic network, this logic does not hold for the characteristics of each class. 4. Fuzzy classification allowed to obtain locally optimal classes that maximize the use of the initial information and minimize errors due to subsequent interventions based on external information. Besides, this method has an important disadvantage: the difficulty of visualization (due to overlapping) and its validation. 5. To define the limits of the criteria that ultimately characterize each class, a useful approach is to calculate the mean and standard deviation of the features for points in each specific class. The comparison of these values with the corresponding values for the sample of points that are used for training shows the degree of "coercion" that puts the system as a prerequisite in the process of classification.