Welcome dear students, welcome back to this course on Geo Spatial Analysis In UrbanPlanning. So, today we shall look into Raster Operations, we are in the module 2 regarding Imean we are going to look into the GIS functionalities. And spatial analysis basics we havedealt with the vector analysis, today we have also dealt with 1 module related to rasteranalysis.
Today, we are in the 10th lecture where in we are going to discuss about the raster operations.How the terrain can be visualized if we have a digital elevation model which is also a rasterdata? How a terrain can be visualized? What are the different ways to visualize it? And howwe can do a raster data classification? So, I mean a brief into all these would be discussed intoday’s lecture.
(Refer Slide Time: 01:15)
So, the concepts that we are going to cover today are the raster operations, we shall look intoa measure of the physical distance. We shall look in to allocation and direction, we shall lookin to data raster data operation regarding data extraction. We are also going to cover differentaspects of buffering of raster data and see how it is different from your vector buffering whichwe had already I mean done earlier.
We shall also look into operations for analyzing and visualizing terrain using differentapproaches. Like, one of them is to calculate the shaded relief, I mean we can also do a hillshading or it is also known as hypsometric shading and then we shall calculate the slope andthe aspect. And finally, we shall look into concepts of raster data classification the variousapproaches and how we can calculate error when we are doing a raster data calculation.
(Refer Slide Time: 02:26)
So, I mean regarding the this data operations we had talked about in our earlier I mean lectureregarding the data being analyzed in a local context in a neighborhood or context or in aregional or a global context. So today, first we are talking about the measure of physicaldistance which is I mean you already know how we can measure distances using vector, butwe can calculate distances using raster’s as well.
So, these distances we can calculated as physical distance or as cost distance. Now a physicaldistance measure is a Euclidean measure it is a measure of the Euclidean distance that we hadalready seen and it is basically a measure of the straight line. And cost distance is a function ofthe physical distance it is also a function of the speed on a given segment of a road or also onthe road condition.
So, these factors would be governing the cost of traversing a link say suppose, we have a roadsegment which is shown as a line either vector line or a raster line we can work out thephysical distance as well as the cost distance. So, in today’s I mean context we are going totalk about finding out the distance with respect to a raster data set.
So, I mean we generally measure the physical distance as I mean as a product of the resolutionof each of the cells that is the spatial resolution what is the size of the raster cells in yourparticular raster data and we measure I mean multiply it with the distance I mean that is thesquare root of the centroid of the pixel origin pixel to the destination pixel.
So, I mean it is used of the distance is measured using your Euclidean formula to work out thephysical distance in case of a raster what the difference between the vector data distance andthe raster distance is that in vector you have the specific point coordinates which is moreaccurate in terms of measuring the distance wherein in case of a raster we generally measurethe distance from the centriod of the pixel. So, there could be some inherent limitations interms of the exact measure when we are measuring the distance using raster data.
Now, the physical distance measure it buffers source cells and it would create buffers to thespecified maximum distance and it is it could be a also neighborhood operation or it could be Imean a global operation. Now, in case of raster I mean data operation where we aremeasuring the physical distance we would have to reclassify the data or we would have toconvert or regroup this physical measure because since this would be a continuous distancemeasure in case of a raster; so we would have to discretize the distance zone.
So, we can do this there is a operation it is known by different name’s in different software’sthis operation is known as slicing which can divide a continuous distance function of rasterinto equal-intervals or equal-area distance zone.
Now, you can see this particular example wherein we have this particular raster where wehave I mean these 2 cells which are connected by lines. So, we measure the distance centroid
and then we I mean from the origin to the destination cell and then we calculate the distanceusing the equation that we had already talked about.
In the second image, you can see that there is a continuous distance measure from a streamnetwork. So, in this case a slicing has been done to convert the continuous distance raster intodifferent discrete distance zone. So, this is very important whenever we are measuring physicaldistance using raster data operation to slice the distance, once we have measured the distancefrom a linear feature.
(Refer Slide Time: 07:12)
Now, talking about allocation and direction we can measure the physical distance and produceallocation and direction rasters; for example, I mean we can have the cell value in theallocation raster which would correspond to the closest source cell for that cell we shall see in
due course through an example. So, the cell value in the direction raster it would generallycorrespond to the direction in terms of degrees that the cell is closest from the source cell.
So, I mean we can use the compass direction suppose I mean; we take a clockwise measure.So, 90 degrees would be your east, 180 degrees would be your south, 270 degrees would bethe west, and similarly 360 degrees or 0 degrees would be the north.
So, in this particular example you can see again that in the first image there you have the rastercells, which shows the physical distance measure in cell units from each cell to the closest cell.So, you have these particular cell values which are value 1 and value 2; the value 1 is thesource cell and 2 is the destination cell. Now, the values in this particular cell; gives you ameasure of the distance from the centroid of this particular cell to the neighboring cells.
So, from 2 to go to this particular cell you will have to traverse a distance of 1.4 because thatis square root of I mean 1 squared plus 2 squared. So, that would give you root over 2 whichis equivalent to 1.414 so, which is I mean rounded up to 1.4. So, you can likewise see thisparticular cell that this cell is displaced from 2. So, we find out the distance from each of thesecells to the nearest pixel.
Now, in the second image you can see that it shows the allocation of each cell to the closestsource cell. So, wherever I mean you have this source cell 1 and 2 this white cells it showsthat they are allocated they are I mean in closest proximity to the source 1 and the cells mark 2are in closest proximity to cell 2. Now here it is important to note that the distance from thecentroid of this pixel 2 the destination cell; to this particular cell is 2.2.
And distance this distance is also same from the centriod of the source pixel, if this is thesource and this is the destination pixel you see that this distance is same from both this sourceand the destination pixel. So, we have to allocate it to one of the either of the cells, so in thiscase there could be some kind of an anomaly.
Now, in the third image see you can see that it gives you the direction in terms of degreesfrom each cell to the closest source cell. So, I mean you would measure it from the I mean
from 90 degrees and then we have said we can measure it clockwise. So, this would give youthe directions in terms of degrees from each cell to the closest cell.
So, the cell in the dark that is shown in row number 3 and column number 3 it has we had saidthat the distance is 2.2 it is same from both these 2 cells I mean and the direction 2 to the 43 tothe it is from with respect to the source cell 243 degrees is with respect to the source cell.
(Refer Slide Time: 11:22)
Now, talking about raster data extraction there could be cases where we want to extract theexact pixel value of a raster location; say suppose, you may have say temperature values orprecipitation values of a continuous data set over India and you may want to extract the datavalue for a particular city or a given location. So, we might need to do a data extraction of aspecific points or regions; so in that case a raster data extraction operation would create a new
raster by only extracting data from the existing raster I mean and it is this operation is similarto a raster data query.
Now a data set or a graphic object I mean we can use a query or a expression to define theareas to be extracted you can have a very complex query when you are running it or you canalso put it in a equation or you can see if the data so you can have various ways of giving aquery expression when you are doing a raster data expression extraction.
Now, the I mean it would extract the point locations I mean for example, I mean you can havea bilinear interpolation we had already talked about bilinear interpolation when we werestudying the sampling approaches resampling approaches so the nearest neighborhood I meanoperator the bilinear interpolation, the cubic interpolation, so we had seen what is the what arethese types of interpolation.
So, we can extract the value using to I mean interpolation techniques such as bilinearinterpolation we can use other interpolation technique as well and it would attach thisparticular value to the new I mean new field in the new layer. So, it could be a point feature Imean we can extract create a point vector data set and we can extract the values from theraster data set and tag it as an attribute to the point attribute feature table.
Now, the data set could be a raster layer that is the input layer or it could be a polygon featurelayer and extraction tools would basically extract the cell values which is defined by the rasteror polygon layer like; in our earlier point we were talking about extraction using point featurewe can also do the same operation using a raster feature or a polygon feature using a vectorpolygon feature or raster area feature.
So, what we can do is we can extract that cell values within a defined cells of raster which islike a mask or a polygon air polygon layer and assigns no data to cells which are outside thesemask layers of the raster or the polygon feature layer.
(Refer Slide Time: 14:38)
We can also do buffering like we had done the buffering in case of vector layer where in thebuffering of a operations would give you the physical distance measure the physical distanceand it is also similar to I mean both the vector and the raster operations where we can measurethe distances from selected feature. Now, when we were talking about vector bufferingoperations we had used the x- and y-coordinates while measuring the distances.
Now, we can also I mean in the case of your vector buffering we had seen that we can createvery accurate buffer zones in comparison to a raster buffer I mean when we are doing a rasterbuffer we cannot have such accurate buffers because your line coordinates would be veryspecific or point or polygon coordinates would be very specific, so you can create veryaccurate buffers, but with raster data sets you may not be able to create such exact buffers.
So, the vector buffering operation was more flexible and it gives more options in terms ofcreation of even multiple buffers zone not just one single buffer, but multiple buffer zones. So,I mean we had seen that we could be we could create separate buffer zones for differentselected features or we can dissolve the buffer zones in the boundary for the all the selectedfeatures.
So, I mean these options are available in the vector buffering operations; but when we aretalking about raster buffering it uses cells in measuring the physical distances and generally itcreates continuous distance measures. And in the measure tool we had seen that we generally,splice those distance measures I mean when we have these continuous data values we wouldbe splicing those and grouping it into different distance regions.
So, we said that slicing is required to define the buffer zones because this operation wouldgive you a continuous distance measure. So, it is difficult to create or modify separate distancemeasures like in vector I mean buffering mode we had said that we can create separate bufferzones for each select feature. But in case of raster data it would be difficult for us to createseparate distance measure using raster based operation.
(Refer Slide Time: 17:22)
Now, let us talk about I mean representing data or analyzing and visualizing data in case of aterrain where we are talking about height information. So, we had said one such example ofdata could be a digital elevation model or it could be a digital terrain model so generally, Imean we can visualize the terrain using a shaded relief model.
So, in this we basically work out the ratio of the amount of direct solar insulation or directsolar radiation received on the given surface and it is generally work it can be worked out interms of your radians values. So, you can see that this method could be very well used forworking out the physical I mean quantities of radians values.
So, this is the very interesting tool wherein we create the hill shading which would simulateshading due to the effect of sun on terrain because there could be changes in the terrainelevations, so you may have hills or mountains and you can see those the effect of the hills in
the image and you can perceive the third dimension because of the this aspect of shaded relief.It would help viewers to recognize the shape of the landform features.
Now, there are four factors which would generally control the hill shading the first one is thesun’s azimuth or the direction of the light I mean from which the sun’s light is coming to thegiven sight. So, as we had said earlier the convention is to measure I mean we assigned 0degrees to the north and we measure the direction in a clockwise fashion. So, again I meanyour east would be 90 degrees, south would be 180 degrees, west would be 270 degrees andagain north would be 0 or 360 degrees.
The next factor which I mean controls your hill shading or shaded relief would be the sun’saltitude that is the angle of the incoming light with respect to the horizon there is also theeffect of slope the topography may have a kind of a slope. So, depending on the nature of theslope the surface would be affected and also the aspect the direction in which the slopeextends. So, we shall see in our later slides the different concepts of the slope and aspect.
(Refer Slide Time: 20:17)
Now, we can also do a hill shading so this is done using this particular equation I meanwherein you have this factors which are R f is the relative radiance values, a is the facet or araster cell or a triangle. Now we have A f which is the facets aspect I mean we can have afacet when we are having TIN a triangulated irregular network, so we can have a facet or itcould be a raster cell also so we can; we can; we call it as a facet your A s is the sun’sazimuth, H f is the facet’s slope and H s is the sun’s altitude.
So, these values are put in this particular equation to work out the value of R f which is therelative radiance values of the raster cell. So, this I mean equation can be run to generate theradians values of the raster cells or it could also be run on a triangulated irregular network orTIN data set which is a vector data set to calculate the effects of hill shading.
(Refer Slide Time: 21:36)
Now, talking about aspect and the slope we were talking about slope and aspect been one ofthe factors which basically has an impact on your hill shading. So, talking about slope first it isthe first derivative of the elevation that is it is the rate of change of elevation with respect todistance. So, I mean it is expressed as either percent slope which is 100 times the ratio of therise in terms of your vertical distance to the horizontal distance that is also known as run.
Now, the second way which in which your slope is expressed is degree slope in which wecalculate the arc tangent as the ratio of rise over run. Now, we can also work out the aspectwhich is the directional component of the slope say suppose; if you have a pixel or rectangularcell and it has a particular inclination if you put a drop of water at the centroid of thatparticular pixel the aspect would be the direction in which the water would flow, the drop ofwater would dyne or flow.
So, it is basically the it gives you the directional component of the slope because from slopeyou can only measure the first derivative of elevation that is the rate of change of elevation,but it is only the aspect which will give you the directional nature of the slope. So, again weuse the same convention that 0 is the north and we measure it clockwise.
Now, yours aspect measures can also be converted I mean you can easily convert it into linearmeasure since these are these terms are in degrees. So, you can either take a sine or a cosineof these particular degree values and you would get values ranging from minus 1 to plus 1. So,we can also convert the this aspect values into linear measures. So, I mean we generally useslope and aspect extensively whenever we have DM data and we can run it on different typesof analysis.
So, it is used extensively when we are studying watersheds, when we want to generatewatersheds from a digital elevation model we can use the slope and the aspect. So, I hope allof you are aware of watershed, if you are unaware of watershed please look into the definitionof what a watershed is. So, we can also work out the landscape units, we can also domorphometric measures, we can find out I mean this morphometric measures can be used forstudying soil erosion.
This slope and the aspect can also be used for site suitability analysis in case of a urbanplanning job. So, I mean we have the this visuals in which you can see that the aspectmeasures are grouped either into four or eight principle cardinal directions. So, in the first topimage you can see it is grouped as north east west and south and in the next one you can seethere are eight principle directions in which you have sub element of north east, south east,south west and north west.
(Refer Slide Time: 25:21)
Now, let us go into image classification that is the last part of this lecture. So, when we aredoing the raster analysis this is a very important tool which would give us some kind of slicingof the data based on the input values. So, these can be done on a statistical data sets, it couldbe done on multiple thematic layers and you can classify it all the different data sets and theattribute values based on the inputs.
So, you can run this I mean the same algorithm the same concepts on vector data as well asraster data. So, in case of raster data we generally do image classification of remote sensedimages, so it generates the land use and the land cover maps. So, I mean whenever we arerunning the classification it generally would assign different classes to the input pixel values.The assignment the way this image classification is done is based on different algorithms.
So, some of these algorithms are either supervised or some of the algorithms are unsupervised.So, we can find out the spectral reflectance values and distance from the class means whichare used as a guiding tool while we are running this classification algorithms. So, I mean wewould choose a class which has the same reflectance properties of the class means in thewavelength bands.
I mean when we are talking about remote sense image we have different wavelengths or bandsin which the data is captured or stored. So, we can have your the visible bands, we can havethe near infrared bands, or we can have your thermal infrared bands, or the microwave bands.So, we can have different bands in which the image can be acquired and then we can choosean algorithm to classify the data.
Now, there are different ways in which the classification can be done that we had talked about.It could be either done in a supervised mode, or it could be done in a unsupervised mode, or itcould be done in a way in which we can combine both these two methods which is also knownas a hybrid classification approach.
(Refer Slide Time: 28:01)
So, generally in supervised classification what we do first is to we create a training set inwhich we have a raster data. And we select some regions or areas where the groundconditions are known to us or the ground cover is known to us, so we use that as training setsand then we create a signature file. So, the from this signature files we try to calculate thestatistics of your mean, variance, standard deviation.
And we create the signature I mean values for these I mean statistical parameters. Then whatwe do is, we try to take each and every pixel and try to measure the distance to the means ofthese training samples or classes. So, wherever the distance is minimum it is used I mean thepixel is basically assign to that particular class, so that this is the last operation that we do onthe data set.
So, I mean in this case u as a user would guide the classification process. So, it depends onyour ingenuity of how good you are at interpreting the data or about your knowledge of theground conditions. Now, I mean we had already talked about acquiring the training sets Imean it has to be representative of the entire image. So, that the pixels that you are using astraining sets, I mean would have representation on the entire image.
So, I mean we can assign classes and create a signature file. So we can create a name or wecan assign a color to the signature samples that we had taken. So, those output pixels wouldalso have similar color; they would have similar same pixel values as your class number inwhich you have acquired the signature training sets. And it will also have the same name;either name that you are specify it that same nomenclature would be retained in the outputimage.
Now, the training set is used by the software to identify the classes and I mean we havedifferent types of algorithms for classification which is used in the software’s. So, you can readabout parallelepiped classifier, you can I mean read about Gaussian maximum likelihoodclassifier, you can we also have the minimum distance to mean classification scheme, or wecan also have a principle component.
(Refer Slide Time: 31:07)
So, there are different types of algorithms which are used or approaches which are used whenwe are doing a supervised classification. So, in the maximum likelihood classifier we try toidentify pixels and find out the minimum distance to the mean of the different classes to thatpixel value. And wherever the highest probability is there based on the statistics for each classin each band the I mean it is assign to that particular class.
And in this case the basic assumption taken is that the data is normally distributed. So, if yousee the I mean histogram of a satellite image generally you would see an kind of a invertedbulk of. So, which is representative of a normal distribution and that is the basic premise ofthis supervised maximum likelihood classification algorithm that the data is normallydistributed.
Now, the second one second approach is the minimum distance. So, I mean we create theclasses based on the nearest class and we I mean calculate the mean vectors work out theEuclidean distance. So, there are different distance approaches we have different names fordifferent distance and your equations. One such way of finding out distance is named on ourgreat statistician that India has produced Professor Mahalanobis.
So, it is known as the Mahalanobis distance which is also used to do the supervisedclassification. Otherwise you have different types of distance such as Euclidean distance. Wehave the next one which is known as parallelepiped in which you specify a box which isbasically the bounds of a particular class the upper and the lower bounds of a particular class.And it is based on the mean and the standard deviation of this particular parallelepipedtrainers.
So, I mean if the pixel has values in different bands and it falls within the parallelepiped we callit has a parallelepiped, because say suppose you have two bands in your input raster layer youmay have two bands. So, if you plot that pixel value in x and y you will get a feature space youwill get the pixels will lie in that particular feature space.
Now, you can create parallelepiped parallelograms, so and in case you have more than twobands, so your data would extent to the third dimension. So, in that case you can extrudethose parallelograms into parallelepiped, so that is how the name is derived. So, I mean wewould work with multidimensional data and this is a approach wherein we give the bounds theupper and the lower bounds of the training data values.
And wherever the pixel if it falls within those parallelepipeds, so it is assigned to the particularclass and it may so happen that the training parallelepiped classes may not be adequate tocover the entire range of pixel values. So, there would be some pixels which would remainunclassified in such a case. So, I mean it may also happen that the bounds of the parallelepipedcould be overlapping one another.
So, in case a pixel falls within such an overlap it would put it in one of the two classesdepending on the statistical mean and the standard deviation values or it could also be put intoa overlap class. So, we said that if the pixel does not fall in any of these parallelepiped classes,then it is classified I mean it is given or assigned as unclassified or it is assigned as a null class.
So, among these the parallelepiped classifier is the least computationally intensive and thequickest. So, I mean if you want to do run I mean quick classification then in that case theparallelepiped classification would be good. But, it has its inherent limitation in terms of pooraccuracy and that your many pixels might remain might be assigned to the null class or it mayremain unclassified.
So, I mean we see these this particular image which is the feature class which is the featurespace of band 1 and band 2 data values. So, we have these class boundaries in which you see ablue class, red class, and a green class. And this arrows give you the distance vector meanvectors which gives you the distance from the mean of this particular data cloud.
(Refer Slide Time: 36:35)
Next we can also use a unsupervised classification method. So, in which the data would bedivided into number of clusters depending on how many number of clusters you want and lateron you can edit those clusters. So, I mean there are two basic steps wherein it is used forunsupervised classification.
So, first is we generate the clusters and the second is we assign the classes of the pixels to thisparticular clusters. So, some of the I mean algorithms which come under this unsupervisedclassification are the K means or the ISODATA classification which are basically iterativealgorithms. That basically your the entire data of a particular band would be partitioned intothe number of classes that you desire.