 Hello everyone, welcome to the next lecture in the topic land use land cover monitoring and change deduction. In the last lecture, we got introduced to the concept of what land use land cover is, what change deduction is and what are like the different ways in which we can on the basic introduction to how remote sensing is helpful to do land use land cover mapping. In this lecture, we will see like few commonly available land use land cover datasets and few examples to tell us how this can be applied for real life issues. Whenever we want to do land use land cover classification, we need to remember one thing, LULC classification scheme. This will be really important whenever we want to create some LULC map for ourselves or when we want to interpret some already existing data or when we want to do change deduction ok. So, simple example is let us say I want to compare the changes in land use land cover from the year 2000 to 2020 for some cities in India. When I want to do that, let us say I have data from some source like somebody else has created a LULC map in the year 2000, I get it. Now, I am going to sit and create an LULC map by myself. If I want to compare them and come to some sort of meaningful conclusion, then definitely there should be some correspondence between the different classes right. Say whatever a classified as like forest, same classification should be followed in both the maps ok. This is forest like I might have classified something as like just forest. The old map may contain ok, this is like nevergreen forest, this is like a deciduous forest and so on. But anyway they correspond ok, this is forest, this is forest. I can just combine the 2, 3 different forest type into one forest. I can do some analysis to reclassify them that is possible. Say I classified something as forest, but according to that old person, if the same definition goes to like a shrub land, then I cannot compare it. So, there should be some sort of uniformity between what we define as one particular class. So, for helping us to do maintain some sort of consistency, people have developed what is known as classification scheme. So, a classification scheme will tell us or will define what a certain class is ok. If a class can what to say satisfies certain criteria, then the pixels corresponding to the pixel should be categorized as this particular class. So, an LULC classification scheme will define such things. So, that is like the major advantage and they will be like consistent. If we use the same classification scheme for like say multi-temporal images, there will be consistency ok. If whether I do LULC classification or if someone else does the LULC classification, if we follow one particular classification scheme, there will be a consistency ok. Forest should have this definition and urban area should have this definition like that everything will be defined properly. This is one thing. So, a classification scheme contains taxonomically correct definition of clauses organized according to some logical criteria. So, that is really important taxonomically correct definition according to some logical criteria. It should not be illogical right, a forest should not be classified as illa like a cropland and defined right. So, there should be some sort of logical criteria and everything should be defined correctly. So, it will act as some sort of like guideline to us, whoever does the classification. So, what should be like the major qualities of a classification scheme? A classification scheme should be mutually exclusive that is there should not be overlap between different different classes. Say a same class ok this is like a shrubland, this is like a desert vegetation. There should not be two classes, there should be some clash because some shrublands tend to have like vegetation growing in desert. So, that this is like a mutual overlap or kind of like a clash. Such clashes should be avoided, the classes defined in a classification scheme should be mutually exclusive, there should not be any overlap. Then it should be exhaustive, exhaustive means it should contain whatever classes we need for our purposes. Say our goal is to identify the major classes present within a given area, all those things should be present there within the classification scheme. Say I want to identify like say some particular land cover will have what is known as like marsh lands, salt pans and so on. So, these are different categories of land, but if I am going to use like a classification scheme which does not have a definition for what a salt pan is, I cannot use it right. So, I have to improve it, I have to add ok a salt pan should be defined like this. So, for our own purpose the classification scheme that we are going to select should be exhaustive, it should contain to the maximum extent possible what we need. Then it should be hierarchical, hierarchical means there can be like a broad definition within that there can be subclasses. Say forest, within the forest evergreen forest, deciduous forest evergreen broadly forest like that. So, there should be some different level or even if it is like urban area, within urban you can have commercial residential and so on. Ok, so you should have some sort of like hierarchy. So, a classification scheme should be mutually exclusive, exhaustive and hierarchical. Say this is one such example, this is example taken from what is known as an IGBP classification scheme, International Geosphere Biosphere Program. So, they have defined 17 different land cover classes and so on this is like the definition. If you want to define something as an evergreen needle leaf forest, it should satisfy this criteria. If you want to categorize something as like an urban and built-up plan, they should categorize this criteria like this. There will be like some sort of criteria defined and if we look at those definition then whoever does the classification, they will follow this ok. This is the classification scheme we have adopted for the project, we will go by the same definition, there would not be any confusion. So, this is sort of one example. Then the next example is given here, this is like again a general LULC classification scheme developed by Anderson and others in the year 1976, which is still used by many. So, this is developed by United States Geological Survey where you can see like two different levels, level one, urban or built-up land, within that you have residential, commercial, industrial, transportation and so on. So, if someone stops with just say this is maybe this is my image, I just identify this is urban, this is agricultural land, this is forest land. If I identify only these things, then I call this map as level one map. Then if someone does this is urban, within the urban this is like residential area, this is commercial area. If someone further goes into this, this we call it as level two classification. So, this map or this classification scheme gives us two level. Some classification scheme may have three or four levels, say within residential you can have like a single income residences, multi income residences and so on right. So, this is level three. So, the classification scheme is this is an example for how hierarchical in nature classification schemes will be. This is really important for even before we start a project, whenever we want to start a project or like some sort of like application activity, we should be really careful. Either if we do classification on our own or if we take data sets from others already existing maps or images, we should be first thoroughly understanding what is the LELC scheme follow there, what are the definitions and so on. Then only our all our goals, project goals can be achieved. And as I told you earlier, if we have two LELC maps and if they follow different classification schemes, direct change detection is not going to work. We have to somehow bring both of them to the same classification scheme and then only we can do change detection analysis. So, this is extremely important for us. Some LELC data sources that are available to us. So, MODIS sensor produces a yearly land cover product. So, you can just look at like MODIS land use land cover product and get it. It is produced every year at 500 meter spatial resolution and they use different classification schemes. So, whatever based on the needs, there are like multiple classification scheme available, we can just download the image and use it. European Space Agency produces a land cover product. So, they have like a different resolutions and they are also producing at say 5 year intervals. In the recent past, they are producing at 1 year interval and so on. So, that is available. These are like global products available across the globe. Then there is another product called Globe Land 30, which is again like widely available. So, this is produced using like lot say Landsat datasets, but this is not again operational product. They have just produced different instances of map which we can download and use. Then for India, there was like what is known as a Decadal LELC database produced by Roy and others in the year 2015, where they had LELC map at every 10 year gap, say 85, 95, 2005 and all, 1985, 1995, 2005. So, this is just one such example of an LELC map taken from Roy and others published in the year 2015. So, this again is openly available, but I think the data is available only till 2005, if I remember it correct. And also there is a global forest change map. So, forest change, this is not like an actual LELC map, but this will identify how forests have changed with 2000 as the base year. From with respect to 2000, how forest are changing. So, this is like done at 30 meter pixel level and this is again like the link is provided here in the bottom of the screen. So, this is one of the widely used dataset, but this will just tell forest and its change from year 2000 how forest changed. So, this is like a forest change map. So, these are some of like different global data sources and also Roy et al is only for India. In addition to this, in the lecture about like data portals, I told about Buven data portal. So, the Buven has LELC map for India for at different scales. So, that also we can use for our application needs or research in. So, Buven has but these are downloadable datasets which we can use for our projects. Now, we are quickly going to see few examples of how this LELC map can be used for real life application. So, the first example what we are going to see is monitoring urban expansion. So, when we do as I told earlier we can do or we can create LELC map over different time intervals and then do the change reduction. So, this is post classification change reduction or we can do change reduction directly. So, here we are going to talk about one example where the authors have analyzed and the growth of Chennai city using like a simple classification scheme. So, this is how the LELC map for Chennai city looked in the year 91, 2003 and 2016 with each color representing the class given here, built up urban agriculture land or in like green, dark green represent general vegetation, barland, barren land is like which is not being used and so on. So, we can see how the urban area is kind of like expanding with this right. We are clearly able to see from the year 91 to 2003 to 2016. Similarly, we can see how agricultural land is expanding maybe like natural vegetation are kind of like removed and it has become like agricultural land. Maybe how water bodies are shrinking that is also maybe we can see some small water bodies might have gone. So, this kind of LELC classification and comparing them at different time intervals will help us to understand how different LELC features are changing say this is corresponds to urban growth. A similar example has been carried out over like Hyderabad city. So, this is over Hyderabad you can see from starting from January 2005 to January 2016. We can see how built up area have grown tremendously right. So, this is again like example for using LELC maps for understanding urban expansion. This is really necessary like when someone wants to do some sort of like planning to develop a township say within last 20 years the city might have expanded drastically which will put a several constraint on the resources available say water, land, air quality problem and so on. So, the government may think okay the city has grown tremendously we may need to develop like a satellite township at the outskirts such what to say policy interventions may come from the government side. So, in this particular paper they have used like a very simple classification scheme it is not like really complex they have taken like multi temporal satellite images they have like norm what they have done is bought all the data to like the same spatial and radiometric qualities then they have done like unsupervised classification that is they have as the computer do clustering first and then using other information they have categorized temporal land use land cover maps they have validated those LELC maps like whatever map we produce we have to check the accuracy with respect to ground data. We are not going to cover accuracy assessment but I just I am telling accuracy assessment is one of the important tasks in land use land cover mapping and change reduction. So, they have done validation and then they combined other data sources in qualifying the LELC change. So, this is basically they have done post classification change analysis. So, the important steps are they have like identified images multi temporal images they have bought all the images to same spatial radiometric qualities done classification validated and done accuracy assessment and then the change assessment or change reduction. So, this is like the broad steps involved and this is like whoever want to do can follow like a similar steps this is like the conceptual way of how to do this but the exact algorithms we need to use the exact way we need to process the satellite data we may not be able to discuss now but this is like the broad scheme of things in which one should work. We need not even do LELC classification say if our interest is in identifying what is the change in green spaces over like a city even like a simple NDVI map and careful analysis over such data will provide us huge information like one such study is being like cited here one of our own students from IIT Bombay has done this work so this is like urban green space map for the Mumbai city over at year 2001. This is like the green space map over again Mumbai city at 2011. So, the urban green space are like defined using NDVI map simple NDVI images no LELC classification are being done under using simple NDVI and using identifying these all green spaces may be parks, grasslands, forests and so on. So, these are all classified as urban green spaces. So, just you can refer to the paper to deal or to know more about like the definitions used but as a simple example I am telling. So, within each boundary here is kind of like a census ward or a census section within Mumbai city. So, within each census ward or census section how much is like the green cover present in different years and in those 10 year period from 2001 to 2011 how this has changed. So, you can see most over most of the places it is kind of like a negative change. So, this actually suggests in almost in most of the census section there is like a decrease in the green cover. So, which is kind of like not really good scenario we have to improve the green cover for a sustainable and happy living. So, this is like a very simple example of how to use remote sensing data sets no need to even do classification but this is like a very simple change detection analysis you can think it off. Create like NDVI maps properly defined urban green spaces once everything is like properly defined and classified this is like a green space and all we can do simple change detection technique and say okay the land cover has changed it like this. So, this is one such example of producing really important study using like careful analysis of remote sensing images even using simple techniques. Next major thing we are going to see is how LULC change map will tell us about deforestation effect. So, in this particular slide we have like a true color image or a natural color image whatever you can call obtained over like Amazon forest in Brazil. So, this is in the year 2000. So, just by visual display we can see these are all cleared lands and the rest of the green patches are like Amazon forest. In the year 2012 it became like this the cleared lands remained as cleared lands but whatever is like labeled as forest here they are now became like they have been now cleared. So, this even by just visual looking we will be we are able to tell okay such huge area of land has been kind of forest has been removed for some other purpose. So, we can actually quantify what is the area of decrease in forest how much time it took. So, these are very simple example this is not again we have not done any sort of LULC classification you can just look at like the spectral characteristics maybe you can take simple differencing in maps and say okay this is how forest has changed this is how deforestation has occurred within this particular country. So, this is an example of a simple visual change reduction and when you do it in a computer and do perfect statistical analysis you will be able to tell what is the range of deforestation or the extent of deforestation that has occurred in that particular region. Till now we have seen examples using optical datasets like the LULC maps were all produced using visible and NAR datasets or even like the simple change direction examples I told you are all produced using visual and NAR bands or visible and NAR bands but even we can use thermal infrared remote sensing for identifying large scale transformation in like the land cover pattern and one such study we are going to discuss now. What is known as like a yearly land cover dynamics by combining LST and NDBA here LST means land surface temperature and NDBA is you know it is like a simple spectral index which will tell about the vigor of vegetation. Say the study is very simple conceptually to understand we will normally have multi temporal images within a year over like every pixel right. Say we to I already discussed same pixel will be repeatedly imaged by the satellites what we call it as like the repeat cycle. So, we will be getting land surface temperature and we will also be getting vegetation. As the land cover changes the way in which the LST and NDBA for that particular pixel orient will change that is let us say this is like the LST and NDBA for one such pixel in somewhere in Sweden. Say for each image for that corresponding pixel they have plotted the NDBA value in the x-axis, LST value in the y-axis. So, it will define those points. Based on the features present there and if you perform like a regression analysis based on the feature present there the regression line will take one particular angle with respect to horizontal and one particular slope either in positive direction or negative direction. Say this is like a what this tells this tells then NDBA increases LST basically increase. So, here LST is like normalized this is normalized LST normalized means with respect to some minimum and maximum they have like normalized LST minus LST mean divided by LST max minus LST mean some sort of normalization. So, we are converting all the LST to between 0 and 1. So, what this particular image represents for that pixel if NDBA increases LST increases. So, what this tells as temperature rises like as crop grows temperature which is tells us ok that particular pixel is vegetation growth there is controlled by temperature like some areas they can be in winter they can be covered with snow and all. So, vegetation will not grow there will be very less temperature sunlight and everything but during summer time vegetation will what to say grow healthy. So, that means as temperature rises when winter resides as temperature rises vegetation will grow. So, such pixels are controlled by the vegetation growth is controlled by temperature over places like India like semi-arid those kind of places water availability will control the vegetation growth. So, for such pixels especially some pixels over India semi-arid region the graph may look something like this this is NDVI this is LST cap that is whenever there is like a dry bare surface LST will be high NDVI will be low but over vegetated surfaces LST will be less. So, this will signify vegetation is actually doing cooling of the surface. So, this will this sort of pattern will be seen over semi-arid regions over deserts NDVI will be more or less constant will be in the lower band but there will be different values of LST. So, it will be very steep straight line over again humid forest like Amazon rainforest which will be green always the graph may look something like this like very high NDVI value and LST will varying only with respect to that zone. So, these are some examples of how LST and NDVI will orient themselves for the same pixels here people have done multi-temporal analysis like for the same pixel get save 8 to 10 images per year or more than that and plot them together. So, how these points are orienting in the LST NDVI space whether it is positive slope or it has a negative slope or if the slope is just like a straight line with 90 degree angle from horizontal all these things will tell what sort of land cover is there. So, basically here so these are like as I told semi-arid lands or cropland kind of thing these are like places where the slope was very steep and where the slopes are really steep. So, this actually signifies there may be deserts or something like forest and so on. Say wherever the slope is between minus 15 to plus 15 actually those places show like a very good temporal change in LST and NDVI. So, this is like I am not elaborating everything in detail but by looking at the angle of the slope of this regression line by looking at the variation how much LST changes over one particular region all these things will tell us really good amount of information about the land cover presenter. So, this is not actual land use land cover classification but this will tell us conceptually okay that is a humid zone that is like a desert that is like a vegetated area where vegetation growth is controlled by water availability may be over certain regions like over here and the higher latitudes vegetation growth may be controlled by temperature if temperature increases during summer vegetation will grow during winter all the vegetation will shed the leaves and they will go into some sort of like hibernation about these things. So, these are really good examples of understanding about the land cover pattern by combining thermal information with NDVI information. So, this sort of maps has been produced every year. So, this we can create this do this sort of analysis for every year or even across the globe this has been done and how each pixel has changed let us say a pixel was for us before. So, if it is like a dense forest for that particular pixel the pixel may look something like this like all NDVI values were very high and LST was oriented like this the relationship can be like this with like a very steep angle that can if the pixel is like a dense evergreen forest suddenly if deforestation has occurred and now it has transformed to kind of like a cropland then the next year or whenever deforestation has occurred that pixel may look something like this like if it is cropland when you put water vegetation will grow thereby reducing the temperature let us say water availability controls the temperature there some pixel here actually tropical region. So, the slope has gone from like this very steep slope to a negative slope value this will indicate ok some drastic land cover changes happened over that particular pixel. So, this will tell us the change in broad classes of land cover across the globe. So, this is what people have defined as yearly land cover dynamics observed from LST and NDVI maps. So, as a summary in this lecture we have discussed about LELC classification scheme and we have seen few examples of how this LELC classification or in general remote sensing information will be helpful for understanding land use land cover as well as understanding the changes happening in land use land cover. So, with this we end the topic of LELC application like here I have not talked anything about like the algorithms what sort of rules we follow and all those maybe like has to be dealt under like image analysis course which we are not going to concentrate now there that is out of the scope of this course. But I just highlighted to you what land use land cover classification is and in what all different ways this can be put to use. So, all the corresponding references were given in the slides within the video itself you can identify the references and you can learn on your own and about your interested topic. So, with this we end this particular lecture. Thank you very much.