 Yes, good morning everyone or good afternoon actually depending upon what time it is in your local time zone. Welcome to this webinar this is the first one this year that we are organizing as a joint collaboration between the water channel and IHG. This is the first in a series of content that we will be producing jointly over the course of the next one year, which will include podcasts webinars and blogs. If you have registered for today's webinar will keep you updated regarding all that regarding that line of content. The discussions today will be focused on refugee communities and their interaction with the environment around them. We live in very unfortunate times when almost 30 million people around the world have had to flee their homes and live as refugees in temporary communities having to lead the sort of temporary lives. And 30 million might not sound like a huge number compared to the global population and as against the global scale but these 30 million refugees are concentrated in certain parts of the world certain countries or certain reasons that for some reason or another form like a natural receptacle for people fleeing war and conflict. And when you're living in a temporary community relationship with the land with the water. It is different compared to if how it would be if you were living there as a more permanent resident of that place of that area. But how so exactly how like how different is that relationship exactly how exactly to refugee communities interact with the natural environment around them. How different is it compared to like more permanent communities. Thankfully we have GIS data going back several decades which can help us study this and thankfully we have with us. She won Kevin Rooney who has done exactly this kind of study. She won is from Uganda. She has recently graduated from the water science and engineering program from IHEL. She has a background in civil engineering and she has been working as a civil engineer for Uganda's federal government since 2016. Moving over the proceedings to Sivan. I would just like to encourage you to please post your questions and comments in the chat box. We will keep collecting them throughout and we will discuss them after Sivan has completed her presentation. So Sivan it's over to you. Thank you so much. Abraham. I hope you can get me well everyone. Yes, my name is Sivan Keburunji. Allow me to welcome you to the seminar. Good morning. Good afternoon. Good evening. I'm presenting a seminar on the topic can quick long term assessments of refugee impacts on the hosting natural environment be conducted from freely available remote sensing data. And as he mentioned, I'm a recent graduate of IHEL water Institute for water education. And these were my mentors and supervisors for my thesis study, based from this seminar topic. So the presentation content includes an introduction with such objectives and questions methodology results discussions that are limitations and conclusions. 26 million people have been displaced globally as of the end of 2019. This is according to UNHCR and environmental degradation is expected, of course, as refugees or existing camp areas. The monitoring refugee impact on the natural environment is necessary to inform countermeasures and remote sensing has registered increasing usage in the recent years in this field of refugees. In recent studies, the areas that have been tackled include extraction of refugee tent areas using remote sensing, mapping refugee camps, monitoring land cover changes, monitoring urban expansion. And the area that we thought the research could tackle is based on these issues. One is that fewer studies exist with significant temporal coverage. Most studies use high resolution data that isn't freely available and also has a low temporal coverage. And also most studies use sophisticated methods which can be time consuming. And then we have more studies conducted in varying climates necessary or recommended from a number of these studies. This research was based from the Kester Disney Ethiopia, which is home to about 779,261 refugees and asylum seekers as of August 2020. And these are contained in 26 refugee camps, three hosting communities and six settlements. Some of these camps are over 20 years old and they present a diverse historical, climatic and social context. And we have refugee entry sustained in the east mainly by the political unrest in Somalia and also droughts in Somalia with refugees coming in through the east and the southeast. And in the west, we have the political situation in South Sudan that intensified actually from 2013 onwards. And also in the north we have a political situation in Eritrea that causes refugees to flow into the country. So as you can see the neighboring countries are all having refugees into Ethiopia, which has become a home and has created an enabling environment for refugees. There is such objectives included to investigate the impact of refugee camps on land cover vegetation and surface water resources and corresponding relationships with population increase. And also to assess the methodological use of indices from open access remote sensing data to monitor rapidly refugee camp influence on the surrounding natural environment. My guiding research questions were one, what is the impacts on land cover and vegetation? What's the impact on surface water resources? What are the relationships between the impacts I find and the population? And what are the methodological pros and cons of using indices? So why did we choose to use indices and what are indices? Indices are computations between spectral bands of an image using established formulae. They are quick and easy to apply and they are favorable for large temporal analysis and that's why they were prioritized. However, we have other approaches like supervised classification that yields better results for land cover analysis, for example, but is time consuming. And the selected indices include the normalized difference vegetation index, the NDVI, which utilizes the near infrared band and the red band. And this has been extensively used to study changes in vegetation biomass and vegetation health and ground to examine the growth of plants according to the phenological stages and how the biomass is changing. And this index ranges from negative one to one with low negative values for water and high values for healthy vegetation. I also use the enhanced vegetation index, which improves on some of the weaknesses presented by the NDVI, some of which include influences from soil color and soil brightness and atmospheric conditions. So the enhanced vegetation index, the EVI improved on that and I use the two band enhanced vegetation index, which also uses the same bands near infrared and red and also ranges from minus one to 1.25 with low values representing water and high values representing healthy vegetation. The normalized difference water index or the NDVI utilizes the green and the near infrared band. And this was specifically used to study the surface water resources and it renders from minus one to one with low values representing land and positive values representing water. And how was the index, how the index approach applied in this, in this study. We used it for land cover classification. And four classes were envisaged under the study where we have water representing open surface water areas, sparse vegetation representing built areas, bare soil, rock and unhealthy vegetation, moderate vegetation representing moderately healthy vegetation. And dense vegetation representing forest and very healthy vegetation. And indices because they pick on the near infrared, which is reflected by green plants is actually affected by climate and as you can see, a healthy plant reflects a lot of near infrared and very little red light, while an unhealthy plant reflects less near infrared light and more red light. And this causes a difference or a variation in the index values that are registered. And so it was important we tailor the analysis to a wet climate which is expected to have healthier vegetation and a dry climate which is expected to have less healthy vegetation. So the analysis was tailored to two climatic contexts in that regard. And the workflow followed obtaining class thresholds, and then using them to conduct temporal analysis, and also to compute an index anomaly, which gives a deviation of indices from the way from the values before the campus was established, and to values after the camp was established. And so what was the impact we targeted to identify vegetation loss and deterioration, as well as declining surface water areas. And I hope to do this by detecting areas that had increasing sparse vegetation with decreasing dense or moderate vegetation, and also decreasing surface water areas. And when was it opportune for me to detect this periods that had population increase and prevailing wet weather conditions so that we could pull out the impact of the human influence. And the wet weather conditions were determined based on the drought indicators the SPI and the SPI. We have studied where eight of them fall on the western side of the country and this is the wetter region, the wet areas and it receives more rainfall than the eastern side that has predominantly arid climate in the eastern side of the country. The selection was based on climate population and location of the camps. And also this, these eight camps consists comprised of 35.3% of the refugee population as of 2019. The topography varies from 189 meters to 17, 7200 meters. And the study period was from 1984 to 2020, basically because that's when Landsat imagery is available from this range, but it varied from best on the edge of the camp. And I consider January and February month spire, because that's when we experienced the least rainfall in the country, and we hope that would be the least interference from precipitation events. The wetter that was used is Landsat 578 imagery and a remote desk study was conducted so there was no field studies or investigations done due to the limitations presented by the COVID-19 pandemic. So the first task was to set the index thresholds for each land cover class. And this was conducted in both the wet and the dry climate context. And it was conducted through the Google Earth Engine platform using JavaScript language. And the process followed me obtaining Landsat images. This is surface reflectance images from 2014 to 2020, filtering them based on the January and February months on cloud cover of 20%. I get a median clip to the study area. I draw training polygons for each land cover class. And this was enabled by visually inspecting the images and also using various band composites like the color infrared and the shortwave infrared composites that enhance information from each image. And then from this training polygons, I transferred them to QGIS and obtained class thresholds. That's the range of each class. And then I computed the indices. And I also performed supervised classification. This was using a random forest classifier because it has registered higher accuracy. This supervised classification methodology. I hoped to compare accuracies from the indices to the image that was developed using the supervised classification methodology. Then after thresholds were obtained, temporal analysis were conducted. And this was to identify land cover changes over time. Also done using the Google Earth Engine platform. For each year I got Landsat images, surface reflectance images. I filtered based on the months and the clouds. I got the image clip to the study area. I compute indices for each of these years. And I applied the thresholds that I developed and then I compute class areas for further analysis. The last method or step that I had to do was to compute the index anomaly. And this helps to determine spatial extent of possible human influence. This was also done in the Google Earth Engine platform. And I got index images, PAYA, which is the median image for the John and February months. I got images before the camp, got their median image, and that was my reference image. Then for each of the years after the camp was established, I subtract this reference image. Then I put the camp clip to the study area. And that gives the index anomaly, which is the deviation of the index from conditions before the camp was established. And so for the results that I got, the index thresholding. For the wet climate, I consider the Pugnido camp area, which is in the west. It actually consists of two camps, as you can see. The map shows a black square in the middle. That's a larger content that shows the area I originally considered. And then I had a buffer area offset of five kilometers to give me a wider range of analysis. That's seven. When I compared accuracies with the supervised classification image, NDVI gave me 0.727. And for EVI 2, it was 0.744. And for the NDWI, it was one. For the Landsat 8 imagery, the accuracies obtained from the NDVI were 0.766 for the NDVI. EVI 2 gave me 0.776. And for the NDWI, the accuracy was one. And for the dry climate area, I consider the Hilawayn camp area, which is in the southeast. And for the Landsat 7, I got an accuracy of 0.784 for the NDVI and 0.92 for the EVI 2 and one for the NDWI. Here I only considered Landsat 7 imagery because all the other dry area camps could not utilize the Landsat 8 imagery, which is only available from 2014, and has varying spectral wavelength ranges from Landsat 7 and Landsat 5. So the key points from the thresholding procedure was that accuracy obtained was satisfactory for the accuracy computations that I conducted. The EVI 2 presented higher accuracy than the NDVI for vegetation. And this was prioritized for temporal analysis of land cover. And there are lower thresholds in the dry climate camp area, which indicate lower vegetation health, which is expected for arid regions compared to wet and usually more vegetated areas. And the EVI 2 thresholds applied for other camps were applied for other camps in similar climatic contexts with the understanding that the EVI 2 values were comparable. And this is because we're focusing on quick and rapid analysis of large temporal coverage. And because of this, we applied these thresholds across time and across camps with a similar climatic context to see if it is helpful for us. Of course, we know climate will be a big factor, but we wanted to test that. For the temporal analysis, the first camp analyzed was Bambasi, which was established in 2012, and with a population of 17,653 in the eastern, in the western region of Benishangu, Gumuz. And as we can see in the original image, this black square in the middle is the camp extent that was just drawn from Google Earth imagery visual after visual inspection. And the five kilometer buffer area was offset from that to give the study area for this camp. We can see that from 2014 we have sparse vegetation increasing in the EVI 2 classification. And this is in a plan form that is synonymous with the actual camp establishment when you look at the original image. We can tell that there was establishment of the camp and we can see the footprint of the camp from this analysis. We see it in 2014, but in 2017 and 2020, within the camp extent, which is the small black box in the middle, we see these effects fizzling out or being lost. And we can see that we have dense vegetation as we move towards 2020. And because of that, we see sparse vegetation being reduced and also moderate vegetation actually in reducing to dense vegetation mostly. So we see that this impact is lost from 2017 to 2020. But in 2014, we can see this camp footprint in the middle of our study area. When we get a closer look to this camp, we see the actual camp in a zoomed in image from Google Earth imagery. And there's a river that is crossing the study area. This is the Davos River. It's flowing north. At the northmost point, the catchment area is 11,734 square kilometers. And to better understand what was happening, we analyzed the areas that we developed from the classification. And we can see that from 2012 to 2015, in the camp extent, which is the first graph here for the EVI-2, we see an increase in sparse vegetation and a decrease in moderate vegetation. And in this same period, that's 2012 to 2015, there were no droughts. There were no prevailing droughts in this area, severe droughts. And this is the SPI analysis based on the 12 month time scale. But we see at the five kilometer buffer scale, this effect is not visible. And instead, we see dense vegetation increasing as we move to 2019 and 2020. And what can we conclude from this? We see that from 2012 to 2015, there's an impact at camp extent scale based on the land cover analysis. And based on the NDWI analysis, we see that there's a declining trend in the surface water areas from 2008. Detailed hydrological studies were not conducted to better understand why we have declining water areas from 2008. But they would be recommended because the strength is quite alarming. No water was detected after 2008, and this could be due to declining quantity or quality. And also, it's unclear if it's due to human influence or from the refugee camps or from further upstream. Based on the index anomaly study, we see that within the camp extent, there are regions, there are red regions which indicate declining index values or decreasing vegetation health within the camp extent. And we can see that the camp is actually an increasing southward. I compared these results to a 2016 European Space Agency land cover map that was obtained. And we can see that, yeah, within this area, there's a bit of cropland. This map provided more discrete land cover classes to better understand what's on ground. In the east, in the west, sorry, there's a town over there and we can see the increase of that town tending to the northwest with increasing regions of red or declining index. We also see some red sections along the river that show that some human activities could be ongoing or natural processes could also be ongoing. So investigations are needed to further confirm if this is due, if the declining index regions are due to human influence or natural processes. When we analyzed another camp called Hila Wain, which was established in 2011 with a population of 33,936 in the region of Somali in the southeast. This is the third climate and from 2016 to 2019, the original image shows a carbon footprint within this camp extent box that's in the heart of the images. But when it comes to the EVI to classification, there's hardly anything that can be seen because there's predominantly sparse vegetation in this area. So it was not possible to tell the any impact of the refugee camp based on this visual analysis and on this EVI to classification. When we come to a closer look into the camp, there's a river also crossing this area. There's a general river that flows south, which also presented declining water trends from 2011, and it has an estimated catchment area of 66,645 square kilometers. When it came to further analysis of the areas, virtually not much information could be extracted because it's predominantly sparse vegetation. So we can see that this methodology was not very good for the dry or arid context. And the key result we got from this camp is that there was declining water areas after 2011 detailed hydrological analysis are needed though to confirm the extent and also the causes of this decline. So we cannot confirm that this is due to the refugee camp or even surrounding population because of the large catchment area contributing to the water that we are seeing. And for the anomaly, we only see areas of decreasing vegetation health along the river. That's those are the areas in red. You know, those areas are being used for crop land, a bit of shrub land, and a bit of grass, and not much within the camp extent could be determined from this analysis. And for the other camps, the other six camps that were analyzed, both in the wet and in the dry climate context, there's no apparent human influence in the environment that was detected using this methodology. But for the EVI2 anomaly, all the other camps showed areas that had decreasing vegetation index, indices, except for this camp called Barathe, which is in the north, that did not present any areas with decreasing index. Again, these areas show potential human impact. They require field verification to confirm if it's indeed human impact or maybe other natural processes like soil erosion. From other databases, six camps presented forest loss. That's from the Hansen Forest Change Database. Burnt areas were observed in two camps based on the Modis Burnt Area Database. And based on the land cover maps, we realized crop farming and livestock rearing was the most predominant livelihood opportunities that these guys were using. And because of that, the wet areas have more opportunities for livelihood or for survival because they have shrubs, they have grass, they have rivers flowing, they have good climate, while areas in the arid regions were less suitable for survival because they have large vast areas of barren land. As a discussion, temporal analysis were affected by meteorological conditions. The main discussions were drawn also from the prevailing drought conditions, that's SPI and SPI. A number of consistencies in the land cover classification was noted. And that's why we had increasing moderate or dense vegetation in drought conditions and increasing sparse vegetation in wet conditions without human influence. Factors causing this inconsistency is mainly with the types of timing in relation to the image acquisition deaths for the satellite images, the type of vegetation and phenology stage as it varies across years, and also significant river flow into the study area that supports irrigation and wetlands sustenance. Detailed hydrological studies are required of course to further understand the declining water trends that were observed. So the pros for this method is that it was quick and easy to apply visual inspection of land cover maps was generally satisfactory for most camps and the EVI2 anomaly showed preliminary areas with potential human influence requiring field verification. And for the cons we have that this methodology is very dependent on thresholds applied, is influenced by prevailing meteorological conditions, is not suited for areas without significant vegetation, as we saw in the wet in the dry area camp. And discrete classes like trees and wetlands could not be adequately determined without ground proofing data, which was not available. Alternative methods tested, we tested the automatic thresholding techniques for the indexes. This is the OTSU thresholding technique, it was not satisfactory for what we wanted. The unsupervised classification method also requires significant image post-processing, supervised classification requires significant image pre-processing, which are all time consuming and are not good for rapid analysis. Suggested improvements include setting land cover index thresholds for each image seen using ground proofing points in the future if we have access to those. And further research suggested is for quick and automatic index thresholding approaches for land cover classification. The data limitations that we had was ground proofing data which wasn't available to refine the land cover classification process. The actual camp extents on ground were not known, so visual inspection of the images is what was used to establish these extents. And in some cases, large camp areas were considered for analysis than maybe what may be on ground. And the population for refugees and surrounding communities over time was not readily available. The conclusions is that the index anomaly provided proved effective in preliminarily mapping out locations of potential human influence on the environment, pending field verification. The applied land cover classification methodology is not effective in monitoring the refugee impact on the natural environment with medium spatial resolution imagery. The use of high resolution imagery could perform better. And further research into quick and automatic index thresholding approaches for land cover classification is suggested. Thank you. Thanks a lot, Shivan. Can you hear me? Yes. Thanks. We'll now get to the questions while you're fired up. We have been compiling the questions and comments that have been coming in. Compiling all the questions. Could I ask one more question? Yeah, in your study, did you spend any time thinking about how different is the effect of a refugee camp on the surrounding environment as compared to the typical impact of, say, a village or a small township? Do you have any initial thoughts on that? I did not consider that much because most of my study was on the technical perspective, but we expect a more adverse effect of refugees because there are no systems in place. At least for a village, there's usually quite some support from the government and systems to ensure natural resources are managed to some degree. But with refugees, it's all about survival. And so the difference is that the impact is expected usually to be adverse than that compared to a village. Okay, I hope you can see on the screen now some of the questions. Yes. A question from Peter Riyad as which softwares or languages were used to get there in pieces. Okay, should I answer question by question? Has he read? Yeah, if I, if I, let's just read through the questions for the audience also has a few seconds to absorb the question before they can hear your response. Okay, which we start with Peter Riyad, which software languages were used to get the indices. Yes, I use JavaScript and this was implemented in the Google Earth engine platform. And I also used results from that script in the QGIS software for further analysis and production of maps and other statistics. A question from Mr. Endalchu Kabede is how many training points did you take for each land cover class? And do you think the land cover classes four will represent very well the actual land cover in the area? Okay, so how many training points I took? I was using training polygons and I had about six for each land cover class which would give me roughly between 800 to 1000 pixels per class. And do I think that they represent very well the actual land cover in the area? Not very well, definitely. With remote sensing, we always need ground-truthing data, always need ground-truthing data to ascertain what we see. But we know they are very representative of the vegetation health, which is essentially the backbone of the methodology based on vegetation health. Could I please ask the audience to please mute your microphones just to make sure that we are not speaking over each other? And long could you please help me with that? Could you please put on mute people whose microphones you noticed are on? Okay, thank you. Can you hear me now, Shiva? Yes. Yeah, so the next question is also from Mr. Kabede. Do you think the decline of water in the Somali camp is related only to, in the Somali region is related only to the refugee camp because the area is highly vulnerable to climate variability or climate change? Of course, I was saying as I presented that detailed hydrological studies were not conducted, so we cannot ascertain the extent of the causes. Definitely from both rivers, they had large catchment areas and our study area is just a speck of that catchment area. We know there's upstream events that really affect what's happening, what we are seeing in the area. So detailed hydrological studies would need to be conducted and also it's really to an insignificant extent the impact of the refugee camp because, like I said, there are very large catchment areas and I can't ascertain if the decline is really just from the camp itself. The next question from Neeraj is how much of a difference is there between the data type slash data quality from 1984 and let's say now 2021 and what are the implications of compiling together data with such differences? Thank you so much for that question. So the study was just based on Landsat data which is the one I'm just going to focus on. And we have Landsat 5 which is from 1984 to 2012 and Landsat 7 which is from 2000 to 2020 and Landsat 8 which is from 2014. Landsat 8 has narrower bands and so the images produced are more representative of really what's on ground. They are clearer and they are preferred to use but we cannot combine them with Landsat 5 and 7 because of the difference in these spectral wavelength ranges. Landsat 5 and 7 have the same spectral wavelength ranges and therefore I combined these two because really there was no effect of doing that. Landsat 7 has had data quality issues because of a technical problem they had in 2000, in the early 2000s and so very many images were discarded from the analysis because of those quality issues. I don't know if I've answered your question. I hope so. If Nidhaj is not yet satisfied, he may please follow up his question in the chat box. Thank you. We have one more question which is from Thais Begsma. If possible do you think there would be any value to doing ground research validating your research? If yes, what results will you look to specially double check and how? Okay, thank you so much for that. First of all, if we had ground research we would need to adjust the methodology because the methodology was based on inadequacy of data. If we adjust the methodology we definitely will get more refined results and we will have more discrete land cover classes for example instead of sparse vegetation and dense vegetation, we'll have forest, we'll have grassland, we'll have shrubland. And this is much easier to study and to analyze across time. And so that's the first thing. And then the second is that the ground research also would validate the results from the index anomaly, which I kept saying we cannot confirm if it's due to other natural processes. Like I said, soil erosion is one of them, or if it's due to human influence. And if we can tell on ground what is happening, then we can confirm that this is due to human influence and there can be some measures put in place to cover that. So in those two lines, I think this would be very helpful, but the overriding factor is if there's ground research, then there will be a change in the methodology as I had already suggested also in the presentation. Thanks a lot. With this we have reached the end of the questions that have come in so far. If somebody has a question now, because we still have a few minutes. If somebody has a question now, we'd be happy to have them ask Sivan by sort of unmuting the microphone. So does somebody have a question that they would like to ask Sivan directly. Please raise your hand and we will activate your microphone. If somebody has a question now. Yeah, there's one, there's one in the chat. Yes. What do you think the impact of this research to, to government thing. To government planning. Thank you so much for that question. The main approach of the research is for technical advancement methods that can do rapid refugee assessment of a time. I tested the use of indices with minimal ground with minimal data, you can say. And so in that regard, it's helpful to know that this method is not effective and to also present why and the challenges that I first. That is in the technology perspective. In terms of government planning. The main output, which is helpful is the index anomaly, which showed areas that had potential human influence. This would need a team to go on ground to confirm these areas. This is changed because of human influence or other natural processes and in the event that they are confirmed to human influence. Then the government leadership and the administration in the local areas can put in plan measures to, to curb this or to protect the environment from further degradation. Yeah, so mainly that's what I can say. But also to throw something out there is that we wanted UNHCR to have an approach to conduct rapid assessments without having to go to the field. And we wanted to see if there's a platform that they can do that. So it also in part helps to, you know, throw some ideas to UNHCR to see if they can, if they can advance this further. Thanks so much, Sivan. I think with that we have truly come to the end of the proceedings. Thanks for your presentations, Sivan and for for participating in the discussions and thanks to all of you the audience for turning up in good number and for your questions and comments. The recording of the session will be available by tomorrow on the water channel, which is let me type it out over here. www.thewaterchannel.tv. It's not a clickable link. Let me try and put a clickable link over there. Long could you please do that. The recording will also be available on the AAT website and on AAT's YouTube channel. Let me paste the link to that over here. Yes, this is a clickable link. So yeah, with that, I would like to say goodbye. Thanks again all of you and see you at the next webinar. Alright, thank you. Thank you.