Climate Change Projections and Droughts

0:00 Speaker 1: First, let us look at natural climate variability and change. They result from natural and non-anthropogenic causes. And these causes include ocean atmosphere and land interaction, variability in incoming solar radiation, volcanic activity, and Earth's orbital variation. These causes make climate vary on a variety of time scales. and some examples are a linear southern oscillation with 2 to 6 year period or pacific decadal oscillation with say 10 to 12 year oscillation. Here you can see this blue curve which is schematically showing natural climate variability.

0:46 Speaker 1: One thing to notice here is that there is variability but it occurs around the same mean, so mean is not changing. If you look at this black curve which is observed you can see that there is variability, but also the mean is changing and there is a trend. And this is because of anthropogenic climate change. So these causes are increasing greenhouse gases and aerosols in the atmosphere due to fossil fuel burning, industrial waste and deforestation. So that's what we mean by natural variability where mean is not changing, whereas climate change is where there is trend and climate mean also is changing.

1:33 Speaker 1: Global climate models, they are designed to capture both natural and anthropogenic climate variability and change. Most global models are grid models shown here. They have horizontal grids as latitude, longitude, and in vertical, they are certified in height or pressure. And at each grid point, Detailed physical processes are models. They represent interactions among different components. So these are complex models. They require supercomputers and big data management. Also, it is important to keep in mind that climate models are often designed to capture global signals.

2:16 Speaker 1: So at regional scale, there may be some biases, so they have to be examined. Observations are an important part of model development. for initial and boundary conditions and for validation and bias correction. So although models provide a lot of information and predictive capability, they do depend on observations for initial and boundary conditions and for validation. There are more than 100 Earth system models or ECMs from more than 50 modeling institutions. Modeling institutions are constantly improving their models, such as making them higher resolution, improved physics and chemistry, and adding more processes.

3:04 Speaker 1: Leading modeling groups have coordinated simulations of climate change scenarios with the auspices of the Coupled Model Intercomparison Project, or CIMIP, which is currently in Phase 6. And CMAPE program provides important diagnostic and evaluation information for each ECM. Since all ECMs have different way of treating physical and chemical processes, have different resolutions, their outputs also vary. And so the goal of CMAPE is to compare them and come up with and evaluate how they compare with each other.

3:45 Speaker 1: Specifically, let's look at CMIC 6 and climate change projections. So, the CMIC program is organized by the World Climate Research Program, currently on phase 6 as we saw, with the goal to understand past, current and future climate change occurring in response to natural and anthropogenic causes using multi-modal framework. The program was started in 1995 and has evolved in which multiple models participate with the goal of assessing model performance during a historical period that is between 1860 and 2014 and produce future climate projections.

4:30 Speaker 1: All models use common experiments and forcing data and a major goal is to quantify and understand the spread found among the model suggestions. The schematic here shows basic CMIT program and then the CMIT 6 components here with response to forcing, systematic biases among models, and then variability, predictability, and future scenarios. And all these are the topics related to that. So, now let's look at what is climate prediction and projection and what's the difference. Here is the report that provides detailed information but basically climate predictions are estimates of future natural conditions.

5:25 Speaker 1: So, for example, In the last session, we saw S2S prediction system, which is climate prediction system. It started with observed initial conditions and then model was integrated in time. Whereas climate projections are estimates of future climate under various assumptions of future human-related activities in socio-economic and technical development. So, there are assumptions added to the natural conditions. So climate projections have less certainty as they are based on assumptions which may not evolve as expected.

6:03 Speaker 1: So what does modeling and projecting human-driven climate change involve? So the foundation of climate change projection comes from simulations that capture human influence on climate systems. It requires radiation physics, atmospheric dynamics, chemistry, oceans, biosphere, cryosphere and human driven shifts in emissions and land use. So what that means is a model has to have natural variability in it that includes all these processes that shown here that changes atmospheric composition as well as land surface.

6:42 Speaker 1: But on top of that, human driven factors have to be included such as industries, urbanization, land use, land change and transportation, etc. include human-driven influence in the models. IPCC, which is Intergovernmental Panel on Climate Change, they have come up with scenarios and storylines. Scenarios is something that describes a future world through a plausible and internally consistent set of assumptions, potentially including greenhouse gas and aerosol emissions, land use change, socio-economic development and technological change.

7:30 Speaker 1: IPCC notes that scenarios are neither predictions nor forecasts but are used to provide a view of the implications of development and actions and how to include them in the model. Storyline is a way of making sense of a situation or a series of events through the construction of a set of explanatory elements Usually it is built on logical or common reasoning. Storylines can have societal and physical elements, such as the physical implications of a given amount of global warming, potential impact of increasing population, the ramifications of a given policy or new financing being implemented.

8:17 Speaker 1: So all these factors, they can be used to explain a storyline or develop a storyline.

8:26 Speaker 1: defining some scenarios here. SRES, or Special Report on Emission Scenarios, they were developed by IPCC way early in 2000, a third assessment report. Then there were representative concentration pathways, or RCPs, and now they have SSP-RCP, which are shared socioeconomic pathways and representative concentration pathways. Also, there are adaptation strategies which define global warming levels and nationally determine contributions. So if you look at the figure here from Assessment Report 6, these are surface temperature anomalies from 1750 to today and then projections up to 2100.

9:15 Speaker 1: The numbers are shown here. Darker red is much more global warming, so higher anomalies. Anomalies are calculated with respect to 1850 to 1900 period. That's the period used for calculating mean temperature. So you can see that as we get closer to this era, the current, we are in the warming region here. But what we want to see is that different SSPs, which are shared socio-economic pathways, how they work in adapting so that CO2 emissions are either net zero, they're half, or they're peaking. Depending on that, temperature anomalies with this sort of global warming will result by 2100.

10:11 Speaker 1: So low, very low SSP, one to 1.9 here, that CO2 is halved very soon and then net zero just after 2050, then the anomalies are lower. If you go above, then if you go here in the middle, then half is by late in the 21st century. And then if you go up above, there is really no net zero. You don't see anywhere. So this is very high. So these are the scenarios where you have much more global warming. Global climate models, they use scenarios in which both SFP and RCP are used. So, SFPs, they primarily describe social economic development and RCPs describe greenhouse gas concentration.

11:09 Speaker 1: These can be related and mitigation can create unique combinations. Each SSP leads to a given RCP without mitigation, but mitigation can lower RCPs. So initially in earlier IPCC modeling comparison and CMIP comparison, RCPs were used, different emission scenarios were used. Now, shared socioeconomic pathways are used, which include mitigation strategy, So for example, this is by 2100 forcing level in watts per meter square, higher the forcing, more global warming. And so tier one and tier two are just different experiments.

11:54 Speaker 1: These are SSP1 to SSP5 and these are RCPs. So say here in this region, SSP1 is such that mitigation is promoting sustainability and there is low RCPs. As you go up here, there is really fossil fuel development, so there is really not much mitigation going on. Nothing has changed. And then the higher RCPs here, so more emissions, and that results in more warming or more forcing by 2100. So with

12:32 Speaker 1: that background, we'll move on to overview of next GDDP, which is a downscaling project. downscaling of global climate model output. Now, downscaling is an analysis designed to bring global model information to final resolution, including representation of final scale features such as land use, mountains, and coastlines, and physical processes associated with final resolution dynamics. Downscaling of global climate model output is necessary for assessing climate impacts on local and regional scale.

13:12 Speaker 1: So most models included in CMIP6, they have low resolution, lower than one degree latitude, longitude. And so to resolve local scale phenomena such as tropical storms, hails and tornadoes, dust storms, monsoons, heat extremes and droughts, To look at regional or local scale, these data are low resolution, and so downscaling helps in looking at finer scales. So with that objective of downscaling data, NextGDDP was initiated in 2015, initially with CMIP5 project, and now it is updated with CMIP6 with newer climate projections.

13:58 Speaker 1: There are 35 CMIP6 global models included in the downscaling effort. and a number of geophysical parameters are downscaled from most of these models. So here you can see list of global climate models and these are parameters available from these models. These models are all acronyms and of course these parameters are humidity, relative humidity, specific humidity, precipitation, temperature, temperature maximum and minimum, and these models you can find description in this paper here. Statistical downscaling methodology used in NextGDDP uses a daily variant of the monthly bias correction special desegregation or BCSD method.

14:53 Speaker 1: Observational data and reanalysis data at quarter degree scale are used for bias correction. After the bias correction, special desegregation merges observed historical climatology at each time step with changes from GCM simulations to produce final downscale projects. And here is that BCSP code which is used to generate downscale output which is available in GitHub. So the link is provided here, so you can also download and look at this code for bias correction and specialty segregation. The downscaled GDP data are available for the following climate projection scenarios.

15:43 Speaker 1: Historical runs, they provide data for 1850 to 2014. And for these shared socioeconomic pathways or SSPs, ranging from 1.26 to 5.85. The data are available at daily timescale with quarter degree latitude, longitude, special resolution, and they are quality checked. Now, looking at these different projections, we talked about these SSPs earlier, but SSP 1 has low challenges for mitigation. That means it's resource efficient, And for adaptation also, there's rapid development in technology, but perhaps green technology.

16:28 Speaker 1: And as you go down, SSP5 is the high challenges for mitigation and low for adaptation. So there is rapid development, but it is dependent on fossil fuels. So bottom line is that here, the resulting global warming would be lower about 1.8 degrees Celsius. There's a range, but that would be the maximum. And here it would be greater than five, about 5.6 degrees Celsius or so warming by 2100. So now how to select climate change projections? So a number of factors have to be considered as listed here.

17:10 Speaker 1: First of all, understand the differing needs of mitigation, adaptation and risk management applications. then recognize the main components and distinguishing factors of climate protection sets, recognize the benefits and trade-offs of different climate protection sets and versions, and consider selection of the best climate protection set for various application needs. Just to note here, RSAT does not recommend any particular model or scenario for climate change, but there's a general guideline that we provide to access and analyze TCM data to facilitate your efforts engaging climate change in the region of your own interest.

17:54 Speaker 1: And here is the asset training I mentioned earlier as information about selecting climate change projection. There are additional factors to be considered when selecting climate change projections. It also depends on application and decision-making needs of a particular region of interest. So for a number of applications shown here, say from human health, agricultural, all the way to energy resources, different application-ready parameters are required, listed here. And local and regional physical and socioeconomic conditions also matter in decision-making.

18:37 Speaker 1: So interacting with climate scientists in that region, they can help users in deciding which model projection, which model and which model projections are best for use, including spatial and temporal resolution, what kind of bias corrections are done. So these factors can also help in deciding which climate projection to choose. Once you choose climate change projection, then you can download or access the data. from Amazon Web Service. That's where they are available. So here is the information about resources on AWS.

19:20 Speaker 1: The data are in NetCDF format. Resource type is S3 bucket, and this is the Amazon resource name. Licensing information is given here in this document. The link is provided. An important thing to know here is that

19:42 Speaker 1: As of September 2022, all next GDDP-CMIP6 outputs are made available under a blanket Creative Common Zero license. These data are managed by NASA, and you can contact support at nccs.nasa.gov for more information or if you have any questions. Here is how to cite this data if you download and use. And there are also usage examples, tools, and applications at this next GDDP CMIP6 dashboard by NASA. And there are publications with more information. There is a subset of data available from AWS. And there are two models, GFDL, which is a NOAA global climate model, and GIS, that's a NASA global climate model.

20:39 Speaker 1: Two scenarios are available, SSP245 and SSP585. And the data are available in cloud-optimized GeoTIFF format. The data are divided into three subproducts. Daily, so source data with daily timestamps. Monthly, so these are monthly aggregation across both models. And crossing, it's a single data product for each SSP indicating the number of years until the average daily temperature crosses two degree limit from the Paris Agreement at each position on Earth. So this subset also can be used and available from AWS.

21:20 Speaker 1: The information can be found here. These data are also available from Google Earth Engine. So what we are going to do is next have a demonstration of how to access these data easily in GEE and also analyze to look at cloud conditions evolving over next several decades. So we'll start with the demonstration. Let us start with this demonstration of analysis of NextGDP climate change protection data using Google Earth Engine. As you know from part one, Google Earth Engine is a cloud-based platform and uses JavaScript for coding and for data analysis.

22:15 Speaker 1: So in this demonstration, we have provided a JavaScript code and we are going to walk through it block by block to see how to access and analyze next-generation data.

22:29 Speaker 1: The objective is that we will examine near-surface temperatures and precipitation as drought indicators, as we find in GDP-CNPC models. And we will select the data for the area of interest and see how temperatures and precipitation evolve over the 21st century. In the training web page, you will find the exercise for today, which is the hands-on exercise where you can use this code and examine GDDP data. So you can go there and click on the link right now and you will be able to access the code and follow along.

23:14 Speaker 1: You will need to have an account on Google Earth Engine to run this JavaScript code. To access NextGDDP data on GEE, we need to find out information about how to access it. And for that, we first go to earthengine.google.com, go to Datasets, and search for GDDP CMAP6. So this is next GDDP CMAP6. Once you click on that, you will find out how to access the data in the sense that this is the name of the image collection in Google Earth Engine, and this is the string that you will have to use to access the data.

24:11 Speaker 1: It also tells you what are the data set availability, who is the provider, and different tags are given here. More importantly, that is description of the data set. And the links are available for the paper. We saw that in our slides also. There are bands given here. So in bands, you will find all the parameters available in the GDDP models. Units of these parameters, minimum and maximum values, and description. So we want to look at surface temperature and precipitation. So TAS is the daily near surface air temperature.

24:57 Speaker 1: given in Kelvin, so we need to know that string name. Also, we need to know precipitation or PR, which is in kilograms per meter square per second. It's the daily mean precipitation. There are also other parameters such as relative humidity and specific humidity near surface. There are radiation components and surface winds also available. and then daily minimum and maximum temperature. For example, we are going to focus on TAS and PR, but you can also look at other parameters by choosing by these strings.

25:38 Speaker 1: Next, we see image properties. In image properties, you will find all the models available in this dataset. So GDDP, as we know, CMIP6 has about 35 models or so participating, and GDDP downscales this model data. The models which are included in this GEE data set are given here in different strings. Again, we saw the list of these models in presentation slide, and the reference given in there has description about each of these models. But what we are going to do is focus on one of these models and that is GISS or GISS.

26:25 Speaker 1: This is NASA Global Climate Model. So we need to know this string GISS E2 dash E2 dash 1 dash G. So we have picked that. There are other models too. GFTL is a NOAA model. Mirok is a Japanese model. Headley Center model, a UK model and you can get more information about it from the reference that we saw in the slides earlier. Also, another piece of information is about scenario. These are SSPs, Shared Socioeconomic Pathways that we talked about earlier. So in this dataset, historical data are available, historical model runs are available.

27:10 Speaker 1: SSP 245 and SSP 585 are available. So this is the information we need to know before we can select and analyze GDDP data. So going back to our code now,

27:27 Speaker 1: the model outputs are available for two scenarios, SSP245 and 585, as we saw. And what that means is SSP245 is the middle pathway of future greenhouse emission. projecting radiative forcing of 4.5 watts per meter squared by 2100, whereas 585, as we saw earlier, has additional radiative forcing of 8.5 watts per meter squared by the year 2100. So this is the upper boundary of the range of scenarios we saw earlier. Now for this demonstration, what we've done is we've selected one model and we're going to select one scenario.

28:13 Speaker 1: So what we will see is how to select a model from GDDP model ensemble that we saw in the image properties, how to pick a scenario, how to select parameters, temperature and precipitation. What we will do is we collect daily global data for 2020 2050 and 2100. We'll look at these three years, just one is 30 year apart and an eight year apart. So we just to see how things are changing over the century. Then we'll clip the global data to the country of our interest and we'll define the country in a variable AOI later as we will see, or area of interest.

29:03 Speaker 1: We'll find annual mean of near surface temperature and precipitation for the three years that we have picked. Then we will map these parameters for AOI and we will also find differences. So how temperature differs between 50 and 20 and 2100 and 2020. So we'll map those as well. And finally, we will examine annual mean near surface temperature. What is the trend? How warming is occurring over the AOI we pick between 2020 and 2100. So that's the exercise. So now this segment talks about how to select a country.

29:53 Speaker 1: And for that, we've used the Food and Agricultural Organization FAO's global administrative unit layers or GAUL and Sean McCartney demonstrated this in part one as well, but we have imported this GAUL levels. So in here level zero is the country level shapefile. So all the countries in the world are included in here. If you go to level one, these are sub regions. What we have defined is this as country, so this is sub-region. So if you pick level zero, it will be one country you can pick. If you have level one, you can have states within the country, you can pick a particular state as well.

30:43 Speaker 1: For this demonstration, we will use level zero, and so the variable we will use is countries here. We have already imported this data, and this is the block that explains all that. Now this segment defines all the variables used in the subsequent code. When you go through the exercise, you can look at this for the definition of variables. What we are going to do is we'll define each variable as we come across as we work through the code. So starting with our analysis now, for the demonstration purpose, as an example, I've picked country of Ethiopia.

31:27 Speaker 1: And the reason for that is that if you go to this World Atlas site, according to their projection, by 2050, a large part of Earth would have a really deficit of water and drought-like conditions. But the top 10 countries, most of them in Africa and Asia, they have been identified as most affected countries by drought. like Morocco, Uganda, Somalia, Iran, Pakistan, China, Afghanistan, Eritrea, Sudan and Ethiopia. So based on this report, for this demonstration I'm focusing on Ethiopia. So here we are picking, defining the variable AOI and specifying the country of our interest which is Ethiopia in this case.

32:23 Speaker 1: When you want to choose country of your own interest, all you have to do is replace this name by the country of your interest. So this AOI will define that. We define a variable myAoi. What that does is save this country feature connections of gaul that we imported, takes that and filters AOI and assigns that shapefile for the country that we picked. This map.center object, it focuses the map on the region that we picked. So in this case, it's my AOI. And this is the zoom factor. And so one is the whole earth and 25 is the small segment.

33:17 Speaker 1: So we are picking zoom level of 5. And map.addLayer

33:25 Speaker 1: the shapefile to the map pane and when you run this code, it shows up here as an example. We did that already so you can see that Ethiopia shapefile is showing after that. So, this is what we have picked for the analysis. Alternatively, you can, if you have another region of your interest and you have a shapefile you can upload it to this code. So you go to this left window and go to assets, new, and here you can upload. So you can select shapefile or CSV file, which you already have on your computer for a specific region of your interest.

34:17 Speaker 1: You can upload that. And then you will have to import as AOI. that is very important. Or you can even go to this map pane and you can draw a polygon or a square in any area of your interest. I'm just showing an example. So, you create this geometry and go to this view symbol and you have to name this as arrive and then say okay to import it. We're not going to do that. We're going to delete that. So, we will see Ethiopia as our country of interest, but you can have a variety of ways of choosing your own region.

35:09 Speaker 1: Next, we start working with the GDDP data. First, we start with the Google Earth Engine image collection NASA slash GDDP CMYP6 that we saw earlier. From that, we are filtering the data for a particular model that is NASA GIFS. One scenario SSP245 and for the date range 2020, January 1st, 2020 to December 31st, 2100. So we define a variable data set. We start with the global daily image collection in which all the parameters are there, all the models are there, and all the scenarios are there. From that, we are filtering by date first, then by model width, and by scenario SSP245.

36:08 Speaker 1: So, dataset will have global data, but just one model, one scenario, and still it has multiple parameters. Here, we have a print statement for dataset, and it prints first 120 elements from the image collection. But that is, when you have a print statement here, you will see that it prints here in this console window. So for dataset, it's an image collection and it tells you what's in there. So this is the first day here. This is January 1st, 2020. then it has all the variables. So now what we are going to do is go to this segment.

37:00 Speaker 1: Now we are focusing on three years for the demonstration purpose. You can do this for many all the years but let's see we'll start with 2020, 2050 and 2021. So we'll select parameter CAS which is the surface at temperature as we saw earlier. And we're going to click that to the Ethiopia shape file. So we define a variable airT, surface air temperature, but take that data set image collection, which has all the parameters. From that, we just pick surface air temperature. And this is still global data.

37:42 Speaker 1: So now we clip it to our area of interest, that is Ethiopia. define a new variable, airTemperatureRegional, or airT underscore ridge, which takes that airT and clicks to myAOI. Again, the print statement is just for checking of the data set.

38:09 Speaker 1: So now we have all 2020 to 2100, all the years are there. By picking these three years, for each year, we want to make annual new temperature maps. So we'll start with year 2020, and start and end date, these statements, they just pick one year. So daily data is picked for one year, and a new variable, TYY, is defined, which takes that surface air temperature regional, Ethiopia that we created up here and filter it just for one year and in so that we filter data for one year and T2020 now we are finding mean so this reduce reducer mean takes all the daily data for 2020 and you find mean and so we have one map for the temperatures of T2020.

39:17 Speaker 1: So, this is how we find annual mean and then these are the spring statements again just to check that our calculations are correct. or not correct. And you can go through this in the console window to see exactly it is as you expect. So now we are repeating the same step for 2050 so at the end we have the 2050 and 2100 so we have T2100. So we have three years, annual mean temperature. Now we are looking at differences between temperature, surface air temperature. So here we define Tdif 2050 minus 2020 which takes annual mean surface air temperature for 2050 and subtract annual mean surface temperature for 2020.

40:18 Speaker 1: And this is similarly for 2100. So now we have differences in temperature with respect to 2020. So how far it can be in 2050 and then 2100. So we have three years of data, annual mean, and then two differences. And in the next segment, we are visualizing all these parameters. and we can.

40:50 Speaker 2: Specify this is the variable we have defined here just for visualization minimum and maximum temperature.

41:00 Speaker 1: Range in Kelvin and we have defined color palette with colors here and for the difference this is the palette we used minus 0.7 to 2 these were somewhat randomly or by looking at the map and we'll go through that procedure to do that. Then we plot, actual plotting is done here. So map.addLayer for the 2020, 50 and the 2100 and then these two differences.

41:35 Speaker 3: 2050 minus 2020 and So at the end of this whole segment, we have temperature analysis.

41:55 Speaker 1: The analysis for precipitation, we follow the same procedure as we did for surface air temperature. So we start with the dataset image collection with all the parameters and we select PR from that. And so we have global precipitation data just no other parameter, but PR for one model is and one scenario RCP245. Then we clip it to my AOI, which is Ethiopia. So now from global data, we have regional data of precipitation. Again, this is all year 2020 to 2100. Also note that here we are defining a constant, which is seconds per day.

42:52 Speaker 1: In the GDDP data set, precipitation is given in the units of kilograms per meter squared per second, which is equivalent to millimeters per second. And if you recall, in last week's S2S prediction QGIS analysis, we did unit conversion from kilograms per meter square per second to millimeters per day. And that's exactly what we're doing here. So we're taking the precipitation value and then multiplying by this constant. So now we are going to find annual mean rain rate for 2020, just as we did for temperature.

43:34 Speaker 1: So we'll start with the start date and end date, and we'll find image collection in which we have daily data just for 2020 for precipitation. And then this reducer mean that then provides mean for that year. So here we have PR 2020 annual mean and we're multiplying to convert the unit. And these are just the print statements like before just to check the data. We do exactly the same procedure and get annual mean rain rate for 2050 and annual mean rain rate for 2100. Similar to the surface air temperature, now we are finding differences in precipitation between 2050 and 2020 and 2100 and 2020.

44:26 Speaker 1: So these are the variables defined and the subtraction done here. using the annual mean precipitation. Now, we are defining visualization parameters here for the maps. This is for the precipitation, and this is for the precipitation differences. And here, actual plotting is done in map.addLayer. And so we are plotting mean precipitation as well as the differences. And when we run, we get all these maps. So let's first look at the map that we have and then we can proceed further. That we have all the layers are generated.

45:19 Speaker 1: We start with air temperature. So this is 2020 surface air temperature. And if you click on this view, you will be able to see the colors here and the range. This is what we had specified. but what we can do is go to custom and stretch 100% and then it will come back with exact values and now you can apply it so this is the actual range now now if you want to make this permanent you can go back to the code and change median max values But we are doing this just for visualization. So let's not do that.

46:13 Speaker 1: We'll just check the numbers here. Here we have a hot, quite warm region where it's yellow and red and green and the blue and purple and dark blue, they are the colder parts. So this is more 280ish and this is about 300. So that's how it is. You can look at air temperature for 2050 and 2100 and then let's look at differences as well. We can do the same thing. Let's go to 2100 and do the same thing and apply. It should come back. It's arranged. It's working on it. So, you can see that the values have gone higher.

47:19 Speaker 1: In 2020, it was 280 to 305. It's 282 to 307. And clearly, we can see that larger parts of the country, they are warmer now in 2100 compared to 2020. And you can see that in the, let's see, difference. This is 2100 minus 2022. And here, we can, again, do the same thing. Apply. You should be able to see the numbers.

48:03 Speaker 1: So here everywhere differences are positive implying that 2100 has warmer temperature everywhere compared to 2020. So obviously there is warming going on. So this is temperature map and let's quickly look at precipitation before we look at long-term trends in temperatures. So looking at the precipitation now, this is precipitation in 2020. And we can check the numbers here.

48:55 Speaker 1: So it's 0.1624 millimeters per day. in the western part of the country there is more rain than in the eastern and southern part. That's the pattern we see and if we look at 2021 precipitation also shows pretty much the same pattern. If you look at the difference between 2100 and 2020 precipitation, there is interesting pattern. In case of temperature, there was uniformly increasing temperature in 2100 But in precipitation, it looks like there are regional patterns. There are places where 2100 has more rainfall than 2020.

49:48 Speaker 1: And then there are places where 2100 has less. So local processes may be playing a role. We are actually just looking at one model and one protection. Ideally, we should be looking at many models and comparing the patterns and magnitude and taking ensemble mean of all models perhaps. So we'll conclude our spatial analysis here and last part we will be looking at is temperature trend in the country between 2020 and 2100. From our temperature analysis, we do see that temperatures are getting warmer between 2020 and 2100 over the entire country of Ethiopia.

50:43 Speaker 1: And so here we are looking at temperature trend between 2020 and 2100. So we first find annual mean for each year. And then we find mean over Ethiopia. So we have annual mean, country mean temperature for 81 years. And we're going to plot that as a time series. So in this segment, we are just coming up with a list of years. And in the print statement, you can check. that this is 2020, 21, 22. So there are 81 years from 2020 to 2100. And so for that list now, we are finding annual mean from this air T. So this is global surface air temperature we calculated or extracted above from the GDP collection.

51:49 Speaker 1: So we have this T annual image collection, which is temporal mean for each year in this list date. So for each year, we have a map over Ethiopia with annual mean temperature. It's very similar to what we did for individual years like 2020 and 2050, 2100. This will be for each year now between 2020 and 2021.

52:22 Speaker 1: Once we have that, here we are now taking special mean over the entire country. So we'll start with the annual image collection and we average, we find mean over my AOI, that is Ethiopia. Scale here is 25,000. So this is in meter. Since the resolution of GDDP data is 25 kilometer, The scale here is 25,000 meters equivalent to 25 kilometers. And once we find this special mean, then we have for each year, we have one number that is country average temperature, annual average temperature. And so that's what we are plotting here.

53:13 Speaker 1: So when we print time series, we get a chart here in console. And let's move down. And here is the arrow to make it big. So this is, you know, from model GIST. So GIST temperature 296 to 301 Kelvin. Here are the years. And clearly you can see that there is a trend between 2020 and 2100. You do see variability, but there is clearly a trend. You can also download this as CSV file and do further analysis in Excel if you like. Suppose you want to fit a line or actually want to find slope and intercept, you can do that too.

54:09 Speaker 1: Finally, this segment shows an example of how to export an image or data to Google Drive. An example we use tp. 2050, that is the annual mean surface air temperature that we have calculated above. For exporting this to drive, you have to use these lines to prepare the data. And when you run the script, in this window on the right-hand side, you will see option Start. When you click on that, you will see this unsubmitted task that corresponds to the data prepared here. You click Run here, check everything for initiating image export.

54:59 Speaker 1: You can keep the file for that as geotiff if you want to do further spatial analysis with GIS or other software, and then run it. Click here to run, and then the export to Google Drive starts.

55:15 Speaker 1: So depending on the size of the image and data, it may take a while to export it to Google Drive, but this is the way to do it. This is just an example. So this concludes our demonstration for today. And we just want to note that every effort is made to ensure the code is free of errors, but there is no warranty for the maps and their features. to be either spatially or temporally accurate or fit for a particular use. This code is provided without any warranty of any kind whatsoever, either explicit or implied.

55:54 Speaker 1: So with that, we summarize today's session. We learned that climate is the long-term mean of characteristics of geocache physical parameters. By long term, it means 30 years or longer. Climate variability and change, they are due to both natural and anthropogenic causes. Natural causes include variations in solar radiation, Earth's orbital variations, and ocean atmosphere land interactions. And anthropogenic causes include increase in greenhouse gases and aerosols due to industries, transportation, and land use change.

56:33 Speaker 1: Global climate models capture both natural and anthropogenic climate variability and change. And CMIP5 allows intercomparison of model outputs to understand differences among them. CMIP is now in Phase 6. Next, GDDP uses statistical methodology to downscale outputs for five different climate projections of several CMIP6 models. at quarter-degree latitude, longitude, special resolution, and daily temporal resolution. NextGDDP data are available from Amazon Web Service, and they're also available from GEE, as we saw in the demonstration.

57:15 Speaker 1: As an example, NextGDDP surface air temperature and precipitation changes over Ethiopia were examined between 2020 and 2100 in GEE. We used one model, NASA GIS, and one projection, SSP245. Our

57:35 Speaker 1: last training session will be on Thursday, on 1st of August. This will be part four, and that will be on overview of regional use to facilitate drought monitoring and assessment. We will have demonstration of Navajo Nation Drought Security Evaluation Tool, or DSAT, sustainable forest management and information system or SFMA. Both these tools are regional tools and with some efforts can be customized for other regions. There will be homework posted on our website on 1st of August. An answer must be submitted via the form.

58:20 Speaker 1: The homework will be due on 15th of August. The certificate of completion will be awarded to those who attend all four live webinars and complete the homework assignment by the deadline. Then the certificate will be awarded to you email approximately two months after the training is over. With that, we have some time now for you to work on the exercise. You can take the Google Earth Engine course, link is given to you and you have instruction about where to change

59:01 Speaker 1: the region of your interest and then run the course and you will get images and time series just as we saw for Ethiopia. So, we will stay online and you can work on the exercise. So, question one is, How do you calculate the temperature in 1750? So there are paleoclimate data and past data reconstruction from proxies such as tree rings, sometimes air bubbles caught in ice layers. These are used for temperature, they're calibrated with temperature. So that's paleoclimate proxies are used. Question two.

59:59 Speaker 1: Just one moment, please. I want to make sure. Okay. So question two, is the glossary available for these abbreviations and terminologies? So we've tried to define most acronyms, but for IPCC, specifically climate-related glossary, you can use this IPCC document that the link is provided.

1:00:47 Speaker 1: Question three, slide 22, next GDPC matrix. Can this data be used to calculate wet bulb temperature? I know there are a lot of uncertainties with the results as the same model. So the answer is yes, you can estimate wet bulb temperature because in GDDP, you have both like relative humidity, specific humidity out there and winds are there as well. So you should be able to calculate wet bulb temperature. Question four, is climate change impact uniform across the globe or does it its intensity varies from the North Pole to the South Pole.

1:01:35 Speaker 1: So yes, it varies almost everywhere. It depends on the local processes and regional processes. So that's why downscale models are preferable than global models with higher special resolution. But yes, climate change impact is not uniform everywhere, no. Question five, is downscaling only done in the temporal component and not in special one? Will there be downscaling below one kilometer square? So currently, I'm not aware of anything where downscaling is at that high resolution, like one kilometer square or less.

1:02:20 Speaker 1: If you look at CMIP6 models, most of them have low resolution. It's like 100 kilometer to 250 kilometer resolution. So GDDP provides data at 25 kilometers square. So that's at least a higher resolution than most global climate models have. If you want very high resolution downscaling, you have to have observations at that scale too, to first of all develop downscaling coefficients and also for bias correction and everything. So I believe that's a very high resolution. What people do is that they take downscale data and then use as forcing to say cloud resolving or very, very high resolution model, depending on the application.

1:03:14 Speaker 1: Question six, doesn't next CMBip6 also use CDF to bias correct? Why are both correction methods used and what is the specific purpose of each? So I don't think I can answer this in detail. We'll try to find out. But what I know is that CMIP6 has all the models. They do use bias correction in CDF way. But for GDDP, which is your downscaling at higher resolution, so further bias correction has to be done. at that resolution. And so, yes, BCFP is used in GDDP.

1:04:04 Speaker 1: The next GDDP symmetric model can be used worldwide in latitude range or continent? These are global models. So, yes, they can be used worldwide. This question 7. Question eight, is the result code just showing the map or can we get a CSV file for daily temperature or precipitation in these years of scenarios? Yes, you can definitely download CSV files. If you are looking at a big region and you are looking at maps, daily maps, maybe there'll be huge files for CSV. Maybe you can download as

1:04:47 Speaker 1: on your Google Drive and convert them to CSV if you like. Time series, as you saw, there is a way you can download CSV file. Question nine, are there any examples or case studies whereby this analysis was used for decision making by an organization or a country? So this is really a very good question. We do not have specific answer for that this country or this organization has used for decision making. These projections are used as a guideline by most organizations and in countries. Now, we will find out if we have any specific examples what you do here is that, say, insurance companies, when real estate companies, they do look at these projections when they are thinking of future, where they're going to build, say, property or buildings, what kind of projections are there.

1:06:05 Speaker 1: I don't know if that's the only factor for decision-making, but these companies do look at how climate might change in a few years at a given location. But I do not have any specific organization or country that I can quote here. Question 10, is the GEE data bias corrected or downscaled for further analysis? So what you mean, GDDP data or the one data are in GEE, but it's the same. So yes. GDP data, they are bias corrected. I would say that if you have observations in the region of your interest, you can compare with observations and see how good which model works in your region.

1:06:57 Speaker 1: You can see other biases left or. So question 11, can you provide training material to download the GRACE data from the new platform, downscale the data at a quarter degree, and compute groundwater storage anomalies. So I will provide you a link of GLDAS model. I'm going to make a note here. They assimilate GRACE data, and then the model has higher resolution than the original GRACE data. I'll be sure to post this answer once I have the information.

1:07:45 Speaker 1: This is question 12. Can you provide the code for Celsius degree, please? So yes, I think we can do that. We can just add one line, subtract 273 from everywhere. I think you can modify the code. When you are looking at the differences, they are equivalent to Celsius. But when you're looking at temperatures in Kelvin, we can perhaps add a line of gold there. Question 13, how can we get the final product of the precipitation scenario in net CDF? So you can go to the original data set. And from Amazon Web Service, they are in NetCDF format.

1:08:37 Speaker 1: If you are going to GEE, best thing is to save as GeoTIFF and then convert to NetCDF using Python or GDAL. I'm not sure whether you will be able to download next PDF from GEE. I will have to check on that. Question 14, how is data accuracy ensured in Earth Engine scenario modeling? Or are calibration procedures implemented in Earth Engine to enhance the reliability of scenario output? So Earth Engine is actually not adding anything, it's just taking the data and making it available through the cloud platform.

1:09:23 Speaker 1: So whatever the original data are, they are in Earth Engine.

1:09:29 Speaker 1: So there is no additional calibration procedures used in GE.

1:09:36 Speaker 1: Question 15, how can I apply the code to calculate environmental variables such as NDVI, albedo, and evapotranspiration. So I am not sure whether this is specifically related to modeling or you're asking in general. We had Sean McCartney demonstrated in session one. So code is available for NDVI calculations. albedo and evapotranspiration, these products are available. So I think we need some more specific question here about calculating environmental variables.

1:10:32 Speaker 1: Question 16, are there any training resources available on soil moisture prediction estimation Yeah, so I think our coordinators have provided links to RSAT soil moisture trainings. Question seven, could you tell me how break number index in JavaScript, I mean, after line one and line two, the number of line will restart again. How break number... I'm not sure I understand the question. We'll restart again. I think I'm not familiar with that. And we'll check with people who are experts in JavaScript.

1:11:40 Speaker 1: Question 18, can we select more than one model and get an ensemble to get more accurate results? Yes, definitely. I think ensemble mean has less inaccuracy. But as I explained earlier, it's also important to look at more models. Individually also, maybe for some region, a particular model may have less bias or can be used. But yes, ensemble mean is that's what people use mostly. So question 19, can you explain why the two bias correction methods, both CDF and BCST are used? I think we did go through that before.

1:12:28 Speaker 1: CDF, CMIP6 models, they use CDF. But when you are downscaling, then again, you have to consider biases because just to see you how, that's why BCSD is used. Question 20, how do you downscale the CMIP6 data using ground data? So that is a whole, that's not a simple answer. We can provide an area of research and people use different methodology, using ground data to do downscaling. And that BCSD, if you look at the code where you are doing special desegregation, so you can look into that, but it's a complex, there are many, many methods used for downscaling.

1:13:34 Speaker 1: Question 21, these models have some area preferences or are they good for every part of the world? They have area preferences in the sense that some model works better in some area, some other model works better in some other area. That is possible.

1:14:05 Speaker 1: And so they have tried to correct biases and make them as best as they can, depending on a lot of factors. But whether they are good uniformly everywhere, that is in question that has to be checked. Okay, question 22. It's pretty much the same as before, GDDP in GEE. Is that based on both BCST and CDF? So GEE, it just takes GDDP data and puts in their own cloud platform. There is no further analysis done on GEE for bias correction. Question 23, how can I unite two countries that share a border on GEE?

1:15:01 Speaker 1: You can draw a polygon. That's one way to do it, covering both the countries. Or you can define a shapefile where you can merge shapefile, two country shapefile and use as one.

1:15:28 Speaker 1: Question 24, how can I use this code to compute the SPEI index? So this code cannot be used as is to calculate SPEI index. You have to add that segment.

1:16:03 Speaker 1: Question 25, how do you run the code using all available models? So you have to go back and extract data from each model with the string. There is a way to just get all the models. That means if you don't specify that model, then all the models are used.