NASA ARSET Applications in Urban Heat Island Mapping

0:02 Speaker 1: Welcome back to our rset training series Introduction to Thermal Remote Sensing and Applications in Urban Heat Island Mapping. Today is part two of our training series Applications in Urban Heat Island Mapping. My name is Savannah Cooley. I'm a researcher at NASA AMES Research Center with the Bay Area Environmental Research Institute in Mountain View, California, and I'm a trainer with the URSET Ecological Conservation Team. We have two instructors for part two, myself and our guest instructor, Glenn Holly, who is a research scientist in the Earth Science Section of NASA Jet Propulsion Laboratory.

0:47 Speaker 1: For Part two, we have two objectives. First, participants will be able to filter and visualize ecostress land surface temperature data using provided our based data processing workflows. Second, participants will be able to downscale native seventy meter ecostress LST data to a fine ten meter spatial resolution using a random forest machine learning model implemented on an interactive Google Earth Engine interface to analyze neighborhood level urban heat patterns. If you have questions during today's training, you can put them in the Q and a box within WebEx at any time.

1:35 Speaker 1: We encourage you to put your questions into the Q and A chat as they arise. The earlier you share your questions in the chat, the more likely it is that we will be able to respond to them in real time during the Q and A session. If there are questions we do not have time to address during the live Q and A session, we will make sure to collect these and post the written answers to the training page within our week after the training. In other words, we will try to address as many questions as we can today.

2:06 Speaker 1: All of these questions and answers will be posted, along with the additional questions and written answers to the training page within a week.

2:13 Speaker 3: After the training.

2:14 Speaker 1: Before we delve into our next section, let's review some of what we learned in part one. Remember, satellite sensors measure radiance and convert it to brightness temperature, which is defined as a temperature A perfect black body would have to be to emit the radiance we observe. The problem is that real surfaces aren't perfect black bodies. They have emissivities that are less than one. If we assume everything has an emissivity of one, will systematically underestimate surface temperature. Let's do a knowledge check with a numerical example.

2:56 Speaker 1: What is the radiometric temperature of a sand desert surface with amissivity equal to point nine and kinetic temperature equal to three hundred and twenty kelvin. Take a minute or two to work this out on your own before I walk through the answer. With a sandy desert surface having an emissivity of zero point nine and a kinetic temperature at three hundred and twenty Calvin, we can calculate the radiometric temperature as follows. We would raise the emissivity to the power of zero point five as shown here, multiply that by three hundred and twenty Calvin, and then the result is three hundred and eleven point seven Calvin, which corresponds to answer D here in the set of answers.

4:09 Speaker 1: This implies that there is about an e Calvin underestimate of the surface temperature if amissivity is ignored. The key takeaway is that the lower the emissivity, the larger the gap between what a sensor measures and the service is true temperature. This is why accurate amissivity is so important and why we will spend time delving into the topic in this training.

4:39 Speaker 1: With that, I will turn it over to Glynn to tell us about Ecostress land service temperature data.

4:51 Speaker 2: Okay, good morning everyone. My name is Glenn Hally.

4:55 Speaker 3: I'm a research scientist at NASA's Jack Propulsion Lab. We have spent the past twenty also specializing in thermal infrared remote sensing and also spectroscopy applications and science. Today, I'll be walking through how we sharpen also sometimes called dancescale Ecostress land surface temperature from its native seventy meter resolution dwance ten meter resolution using a random forest machine learning approach, and we'll be doing this within Google Earth Engine API. This technique opens up a whole new range of applications for therm or remote sensing, specifically in urban settings, and by the end of the session you should be able to run the pipeline by yourself and create your own high resolution maps over any city in the US using data since twenty eighteen when Ecostress first started taking observations.

5:50 Speaker 3: Okay, so I'll be first going over some introduction on Ecostress, followed by urban heat islands, and then we'll go into some of the theory of the downscaling. So Ukostress was launched in twenty eighteen. It's stocked on the Japanese Experimental Module on the International Space Station and it's been delivering film I've read observation since about August of twenty eighteen. We're currently funded through about twenty twenty nine, which is really great considering that the initial mission was only supposed to last about one year, but we've been extended due to the credible success that we've had in science and applications from the data from the instrument.

6:35 Speaker 3: The native pixel resolution size is seventy x seventy meter resolution and we acquire data on average three days over most parts of the US and also the globe. And also one unique thing about Ecostress and then we'll coverver the next few slides, is that we can take these observations over the entire diurnal cycle. We also have spires five spectral bands in the filmal infrared, which is also unique amongst current spaceborn thermal instruments. So moving on to some of the actual characteristics of the instrument, what I've shown here is an actual example snapshot of what we call a scene or a granule that we observe from the space station.

7:19 Speaker 3: Each of these snapshots. Think of a camera snapshot is around four hundred by four hundred kilometers wide, so really wide snapshot, and within that we have seventy meter resolution pixels, which is about the size of a small football field. And this is currently the highest resolution thermal data from space now with the precessing orbit of the ISS meaning it doesn't come over the same place on Earth every single day like Lancet and ASTA for example. This image shows the number of observations we get over a three month period over the dinal cycle, which is shown in UTC time on the left y axis.

7:57 Speaker 3: Yet so you can see that echo Stress in blue captures a lot more observations than both lances at eight and nine and also throughout the dininal cycle, whereas Lansad only captures at a single time around the early morning overpass every day. The five thermal bands, which you've probably learned about in Savannah's tutorials are located between eight and twelve micron range. This is the classic long wave thermal infrared region where we can retrieve the surface temperature.

8:26 Speaker 2: From and these five bands.

8:29 Speaker 3: The more typically the more bands we have in this region, the more accurately we can derive the surface temperature and also the spectral emissivity from the thermal infrared measurement. So just some quick latest news from Ecostress. Currently, the Collection two products are available from NASA, but we are delivering Collection three science products, specifically the Level two Land Service Temperature product as an imminent release of the alp DC, and there's a number of improvements in that product products.

9:00 Speaker 2: Specifically, we have a new sea surface temperature.

9:02 Speaker 3: Algorithm and a lot of improvements to the data quality and of the cloud masking. Also very exciting for us this year is that, and I'll be showing you that. Of course, the demo in Google Earth Engine is that the entire Ecostress Land Surface Temperature archive is now being ingested into Google Earth Engine over the entire continental US. So this is fantastic news for us and opens up a lot of data and science and applications that we can use with other data within Google Earth Engine. We have more than one thousand per review publications so far, and since earlier this year, we've acquired more than six hundred and six thousand of these four hundred buffour hundred kilometers snapshots I showed you earlier.

9:43 Speaker 3: We also have a new applications page referenced here. You can see that we support a diverse discipline of different science applications. The one of course we'll be focusing on today is urban heat and urban resilience. So what makes ecostress unique. As I mentioned earlier, most thermal sensors in space have a sunsynchronous orbit, meaning that they cross overhead in the mid morning hours, in this case for LANSAT and Assets around ten or eleven AM.

10:16 Speaker 2: And you can see in this Los Angeles temperature.

10:18 Speaker 3: Profile that I've shown that the morning overpass, which is this white line going across the screen, typically misses the warmest times of the day and the hottest times of day in the peak afternoon hours. And this is critical, critical for urban heat mapping because you will miss the maximum temperatures during a typical urban heat urban heat extreme event. If we look at ecostress, we sample across the entire full dimnal cycle and seasonal cycle, which is exactly what makes it uniquely valuable for heat stress and urban applications.

10:56 Speaker 3: So then moving on to urban heat items, just I'll give some background and ourb and here is what we've done so far with ecostress and then we'll move on to the fun part, which is the downscaling.

11:08 Speaker 2: So why does this matter to us?

11:10 Speaker 3: Well, heat waves, a lot of people don't realize this, are the single biggest weather related killer in the United States. They more than double that of floods, torninators, and hurricanes combined. Even so, as you can see on the chart on the left, this is the No and National has its statistics chart. Heat fatalities are running by roughly two hundred deaths per year over the last decade, and a lot of people don't realize this because you can call them the silent killers. They don't cause as much physical destruction as other natural disasters, and so you don't care about them as much.

11:44 Speaker 3: And the downstream impacts go well beyond just mortality. For example, there's increased risk of heat stroke or hypothermia. These are the most common era visits and a lot of these are typically make a lot of sense. Increased electrical demand from the use of AC, higher wildfire risk, especially in the southern California region in the later parts of the year when we have Santa Ana events, and then of course degraded eir quality due to the increase in neo surface ozone, and these health effects also exacerbate, for example, other underlying conditions like cardiovascular and also mental health and cognitive function.

12:28 Speaker 2: Over the last few years we've spent a lot of time with a lot of very.

12:31 Speaker 3: Enthusiastic summer students at JPL where we've now mapped most of the extreme heat events with eucostress and cities across the world. So what you've see here are snapshots from these images in European and US cities from Prague to Loose all the way to Birmingham, Milan and Sharp and ecostress data have been featured in both BBC articles and also the European Space Agency typically features an article every year on how cities can adapt to heat waves, particularly during the nighttime, and it's also increasingly increasing by city planners, public health teams.

13:08 Speaker 2: Worldwide and a lot of the media within the US.

13:14 Speaker 3: So I'll go through some examples and why ecostress is unique in its ability to capture urban heat islands and urban heat in general. So this is a case study over the what it is called the Cherbis heat wave in July twenty twenty three broke wall kinds of records across Europe in terms of heat. This is over Athens in Greece. On the left, we have a daytime ecostress scene and the rider is a nighttime ecostress scene at eleven PM. And so in the daytime scene you can see an actual urban daytime cool island effect which most people don't realize is that during the daytime in certain cities, the urban area can actually be cool.

13:53 Speaker 3: Then the surrounding agricultural area that is driven by very hot, bare rural soils during the summertime we have generally less healthy vegetation. But at nighttime that picture flips. So the city retains its heat due to a high thermal heat capacity during the day, and then that stays significantly warmer than the surroundings at nighttime. And it's that nighttime urban heat island that is really drives up heat related illness in cities around the world. And it's only really eco stress that can capture these type of dynamics and changes in the urban heat island.

14:29 Speaker 2: This is another fantastic example from Macy Runkel at Chapman University who was with us last year looking at how urban heat island intensity changes of a full twenty four I period in Delhi India in this case, So each circle here represents the UHI, which is the difference in the rural and urban temperature every two hours throughout a typical summer cycle in Delhi. So you can see that the urban heat island is at maximum about ten pm at night, and then throughout the early morning hours it slowly diminishes as we go clockwise around the images here, and then during the daytime you can see that that flips around ten or twelve pm, that flips to a cool a cooling urban heat island effect, so the city is actually cooler than the surrounding dry urban rural landscape.

15:25 Speaker 2: Another real striking example here that got a ton of news media attention back in twenty twenty four was led by Ashley Agatet, also from Chapman and An a summer student of ours in twenty twenty four.

15:36 Speaker 3: Actually in twenty twenty three as well, there was suddenly a very sharp spike in hospitalizations from people in the Phoenix area from severe contact burns, and these were mostly due to several extreme heat events that occurred over those two years and actually resulted in almost fifty fatalities. For example, the gentleman here on the left actually lost his leg when he fell on the hot sidewalk walking out of a grocery store and wasn't able to get up because every time he put his hands on the sidewalk it'd be so hot that he wouldn't be able to push himself up, and he actually lost a leg from third degree burns as a result.

16:14 Speaker 3: So what we did here was we mapped second the second degree burn risk by downscaling ecostress temperatures down to ten meters resolution over the city and extracted it over the sidewalks and paved areas, and then we correlated that with the time and would take to get a second degree burn if someone came into contact with that surface. So as an example here you can see over most of the Phoenix area, specifically in the poorer regions in Maryvale, for example, where the surfaces succeed one hundred and fifty degrees fahrenheit.

16:47 Speaker 3: That's almost hot enough to cook an egg on the sidewalk. By the way, it would take less than three seconds to get a second reburn risk if you fell on the sidewalk or the paved area. And this work was what's really popular and was featured in Scientific American and La Times, New York Times. It even made the Nature Image of the Month for twenty twenty four and August I believe, And so it's this kind of public health communication that thermal remote sensing can directly support, especially for city health officials.

17:23 Speaker 3: So in the next part, we're gonna Savannah is going to lead you through a hands on exercise with land service temperature in LA County where she's going to step through how to visualize ecostress LST using a number of different software. We're going to use the peers to download the data and then visualize it with QJS. And also are.

17:48 Speaker 1: Thank you Glynn for that excellent foundation on urban heat, islands and ecostress land service temperature. We are now transitioning to a hands on exercise exploring nighttime land surface temperature in Los Angeles County. Nighttime LST is a metric that is often just as critical for public health as daytime peak temperatures, because sustained nighttime heat prevents physiological recovery from heat stress. By focusing on the nighttime, we can accurately quantify the thermal gap created as impervious urban surfaces slowly reradiate trapped heat while natural and irrigated spaces provide cooling through vappa transporation.

18:38 Speaker 1: In this demo, we will use R to define and determine if these natural cooling benefits actually persist as a heat wave intensifies, allowing us to identify the most effective locations for thermal interventions like tree canopy investments.

18:59 Speaker 1: The two objectives for this demo are to access, filter, and visualize ecostress LST data using appears, QGAS and R, and also to interpret LST characteristics for assessing urban thermal resilience.

19:18 Speaker 1: The activities for this demo include the following. There are some preparation steps, including cloning the GitHub repository or just simply downloading the R script. Second would be to make sure that you have R and qgis installed if they're not already, and then installing any required libraries that the R script lists. The script processes ecostress land surface temperature products by converting Calvin to celsius and applying mandatory QA and cloud masks to ensure only reliable surface temperature pixels are analyzed.

20:05 Speaker 1: The script then defines and isolates specific geographic boundaries for the Chatsworth Nature Preserve and an adjacent commercial area as an example, which allows for a direct comparison of different land covers. In particular, we will evaluate nighttime cooling dynamics by analyzing the differences in heat release between natural open spaces and impervious urban services during the September twenty twenty four Los Angeles heat wave. We will assess thermal resilience by determining if the cooling benefits of urban green spaces hold or erode as heat accumulates over multiple days of extreme temperatures.

20:54 Speaker 1: By extracting individual seventy meter pixel values into a tidy data format, the script generates mean temperatures and standard deviations to quantify the thermal behavior of each site. Finally, the analysis will produce a suite of sophisticated plots, including side by side LST maps, violin box plot distributions, and a thermal gap bar charts to visualize how the temperature difference between green and gray spaces evolves over a twenty three hour window. One thing I'd like to flag is that the nature preserve we focus on for this example comparison with the urban commercial area is in large part possible due to irrigation.

21:44 Speaker 1: So you will notice that in many of the unbuilt lands surrounding the urban area in the hills, for example, outside of Chatsworth, the temperatures aren't much lower than what we see in the urban air areas. This is because although there is vegetation in those areas, there is no irrigation, so the evaporative cooling capacity is much lower relative to the managed parks and other gardens that are well irrigated. So water helps plants stay cool, and if water is limited during a heat wave, plants will close their stamata to conserve water at the cost of having a higher temperature.

22:31 Speaker 1: I won't discuss this in the demo itself, so I wanted to just state that now as something to consider when we think about temperature impacts of green spaces. With all that said, let's start the demo. One thing I'd like to note is that these data from ecostress we're obtained through an appears request, and Appears is the application for extracting and exploring analysis ready samples. It's a tool that has a number of different data sets where through the extract feature you can get specific points or areas that are also able to be bounded, both specially and temporally.

23:25 Speaker 1: So I'm going to go ahead and log in here and just show briefly the request process. There is a more in depth instructional video for accessing ecostress data and other data sets with Appeers that will be linked to in this training resources page for the purposes of demonstrating here, I'm just going to show how we submit an appears request.

24:00 Speaker 1: So I'm going to give this request a title, and then i have an existing area of interest in Los Angeles County in California that will be the focus of this request, this data request, So I'm going to go ahead and upload that file here. I'm adding here a

24:34 Speaker 1: doojson file that is defining the area of interest over LA County for which the data will were requested for.

24:44 Speaker 1: And this demo will focus on one of the hottest summers on record, which was the summer of twenty twenty fourth.

24:55 Speaker 1: This in early September, and so the start date is going to be September third, twenty four and the end date will be September ninth. The product we're interested in is Ecostress land service temperature. The service temperature, this is the one that we will po us on. We will also want the quality Control data Layer QC file as well as the cloud mask. So those are the three data sets that will be requesting here and the file output file format will be GeoTIFF. Optionally we can use calendar, date and time for output file names, and then I'm just going to suggest include geographic, which is the WGS eighty four EPSG four three two six coordinate reference system.

26:31 Speaker 1: With that, we'll go ahead and submit the request. Appears requests, depending on the size of the requests and the amount of data that are being requested, can take some time to process, from hours to even a couple of days, and so to speed things along for the purposes of this demo, I already have an existing request that I submitted ahead of this demo and have downloaded that data from the email that is sent automatically. When the Appears request is ready to be viewed and downloaded the way that I will look for these data.

27:19 Speaker 1: If you're already logged in, you can click explore. So because this was a small area and number of files, the Appears was able to process it relatively quickly, and so we have the results of this request already to download. Here are the files that are included, where we will just go ahead and download all of them, and this may take a little while. And notice that for the purposes of this demo, I'm only using six out of the twenty two files here, which ended up being the ones that had the fewest number of clouds and kind of were the covered the most area in the study area.

28:25 Speaker 1: So again it's okay to just use the data directly from Zenodo, but I just wanted to illustrate how those files were obtained.

28:44 Speaker 1: And the download is now complete, So this is how the data were obtained from appears. Now we will go to QJA yes and inspect the data visually. Let's start by adding a basemap. I'm using the hcmgis lug in to ring in a satellite base map. Let's zoom into southern California and add our data. For now, we will focus just on the land surface temperature data. We'll do the filtering with the quality control flags and the cloud mask later on. Let's go ahead and visualize symbology that shows a color ramp.

30:12 Speaker 1: Notice that the values here are in kelvin, so we have yet to convert these into celsius, which we'll do later in our but for now, just to get a sense of where the hottest and coldest temperatures are. We can go ahead and apply this color ramp. Feel free to change if you have a different ramp that you'd like to look at. There's lots of different options here, and we'll do the same for this second day that we'll be looking at, which is September fifth.

31:02 Speaker 1: So notice that these two ecostress acquisitions within these two different days have some amount of spatial overlap, but are not fully spatially overlapped, and correspond to almost twenty four hours difference between these two acquisitions. So just by visually inspecting, we can see that this area of Los Angeles County has a mix of impervious services kind of built up areas as well as more vegetated areas and green spaces. And just through visual inspection, we can see differences in temperature in these urban areas where there are larger green spaces.

32:03 Speaker 1: So let's start by looking at the Chatsworth neighborhood in northwestern corner of LA County, about thirty miles from downtown,

32:19 Speaker 1: and it will zoom in here. What you're looking at is the Chatsworth Nature Preserve on the left, and then an adjacent commercial area which is immediately east of the Nature Preserve. These two areas sit almost side by side, so very similar weather and both experiencing the same heat wave. The question we're asking today is what other the surface underneath makes a measurable difference in nighttime temperature, and whether that difference holds up after days of sustained extreme heat. So after the first night of the heat wave on September third, Ecostress captured this image.

33:19 Speaker 1: And note that the naming convention here is in UTC, which in our case we're looking at September third, at about eleven pm Local times, a Pacific time. Even just at a glance, you can see cooler tones in over this nature preserve compared to the surrounding area, with the built areas showing warmer temperatures compared to the nature preserve. And then we can also be curious about the the second overpass, which is captured almost exactly twenty four hours later on September fourth, around ten thirty pm local time.

34:10 Speaker 1: So let's go ahead and check that and just see Note that there are some differences in temperature patterns here, but the overall kind of takeaway is that there seems to be a consistent difference even after another full day of the heat wave, where there are lower temperatures in this nature preserve compared to the surrounding built areas, and by this point La had baked through a full day of extreme heat. Burbank hit one hundred and eight degrees fahrenheit that afternoon, so these surfaces had been absorbing solar radiation for fourteen hours before the satellite passed Overhead said, the general spatial patterns do look similar.

35:03 Speaker 1: The preserve still reads cooler than the commercial area, but is the magnitude of that difference the same. Has the heat accumulated enough that even the preserve can't recover overnight? That's what we're going to quantify in our So first we will navigate to the GitHub repository for this training, where there is a read me file detailing an overview the prerequisites, data requirements, downloading process, and a few notes on setup. We will go ahead and download the r script here. Go ahead and save this file to a location on your local computer and open the file in our studio.

36:10 Speaker 1: Here we have the r script loaded. A written description of the focus of the script is here, which we also discussed in the slides leading up to this, so I'm gonna let participants review this in your own time. The first section here, Section one is to load all of the libraries. Go ahead and load them. If there are any libraries here that you don't already have installed, make sure to install them. And now we will go ahead and obtain the data for this demo, which is stored on Zenodo. So make sure to update this for repoder variable, which is where the data will be downloaded.

37:19 Speaker 1: Here I've made a folder twenty twenty six thermal rs Urban Heat Demo, so that's where I am keeping the data for this demonstration. If you're following along, go ahead and update this variable to wherever you'd like to store the data. This is the link to the data set which has all of the ecostress files. And then we will go ahead and in this portion of the script, it downloads and unzips the data as long as it's not already there in the first place, and I'll call you a t to the two files that were focused on.

38:05 Speaker 1: One is like I said, the first night September third, around eleven fourteen pm Pacific time, and then we have night two, which is September fourth, at ten twenty six pm Pacific time. So let's go ahead and run this section of the script. All right, that ran successfully. Section three focuses on is the processing function for processing ecostress data.

38:46 Speaker 1: One important note I want to flag here because it's common point of confusion and ecostress workflows. If you look at the level two algorithm theoretical basis document known as the ATBD or older tutorials, you'll see a point zero two scale factor applied before the kelvin to celsius conversion of the data. And that scale factor is correct, but it only applies for the native HDF five swath product, where integer storage requires that rescaling step. What we're working with here are the geotifh tiled data, which are analysis ready products from a PEERS where the scale factor has already been applied and values are stored as float thirty two in kelvin during a PEERS processing.

39:41 Speaker 1: So we will skip the point zero two multip multiplication entirely and go straight to subtracting two hundred and seventy three point one point five, which is what we do to convert from kelvin to degrees celsius. And this is the line that applies the quality control flag, where we're only extracting bits zero and one, and so this is the other thing that I wanted to highlights, just the QC masking, where the QC layer is a bit packed integer where multiple quality flags are encoded in different bit positions within the same value.

40:27 Speaker 1: So bits zero and one carry the mandatory QA flag, where zero means high quality and one means good quality. Bits two and three mean poor or cloudy. If we just wrote QC greater than one as a raw integer, comparison, higher order bits unrelated to quality could flip us into masking pixels that are actually fine. So the modulo four operation here extracts only the two least significant bits, which is the standard bit manipulation approach for this kind of packed flag field. And finally, the cloud mask.

41:18 Speaker 1: The L two user guide was updated in twenty twenty four to explicitly note that in collection two, cloud information is not propagated into the QC bitfield, So that means that QC filtering alone is not sufficient to filter out all clouds, so you must apply the separate cloud mask layer. Our function here does both in sequence. So I will now run section three of the code. Night one comes back at roughly thirteen to twenty nine degrees celsius, and night two goes from about twelve to thirty five degrees celsius.

42:18 Speaker 1: That upper bound climbing from twenty nine to thirty five already tells a story. After a full day of heat loading, the hottest urban surfaces are retaining substantially more heat by the second night. A quick note I'll say on the coordinate Reference system situation. So the file name contains eleven N, which refers to UTM zone eleven North, but the JUDITI files themselves are stored in geographic coordinates WGS eighty four longitude and latitude. The eleven N is part of the Ecostress tiling grid naming convention, not the store projection.

43:01 Speaker 1: So to avoid any CRS resolution issues, we keep everything in WGS eighty four, which is EPSG four three two six, and we define our aois directly in decimal degrees. So this is our default AOI that we're looking into. If you remember what we saw the map on qgis we have the latitude and longitude bounding box coordinates for the Nature Preserve area in Chatsworth, as well as a coordinates box around the extent of the commercial area that's adjacent to this nature preserve. So for the homework you all will be looking at this other area of interest to do a different analysis in a different part of La County.

44:03 Speaker 1: But for today, we're going to go ahead and so you can see this part is commented out, So I'm going to go ahead and run this section for here.

44:20 Speaker 1: Section five will produce a side by side map comparison where we can see these two Ecostress acquisition dates side by side with the two areas of interests highlighted. So I'm going to go ahead and run section five, where you can see we're using the gg plot package for visualization.

44:59 Speaker 1: You can see that. But these two maps are helping us visualize the data using an identical color scale, which is intentional and important because when you're comparing two images temporally, you need the colors to mean the same temperatures in both panels. If each panel were auto stretched to its own range warmer night two pixels would look the same color as night one pixels, and you'd lose the ability to make a visual comparison. So the green and orange outlines mark our two study areas. Even from the map alone, you can see the Preserve tends toward cooler end of the palette in both nights.

45:45 Speaker 1: What we can't see clearly from the map is the magnitude of this gap, or whether it changed between nights. That's what the next two figures will quantify. So in section six here of the code, we are pulling every valid seventy meter pixel that falls within each bounding box for each night. Each row in the resulting table is going to be is one pixel, so one actual satellite measurement, and this gives us the full distribution, not just a single summary statistic, which matters because urban services are spatially heterogeneous, a parking lot and a rooftop can behave quite differently even within the same bounding box area.

46:38 Speaker 1: So let's go ahead and run section six of this script. This provides a summary statistic table stored in the variable DF summary, where now we can quantitatively look at the mean land surface temperature in degrees celsius, along with the standard deviation and the number of seventy meter pixels included in each of these areas. So we can see from this numerically that the Nature Preserve on night one was cooler on average than it was on night two. Which is expected. The standard deviation remains pretty close on both nights, although slightly lower on night two, which is interesting.

47:41 Speaker 1: And then similarly in the commercial area we see an increase in temperature from twenty three point seven degrees celsius to twenty six point six degrees celsius on night two in the commercial area. But now let's look at this from a full pixel level distribution for each AOI, so that brings us into section seven here of this script, which is which will show a violin plot with raw pixels of the data values underneath. Let's go ahead and run this so the points you see are the individual seventy meter pixels and the violin shows the overall shape of the distribution, where the box shows the intercortile range and the raw points show you exactly where the data actually are.

48:49 Speaker 1: A few things to notice here. First, the preserve distribution is consistently cooler than the commercial area on both nights, which we saw in summary table as well. That gap is real and visible. Another thing to consider is the key comparison you're thinking about. Does the gap between the two aois change or areas of interest change from night one to night two? Hold that question while we look at the next figure here. So in section eight of the code, our core question is, after a full day of extreme heat on September fourth, do we still see the same thermal gap between the preserve and the commercial area on night two?

49:45 Speaker 1: So we'll calculate the mean LST per area of interest per night and plot the absolute temperature for each area of interest across both nights, which shows whether each land cover temperature warm from night one to night two, and then also the thermal gap, which is simply the subtraction of the temperature of the commercial area minus the temperature of the preserve on each night. If this gap shrinks, the cooling benefit of the preserve is being reduced by accumulated heat stress. But if it persists or grows, the preserve is maintaining its function even under sustained extreme conditions.

50:30 Speaker 1: So again we're using the library to make these plots, and I'm just going to go ahead and run section eight. Increase the size here for easier viewing. Panel A shows the mean LST trajectory for each land covered type across the two nights, with error bars showing one standard deviation. Panel B converts that into a single number. The thermal gap defined as the commercial area mean minus the preserve mean. So in this case we see the gap remains very similar on night two, with slight point two degrees celsius decrease from four point five degree c to four point three degree c.

51:29 Speaker 1: This suggests that the preserves of appotransporration and long wave emission are not overwhelmed by the heat event, at least not after two nights from the start of the heat wave. This supports the argument for urban green space as heat resilience infrastructure, not just as an amenity. Another possible outcome for subsequent analysis might be what to evaluate whether the gap substantially narrows on further nights into a heat wave. And for this particular heat wave, there weren't ecostress data available around similar times after these two days for comparison, but that's something to flag and to consider if you're interested in studying another heat wave in a different location.

52:18 Speaker 1: So it'll depend on what data are available, but it's something that could be interesting to follow up with looking at longer term multi day heat waves and whether even certain natural areas might have a threshold beyond which sustained heat stress suppresses their cooling function. So this would be interesting to test in future work in other places. Like I said, if a more prolonged period of the heat wave, the temperature difference between the two areas might actually decrease.

52:53 Speaker 1: So this is really one big urban heat island resilience question, and it's why night time ecostress data can be valuable for urban planning, not just because it captures the heat island effect, but because it lets us evaluate whether specific green infrastructure actually delivers measurable cooling when it's needed the most. These are the kinds of analyzes that urban heat emergency planners, parks departments, and climate adaptation offices could use for their local and regional needs.

53:26 Speaker 3: Data like this.

53:27 Speaker 1: Identify cooling refuges, evaluate wor tree canopy investments will have the greatest thermal return, and map residential neighborhoods where nighttime heat exposure is most dangerous, particularly for people without air conditioning. Now I will turn it over back to Glenn.

53:53 Speaker 3: All right, So moving on to section four now, which is land service temperature downscaling. This brings us to this central idea of today's talk, thermal sharpening, or it's also sometimes called downscaling.

54:08 Speaker 2: It's often used interchangeably.

54:10 Speaker 3: The basic concept here is to take course resolution thermal data will course here meaning seventy meters is actually that is the highest resolution available from space, but still it's all pretty coarse. If you look at the urban environment and then combine that with higher resolution data that is well correlated with temperature, and that's the key here, and by doing that we can recover the thermal detail in a much finer spatial scale. In this case, we're going to go all the way down to a ten meter resolution.

54:40 Speaker 3: So this is a nice example that illustrates that where we can go from what looks like gloves of cooler and warm temperature features in the environment. Here this is other LA by the way down to ten meters, where we can now start seeing very fine scale detail in the temperatures. We can see the differences between these hiking trails and Griffith Park, golf courses, freeways, building roofs and so forth. So this opens up just a huge capability for us over specifically urban areas, but also in other natural vegetation areas as well.

55:17 Speaker 3: So as you can see, these are two more comparisons over different parts of LA. The one I just showed Griffith Park on the top and the bottom is of the Long Beach Harbor area.

55:28 Speaker 2: But as I said before, we can start really.

55:31 Speaker 3: Discriminating different urban features from one another. We can see, for example, the hot costco roof versus the coolest cemetery. You can see the difference between the freeway and the much cooler fairways on the golf course, for example.

55:46 Speaker 2: So this animation, let's see if it works.

55:49 Speaker 3: Yep. This shows you the kind of detail that we'll we're revealing from the seventy meter down to the ten meter level. This is of a Pasadena and you can see the Rose Bowl highlight again the top left corner there, and after sharpening you can see that we can resolve the distinctive sort of cool areas north of the Rose Bowl and the golf course, and also the hottest parking lots and roads to the southern part of the Rose Bowl. And we'll be going over that in quite a bit more detail in the tutorial because this is actually going to be one of the venues for the upcoming Olympic Games in LA and so this kind of high resolution thermal information is critical for heat intervention planning and also planning for shade interventions in areas like this.

56:34 Speaker 3: This is I think just a fantastic image. This is also created by a summer student of ours last year, by Ashley Agatap, and it shows a dance scale image over Paris during extreme heat wave in June twenty twenty two, where they actually had to close down a lot of the public spaces and museums within the city because it was so hot. This is sharpened and then and overlaid on Google Earth, which is why you can see the structural features of the actual urban area come into play here. And I'll go through this a little bit.

57:08 Speaker 3: I'll show you how to do this in the tutorial as well. And also it has it has its caveats too, which I will go through. But you know, you can see individual neighborhoods, parks, you can clearly see the same river and other infrastructure standing out with the temperature. And it's this kind of intraurban thermal detail like you simply cannot resolve at the seventy meter or higher pixel scale as I mentioned before, and looking ahead, the Olympic Games is coming up in twenty twenty eight in La This is a sharp and echo stress temperature map similar to what we did over the Phoenix area where we pull out the temperatures at the street level.

57:47 Speaker 2: And what I've highlighted here.

57:49 Speaker 3: Is the actual venues of the Olympic Games, and there's just tremendous potential heir for designing, for example, heat mitigation approaches, planning where to add shade, both natural and man made shade structures, and also cooling it outdoor events and watch party locations during the Olympic Games, and even extending these techniques to the coastal zone coastal water temperature and water quality monitoring. So I will be going over some of these areas that we'll be focusing on also for the Olympic Games in the tutorial.

58:20 Speaker 3: So moving on to some theory of the actual downscaling. Statistical downscaling would call sharpening. So what we're doing is we're just aggregating a coarse scale thermal pixel into its subpixel components. And the trek here is to use auxiliary data that has a higher space resolution that is statistically correlated with the land service temperature, which is our target yet and so this illustration shows that quite nicely. It's from chenne Bel and colleagues and shows how a single course pixel A one in this case gets disaggregated into four subpixels B one through before using the relationships learned from the higher resolution and covariates.

59:02 Speaker 3: And we'll actually be doing this at a finer scales, so we'll be taking one ecostress pixel and dividing it up into fifty different individual pixels. So that's seventy meter to ten meter resolution. So at the highest level, the workflow looks like this. We start out with ecostress at seventy meters as our target variable, and then we assigned predictive variables consisting of Sentinel two. This is a European Space Agency sensor that is launched in twenty fifteen and generates ten and twenty meters high resolution visible shortwave infrared data, so.

59:40 Speaker 2: That is our primary predictor.

59:42 Speaker 3: Here is the reflected surface reflectants from Sentinel two. We also include land cover types built up fraction and also the elevation. The output then is a We run that through a random forest regression. We train the data at seventy meters and then we predict the final or sharpened LST at the ten meter scale. So you may ask, well, why do we use sentinel to the spectral reflectance bans. Firstly, we use we could use that there are some twenty meter bands as well that we have been using in the past, but in this tutorial we'll be focusing on the ten meter bands, which gets us down to that sort of street level resolution that we're after.

1:00:26 Speaker 3: So the bands are located in the visible part of the spectrum, so between four hundred and seven hundred nanometers, this is the part of the spectrum we actually see with our naked eyes. We have three bands there, bands two, three, and four, blue, green, and red. And so these bands, the surface reflectance of these bands correlate with temperature because the more reflective a surface, the less the surface will be heating and therefore the cooler the surface will be. And this is a bit of a generalization that doesn't always occur, but it's a good statistical correlation with the temperature.

1:01:00 Speaker 3: And we also then at band eight, which is the near infrared band. This is right on the edge of the infrared spectrum that entsed around nine hundred and animeters, and this band is actually the opposite information to the other three, so it's in it's sensitive to vegetation, health, chlorical content, water content, and it's one of the strongest predictors in fact of LSD special variability in urban rural gradients whenever you have vegetation involved. And we'll go over that a bit later as well.

1:01:28 Speaker 3: So why does this work? So what I've plotted here is land service temperature. Each pixel he has an actual temperature on the ground, and I've plotted it versus the NDVII, which is a normalized difference vegetation index. Some of you may have heard of this. It's a band ratio between banded eight and four that I just showed you. And generally high ENDBI values are greener vegetation and are generally cooler, and the opposite is true with low DVII values. So, because service temperature is to sartisically tied to these surface reflectance properties, the boat bear low reflectant services like roads for example, run hearts vegetated bhy NDBI services stay cooler.

1:02:11 Speaker 3: And then on the bottom end of the spectrum here we have high albedo services like white roofs for example, or white roads are also cool and so there's a sort of triangular nonlinear relationship between surface reflectants and land surface temperature. And these are the exact nonlinear relationships that our machine learning model is going to be learning at the seventy meter scale and then applying down to the seventy meter scale. So why sharpen ecostress down to these levels. What are the primary urban applications and what are our reasons for doing this?

1:02:48 Speaker 3: Well, there are six concrete reasons, heah. The firstly, we're able to of course resolve urban features from the seventy meter down to the ten meter scale. Streets, parks, rooftops blurd together at the seventy meter scale and be able to resolve those. It lets us valid our heat interventions at the scale at which they're actually deployed, so think of fool pavements being applied over city's cool roofs. It also dramatically reduces our mixed pixel error, especially when we're in sort of heterogeneous neighborhoods.

1:03:21 Speaker 3: It forthited lives at the parcel level planning data that the city agencies actually already use. It sharpens heat vulnerability mapping and undeserved poorer communities, which is critical for example downtown LA. And the last part here is really critical is that we apply this residual correction, which we will go over in detail, which means we actually preserve the radiometric accuracy by normalizing to that of the ecostress instruments. So we're adding spatial detail, not inventing temperatures, if you will, and I'll show you some of these.

1:03:56 Speaker 2: Examples in the next few sites.

1:04:00 Speaker 3: At seventy meters you can think of we have a like a small city buck scale temperature, and then after the sharpening, we end up with ten meters, which equivalent to forty nine pixels, and so we have a fifty times improvement in the space or resolution using this methodology. So why random forest for this task. Well, it's inherently available within Google Earth Engine, which is great. And the basis for which it works is you take an ensemble of these what we call decision trees. You have outlined m tree one through hand.

1:04:33 Speaker 3: Typically we start out with about one hundred trees, but you can increase or increase that value, and each of these is trained on a random.

1:04:41 Speaker 2: Subset of features.

1:04:43 Speaker 3: So what I mean here by features is that it's a subset of a different combinations of the predictive variables and also the different pixels within the scene. And so we then end up with a final prediction, which is an average across all of those trees. So it's a very powerful technique because it, as I said before, it handles these nonlinear LST and service relationships. It's robust outliers because it's able to generate an ensemble mean which reduces the variance in the data, and it's also transferable across scale, So we train at seventy meters and then we can predict a ten meters.

1:05:21 Speaker 3: Also, something that comes out of the random forest which is really interesting is a feature importance information which identifies which of our predictors were the most sensitive or the most important in arriving at our final result, and we'll go over that in a minute as well. So in from sort of the high level, this is what the sharpening and Google Earth Engine is going to involve. These are the different steps. We start with our training data and our target variable ecastress LST at seventy meter.

1:05:56 Speaker 3: We then train the random forest model within Google Earth Engine at the seventy meter scale, and then we predict the ten meter final temperature and then apply this residual correction, which means we upscale the ten meter prediction to seventy meters subtract from the original seventy meter observation from uka stress, and then we add that residual back to the ten meter result and I'll show you that on the next slide.

1:06:23 Speaker 2: So these are just some cutouts that are made from within.

1:06:26 Speaker 3: Google Life Engine at various steps along the way throughout the sharpening process. This is a cutout over an area heterogeneous urban area here in La So in the middle here you can see the original ecostress the LSD at seventy meters. The next step is the initial downscaled LSD. So this is after applying the model at ten meters, we end up with this result. So you can see immediately find spatial detail coming out within the urban environment that correlates nicely with the actual visible image. On the top left.

1:06:59 Speaker 3: You can also see that the actual absolute values of the temperature here are quite different to the actual observed data in the top we don't see these sort of hot spots in the lower left and bottom areas. So then after the residual correction I just explained, we upscale the ten meter to seventy subtracted from the seventy meter ecostress, and we end up with this residual corrected final LST at ten meters which preserves the actual radiometric information in the original seventy meter resolution image.

1:07:29 Speaker 3: And you can see that these correlate a lot better now, So quickly go through what these predictors look like, sort of at a much larger scale over the entire LA City and County area. This is the area that we'll be focusing on in the tutorial today by the work. By the way, so this is the sentinel to surface reflectance. This is for band eight. This is the near infrared band a ten meter resolution.

1:07:59 Speaker 2: This is.

1:08:01 Speaker 3: Provided by Coperticus and European Space Agency through Google Earth Engine, and so we'll be using the blue, green.

1:08:08 Speaker 2: Red, and ne infrared bands.

1:08:09 Speaker 3: This is the near infrared band. And you can see here immediately areas sort of features that pop out. For example, these wide areas that you see here with high reflectants are in fact green spaces.

1:08:22 Speaker 2: So these are pretty much the golf courses around the.

1:08:25 Speaker 3: LA region which have a high band aid reflectants because they're chlorofully active green types of irrigated vegetation. And conversely, we are very low reflectants over dense urban paved surfaces, mostly human made types of surfaces. This is the NDBIS I spoke about. It's a band ratio between band AID and Band four. High NDBI values correlate also to dance and green vegetation, and conversely varius soils and some nest vegetiation or urban materials have a very low NDDI And conversely with temperature.

1:09:06 Speaker 3: Of course, high NDBI values should have cool temperatures and low DBR values should have harder temperatures.

1:09:14 Speaker 3: The other.

1:09:16 Speaker 2: Predictor, yeah that I'll show is the built surface fraction.

1:09:19 Speaker 3: This is from the European Commission's Global Human Settlement Layer available within Google Earth Engine, and this data captures gradients and urbanization that other land cover land use categories alone are n't able to capture, and especially useful in mixed perio urban regions where you can see the fraction of the build up area changing very rapidly into more vegetator rural regions. And then our fourth predictor is actual elevation. This is from SRTMS that he need a product. We then resemble that ten meters for the training and elevation really matters in fact in this region because the lapse rate alone explains a couple of degrees of LSD variation, especially across these hilly areas of LA.

1:10:06 Speaker 3: You think of the Santa Monica Mountains, see are the Saint Gabriel's and we want the model to learn that effect explicitly rather than sort of baking it into the residuals. And then lastly, hear the star of the show. This is the Eker Stress Land Service temperature at seventy meter resolution.

1:10:24 Speaker 2: This is the actual image we'll be working with in the tutorial.

1:10:27 Speaker 3: Next. This was I believe in the fourteenth of August twenty twenty. This is an extreme heat wave events in LA. You can see the extremely hot temperatures here in the San Fernando Valley in the northern part here that is insulated from the cooler ocean breeze. You can see strong gradients here along the coastline having much cooler temperatures and then increasing too much harder temperatures.

1:10:52 Speaker 2: In the interior urban areas.

1:10:55 Speaker 3: This was quite a four PM in the afternoon, so one of the hottest parts of the day. I also notice that there's no cloud artifacts in this image, so there's a very nice clear sky day, which is why I chose this image.

1:11:08 Speaker 2: Also noticed that very distinctive cool.

1:11:11 Speaker 3: Islands here you can notice these sort of cool little areas within the urban environment. And again these are these mostly golf courses, parks, green spaces that are twenty thirty degrees cooler in fact than the surrounding urban landscape. And then I mentioned the feature importance outcome of the random forest, and this is really interesting and something that I will show you how to derive from the data and Google atventionine as well, and it's how the predictors actually rank in the train model in terms of feature importance.

1:11:44 Speaker 3: So in this case, elevation actually came in at the top at seventeen point six percent, followed by the four sentinel bands or plustered around seventeen percent, so that these four, these five predictors are driving most of the information that is provided in the random.

1:12:01 Speaker 2: Forest getting us to that ten meter scale.

1:12:04 Speaker 3: The built surface actually contributed about eleven point four percent, so pretty significant.

1:12:09 Speaker 2: But the actual land use land cover map.

1:12:13 Speaker 3: Which I included in this example had a very low information content, and that's actually because it's derived from the sentinel to reflectant data. So essentially you don't want to add in variables that are correlated with each other, so we should probably remove the land use land cover from this case. All right, So that's it on the theory. We'll now get onto the fun pat here, which will be the actual ads on exercise where we'll be diving into Google avengine and actually showing you how to produce these really nice striking maps of the environment.

1:12:49 Speaker 3: Before I get started here, I just want to preface this by saying that we do have everything I'm going to go through here. Literally every step and click I'm going to make is actually available on a tutorial that we created in Microsoft Work. So we'll make these available so that you don't have to, you know, look through the video tutorial again and listen to EGO going through it all. You can actually follow along step by step, every click and every move I make on this document tutorial that you can reference for later use.

1:13:26 Speaker 3: So starting up from the beginning, we're going to open a web browser. For example, it tells you here go to code dot earth Engine, Google dot com. You'll also need to sign up for a free Google account, and you can do that going by going to this website earth Engine Logogle dot com sign up, slash sign up. We also have a link to a YouTube video in our resource page which actually shows you step by step on how to actually sign up for a free account of Google Earth to start and be able to do this tutorial.

1:14:00 Speaker 3: So this is what you'll end up with if you start your Google Earth Engine page. On the left are your various scripts or pieces of code that you can stall. We then have a docs page. You know, I need to look at that. The assets are actually data that your code can operate on. We won't go into that either. So what you'll be doing is you'll be clicking new here it's a bigger red icon.

1:14:26 Speaker 2: Okay, click new. You'll go to file, give it a name.

1:14:32 Speaker 3: I'm just going to call it sharp that you can call it any name you want, and or create a new file which you then can paste directly the code that we're going to provide for you, which is the Eco sharp V one point one code which you're seeing right here. And so that's that's how to get started. So you'll click save and then you have your JavaScript Google Engine code ready to go. So a couple just orientation things here first and center. Of course, you can see the actual code. This is JavaScript code.

1:15:06 Speaker 3: You don't need to even look at this if you don't want to. But you'll see the buttons on the top you're showing how to run the code, reset the code. On the file right panel.

1:15:16 Speaker 2: You can see something called the console.

1:15:18 Speaker 3: This outputs various types of information that the code runs, and we'll see that in a minute. For example, some stats on accuracy, what data was used at what time. Tasks will also go over when we're going to finally export this data to a geot file and open it in qhis, you'll be able to run a task here to export it directly to your Google Drive and we'll go over that in a minute. Okay, so getting started, we're going to start over the Los Angeles region. Of course, I said, we're going to start with that image that I just showed you.

1:15:52 Speaker 3: If you grab this little toggle heah, you can actually move the map up and down to make it easier for you to actually see the data coming in. So I'm going to move it right up for now, and we're going to hit run to simply hit run, and what we end up with is a user interface panel on the left, and you use interface panel on the right. What you'll see on the left, let's start of the left. You'll see a date range here. So in this case, I've put in a single date, the fourteenth of August twenty twenty, where I knew we had this scene over Los Angeles during the heatwave events.

1:16:28 Speaker 3: If you had to change this and say, for example, go from June till July twenty twenty, what the code will do is it'll take all the ecostress observations over their time and generate a mean composite image over a date range. So we can either look at a single date for a single image or create a composite from a date range. You can also see additional summer years. If we clicked on those additional years, we can create a composite over consecutive years.

1:16:57 Speaker 2: Over a fixed date range.

1:16:58 Speaker 3: So if we want all this summertime average temperature across the entire mission period, can click on all of these years and generate a mean composite average. And so this is often useful for city planners and managers. They want to know what is the baseline average summertime temperature that we've experienced over the last five years. The next part here is the local time window, so this is in hours of day.

1:17:22 Speaker 2: What I've done is years.

1:17:23 Speaker 3: I've converted this you can convert to different time zones around the world, and that makes it easier where you don't have to deal with UTC times. So we're in Los Angeles, We're going to click Los Angeles time zone, and in this case we're going to look at all hours of the day. We know there's one scene here. You can change this to if you just want the afternoon hour, say twelve till eighteen. We'll get all the data from the hardest times of day twelve till six pm. But let's go back to zero twenty three.

1:17:55 Speaker 3: The next part here is the max cloud tolerance. So if you want less cloud on the scene, you can decrease this value.

1:18:04 Speaker 2: You can go down to ten.

1:18:05 Speaker 3: I would suggest for a long term composite that you go down to ten percent. So we don't want more than ten percent of cloud on the scene. The cloud is already screened by the way, so it won't be included in the data. But we don't want to combine images where the part of the image is missing, and then we combine it with other images and you get some artifacts coming through. So I would use ten for a composite and use thirty for a single images. Fine, then you have an option to draw custom rectangle.

1:18:38 Speaker 3: So if I click on that and click activate selection, and we just want to do San Fernando Valley, for example, we can drag and drop and run our sharpening over that specific area. So I'm going to clear that because what we're actually going to do here is go to an ecostress tile and we're going to run the code on an entire ecostress tile over this region. So if I activate that selection and I'll click on anywhere in the center of the LA map, it's going to give us the tile boundary for the specific area.

1:19:11 Speaker 3: And I won't go through the details of the tile boundaries, but the sentient we use the Sentinel two tiling structure here, which is based on the n g r S system. This is the Military Grid Reference system where the earth is divided up into these fixed tiles. They're roughly about one hundred kilometers wide, and you can go and Google this and learn more and that if you want. But this is the tile we're working with that covers the whole LA area quite nicely, as you can see.

1:19:40 Speaker 2: And then on the bottom left here you can run and export.

1:19:43 Speaker 3: But before we do that, let's go to the right here where we can see We have some other options here which I really useful, so we can choose exactly which sharpening predictors we want to use. The central two bounds here are required, so that's all is going to be clicked. But we can and click off the land use. As I said, it doesn't have a huge influence on the final results. We probably want to include the surface fraction and elevation those had significant feature importance values. I'm not going to include the NDBI just because that information is already baked into our reflectance spans.

1:20:16 Speaker 3: But you can, for example, use NDBI and the reflectant spans and you can see how what kind of data that provides, for example over ag areas that may be more beneficial. Next, we can choose the the temperature unit here, so we can go from calvin, which is the standard unit the scientific unit that stress comes in. We can switch to celsius and then pharranhyde here. I'm going to stick with pharranhid and then we have the export options. So this is you know, the actual data that we want to look at, and qgis eventually is going to be the LST serventy meter ten meter.

1:20:52 Speaker 3: Don't worry about this model asset as I mentioned before in the top left, if you did export the asset, you'd have this available in your assets here, and then you could apply that random forest model to another scene if you wanted. Later. You don't have to do the training on every scene, but that maybe you should be left to another tutoria.

1:21:15 Speaker 2: Then you can export also the predictor layers.

1:21:17 Speaker 3: If you want to look at, for example, the images I just showed you in the slides with the sentinel bands, Sandy, the island cover. You can export all of those that if you wish too. So I'm going to leave those alone. And if we click run and export, you can see the bottom left here that it's already found all the data. It's already on step two. It's not sharpening the data to ten meter resolution, and you can just see how quickly this operates.

1:21:42 Speaker 2: We already have the LST pulled in.

1:21:44 Speaker 3: This is the Ecostress Landservice temperature actually just showed you in the slides, and while it's running through its sharpening training process. Here, if you click on the layers here on the top right of the image, you can actually toggle between the different predictors and the LST that have been brought in by Google Earth Engine, and also on the right here as an opacity slider, which is kind of cool. You can see how the image fades in and out of the background Google Earth map. So you know, while this is running, I guess you can play around with the LSTs and the other data yet and look at what the data looks like.

1:22:25 Speaker 3: And then okay, let's go through the console here. There's actually a lot of information that gets printed out in the console on the right. Inside here, it just gives the summary of the actual options you use, the date range, the time, cloud tolerance. It'll show you the actual n g RS tile number that Echo stress relates to you. This is the eleven sl T tile, and then it will show you the actual Sentinel two scene that was used. So the code currently looks at Sentinel two data within a month of the observation.

1:23:00 Speaker 3: So in this case you can see we found a scene on the twelfth of July. You can stress scene was the fourteenth of August, so not exactly a month, but close enough. The summertime period vegetation is not going to change too much, for example, over LA, so it's fine that it's within a month.

1:23:18 Speaker 2: The code automatically looks for very clear Sentinel two scenes in this case is barely no cloud on the Sentinel two scene.

1:23:26 Speaker 3: That's pretty important. You should probably check that, but the code will automatically check for cloud pre data. There's some of the information. Yeah, it shows the data that you want to explot and then also actually shows the sharpening accuracy, which is the initial ten meter resolution image that showed earlier. Now, after normalization, this actually dramatically reduces in the bias and rms after we do the radiometrinmalization chief for stress. So these numbers may look pretty big, like eight degrees fahrenheit, but they end up being pretty small in the end.

1:24:03 Speaker 3: It actually might change this so that produces the stats after we've done that radiomentric organization. Anyway, let's go into tasks. This is the important partner. We see our three tasks. Here are the three data sets that we chose to be exported to our Google drive. So I've already done this, so I'm not going to We're not going to wait and run this in real time. It actually takes quite a bit of time because if you click on the ten meta here and head run, it will give you the task name. You can change that if you want, shows the coordinate system, the resolution, and the drive shows the drive folder that it will create for you if you haven't created one in your Google Drive out and then the file name and the file format of Yoursio is a geotep, so it has the special information included within the data.

1:24:54 Speaker 2: And so then head run.

1:24:56 Speaker 3: You'll see that there's a submitted task it we'll be running, and I'll show time and it's taken to run. You do that with the others and then you end up with your data and your Google Drive. So my case, this is the GEE you can stress folder and you can see a list of due to files here. You can see I have quite a few because I've obviously been doing a lot of testing on this in the last few months. And then you can click on the right side. You can see these three little dots on each file. Click that head download and it will download your file to your desktop.

1:25:31 Speaker 3: Okay, so let's move on to the next part here, which is actually visualizing this data in QGIS. So just to recap everything I'm just being through, is on this Google Earth Engine running ekshop version one point one tutorial, So you can go through that step by step at your own pace if you want, and then we'll be available if we're going to be available.

1:25:57 Speaker 2: For questions as well.

1:25:59 Speaker 3: So similarly, we have a full tutorial on the visualization in QGIS. So every click instep you see coming up now is laboriously documented in this file, so you can go through this at your own pace as well. There's little tip boxes, There's all kinds of nice little sympets of information here and it ends with having a quick reference and troubleshooting guide, so really nice reference to have handy while we go through the actual tutorial. So let's open up QGIS.

1:26:35 Speaker 2: I already have mine open.

1:26:36 Speaker 3: There's a link in the resources to how to download QGIS, and I'm pretty sure most of you will have it already from Savannah's tutorials. So let's.

1:26:50 Speaker 2: Let's begin.

1:26:51 Speaker 3: So if you go to the top left, had project and hit new, we start a new project going into project save as and call it something. So let's call it Eco stress sharpening. And I'm just going to say it in my documents folder. So we have it saved, we have our projects saved. We're I'm going to lose all the data that we're we're doing. So next step is we want to we want to this the Google basemp's. This is really important and very useful to have. So go up to right at the top wh is HC, M G, I S. Click on that, go to basemap and click on Google Satellite.

1:27:30 Speaker 3: So now we have our Google Nice satellite basemp. Next we go to a.

1:27:37 Speaker 2: Layer, add layer and raster layer and.

1:27:43 Speaker 3: On the source here you can see I already have mine that i've that I've uploaded. But if I click on the right here the three dots, it's going to take us to folder where mine is in our set, and I'm going to upload the firstly the ten meter lst add close and I'm going to do the same thing for the twenty meter sorry, with seventy meter layer. So we go yeah, and it's seventy meters ad. Okay, Now we have on the bottom left on our layers, you can see we have the seventy meter product ten meter product.

1:28:18 Speaker 3: Google said like that, Now we're not seeing anything because we actually have to zoom to the layer. So if we ride click and hit zoom to layer, we will now see the map center on our image of the Los Angeles and it's all default and gray scale. That's kind of boring. We don't want that. So what are we gonna do is were gonna double click or right click and hit properties on our file. We end up with this page. We start with symbology. Go to the top where it says render type hits single band pseudo color.

1:28:56 Speaker 3: We don't want any y scale. Then for the min max value settings, I'm going to click this errow. And what I typically like to use is I don't like to hard code the men in max because it's going to change for every scene. So I typically like to use between two to five of the lower percentile of data. And then on the hi side, I either go from ninety eight to one hundred, and that kind of varies on the area that I'm looking at. So let's use those values for the color ramp. I actually so this is really a personal preference.

1:29:31 Speaker 3: You can go Magma spectral works pretty well. If you look at all color ramps here, you can see there's a really wide variety of color ramps. I'll show you a little trick if the one that I actually like is if you go to create color ramp, go to catalog CBT city. I know this is a bit much, but qgi us and then all by author on the top rides. I already like this temperature, so let's let's use temperature all right. So if we click and we head apply, now we can see our image nicely scaled in color scale where the blues are the cooler colors red so are out of colors.

1:30:22 Speaker 3: So it's pretty intuitive, and we we're just going to go ahead and do the same thing quickly with our second band. I'll go through this pretty quick already, change our centerur thresholds by one hundred, and we'll go with our temperature color up, which again I have to add I actually need to save this once. I don't have to do this every time. I haven't temperature okay, apply, Okay, Now we have both data sets together at the same color range and more importantly, in the same color ramp. So firstly, what we can do here, Let's go in.

1:31:13 Speaker 3: Let's go into this area in the Sant Banana Valley. Let's look at this little blue island here and see what's going on. So we zoom in here. We can toggle all these on and north, which is really really useful. So you can look at the Google background image. Yep, it's a golf course. So this is a two golf courses in the Sant Banana Valley, and on the right side is an actual park. This is an actual area where people fly remote control aircraft. In fact, this image here, this must have been in the spring, because everything looks pretty green, and in actual fact, during the summer this is pretty barren and bare soil, very hot, and you can actually see that in the echo stressed data.

1:31:51 Speaker 3: You can see the very hot temperatures sere over the park and then much cooler over the irrigated golf course. So looking at ten meters, so we just see two big blobs right at seventy meters, there's not much detail there. If we're going to the ten meter scale, now we can really see the detail pop out. So we can start seeing the actual trees along the fairways of the golf courts that are cooler in the fairways, it's pretty cool. You can see this vegetated waterway going down the center of the golf course.

1:32:21 Speaker 3: You can see roads, you can see specific freeways that are hot. So it just opens up all this detail to what's actually going on in the urban environment. On the right here you can see building roofs and parking lots that are distinctive and have distinctive temperatures from each other. And if you zoom into this sort of suburban area, you can actually see the temperature of individual trees, so this is really cool. You can actually see the cooling impact of trees and green grass yards in this environment.

1:32:59 Speaker 3: You can see to the level of whose roof is the hardest, right, so you can see the dark roof right here you can see my person. This sort of bright red spot is actually someone's very dark black shingle roof. And this cooler area right here you can see it's actually a lighter shingle color, which is a lighter temperature. So really cool in that you can really zoom into very fine scale detail on the temperatures.

1:33:26 Speaker 2: Let's go to another area quickly that I really like.

1:33:28 Speaker 3: I actually used to live close to this area and actually play on this golf course quite a lot. This is the Griffith Park golf course on the edge of Griffith Park, and it's really nice because there's sharp gradients and temperature from the golf course to the freeway into this more industrial complex zone just off the western edge of Glendale.

1:33:50 Speaker 2: So it's seventy meters. Again, not much detail.

1:33:54 Speaker 3: You can see the broad outlines of where the hot spots are, but it's really only when we go to the ten meter scale that you start seeing the actual temperature of hiking trail. So these are bare gravel trails that are hotter than the surrounding landscape. Again, trees lining the fairways. And what's pretty interesting here is let me go to let me change the scale a little bit here. This is where as I mentioned, you want to change the fuminium max of the data depending on where you want to look at to increase the contrast.

1:34:31 Speaker 3: So this looks a lot better for this area. One thing I wanted to point out here, note that this is the one thirty four Freeway running here along the northern part of the image. Right now, you can see how much hotter that freeway is. It's running east west, so it's directly being illuminated by the afternoon sun. There's no intervening trees or shade on the freeway, whereas if you look at the five Freeway south right next to the golf course here, look how much cooler it is. It's actually ten to fifteen degrees cooler.

1:35:02 Speaker 3: And if we actually zoom into Google Earthlier. You can see that there's actually a tree, very tall trees lined up on the court on this side, on the western side of the freeway, which provides a lot of shading. Of course, they have Griffith Park here which provides also even more shading.

1:35:18 Speaker 2: During the afternoon hours, you have a much cooler freeway.

1:35:21 Speaker 3: So it just shows the level of detail that we can get to at this resolution by looking at the actual differences in freeway temperatures, which I think is really cool. Okay, then the last quick thing I want to show you here before we actually get to exporting the data as actual imagery. Let's go to the Rose Bowl. This is actually one of the venues for the World Cup and the Olympics coming up, so this is a pretty important area that they're going to be looking at to try and implement heat mitigation efforts.

1:35:56 Speaker 3: If we look at the Rose Bowl here, we have these huge, massive parking lots right These are n actually going to be very hot surfaces, and this large field is actually gets very hot during the summer too. It's mostly dry sinest grass. So looking at the seventimeter data again, we can see there's some hotspots around.

1:36:14 Speaker 2: We can't clearly see exactly what.

1:36:16 Speaker 3: They are, but it's only when we go to the ten meter scale that you can see yep, these are these large parking lots and roads surrounding the southern part of the stadium, which are you know, twenty thirty degrees warmer than the golf cooks for example, to the north. And this if you can imagine, if there's airflow and wind coming in from the southern, southwest and southern parts of this image, that you're going to get a lot of heat invection from these hot surfaces into the stadium. And so this is these are the kind of areas that we want to target for heat intervention during upcoming events, but also just for the future sustainability of heat in the ARLA region.

1:36:59 Speaker 3: Okay, so what if we wanted to export these images. Before I did that, I wanted to show you another cool trick that sometimes we use, but not very often because physically it doesn't stop making much sense. But if we go into symbology again, all the way down to layer rendering blending mode, it's currently normal if we head overlay, what it's going to do is actually overlay the data directly on Google at the engine background so you can.

1:37:30 Speaker 2: It's pretty cool.

1:37:31 Speaker 3: You can see the almost the three D structure coming through with the temperature overlaid on top. And this is how we generated that paras map that I showed you earlier. But the one caveat here is that you can see the building roofs here which are white, actually come through as looking white, whereas they lose their actual temperature color, which would be the dark red color. The same with green spaces that come out looking green, because blue on green gives us kind of a green. So if you can deal with that, that's fine, but it just provides a really nice, pretty image.

1:38:03 Speaker 3: Okay, let's go back to the normal and then we're gonna the last thing we're gonna do today is going to show you how to actually export a PNG image or an image in any format that you like. So let's let's put the image in this at this scale. You can change the magnifier on the bottom right by the way to any scale you want, and also you can just use your your mask to zoom in an out. I'm going to leave it at this scale, go up to projects. We're going to go to a new print layout or control P. Let's call it LA la shop and you can you can name the print layout anything you wants.

1:38:48 Speaker 3: Really, you end up with this sort of blank paddock here. What are you going to do is add item at the top, add map. Okay, now you're going to get across, and what you want to do is click and drag across to fill this entire blank canvas space. And what it's going to do, it's going to render the map that you see in qjis in this region. So now we can do a couple of things. We can go to add item, add a scale bar. For example, there's a scale bar on the bottom left. Let's add a north arrow is also kind of typical, shows the north up direction.

1:39:28 Speaker 3: You can also add what's pretty important actually is a legend, which shows us the legend of the color scale. I really don't like the legends that QJS comes with, to be honest, So usually I actually do this kind of work in our or math Lab or Python. But this is possible. So now you see twos because we have two layers in our QJI. So we click on that, go to the bottom right here where it says legend items, click auto update. Let's move this up, and let's click on the first one here because this is the ten meter map, and hit this minus red button so we delete that one.

1:40:08 Speaker 3: You can also delete the actual names I believe if you like. But anyway, that's what you can produce with your map as well. And then if we go to layout export as image, you can then export your image in Typically PNG is a pretty good format to do it and preserves the resolution of the image. You can also output in tiff or jpeg and then click save and then export resolution you want to go at least three hundred dpi for looking at for importing these into part points or anything else, hit save and you'll have your image set now.

1:40:48 Speaker 3: The other way we can do this without adding all this other fancy stuff to the image is if we just want to directly export this image right here. Go to project import export export map to image, change the dpi three hundred dpi. Let's save, and again I'll just name it something and you can directly say just this image right here. Sometimes what I do is, in fact, is I crop out the color scale on the on the left here, and I saved that and included with the image, which makes it a little easier if you want to do something in QGIS.

1:41:28 Speaker 2: I think we have a little bit of time left.

1:41:31 Speaker 3: The last thing I wanted to go through here was how to actually do some quantitative analysis within QGI. So this will take another minute or two. And this is all laid out in the tutorial again. So I say, I want to see what is it? What is the temperature difference between this golf course and this sort of bear park area. So what I'm going to do is it's a little bit elaborate, but if I go to layer, uh, let's see vector and layer and layer. Actually I've really forgot how to do this myself, so let's actually this is a good idea.

1:42:12 Speaker 3: Let's go to the QGIS tutorial and let's figure out how to how to do this.

1:42:18 Speaker 2: So create zone polygons. Okay, zone and statistics.

1:42:23 Speaker 3: Yah, we go. So layer, create layer, new shape file layer Okay, layer create layer, new shapefile layer. Let's call it zones geometry type. We want to do polygon and then we're gonna click. Okay, go to zones again. Right click on zones, head toggle editing. It's this with a little pencil mark, and then click this sort of green icon which is adding a polygon feature. Looks like a little cutting green, and then you can start drawing your polygon to any shape that you wish. Let's let's draw the entire golf course here.

1:43:06 Speaker 3: We then right click go ID. Let's call it one, and we have our first one. Let's click on the second one, which is this little really hot zone park right click, call it a different ID number two. Okay, then we have our two different zones. Now what we can do is go into processing toolbox, search for zonal statistics. Double click zonal statistics.

1:43:37 Speaker 2: We have our input.

1:43:38 Speaker 3: Layer as the zones that we just created rest a layer. We want the ten meter data and we want to see let's do the standard deviation the mean that will be important in the mid mat Okay, so run close that. And if we go to zonal statistics and open a tribute table, Okay, there we go.

1:44:03 Speaker 2: We finally have the actual quantitative results.

1:44:05 Speaker 3: Yes, so the ID one, which is the golf course mean temperature of about one hundred degrees faaranheit. The part next to it was yep, twenty degrees hotter on average hundred and twenty degrees, but up to one twenty eight degrees farahnite. So that's that's really hot and the minimum gold called temperature is about eighty seven, so you can see some zonal stats. I usually do this kind of thing in our and Matt lab and packon, but I just wanted to show you that you can actually do this in directly in QGS if you really want to do something to remove that layer, and remove that layer and then see if I can.

1:44:48 Speaker 3: All right, and then if the last thing here is to double click on your file again, if you go to transparency, you can change the transparency let's say seventy percent.

1:45:03 Speaker 2: This is a really cool trick to do as well.

1:45:05 Speaker 3: And I can start seeing some of the background Google of imagery coming in the background, and it just gives it a nicer look. So that's it. That's all I'm going to show you today on this tutorial. Again, you can go through the actual documents which we'll be sharing that will give you the full step by step on how to do what I just did in the QJAS tutorial and the shopping tutorial.

1:45:27 Speaker 2: So thank you very much everyone.

1:45:28 Speaker 3: I hope you enjoyed that, and we'll be here for questions when this is finally presented.

1:45:34 Speaker 1: Thank you, Thank you Glynn for leading us through the downscaling Ecostress LST Demo. I will now provide a recap of our training. Here's a summary covering all seven learning objectives from our thermal remote sensing training. First, the physics of thermal emission all matter above absolute zero AMIDST electromagnetic radiation. Plank's law, Bean's displacement law, and the Stefan Boltzman law govern how much energy is emitted at what wavelengths, and how sens sensitively it responds to temperature change.

1:46:30 Speaker 1: Earth surfaces at around three hundred Calvin peak near nine point seven micrometers, placing land surface temperature retrieval squarely in the eight to fourteen micrometer thermal infrared window, a range of purely emitted energy with no solar contamination. Real surfaces are not perfect black bodies. Emissivity being less than one encodes the composition of and the roughness as well as the mineralogy of these surfaces, and a one point five percent emissivity error propagates to about one kelvin of LST error.

1:47:20 Speaker 1: The temperature emissivity separation algorithm or the tests algorithm used by Ecostress Aster and Modus twenty one simultaneously retrieves both temperature and spectral emissivity, making it far more accurate than split window approaches over bare, arid or geologically complex surfaces, where split window can er by three to ten Calvin. Unlike optical sensors such as Sentinel two, lansat in the visible and neur infrared, which detect reflected sunlight and go dark at night, thermal sensors detect emitted surface radiation around the clock.

1:48:09 Speaker 1: This enables nighttime observations of heat retention, urban cooling failures, and ecosystem thermal stress signals that are invisible to any reflectant. Spased product from AFTERS sixteen day revisit and ninety mid meter pixels to ecostress's diurnal sampling from the International Space Station and Looking Ahead to the Eagles or formerly known as Surface Biology Geology Mission TRISHNA and LSTM. Each mission trades off spatial resolution, temporal frequency, and spectral coverage differently. Choosing the right one depends on whether your application demands long term record find spatial detail or multi time of day sampling.

1:49:01 Speaker 1: Thermal data underpin not just urban heat island characterization, but many other applications, including evapotranspiration estimation, agricultural drought monitoring, ecosystem thermal stress detection, long term climate trend analysis such as lakewarming, expanding hot nights, and surface energy balance modeling. Making accurate thermal and frared retrievals consequential not just for remote sensing, but for climate projections, ecological conservation, and public health decision making. In the hands on ecostress data processing in R, we walked through a complete R based workflow accessing ecostress LST and emissivity collection to geotips via a pears, applying quality control, bitmasking and cloud mass converting Calvin to Celsius, defining areas of interest, and producing violin box plots and thermal gap figures demonstrating how nighttime LST at the Chatsworth Nature Preserve remained measurably cooler than the adjacent commercial area even after a full day of extreme heat during the September twenty twenty four LA heat wave.

1:50:26 Speaker 1: For the component of the second demo, downscaling ecostress to neighborhood scale with machine learning, we saw that because the physics of thermal emission limits spaceborne thermal and for red resolution for instance seventy meters for ecostress versus ten meters for Sentinel two. Part two introduced a random force downscaling workflow in Google Earth Engine that leverages high resolution sentinel to spectral indices DBI and DBI DWI as predictors of land service temperature, exploiting the strong physical relationship between service cover type and heat retention to sharpen ecostress imagery to ten meters and reveal block level urban heat patterns invisible in the native product.

1:51:29 Speaker 1: Reminder that the homework for this training opens today on June second. It is due two weeks later on June sixteenth, twenty twenty six. Upon attendance of both sessions as well as completion of the homework by the due date, participants will receive a certificate of completion for this training. I would like to acknowledge my fellow colleagues in the Arset Ecological Conservation Team who have supported the creation of his training, Sativa Cruz, Juan Torres Berres and Justin Fain. I would also like to acknowledge the RSET program staff who made this training possible, including Selwyn Hudson, Odoy, Maria Maravito, Suzanne Monty, and Melanie Follette.

1:52:19 Speaker 1: We encourage you to check out our website as well as the rset YouTube channel linked here in this slide. You can also join our quarterly newsletter to stay up to date on our latest trainings by sending an email with no subject to our set dash join at lists dot NASA dot gov and then follow the instructions sent in the response. The contact information for myself and our other guest instructor, Glenn Holly, is also here if you'd like to email us directly with specific questions about this training.

1:53:00 Speaker 1: Here is a list of references for this training. These are all great places to start to dive deeper into the material that we have presented.

1:53:15 Speaker 1: Thank you all for your participation in this training. We will now transition into the Q and A. So let's address a few questions now and then the remainder of the questions we'll address later this week. So I want to first just thank doctor Glenn Holly who has joined us as for this training. Thank you so much Glenn for your time and expertise. And yeah, let's dive into the questions. Question one asks, given the possibility of a moderate to strong Almino year, how could al nino interact with long term climate warming to increase urban heat, island risks, and affect glacier mass balance.

1:54:07 Speaker 1: In particular, how might these impacts differ across regions such as tropical cities, mid latitude urban areas, mountain glaciers, and polar ice environments. When would you like to answer?

1:54:23 Speaker 3: Sure? Yeah, I mean, starting off, it's a it's a pretty broad question and interdisciplinary, you know, covering a lot of different science areas and disciplines. But you know, this is our best answer on the fly, is that you know we are getting in Elmina likely this summer it's going to be strong noise, predicting that as a result, we're going to have one of the woman's years on record coming up twenty twenty six, twenty seven. So in general, you know, the all ninios has a positive impact on temperature anomally, so we know expect warmer temperatures overlaying on an already warmer climate, and that is going to ramp up the UHI effects that that we typically experience in urban areas.

1:55:17 Speaker 2: I won't go into.

1:55:18 Speaker 3: Detail about how this affects different regions. I'm actually not a major expert on al Nino and other events like this, but I know, for example, in tropical cities they do, they should expect less cloud cover, so you can expect increased solar loading, increased temperatures, drought, spress vegetation, and combined with higher humidity. That's it's probably going to be subtropical areas they're going to experience the most detrimental effects. Mid latitude urban areas. Think of western coast of California, for example in La just from experience usually results in drier conditions and warmer conditions.

1:56:00 Speaker 3: It's rainfall, higher temperatures. So in general, the Anino events will ramp up and increase the Uhi effect.

1:56:11 Speaker 1: Thank you. Question two asks how does this urban cooling effect during the day differ depending on surrounding land type, for example arid versus agricultural versus forested.

1:56:30 Speaker 3: Yeah, so this is actually nicely demonstrated in the earlier presentation of two areas Delhi and Athens, two quite different cities. But you know Delhi, you think of a typical Mediterranean environment, and what happens there during the summertime is that the surrounding rural areas in Delhi, it's actually mostly agricultural regions, have a very dry, barren sinesse vegetation and an increased exposure to bear soil, and so typically we see a cool island effect during the daytime in cities like that, especially in Dailhi, we see a pronounced cool island effect.

1:57:14 Speaker 3: If you have a city in a tropical region, you know, surrounded by a dance forest, the reverse is going to be treated. The forest is going to be dance vegetation. Typically it's going to be ten to twenty degrees cooler than the urban surfaces, which heat up much faster than vegetation. So you'll see the classic urban heat island effect, which most people are familiar with.

1:57:39 Speaker 1: Great, let's do two more questions. So the question three, why does near surface ozone concentration increase in the summer.

1:57:49 Speaker 3: Yeah, that's that's a good question that most people don't realize. You know what the you know physically, what is happening then, And it's actually a photochemical meaning that it's actual voc so volatile organic compounds from car exhaust, just think pollution in general in the city that reacts with sunlight to produce ozone. And so the hotter it is, the sunnier it is, the morizone's gonna built up especially near the surface if you have high pollutants in the urban environment.

1:58:29 Speaker 3: Thank you.

1:58:31 Speaker 1: Question for which will be our final question for today, asks can ecostress land surface temperature data be integrated with hydrological or watershed models to jointly analyze urban heat of appotranspiration runoff and the benefits of green infrastructure. Also, are there any NASA related application case studies that demonstrate this type of integrated analysis.

1:58:57 Speaker 3: Okay, yes, another pretty loaded question. What I've written down here just some examples that came to the top of my head. You know, the ecostress we do provide a standard about a transporation product which is derived from hydrological models that we actually run at h apl work into the details of those models, but that that standard, those standard data products are available and they are in fact fed directly into a lot of order shared orter balanced models as the latent heat and you can do your own literature search to find those studies for the In terms of the benefits of green infrastructure.

1:59:43 Speaker 3: For the LA twenty eight Olympics, we've actually been working with LA Committee and also other groups from usc U c L A part of the Shade LA organization to see if we can help you use ecostress LST data specifically to try and quite to by which areas would be optimal for implementing, for example, heat mitigation measures, so planning more trees, implementing more shade structures and so forth. And so we've been working on this machine learning model. It's actually a double causal machine learning model that integrates ecostress LST with a number of other different variables to try and estimate what the cooling benefits are or would be of planning more trees in the city.

2:00:26 Speaker 3: So that is work that is ongoing. It's actually going to be sponsored by NASA pretty soon and we're going to be working more closely with the city. Two really good papers that look at the urban bias FERE influenced specifically on urban heat and carbon uptake in general in the urban environments focused on Los Angeles are these two papers right here written by two jpls. I would craze you to go and look more detail at those papers for what is done there.

2:01:00 Speaker 1: I would like to again thank you doctor Glenn Hully for joining us for this our set training series, and thank you to all of our participants for joining us.

2:01:20 Speaker 1: And we will again be posting all course training materials and the Q and A document within a week of this training