<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Burke Greer, PhD on Medium]]></title>
        <description><![CDATA[Stories by Burke Greer, PhD on Medium]]></description>
        <link>https://medium.com/@burke.greer.science?source=rss-8a254eeee51d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dSwn8iRRIiEObqmEliiJDw.jpeg</url>
            <title>Stories by Burke Greer, PhD on Medium</title>
            <link>https://medium.com/@burke.greer.science?source=rss-8a254eeee51d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 22:19:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@burke.greer.science/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How can the space industry better help customers using satellite imagery, IoT data, and Machine…]]></title>
            <link>https://medium.com/@burke.greer.science/how-can-the-space-industry-better-help-customers-using-satellite-imagery-iot-data-and-machine-5a937e08efc1?source=rss-8a254eeee51d------2</link>
            <guid isPermaLink="false">https://medium.com/p/5a937e08efc1</guid>
            <category><![CDATA[space]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[remote-sensing]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[iot]]></category>
            <dc:creator><![CDATA[Burke Greer, PhD]]></dc:creator>
            <pubDate>Thu, 30 Jan 2020 17:27:02 GMT</pubDate>
            <atom:updated>2020-01-30T17:27:02.310Z</atom:updated>
            <content:encoded><![CDATA[<h3>How can the space industry better help customers using satellite imagery, IoT data, and Machine Learning?</h3><p>At a meetup in Palo Alto we were concluding our conversation about satellite data applications when the topic turned to the ongoing difficulties of monetizing the space industry. After all, we had just spent hours talking about how new satellites are collecting the best data ever available, and how machine learning (ML) algorithms and pipelines can find objects in images and detect changes. With better satellites, data, and analytics, shouldn’t this mean we can discover other things that were previously unknowable? Isn’t this an opportunity? So, we began to discuss why satellite, IoT, and ML products are under-consumed.</p><p>While Silicon Valley’s space industry has the technology to do amazing things, beyond the tech itself there are aspects that aren’t usually discussed yet are crucial to the growth of this fledgling industry. Here are a few opinions I have on the topic:</p><ol><li><strong>Discoveries from satellite imagery and ML should be delivered to customers in an actionable context</strong>. The customer using the data needs a one step process for implementing the new information into their work. A stack of images alone doesn’t provide an insight without interpretation. And neither does a change detecting algorithm on its own — or even a map of new building pixels because there is more processing needed for it to be useful. End-of-pipeline customers want to make decisions from insights not do more comparisons, analysis, or work. If you’re not sure how your data provides insights, quantify how it directly impacts something like CapEx or OpEx and in how many steps. If you can’t outline this within a user story, your product is incomplete.</li><li><strong>The value gained from insights must scale positively with the cost of data and analysis</strong>. Having the best insight isn’t enough. The customer needs to gain more money (or save in labor, time, efficiency, safety, environmental impact/hazard, etc.) than they spend on the data and analysis. An insight needs to be worth the cost of generation and added analytical work (see number 1) directly inflates its price.</li><li><strong>Shared data standards for searching and accessing satellite and IoT data on the cloud will speed innovation. </strong>The bleeding edge for leveraging spatial data and ML requires fusing data across different sources on the cloud, but satellite and IoT data standards are not widely adopted or agreed upon. While one data scientist can handle the customized scraping of data from 5 or 10 sources, what happens when the team is 50 people and there are 1000 sources? Or 10,000? And then extend this — how will this model fare in production which is more demanding than development? And across different geographies that require different ML models? As an industry, we have the tech to support this type of data integration, but we don’t have a good system to organize this properly yet. Instead of a common set of data standards, each provider has its own. Common standards will help reduce costs for downstream data acquisition, ML, and implementation at the end of the pipeline.</li></ol><p>There is great hope and opportunity, but we need to make it easier for people to work with data and analysis within real people’s (read as <em>non-GIS experts</em>) workflows. Towards that, I’d love to see more satellite and ML data pitches that go beyond the data and ML and talk about how insights were implemented to drive change. We need to provide real value to customers and not just more data. Also, (standard) data standards please!</p><p>What do you think?</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/0*gSu2Mlrb4FOK1ZjK" /></figure><p>Our Blue Marble (credit: Twin Blue Marbles, NASA images by Reto Stöckli, based on data from NASA and NOAA.)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5a937e08efc1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Interpreting Messy Thermal Data and Why Time Matters]]></title>
            <link>https://medium.com/@burke.greer.science/interpreting-messy-thermal-data-and-why-time-matters-1560dc4c078e?source=rss-8a254eeee51d------2</link>
            <guid isPermaLink="false">https://medium.com/p/1560dc4c078e</guid>
            <category><![CDATA[remote-sensing]]></category>
            <category><![CDATA[thermal-camera]]></category>
            <category><![CDATA[drones]]></category>
            <category><![CDATA[thermal-imaging]]></category>
            <category><![CDATA[thermal]]></category>
            <dc:creator><![CDATA[Burke Greer, PhD]]></dc:creator>
            <pubDate>Fri, 15 Sep 2017 14:47:08 GMT</pubDate>
            <atom:updated>2017-10-20T17:29:28.334Z</atom:updated>
            <content:encoded><![CDATA[<p>The intensity and direction of incoming solar radiation changes for a place over the course of a day. For this reason, when you’re taking images/measurements with a satellite, airplane, drone, or handheld platform, you should consider time as a potentially important variable in your data. Otherwise you may find changes in your images which are not actually driven by the pattern you are trying to sense.</p><p>As an example, in August 2016 I joined a team of scientists and graduate students at a study site on the North Rim of the Grand Canyon where we were collecting multispectral and thermal images using a UAV while also taking coincident climate and physiological measurements in common gardens of Southwestern White Pine. This site is a part of the Southwestern White Pine Project described in more detail <a href="http://www.sega.nau.edu/node/53">here</a>.</p><p><a href="http://ferm.forestry.oregonstate.edu/facstaff/wing-michael">The Wing Lab</a> flew a UAV with a thermal camera three times that day with the first flight near dawn, a second flight mid-morning, and a third flight mid-day. Interestingly, when we looked at our thermal images there was an unexpected pattern where the mid-day data showed lower and more evenly distributed temperatures across the scene than those taken mid-morning at a cooler time of day.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5_7jzQF1LKlppHKNJ5LluA.png" /><figcaption>Mid-morning (left) and mid-day (right) surface temperature. The circles are the stems of individual seedlings.</figcaption></figure><p>These patterns were strange because as the sun rose in the sky we expected that more incoming solar radiation would warm the plants — particularly on the tops of seedlings. Fortunately, we also set up a climate station that measured coincident incoming solar radiation every minute.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*vGyt9i0-mRN9GQbA7f2FiQ.png" /><figcaption>Incoming solar radiation over the course of the morning at 1 minute intervals</figcaption></figure><p>While there is a small trend of increasing solar radiation over the course of the day, notice that there are huge swings over minute time scales. Also, do you see the long periods of reduced incoming solar radiation over the course of the morning and especially after 10:00? Do you have any ideas for what these are?</p><p>It turns out that we experienced the summer monsoon at our study site in August. While we did not have any thunderheads while we were sampling, there were clouds moving around. The time stamp on the mid-morning thermal data was 09:20 which was during a period with higher incoming solar radiation than the mid-day image taken at 12:40 (see the lines below recording around 750 W/m² and 275 W/m² at 09:20 and 12:40, respectively).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*alqAc1BtYQZZlCt9wdpvjQ.png" /><figcaption>Vertical Lines are at 09:20 and 12:40 hours</figcaption></figure><p>So, what is the lesson here? Because of clouds, the lighting conditions were different between flights — this means that we really can’t compare the thermal images from 09:20 to 12:40 because they’re not replicates of the same conditions. Another important lesson is that while UAVs can collect all kinds of amazing data — don’t forget that climate/weather data collected at your site can help down the line when you are trying to interpret what you’re looking at in your images. In your own image work (whether collected from a stationary camera, UAV, airplane, or satellite), you should try to minimize differences in lighting/solar conditions between your images, standardize collection times, or include time as a variable in your study design and analysis if you plan to compare plant images taken at different times.</p><p>That’s all I’ve got for now.</p><p>Extra special thanks to collaborators Michael Wing, Jonathan Burnette, Cory Garms, Jessica DaBell, Chris Still, and Kristen Waring.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1560dc4c078e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Platforms for imagery]]></title>
            <link>https://medium.com/@burke.greer.science/platforms-for-imagery-2f755003ebe0?source=rss-8a254eeee51d------2</link>
            <guid isPermaLink="false">https://medium.com/p/2f755003ebe0</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[drones]]></category>
            <category><![CDATA[remote-sensing]]></category>
            <category><![CDATA[agriculture]]></category>
            <category><![CDATA[science]]></category>
            <dc:creator><![CDATA[Burke Greer, PhD]]></dc:creator>
            <pubDate>Tue, 05 Sep 2017 18:43:14 GMT</pubDate>
            <atom:updated>2017-09-05T18:44:14.502Z</atom:updated>
            <content:encoded><![CDATA[<p>There are many different platforms used to take images of plants. The most commonly used by scientists are satellites, and airplanes, but UAV technology is becoming more available as UAVs and their cameras become more affordable. Each of these different platforms have advantages and disadvantages. For example, there are lots and lots of satellite images that are freely available from NASA (thank you NASA!) that capture large areas at regular intervals, but often at low temporal or spatial resolution. This means that each pixel of an image can be tens to hundreds of meters in size, and could be collected weeks apart — which, depending on what you’re trying to measure, could be your perfect solution, or project killing problem. Also, because satellites are high in the atmosphere, atmospheric interference can cause havoc in the data you’re trying to collect — just ask any remote sensing graduate student or scientist about clouds and sit back and enjoy a 30 minute diatribe about water vapor. And its not just the clouds that are problems — cloud shadows introduce variation as well. As an example, check out these satellite images of central California that were obscured by smoke, and then the remnants of a tropical storm over this past Labor Day weekend (2017). Imagine the trouble you’d be in if this was a critical time for data collection on <a href="http://gilroygarlicfestival.com/">your garlic farm in Gilroy</a>! Some atmospheric interference can be corrected, but these corrections can introduce uncertainty into your data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/674/1*hKK5VrUcTCk-YnaMFmWB_w.gif" /><figcaption>Holy atmospheric interference, Batman!</figcaption></figure><p>Imagery taken from an airplane has less atmosphere between it and the ground than a satellite (planes are closer to earth), but still has atmospheric interference which can change day to day depending on the weather. There are other challenges with taking images from airplanes — for example if you were taking thermal imagery, the temperature of the thermal camera body can change the measurements of the camera which means that altitude and atmospheric conditions on the day of the flight could introduce variation into your data (never mind turbulence!). Cloud cover can still be a problem — but at least an airplane can get below the clouds (though you may now need to account for reduced incoming solar radiation in your measurements than will change measured reflectance). However, aerial imagery generally has higher spatial and temporal resolution than available satellite data. But you would need to hire a crew to make flights, purchase these images from a company, or pilot the aircraft yourself to acquire these data.</p><p>UAVs can provide high resolution imagery for whenever you want to fly — this potentially means extraordinary spatial resolution (we’re talking millimeters) wherever and whenever you want it. But, like airplanes, you need to fly them yourself or hire a company to come do it. I should also point out that there are some regulatory hoops to flying your own UAV professionally. Also, the images collected from UAVs are closer to the subject (like a vineyard), which means that all of the plants in your images are at a more variable perspective than those from a plane or satellite which is overhead.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/857/1*Gt9N3DaID__0sjgFBSxmHw.jpeg" /><figcaption>Sorry I didn’t go to art school, but I hope this conveys at least a little information beyond “Burke can’t draw”. The point is that A is WAY smaller than B: if the satellite is 700 km, and the drone is 30 meters above the grapes, and the distance from the most northerly and most southerly vines is 10 meters, A is ~0.0008°, and B is ~20°.</figcaption></figure><p>Changing and variable perspectives can lead to more variation in UAV images than satellite or airplane-based images. As a hypothetical example, if the sun is brightly shining from the south (and to the right in the above drawing), plants to the south (right) of the UAV could appear to have more shadows because they will be more backlit relative to the camera, and plants to the north (left) will appear to have fewer shadows because they are more sunlit relative to the camera. It is also possible that some plants to the south will also be brighter (ie. have a greater maximum reflectance) because they will be reflecting more sunlight directly at the camera, and some plants to the north will be less bright (ie. lower maximum reflectance) because they will be reflecting more sunlight away from the camera. This can again mean more spatial and temporal heterogeneity (ie. noise) in your data. This is true in satellite data, too, but at a much lesser extent because the difference in angles between the satellite and each plant in your area of interest is much less in a satellite image (A &lt; B) than a UAV image. Across entire satellite image scenes, these types of problems are typically corrected, but are present in the raw data.</p><p>However, despite the broad use of satellites and airplanes, and recent use of drones, I would bet that the most utilized platform across the world for taking pictures of plants is merely a handheld camera — just think of how many photos of fall colors, mountains, and fields, are taken every year. Most everyone has a camera on their cell phone, or a digital camera in their closet. They are cheap to buy, and easy to use and have a huge potential for measuring plants — but are under-utilized. More on this later.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KIYJGf54qIezQ8hm2DY2Iw.png" /><figcaption>Gothic Mountain, Colorado, near the Rocky Mountain Biological Laboratory where I did fieldwork for my PhD (photo credit: Burke Greer).</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2f755003ebe0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Plant Growth and the Environment]]></title>
            <link>https://medium.com/@burke.greer.science/plant-growth-and-the-environment-86694a23f60b?source=rss-8a254eeee51d------2</link>
            <guid isPermaLink="false">https://medium.com/p/86694a23f60b</guid>
            <category><![CDATA[remote-sensing]]></category>
            <category><![CDATA[geography]]></category>
            <category><![CDATA[environment]]></category>
            <category><![CDATA[drones]]></category>
            <category><![CDATA[agriculture]]></category>
            <dc:creator><![CDATA[Burke Greer, PhD]]></dc:creator>
            <pubDate>Wed, 30 Aug 2017 21:52:25 GMT</pubDate>
            <atom:updated>2017-08-31T15:52:32.557Z</atom:updated>
            <content:encoded><![CDATA[<p>A plant’s growth depends, in part, on its environment. Factors like how much sun a plant gets, how much water is available to it, and a plant’s temperature influence the structure of a plant, and how it grows. To understand plant-environment relationships, people measure plant characteristics along with environmental conditions over time and space. However, collecting these measurements can be extraordinarily time and labor intensive, and expensive. For example, plant height is a good indicator of plant robustness, but to measure this someone would likely need to either measure the height of each plant they are studying, or would need to buy LIDAR-based measurements (usually taken by aircraft). Furthermore, to understand how the environment is driving the measured patterns (in this case height), environmental conditions also must be measured and matched to each plant at each time point. These measurements also need to be replicated so that we can tell if we see a true pattern, or just random noise in the data — this is particularly true across space (like a farm field) where plants growing at different parts of the same field could in fact have vastly different environments.</p><p>Imaging technologies have emerged as powerful tools for making scalable plant measurements that are cheaper, and take less time to measure than other approaches, while providing data that is impossible to measure with other techniques. Furthermore, by using successive images of the same plants over time, computer vision and machine learning can identify drivers to plant growth or function (like water loss through transpiration). For example we could capture things like how many tomatoes are growing, how large they are, and how their growth changes in response to weather (like temperature, and sun intensity). We can also sense things like the presence (and even timing of onset) for disease and nutrient deficiency that can be used for management decisions like when to apply pesticides or add fertilizer, or linked to things like environmental stress. Furthermore, because the measurements are from images, there is no destructive sampling, and the only work required is taking the image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vCfkKgcdNaptp2NsJI9Tmg.jpeg" /><figcaption>An overhead image of these tomatoes doesn’t show any tomatoes, but an oblique view lets you look between the leaves. Do you see the tomato?</figcaption></figure><p>Most plant imagery used today is collected using satellites, or using airplanes. These data are great because they capture vast areas while potentially taking high resolution images of plants. However, most of these images are of only the tops of plants which leaves a wealth of data uncollected. For example, if I wanted to count tomatoes in an image to measure how many tomatoes a farm field has, and how big they are, an overhead view would only show the tomatoes that appear at or above the plant canopy. An oblique view would provide a better picture, but using an airplane or satellite for this would be impractical because of atmospheric interference and the difficulty of <a href="https://youtu.be/3Sq7ST6AtX4?t=4">piloting a plane close enough to the ground</a> to get an oblique angle.</p><p>But, in this use case, there are other better tools we could use like handheld cameras (like cell phones), and UAVs. In fact, these platforms offer tremendous potential for collecting high resolution, and oblique imagery of plants because everybody has cell phone cameras already, and UAVs are quickly becoming more affordable.</p><p>That is all I’ve got time to write for now, but look for future posts where I will be exploring how we can better use imagery and camera and UAV platforms to measure plant characteristics and how they change over time and space.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=86694a23f60b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>