Considerations for Planning a Remote Sensing Study of (Invasive) Vegetation
This blog post serves as a companion to the publication "Mapping the Extent of Invasive Phragmites australis subsp. australis From Airborne Hyperspectral Imagery", and assumes a basic understanding of remote sensing principles by the reader. While this post is written in general language that may be more beneficial for people who are newer to remote sensing/vegetation related studies, it contains information that may still be useful for more experienced researchers. The list of considerations presented here is not exhausted, but still provides a good starting point for developing research projects.
The term “remote sensing” simply refers to obtaining information about an object without being in direct contact with it. This concept has been around for a long time, from aerial photographs captured from hot air balloons in the 1800s, to modern-day airborne and satellite-borne sensors. Remote sensing data can be used for a wide variety of applications, such as imaging distant planets, mapping the ocean floor, monitoring wildfires, tracking land use change, and much more.
One area of remote sensing is spectroscopy, which is the science of understanding the interaction between light (the electromagnetic spectrum) and matter. By studying how electromagnetic radiation is emitted, absorbed, or reflected by different materials, detailed information such as the object’s composition or physical structure can be obtained. Spectrographic imagers can acquire spectral imagery at different spatial scales and spectral resolutions (both of which are discussed below), and this data can be utilized for many different applications.
The remote sensing of vegetation is an area of research that has been around for a while. The idea that individual species of vegetation can be differentiated based on their spectral signature is the foundation for many remote sensing studies, such as this blog post’s companion article which investigates the use of airborne hyperspectral imagery to detect and map the invasive reed Phragmites australis subsp. australis. As the relevant technology becomes more advanced and accessible, it is now more possible for those that are new to remote sensing/spectroscopy to utilize these tools for their own studies. For example, this technology could prove beneficial for land management groups looking to identify vegetation species of interest (e.g. invasive species) in order to prioritize areas at risk and form a more detailed and comprehensive management plan. This blog post introduces some important considerations for remote sensing studies of vegetation, native or invasive. The points addressed in the following sections should help to guide the formation and execution of such a research project.
Formation of the Research Question(s)
One of the first stages of any scientific project is to determine the research question, or the goal of the project. You might already have a specific end goal in mind (e.g. “I want to create a map of invasive Phragmites within the park”), so use this to further develop your question(s). For example, using the above statement, the research goal described in this post’s companion paper was refined to “can hyperspectral imagery and a target detection methodology be used to identify and map invasive Phragmites within Îles-de-Boucherville National Park?”. The end result of this question was in fact a species distribution map of invasive Phragmites across the park, which was created by using airborne hyperspectral imagery and a target detection methodology. By narrowing down your research question to be more specific, it will help you begin to plan the methodology and analysis methods for your project.
Examine previous research related to your topic: this will help give you ideas for your more specific research goals, as well as help you understand what research has been conducted so far, and the current scientific trends in that field. It can also provide details for forming your methodology, and for analyzing the data later on. Google Scholar is a valuable tool for reviewing the preexisting literature.
Data Source and Acquisition
Do I need/want multispectral or hyperspectral imagery? Perhaps one of the most important questions to address when forming your research question, it is important to understand the difference between multi- and hyperspectral data, as well as the pros and cons of each. These data contain spectral information regarding a target material: the absorbance and reflection of light from the target material creates a unique signature and contains detailed information about the material’s physical properties. Acting as a spectral “fingerprint”, this signature can be used to identify the target material from other materials that may be present in the scene. Multispectral imagery contains spectral information for a small number of bands across the electromagnetic spectrum, such as red, green, blue, near-infrared, and short-wave infrared. Hyperspectral imagery has a much greater number of bands (a higher spectral resolution), and can therefore reveal much more detail in a target’s spectral signature. However, this increased spectral resolution means that hyperspectral data is much more complex, which can require increased computational ability and additional processing steps in order to handle such large and information-dense datasets. Understanding which type of data will be most beneficial for your research is key, and will determine many of the further steps in your methodology. The availability of different multi- or hyperspectral data will also be important when determining your research plan.
Do I have the necessary resources and equipment to acquire my own data? Acquiring your own data is not impossible, but it does require sufficient resources and equipment. If this is a new field for you, make sure to research what is necessary in order to acquire the data you require. This does not just include things such as physical equipment, sensors, etc., but also other resources such as time, qualified personnel, funding, “know-how”, acquisition/processing software, and many others. Read existing literature and previous studies, and if possible, reach out to established groups that may be able to offer advice or instruction. Be sure to allow for test runs of data acquisition, so you can work through problems that may arise without compromising your data acquisition schedule and data products.
Do I have funding to potentially purchase imagery from private companies/repositories? Will preexisting data work for my research question? If you will not be acquiring the data yourself, you will need to access previously existing sources of data. While there are free/low cost options available such as the USGS EarthExplorer database, there are also private companies that have imagery/data products available for purchase. Be sure to allocate funds for purchasing data, as prices can vary depending on the provider. However, when using pre-existing data, there may be some limitations. Data may not be available for certain locations, or at certain times of the day/month/year. Depending on the sensor platform, it may not be possible for imagery to be acquired every day. For example, the Landsat-8 satellite has an 8-day repeating orbit which means the most frequent imaging of a specific area of interest will happen every 8 days. If you were to need daily coverage, this data may not be sufficient. Therefore, if you need to rely on pre-existing data sources, you may have to alter your research question or methods, or make use of whatever data is available.
Additional Environmental Factors
There are additional considerations to keep in mind, especially if you are going to acquire the data yourself. These can be general environmental factors or attributes that are more specific to your field site/species of interest. If there is no previous experience at the research site or with the species of interest, talk to others who have more experience in these areas or research the area/species in order to have better knowledge to . Examples of such considerations are given below:
Time of year? Certain times of the year or growing season might provide better opportunities for identifying a species of interest. Consider the unique characteristics of your species of interest: for example, does it grow or bloom at a different time than other vegetation species in the community? Distinctly colored flowers or substantial growth could be a key feature for identifying the species in imagery, so it would be beneficial to acquired imagery during this time when these features are most visible. Other factors such as snow cover, how many hours of sunlight, or seasonal flooding would also contribute to determining the best time of year to acquire imagery.
Unique qualities of species of interest: growth characteristics, vegetation communities? Understanding the growth characteristics and nature of the vegetation communities at your site might help determine when/how to acquire data. For example, invasive Phragmites typically grows in dense monotypic stands that are visually and spatially distinct from surrounding vegetation in the hyperspectral imagery. Other species might grow more intertwined, which can make differentiating them in imagery more difficult and could require special analysis techniques. Other qualities could help differentiate the species of interest, one example being height. Phragmites can grow very tall, and can reach heights up to 6 m (20 ft) which can set it apart from the surrounding vegetation. By using ancillary datasets, such as LiDAR (Light Detection and Ranging) which can measure the height at the top of a Phragmites stand, it is possible to add another method of identifying the species of interest.
Ancillary datasets relevant to the field site? As mentioned in the previous point, ancillary datasets such as LiDAR mean canopy height products might be beneficial for identifying and distinguishing your species of interest. Are there any preexisting datasets that could be useful for your analysis? This could include previous vegetation surveys, aerial photos, LiDAR products, weather/climate records, etc.
Site-specific characteristics? Is there some unique characteristic(s) for the study site that might influence when/how you collect data? It could be related to accessibility, or unique biological features. Be sure to think about any possible advantages or disadvantages to your site that could impact data acquisition. You may have insight from previous trips, but if not there might be other people who have been there that could provide some extra information.
Ground Truth Data
Depending on your research question, and if you’re acquiring the spectral imagery yourself or from another source, you might need to collect ground truth data in order to train and verify the imagery. For example, the GNSS ground truth data set described here was collected to support the target detection methodology used in this post’s companion article. Collecting ground truth data can be a great opportunity to build familiarity with your field site and gain a more intimate knowledge of the vegetation communities and characteristics that can help during the analysis and interpretation stages of research. Assuming your study requires the acquisition of ground truth data, here are some considerations to keep in mind:
What kind of ground truth data is required for the study? There are different ways of collecting ground truth data that will depend on the research question or study parameters. Examples include conducting thorough vegetation surveys using a grid system, detailed leaf-level spectroscopy measurements of species of interest, or using GNSS to measure positions that meet certain conditions. For example, the Phragmites study at Îles-de-Boucherville National Park required a ground truth data set that consisted of positional data that were either collected for Phragmites or non-Phragmites vegetation in order to train and verify the target detection algorithm.
How accessible is the area? Are there any restrictions? Is your site accessible by car/walking/cycling/kayaking etc.? This will determine many logistical factors, such as how often you would be able access the site for data collection, what equipment you can bring, the area you can cover, and more. Be sure to note if there are any restrictions to access: for example, certain areas of Îles-de-Boucherville National Park were not accessible as they were areas of active restoration, and there was private land that was not accessible for data collection. There was also a restriction on vehicles for some areas of the park, so ground sampling was conducted via bicycle and walking which meant ground data was collected over a period of two weeks in order to cover the park extent. Knowing the methods of access will help determine the expected time-frame for data collection.
Do I have the necessary permissions to collect data? When choosing your study site, be sure to be aware of any permissions you might need in order to access the site for research purposes. You may need to contact private landowners or official governing bodies in order to get permission, and potentially submit necessary forms or paperwork. Have everything in order well before you plan on accessing the site so that there are no unpleasant surprises the day of.
Do I have access to the necessary equipment and resources to collect the data I need? Think about what equipment you need to acquire data, but also any supporting equipment that may not be as obvious. For example, do you require a wi-fi hotspot to use your equipment? If your equipment needs to be charged, are you able to charge things at the site or does charging need to take place off-site? Do you need to bring vegetation sampling grids of a specific size? If collecting spectroscopic contact measurements of leaves, will they be analyzed in the field via a portable lab or will you need the necessary equipment to transport samples back to the office? Other supporting equipment might include a tablet or notebook for recording data, method of transport (e.g. car, bicycle), and a camera for taking site photographs which can be beneficial as a visual reminder when going over data back in the office. This equipment list might grow after your first field session as you discover additional items that are necessary.
How to avoid bias? When planning your sampling scheme, be aware of potential sources of bias that could appear in the data. For example, Îles-de-Boucherville has an extensive network of trails throughout the park that provide easy access for sampling. However, sampling the areas that are only near the trails would introduce a bias in the ground control data, and many other areas of the park that are not near trails would not be captured in the ground truth data. Therefore, effort was made to go off-trail (with park staff permission) in order to collect data from the areas of the park that were less accessible from the trails. Additionally, the characteristics of Îles-de-Boucherville are different in the northern portion of the park (more trees, agricultural fields) and the southern portion of the park (more grassy areas, low shrubby vegetation). When determining the sampling scheme for ground control points, care was taken to ensure equal representation of points collected over the entire extent of the park rather than being biased to just the northern or southern portions of the park. Another aspect of the ground control points was to ensure an even split of points representing Phragmites and points representing non-Phragmites vegetation, in order to provide adequate training data to the target detection algorithm and also have adequate validation data to evaluate the target detection outputs. If the dataset had been biased either way (e.g. too many Phragmites points and not enough non-Phragmites points, or vice versa) it could have influenced the outcome of the target detection methodology.
Analysis and Interpretation
Once all the necessary data has been acquired, the next step is to analyze the data and interpret the results. Generally, once you have determined your research question during the initial planning stage, the appropriate analysis method(s) will be narrowed down and you should already have an idea of how to proceed. It is also important to have some idea of how you want to analyze your data before you begin data collection, as different methods may require different types of data. However, if you have less experience with these types of data it can seem like a daunting task to figure out the best methods for analyzing and interpreting the results in order to answer your research question (what software is needed, correct settings, processing steps, etc). It’s always a good idea to look at previous research (which should also be part of forming your research question/study goals), as it can help guide your methods and highlight current technology and practices for analyzing these kinds of data. For example, the companion article for this post describes the methodology in detail so that future research can build on it or modify it for their own specific research projects. Google Scholar is an excellent source for finding research articles that cover your topic, and there are many sources online that provide a wide range of software tutorials. There will still likely be an aspect of trial and error to your analysis, but this process can be beneficial in that it helps you better understand the data and processing steps. However, make sure you understand each step of the process and what it actually does rather than just picking settings because they seem like the “right” ones. The concepts behind such processing steps/settings might seem complex, but it is important to understand the underlying theory so you can accurately process and analyze the data, and understand what the end products actually represent.
Although the list of considerations discussed here is not exhaustive, it is a good place to start and can help you think critically about your research/study goals so that your time and resources are used efficiently. It is often said that experience is the best teacher: as you conduct your study it is likely that there will be some aspect of trial and error, through which you will acquire new knowledge that will benefit future work.