Remote sensing provides a powerful synoptic overview of an area. These data can be used to identify different land covers and in some circumstances it is able to identify specific habitats, fragmentation and connectivity of habitats. However, to identify a habitat it must be resolvable, both spatially and spectrally. Spatial resolution should be fine enough to capture the extent of the sub-units to be considered and spectral resolution should be sufficient to identify one land cover (and ideally habitat) from another. Therefore, the data collected must be appropriate to the habitat to be observed, mapped or monitored. This therefore poses a conundrum that balances availability and cost against effectiveness.

An additional aspect to this conundrum is the complexity of data analysis that may be required to extract the detail and information required. Change analysis is simply comparing data sets but co-registration of the data can, in this case, be an important issue. Therefore the source of the data (and accuracy) can be a particular factor (Figure 1). This is ideally described in the associated metadata.  For most local authorities or environmental organisations, straight forward analysis of pre-processed Sentinel-2 satellite data (e.g. for groups with access to the Pan-government agreement, supplied by JNCC) or drone data may also be conducted with minimal outlay with a robustness of processing proportional to the time and expertise available (and the availability of supporting ground data). 

Figure 1b

Co-registration performed to re-align the two scenes, making change analysis possible.

Figure 1b

Co-registration performed to re-align the two scenes, making change analysis possible.

Nevertheless, there are likely to be limits on the level of analysis that can be achieved locally; these limits may be on the accuracy or the detail of analysis and the nature of the data (especially when they extend into hyperspectral domains and to wavelengths beyond the visible and near-infrared). More complex questions, relating to specific vegetation species or vegetation health, habitat identification (i.e. anything to do with the complex assessment of vegetation) and fragmentation / connectivity will be enhanced by spectral information, typically provided by multispectral and hyperspectral data. Most of this discussion considers passive systems that use the sun to illuminate the target area but I note that airborne LiDAR is also very useful for identifying heights, widths and texture and may be essential for some analysis (Fig. 2).

Figure 2a

LiDAR Digital Surface Model (DSM) to indentify elevated features

.

Figure 2b

Identification of individual trees using LiDAR point cloud density and texture.

Similarly, radar also offers specific advantages, not least the ability of some bands to penetrate cloud cover; however, radar also presents specific data issues such as shading and interference (Figure 3).

Figure 3b

Attempt at classification of canopy cover using a backscatter threshold method.

Figure 3b

Attempt at classification of canopy cover using a backscatter threshold method.

The data are acquired from an instrument mounted on a platform. Typically, the level of spatial detail depends on how close the instrument is to the ground and the level of data quality (signal-to-noise, stability and dynamic range of the data) is related to the size of the instrument and the capabilities of the platform on which it is mounted (Figure 4).

Figure 4

Comparison of different remote sensing platforms, including some examples and the associated spatial resolution per pixel.

A camera (typically used on airborne platforms) has three bands and can allow very basic spectral analysis. Multispectral and hyperspectral systems allow further investigation into biochemical and structural composition and therefore may allow indication of health and different vegetation types (Fig. 5). Sentinel-2 is a system with two satellites. It provides data with a resampled pixel size of 10m (in the VNIR). It has 13 spectral bands (with between 10 – 66nm (Full Width Half Maximum; FWHM) depending on the band and the satellite) so each spatial pixel has multispectral information as well.

Figure 5

Comparison between true-colour, multispectral and hyperspectral technology, including the number of bands and examples of instruments.

Another feature of airborne platforms (drones and manned aircraft) is that you choose when they collect the data. There are four aspects to timing: the time of the day, the day in the season and the potential for repeat data within the day or season and the time needed to collect data over a defined area.  The time in the day when data are collected may be critical to match a specific a tidal state, solar noon, when the area is not obscured by cloud or other critical timings. The day in the season may be relevant to gain data in leaf-off conditions, just prior to senescence or when an area is free of snow cover. Repeat data may be important for change detection or to catch a specific event when the date of that event is unknown. Satellites not in geostationary orbit have a specific time when they pass over an area (e.g. Sentinel-2 is 10:30 hrs, local time) but they do have a regular frequency and so a temporal sequence may be analysed (assuming the area of interest is free of cloud). Satellites typically are also very well suited to collect data very rapidly over relatively large areas, when aircraft may take hours and drones may take days or weeks. The advantage of collecting data at the same time is that the atmospheric and ground condition will be the same and therefore allow a comparison. If, for example, there are different illumination or cloud conditions or a period of rain between when data are collected than any comparison may be less robust.

• Drones will typically collect photographs from a light camera, data from a small spectral scanner or simple LiDAR. This is suitable for small areas and relatively simple analysis that can feed into a GIS. Data from drones is particularly suitable for small areas where complex spectral information is not required and they can be deployed for specific dates and times.
• Many survey airborne systems collect photographs using a survey camera and / or LiDAR data. LiDAR is particularly useful for heights and texture (e.g. to identify buildings and trees). Aircraft can be deployed for specific dates and times. (Air)
• A few survey airborne systems collect photographs using a survey camera and also collect multi-spectral or hyperspectral data and / or LiDAR data. Multispectral analysis allows for the identification of vegetation and simple discrimination of different land cover types. Hyperspectral analysis allows for more complex analysis with specific biochemicals being identified and some minerals; this can allow indications of stress and disease to be observed and some automated species identification^*. Multispectral and hyperspectral typical resolve features on the ground ~1m in size. They can be deployed for specific dates and times. (Air+)
• A few satellite-borne systems collect relatively coarse resolution (spatial) multispectral data (at 10s of m resolution, i.e. 10m for Sentinel-2). Most of these data are free. The coarse resolution means that only relatively large features are individually resolved. However, these data are typically available for large areas (e.g. 290km+ across). Data from satellites is particularly suitable where large areas are being studied and typically have a regular opportunity for data collection (based on their orbit). It can be chance where on a data panel your site sits, i.e. the middle or edge. (Sat)
• A few satellite-borne systems collect fine resolution (spatial) multispectral data (at 1-2m resolution) [1] [2] – but charge more for it (e.g. Pleiades, 2m resolution but only 4 spectral bands available for a 20km wide area, i.e. the swath)! Data from satellites is particularly suitable where large areas are being studied and typically have a regular opportunity for data collection (based on their orbit). It can be chance where on a data panel your site sits, i.e. the middle or edge. (Sat+)

The pros and cons of different remote sensing options for a range of common conservation issues.

Table 1

A discrete study site, e.g. an area of parkland or a woodland.

The pros and cons of different remote sensing options for a range of common conservation issues

Table 1

A discrete study site, e.g. an area of parkland or a woodland.

Example areas:

* Partial identification of vegetation species can be achieved using fine resolution images (e.g. photographs) and an expert to manually interpret them.

** Partial identification of vegetation health can be achieved using fine resolution images (e.g. photographs) and an expert to manually interpret them. However, this does not account for any details relating to derived biochemical or cellular structure.

!! Needs manual interpretation by an expert for some analysis

Table 2

A catchment or county-scale area, e.g. a river catchment

Table 2

A catchment or county-scale area, e.g. a river catchment

Generally, data collected from drones coupled with expert interpretation of those data will give you some information at a local scale (individual sites). With the right sensors, training, software and knowledge most organisations could successfully develop an effective monitoring programme. However, there are limits on what the sensor on the drone will be able to provide you and the effort and comparability of collecting data from a large area may be limited, i.e. if two months of good weather are required then the areas may not be comparable or some areas simply not covered. Drone piloting requires training. River catchment scale analysis is most efficiently collected by satellite systems and these data may be freely provided via JNSS or purchased from commercial providers. Intermediate scale data with the potential to collect at both local and river catchment scales can be collected from manned survey aircraft.

Simple manual / visual interpretation of images may be achieved using experts in the issues being investigated. Processing and data analytics to conduct simple analysis may be achieved using commercial and open source GIS and image processing software. The free software tends to be harder to use and need more training but where software is newly purchased the software supplier may also provide limited training. While simple studies of land cover mapping and change detection using those maps (assuming all are perfectly registered!) may be taught relatively easily, assuming the relevant software is available. Complex analysis of vegetation and change detection (using multispectral, hyperspectral and LiDAR systems) may need additional expertise.

In these instances, bespoke external expert analysis may be required. For example, for tracking the development of a tree disease across a river catchment airborne hyperspectral data may be required to obtain sufficient spectral detail and an appropriate spatial resolution to determine individual trees (Fig. 6). These data would then need to be processed to account for the terrain (and therefore ensure that the resampled pixels are correctly mapped on the true surface) and the atmosphere (to account for attenuation of the signal by scattering and absorption by the atmosphere) and then analysed to identify characterised spectra for healthy and unhealthy trees and compare them to a previous data set. The last stage would require the data sets to be co-registered to ensure that the same trees are being identified. The final essential stage of any classification is an assessment of accuracy. This may be achieved via other remotely sensed data (not used directly in the analysis) but is best achieved with a carefully constructed non-bias ground survey that samples small areas (in proportion to the whole river catchment) representative of the variation present in the whole area.

Figure 6

Example of ash dieback analysis along power lines using airborne hyperspectral imagery to locate individual trees, as well as the species, height and health. All these can be combined to produce a final risk map.

In summary,

• What can be identified on the ground is very dependent on the resolution of the data available. Fine spatial resolution data (either acquired from an aircraft or via expensive commercial satellite systems) can be best for identification of features a few metres diameter in size (e.g. individual tree crowns) but typically costs more than data collected from coarser resolution satellite systems where the data cost are paid by the EU, ESA or the UK government. 
• Larger areas typically are most efficiently collected using satellite platforms but these more cost effective systems have a coarser spatial resolution than platforms closer to the ground.
• Once the scale of features to be observed has been identified, additional information may be extracted from spectral channels. Subtle details relating to plant cell structure and biochemicals involved in photosynthesis can be extracted from hyperspectral data. Currently hyperspectral systems only extend to airborne platforms. However, there are individual satellite missions due to launch in 2020+ that are hyperspectral systems. Data from these missions may be subject to ESA data access restrictions and may require purchase and will not have the spatial resolution of airborne platforms.
• Currently, only airborne systems have the capability to collect fine spatial and spectral resolution data over a large local area in a timely manner. These data remain the most efficient way of identifying the position and health of terrestrial vegetation (and associated habitats).

[1] The GeoEye-1 satellite is able to collect images with a ground resolution of 0.41 metres in the panchromatic or black and white mode (and multispectral imagery at 1.65 metre resolution). 

[2] Pleiades is able to collect images with a ground resolution of 0.50 metres in the panchromatic or black and white mode (and multispectral imagery at 2 metre resolution).

ARTICLES

Starting an Internship During Lockdown
Which Remote Sensing Approach is Right for Me?
Life as an Intern at 2Excel Geo