Paul Barber
ArborCarbon Pty Ltd, Murdoch University & Men of the Trees (WA)

Introduction

A wide range of platforms and sensors exist for monitoring vegetation in cities and beyond. These include but are not limited to satellites, fixed-wing planes, helicopters and unmanned aerial vehicles (UAVs or ‘drones’) and thermal infrared (TIR), multispectral (MS), hyperspectral (HS) and standard imagery (RGB). The spatial and spectral resolution is determined by the type of platform used, the height above ground level for acquisition, and the sensor or camera. There are many choices and some of them are freely available and others are costly. The challenge is to select the appropriate platform and sensor to successfully achieve your required objective within the available budget.

The traditional and typical approach to measuring and monitoring urban forest cover and condition involves teams of people, preferably qualified arborists or urban foresters, walking throughout cities and recording numerous attributes including height, crown width and the health on a scale of 0 (dead) to 5 (most healthy). The use of pen and paper has commonly been superseded by field computers with built-in GPS, and more recently by more sophisticated real-time software applications. These tools have ultimately been introduced to improve efficiencies and precision of urban forest monitoring and they have achieved this.

Software packages such as i-tree have enabled managers to utilise field-collected data and input this data into models that produce outputs on many urban forest attributes and services. As with all computer models they have benefits and limitations, and the accuracy of the outputs is dependent upon quality of the inputs and model.

ArborCarbon scientists have been using remote sensing tools for native forest monitoring since 2003 (Evans, Lyons et al. 2012, Evans, Stone et al. 2013) and for urban forest monitoring since 2009. This paper describes how different platforms and sensors have been used for precision urban forest monitoring projects throughout Australian cities and the link to human health.

Methods

Vegetation Cover and Condition

Normalised Difference Vegetation Index (NDVI) data from the Landsat 8 sensor was processed and analysed at a pixel resolution of 30m. NDVI is an algorithm using the visible (VIS) and near infrared (NIR) bands of the electromagnetic spectrum and provides an indication of the presence/absence of vegetation. Spectral radiance data was converted to Top of Atmosphere planetary reflectance using supplied reflectance coefficients, and additional radiometric corrections were applied to enable extraction of NDVI values between -1 and +1 (ArborCarbon 2016).

Four-band multispectral imagery was acquired over areas of interest from a fixed-wing plane. Imagery was acquired at an altitude to provide 40-50cm pixel resolution. The method and timing of acquisition and subsequent radiometric and geometric correction during processing produces a dense 3-D point cloud with resultant Digital Surface Model (DSM) and Digital Terrain Model (DTM). These are analysed to provide a vegetation cover layer stratified into four height-classes (<3m, 3-10m, 10-15m, >15m), enabling the differentiation between canopy and non-canopy. The baseline condition of vegetation was determined based upon an index combining the red and NIR bands, termed the Vegetation Condition Index (VCI). Previous research comparing this index with NDVI showed the VCI to be superior to NDVI for measuring native vegetation. Change in canopy condition was measured by comparing the VCI for each pixel in the baseline image to the subsequent image. Care was taken to ensure that geometric and radiometric variables contributing to unwanted noise in the data were minimised to allow for precise measurement of change in condition.

Surface Heat

Thermal Infrared (TIR) data was processed and analysed from the Landsat 8 sensor selected hottest days over the areas of interest within the satellite orbit cycle. This data is 100m-pixel resolution resampled to 30m. Raw data was converted into surface temperature data using a series of correction equations (ArborCarbon 2016). Mean surface temperatures were calculated for areas of interest such as suburbs and different land-use categories.

In order to identify differences in surface temperature between features of interest (i.e. trees, grass, roads, buildings) higher spatial resolution TIR data was acquired using a FLIR thermal imaging camera from a fixed-wing plane. One project over the City of Perth acquired data at 1m-pixel resolution at nighttime to measure the residual surface temperature within the CBD and surrounds. In a separate project covering the Resilient South region (Cities of Onkaparinga, Holdfast Bay, Marion and Mitcham) in South Australia, 285 square kilometres of data was acquired at 2.0m pixel resolution on one of the hottest days in 2016, exceeding a temperature of 39 degrees Celsius. Concurrent thermal imagery was acquired at ground level at selected locations throughout the region using a handheld FLIR thermal camera.

Childrens Health

Participants (n=300 children) in the Play Spaces and Environments for Children’s Physical Activity (PLAYCE) Study wore accelerometers over 7 days. Time spent outdoors was measured objectively using Radio Frequency Identification (RFID). Personal UVR exposure was objectively measured using polysulphone (PS) film mounted in small cardboard holders attached to a child’s shoulder. A bottom up (sky-view factor (SVF) and top-down (shade coverage from remote sensing imagery) measure of outdoor shade coverage was utilized. The SVF was measured using fisheye photography taken from outdoor play areas. Remote sensing imagery was used to calculate percent vegetation (tree canopy) coverage.

Results

Vegetation Cover and Condition

Satellite-derived NDVI over the City of Mitcham showed highest values in the rural, bushland and peri-urban areas and in pockets of vegetation within the urban area (Fig. 1A). Lowest values for NDVI were mostly confined to major roads and industrial areas located within the urban area. Satellite-derived NDVI is an index affected by both the cover and condition of vegetation and due to the coarse resolution of the data it is difficult to interpret any differences in condition of vegetation between sites. According to the satellite-derived TIR data the lowest surface temperatures are located in the rural forested areas with most of the urban areas high in temperature (Fig. 1B). There is an apparent inverse relationship between NDVI values and surface temperature with a clear urban heat island effect (UHIE) evident.

Using high-resolution multispectral imagery we derived a very precise geospatial dataset of vegetation cover stratified into height classes (Fig. 2) over entire local government areas and sites within. Using this dataset we can very quickly identify where high-density vegetation occurs, and where the largest trees occur. The geospatial nature of the dataset and stratification of height enables us to differentiate between canopy and non-canopy, and to very precisely provide baseline measures (e.g. % cover, m2, ha) of the vegetation and canopy cover across entire cities and any smaller region of interest within (e.g. LGA, parks, streets, commercial, residential etc.) (Fig. 3). Subsequent acquisition of imagery in carefully controlled conditions allows for very precise measure of change in condition to be undertaken, providing an ‘early-warning’ system. The parameters used are highly sensitive to changes in vigour and health of the canopy with any increase in condition showing as blue pixels, no or little change showing as white pixels, and loss in condition or complete death of trees showing as red pixels (Fig. 4).

Figure 1. Satellite-derived NDVI (A) and TIR (B) over the City of Mitcham. Values for NDVI are scaled from highest (green) through to lowest (red) and thermal temperatures are scaled from lowest (blue) to highest (red).

Figure 2. Height-stratified canopy cover layer overlaid onto the true-colour base image (A) and with the base layer removed showing only the height-stratified vegetation cover layer (B). All vegetation has been quantified and stratified into height classes as follows: 15m (red).

Figure 3. Canopy Cover of selected parks throughout the Local Government Area in 2014 and 2015, showing increases, little change and decreases in canopy cover. Names of parks withheld to maintain client confidentiality.

Figure 4. Change Detection Image derived from high-resolution multispectral images acquired in 2012 and 2013 showing increase (blue pixels), no change (white pixels) and decrease (red pixels) in vegetation condition.

Surface Heat

Surface heat can be derived from urban areas using a range of sensing devices. Over recent years we have acquired and analysed surface heat over urban areas in City of Perth and the Resilient South Region in South Australia. The choice of platform and sensor is largely dependent on the required resolution, the objectives of the project and the available budget. Satellite TIR imagery typically produces imagery at 30m pixels enabling the identification of large heat and cool islands in urban areas (Fig. 5A), whereas high-resolution (e.g. 1-2m pixels) airborne TIR can clearly differentiate subtle differences in surface temperature between individual buildings, trees, turf, impervious layers and other materials of interest (Fig. 5B). The higher resolution imagery enables more powerful analysis to be undertaken to determine differences in surface temperatures between lower and higher density developments, and the relationship with vegetation (Fig. 6).

Figure 5. A) Medium resolution (30m pixel) TIR image south of Adelaide city and B) high- resolution (2m pixel) TIR image south of Adelaide city acquired on a day where temperatures exceeded 39 degrees C.

 

Figure 6. High-resolution TIR image of the suburb of Aldinga Beach in SA showing large differences in surface temperature between lower density development with a high abundance of vegetation, and higher density development lacking vegetation cover.

Children’s Health

A subset of the childcare centres where children were monitored for physical activity and exposure to UV were selected. Each of these centres was delineated and these polygons overlaid onto the baseline high-resolution airborne true colour with (A) and without (B) the height-stratified vegetation cover layer. The height-stratified vegetation cover layer was clipped to each of the centre boundaries and a calculation of the area and percentage canopy cover (>3m in height) for each centre was determined. This measure was used as a proxy for shade and protection of children from harmful UV rays.

Figure 7. High-resolution TIR image of the suburb of Aldinga Beach in SA showing large differences in surface temperature between lower density development with a high abundance of vegetation, and higher density development lacking vegetation cover.

Discussion and Conclusion

Our research and work for numerous LGAs over the past seven years has proven that there is a place for remote sensing in precision urban forest management. Remote sensing technologies have traditionally been seen by LGA’s as expensive tools that provide information that is difficult to understand and interpret. This can certainly be the case, and the many choices of platform and sensor can be very daunting for people who are unfamiliar with the technology. It is highly important that careful consideration is given to selecting the correct combination of platform and sensor within the available budget. Equally important after this first step is to determine the best way to analyse and report on the data. Traditionally this has required the skills of experienced remote sensing scientists and expensive software. Recent developments in software, processing speeds, bandwidth and server space have helped to overcome many of these hurdles.

Datasets are freely available that will enable the calculation of greenness and surface temperatures across large areas of urban forest and surrounds. Such datasets will allow for the broad interpretation of greenness across the landscape and the relationship between this and surface temperatures. It is important, however, to understand the limitations of this data and realise that one single pixel of 30m in size may be a mixture of many different surfaces including trees, grass, concrete, water and soil. Likewise, the surface temperature data will be an average value for each pixel of each of these different materials. Another limitation of this freely available dataset is the restriction to the satellite orbit cycles (e.g. 16 days) and greater potential for interference from atmospheric conditions (i.e. cloud, moisture, smoke haze) than airborne systems.

Airborne systems will nearly always be more expensive than satellite due to mobilisation costs. It is therefore very important to ensure the objectives of the project are very clearly understood by those providing the data, and that they can and have previously successfully delivered on such projects. Advantages of airborne systems over satellite include the ability to select dates and times for acquisition of data for minimising atmospheric interference, and to modify spatial and spectral characteristics of the data by flying at different altitudes and changing sensors. We have clearly shown that high-resolution multispectral imagery can be used to very precisely map and measure height-stratified vegetation cover and also to monitor and detect subtle changes in the health of trees across the urban forest. This approach has many advantages over traditional ground-based approaches to urban forest measurement and monitoring. For example, ground-based surveys are much more labour intensive and therefore more expensive, are unlikely to be able to measure all vegetation due to access restrictions (e.g. residential property), and are inherently subjective and qualitative in nature. Remote sensing approaches should not been seen, however, as a total replacement for ground-based assessments as these assessments are important for detecting hazards in trees and therefore measuring risk, can more accurately identify species, and can enable the calculation of monetary values for trees. The two approaches should be viewed as complementary rather than one replacing the other.

Remotely sensed data, if used correctly, can be successfully used for the efficient establishment of key performance indicators (KPIs) in urban forest and green infrastructure management. Some examples of our use of such data has included the establishment of KPIs and trigger values for bushland condition based upon our condition index and canopy cover, the setting of targets for increase in canopy cover across entire municipalities and suburbs within, identifying areas suitable for planting and measuring the success of planting strategies, monitoring change in condition of treated and untreated (control) trees suffering from various tree health disorders, detecting the onset of decline in health of trees for early intervention, identifying trees that have been illegally removed, and relating decline/death of trees to causal factors (i.e. construction damage). Ultimately all of these actions result in improved efficiencies, cost-savings, and the conservation of urban forest canopy.

Many local government agencies are striving to develop management plans of varying names (i.e. urban forest, greening, liveable cities). In order to develop robust plans and set realistic and achievable targets it is very important to accurately determine the existing and trend in temporal change in canopy cover. Our assessments of canopy cover across some LGAs have shown a discrepancy of several percent between our data and the data presented from a modelling approach. One must therefore ask the question how reliable targets for increasing canopy cover by a particular percentage can be set and measured against. A great deal of focus has been placed on the planting of trees throughout the landscape to increase canopy cover. Let’s not forget that it is the existing mature canopy that is our most valuable and it is likely to take many years/decades for the planted trees to reach maturity, if at all. Heavy emphasis is also placed upon the measurement of canopy cover and area, however, I suggest we should be placing a great deal more emphasis on the change in health of our urban forest and consider how to manage this health in the most sustainable manner.

For many years I have spread the message about the need to conserve our urban forest and manage it much more sustainably. I often feel like I am ‘preaching to the converted.’ I have found the best way to reach out to the ‘unconverted’ is to make this message about money and health and link the trees to these two elements. The premise behind our approach is to help local governments reduce resources for management of the urban forest, allowing more funds to be freed up for projects that will conserve and increase the canopy cover and improve the health of our urban forest. I’m very grateful to have been involved in such exciting projects linking urban heat to green infrastructure, and I’m equally excited about the research into the link between children’s health and urban trees. I do strongly believe that an urban forest full of healthy trees will result in a community of healthy and happy people.

Acknowledgements

I would like to acknowledge Resilient South, the SA government, Cities of Onkaparinga, Mitcham, Marion, Holdfast Bay, Joondalup and the Town of Cambridge for their collaboration on data presented in this paper. I also acknowledge SpecTerra Services, Spatial Scientific and Dr Hayley Christian from the University of Western Australia.

References

  • ArborCarbon (2016). Airborne Thermal Imagery and Analysis 2016. Resilient South. : 74.
  • Evans, B., T. Lyons, P. Barber, C. Stone and G. Hardy (2012). “Enhancing a eucalypt crown condition indicator driven by high spatial and spectral resolution remote sensing imagery.” Journal of Applied Remote Sensing 6: 063605-063601-063605-063615.
  • Evans, B., C. Stone and P. Barber (2013). “Linking a decade of forest decline in the southwest of Western Australia to bioclimatic change.” Australian Forestry http://dx.doi.org/10.1080/00049158.2013.844055.