WP5 - Joint Airborne campaigns

Objectives: To execute joint airborne campaigns for atmospheric measurements and combined optical/laser RS

Airborne Technnologies operated aircraft carrying advanced laser scanners over selected study areas to reconstruct 3D shape of trees at the surface by recording the pulsed laser signals at very high resolution. The signal is analysed and transformed into spatially resolved distances between the exact GPS position of the aircraft and the land surface. A flight lasting less than one hour can provide, in this way, very detailed information on the size, height and structure of the trees, leading to an accurate assessment of its biomass and the amount of carbon.

Three main joint airborne campaigns were made during the , as detailed below.

Campaign I: Pellizzano, Italy
Hyperspectral data acquisition: 13 June 2013

Recent developments in LiDAR technology, combined to other available data sources (aerial photographs, aerial photo series by UAVs,…), are now allowing a precise mapping of mountain forest resource volumes and qualities. Integrating this technology will provide an innovative response to the challenges of a precise and robust knowledge on the available growing stocks.

The study area were located in Pellizzano (Trentino, Italy).

Airborne Laser Scanner (ALS) 
Airborne Laser Scanner (ALS) data were acquired between 7th to 9th of September 2012. The sensor used for the acquisition was a Riegl LMS-Q680i sensor mounted on an aircraft flying at a speed of about 100kts at an altitude of 660 m above the ground level. The scan frequency was 400kHz and up to 4 returns were recorded. The point density was of at least 10 points per square meter.

Hyperspectral data
Hyperspectral data were acquired on 13th of June 2013 with an AISA Eagle II sensor. 65 spectral bands were acquired between 400 nm and 990 nm. The spatial resolution of the acquired data was 1 m. 21 stripes were acquired in order to cover the whole study area. The minimum overlap among the stripes was 20%.

Field data
In this study three different kinds of field data have been used: i) plot measurements; ii) field data collection assisted by the remote sensing data; and iii) photointerpretation. 

1. Plot measurements
The Servizio Foreste e Fauna (Province Autonomous of Trentino) provided us the coordinates where to locate a total of 52 plots. The position of the center of the plot was located with a GPS receiver, ensuring the best possible accuracy of the position. All the trees visible from the center position of the plot were measured, until a maximum of 60 trees was reached. For each measured tree the species, two DBHs (along two orthogonal directions), the social position, the health status (i.e. dead or alive), the inclination of the tree and the coordinates respect to the center of the plot (azimuth and range) were recorded. Among these trees, three trees were selected and the height was measured. These three trees were selected in order to be easily recognizable in the remote sensing images (e.g., isolated trees).
The equipment provided by Airborne Technologies are the following:

ALS raw data (standard format *.LAS - *.ALL e *.TXT)
DSM e DTM (derived from ALS data): format ASC (gridascii) AND XYZ,
RGB orto- images : 1:10.000
Hyperspectral data: hyperspectral raw data (format *.raw) + header files (format *.hdr) + time stamps and start of line (*.nav); Spectral range: 400‐970nm

2. Field data collection assisted by the remote sensing data
A specific field data collection was carried out in order to increase the number of samples to use in the classification. This field work was specifically done to collect samples of that species that were underrepresented in the relascopic measurements. An operator with a laptop on which there were visualized the remote sensing data (hyperspectral, orthoimages and ALS) and the shapefile of the ITCs went in the field. He assigned to some ITCs the corresponding species. This was done in areas where the matching among the ITCs and the field trees inspected was surely correct. The samples collected with this method were used only in the classification.

Figure 1. Example of field plot overlaid over the canopy height model (CHM). In blue the position of the PGS receiver. In yellow the points from which the tree positions were measured. In red and orange the field measured trees.

3. Photo-interpretation
To increase the training set to use for the classification of the tree species some ground reference points were added using photointerpretation. The species information was added to some individual tree crowns (ITC) delineated on the ALS data, according to the photointerpretation of the orthoimages of the analyzed area. In particular this was used in order to increase the number of samples of Alnusviridis. These samples were used only in the classification.


Campaign II: Hainich National Park, Germany
Data aquisition: 15 - 20 July, 2014 + Data acquisition: 9 - 10 September 2014

We know that some microorganisms are able to make water freeze at temperatures warmer than any other inorganic particle. Such a propriety, combined with their small size and their ability to be transported airborne, could potentially impact the climate: the microorganisms get updrafted, reach a cloud normally too warm for certain precipitation phenomena to occur and affect it so that precipitation happens nevertheless.

These kind of microorganisms could potentially be a part of the answer to the old question: "does it rain where a forest is or does the forest sprout where rain falls?". Unfortunately, one of the main information that lacks from literature is the amount of microorganisms emitted by different land uses and that's why we tried to quantify them above the forest of Hainich in Thuringia, Germany. The forest configures itself as a potentially powerful emitter of such microorganisms and it gives us the chance, thanks also to a treetop walk, to study both those living on the leaves and those being carried up in the atmosphere.

The Hainich campaign aimed to obtain estimates of the emission rates on a broadleaf primeval forest of Fagus using a sequencing approach to estimate the abundance of total biological particles. The forest, located in the Hainich National Park (Thuringia, Germany) offered a unique opportunity being a large homogeneous fetch with an observational tower allowing to set sampling stations above the canopy top.

A FLUXNET eddy-covariance tower is located nearby the observational one and this permits the retrieval of high-quality and high-frequency meteorological data and CO2 fluxes. The sampling campaign lasted from 15 to 20 July 2014.

With the collaboration of Airborne Technologies Gmbh, we collected air both within and above the canopy with a mixture of ground stations and aircraft-carried probes to investigate this issue... 


Campaign II: Wampersdorf, Austria
Data acquisition: 20 August, 2014

The insurance industry reports a pronounced intensification, at the global level, of weather-related events such as droughts, windstorms and hailstorms.

As an efficient adaptation strategy, improved capacities can be built adopting innovative remote sensing tools to map vegetation damage spatial distribution, to quantify its intensity and its impact. New airborne LiDAR (Light Detection and Ranging) sensors provide high vertical resolution data, which are potentially suitable not only for forest canopies but also for shorter crop canopies monitoring purposes (e.g. corn - Zea mays L. - crop injury and lodging assessment).

To evaluate the potential of LiDAR metrics to map corn canopy height and hail defoliation, a flight campaign was organized in 2014 in Wampersdorf (Austria) in a cropland area affected by a hailstorm. Ground-truth observations were carried out at 16 plots, where defoliation was assessed both visually (observed range from 0% to 70%) and using a biophysical parameter-based method.

The objectives of this study were the following
- to investigate the possibility to use in-situ biophysical measurements (based on canopy height and canopy denseness) as a quantitative proxy for corn canopy defoliation (which is generally assessed by visual estimation);
- to test the ability of LiDAR data to retrieve canopy height of corn crops;
- to analyze the canopy density profile and the LiDAR return patterns in corn canopies with different hail defoliation rates;
- to investigate the ability of LiDAR models -based on both traditional and new metrics introduced in this study- to retrieve canopy denseness and canopy defoliation;
- to highlight the effect of adopting different LiDAR sampling point densities on the ability of ALS metrics to retrieve canopy height and canopy defoliation.

Study area, ground measurements, and field data processing
Canopy in-situ measurements were carried out between the 27th and 28th of August in corn fields with a different degree of damage. For model developing, 16 circular ground-truth plots (radius 5 m) were identified within the study area and the center of each plot was georeferenced with a Global Positioning System (GPS) receiver recording the coordinates of each plot on the ground surface .

To calibrate the LiDAR-based defoliation models, canopy inspection to visually estimate the defoliation in the 16 plots was carried out by researchers of Fondazione Edmund Mach (FEM), and a defoliation value (%) was assigned for each plot. Simultaneously, insurance inspectors of the HaegelVersicherung (HV) insurance acquired a parallel defoliation dataset at the 16 plots, which was used to validate the performance of the models

For the scope of this study, we investigated the possibility to use in-situ biophysical measurements (based on canopy height and canopy denseness) as quantitative proxy variables for corn canopy defoliation, which is generally assessed by the insurance companies’ inspectors using a visual estimation approach. Canopy denseness was firstly tested as a proxy for canopy defoliation. Secondly, a height-normalized indicator of canopy density anomalies due to hail defoliation was introduced. As Freeman et al. (2007) and Gao et al. (2013) observed that plant height is strongly correlated with plant biomass and leaf area index (LAI) at all growth stages, to estimate hail canopy defoliation we considered the ratio between the canopy height and the denseness which was defined as the Canopy Defoliation Index (CDI):

CDI = Hcanopy / Dpanel [1]

where Hcanopy is the average height of the 5 highest corn plants in the plot (measured in-situ) and Dpanel is the average plot canopy denseness (obtained by the SideLook image analysis). The rationale for using such CDI ratio is based on the fact that, if it is assumed that the ratio between canopy height and canopy denseness is rather constant for corn crops, significant anomalies in the CDI may be a much better indicator of hail defoliation than canopy denseness itself. 

LiDAR data acquisition, processing and metrics
LiDAR data were acquired on the 20th of August 2014 with a RIEGL LMS-Q680i sensor (laser pulse wavelength 1064 nm, laser repetition rate 100 kHz, vertical resolution 15cm), with an original (raw data) average point density on the 16 plots ranging equal to 42 points/m2 and with up to four returns recorded for each laser pulse. To obtain such a high average density, the acquisition along the same flight lines was repeated a few times.

Besides the traditional metrics, two new LiDAR metrics -which were expected to be more efficient, in particular for defoliation assessment- were introduced and tested in this work. The first one was named “Canopy Metric” (CM) and is an expression of the shape of the curve of the LiDAR returns distribution along the canopy height. The CM can be calculated as:

CM = [(hc - hr) / hr] * 100 / pr [2]


hc = maximum height value of all LiDAR returns intersecting the plot;
hr = the height corresponding to the maximum percentage value of all LiDAR returns (excluding the 0 - 10 and the 10 - 20 cm layers);
pr = the maximum percentage value of all LiDAR returns for the various 10 cm layers (excluding the 0 - 10 and the 10 - 20 cm layers).

As a first step, this study highlighted the possibility of using a quantitative defoliation proxy (the Canopy Defoliation Index, CDI) which is based on objective in-situ observations and not on visual estimation. The CDI combines the information of canopy height and denseness and is able to detect significant canopy anomalies due to hailstorms.
“Traditional” metrics commonly adopted for forest canopy studies (e.g. 95th percentile LiDAR height, 99th percentile LiDAR height) showed good performances for retrieving corn height information, which can be considered a very important parameter for productivity models, for storm damage mapping, and for drought damage assessment. Considering the fact that the metrics were applied to canopies with very heterogeneous densities -due to hail defoliation-, even better performances are expected in more dense and homogeneous canopies. The performance of the metrics were not shown to be significantly dependent on point density, so height monitoring appears to be possible with low densities (2-5 points/ m2), commonly used for DEM or forestry applications. This result is in accordance with the study of Vastaranta et al. (2013) focused on of Scots pine stands forest defoliation in Finand, and showing that the mapping accuracy of defoliation using ALS data was not overly sensitive to LiDAR point density, and also with the research of Kantola et al. (2013) who found that the nearest neighbor approach with random forest method for the classification of different defoliation intensity of Scots pine in Finland did not appear to be particularly sensitive to point density. More studies are needed to investigate the model suitability at larger scales and its economical feasibility within a monitoring system framework, which is of high potential interest for insurance companies and large-scale farming.

Very distinctive LiDAR return patterns were observed in corn canopies with different hail defoliation rates, according to their leaf denseness profiles. In particular, the information related to the return patterns of the upper part of the canopy showed to be particularly important. While the “traditional” metrics appeared to be not adequate for defoliation mapping, the Canopy Metric (which is based on such “top of the canopy” information) showed a good performance for retrieving canopy defoliation, although higher point densities (starting from 10-20 points/m2) were needed. Further studies are needed to quantify the economic feasibility of the adoption, at larger scales, of a monitoring system based on the proposed approaches for canopy height and canopy defoliation monitoring.