International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

INVENTORY AND MONITORING OF DANUBE FLOODPLAIN FORESTS BY INNOVATIVE REMOTE SENSING TECHNOLOGY – FIRST RESULTS FROM INMEIN PROJECT by

G. Király(1) and G. Brolly(1) (1)

Institute of Geomatics and Civil Engineering, University of West Hungary, Sopron, Hungary ([email protected])

ABSTRACT A methodological overview on Airborne Laser Scanning- (ALS) or LiDAR-based inventory of forests is given. The airborne survey of the Szigetköz / Csallóköz area between the dikes on the Hungarian side and the channel in the Slovakian side has been carried out recently in the frame of a Hungary-Slovakia Cross-border Co-operation Programme 2007–2013, called INMEIN, Innovative methods for monitoring and inventory of Danube floodplain forests based on 3D technologies of remote sensing (HUSK/1101/1.2.1/0141). Our first results on forests investigation are shown on these inundation area determined by the Danube and its tributaries. Keywords: ALS, LiDAR, pre-processing, floodplain, forests, parameter-extraction

1

INTRODUCTION

During the period from 1977 to 1992 the construction of water dam (WD) Gabcíkovo was carried out. The technical implementation caused fall of groundwater levels, which consequently put floodplain forest ecosystems into risk. A hydro-technical project has been implemented to ensure regulatory groundwater regime based on the biological needs of the floodplain ecosystems. State of these ecosystems has been regularly observed in the framework of Complex monitoring of the WD’s environmental impacts since 1992 when WD Gabcíkovo was launched into operation. Data and methods of remote sensing have been used for evaluation of floodplain forests state since 1992. Firstly, infrared aerial images were used, but multispectral ones have been exploited since 2008. Despite the fact that it is an effective method for forest state monitoring, its deficiency is that it does not allow assessing interior structure and quantitative characteristics of a forest stand. This is particularly important when assessing status and development of indigenous forest communities, especially softwood floodplain forests. Recent developments in the field of remote sensing address this deficiency by exploitation of 3-dimensional methods of data collecting and interpretation, such as digital photogrammetry and airborne laser scanning (ALS or LiDAR). However, their application is limited due to lack of technical and software infrastructure and underdeveloped 3-D data processing methodologies. Due to complexity of the problem, it is necessary to address the issues in cooperation. That was the main consideration to submit a proposal jointly.

2

THE INMEIN PROJECT

The INMEIN project is a Hungarian-Slovakian Cross-border Co-operation project, entitled ‘Innovative methods for monitoring and inventory of Danube floodplain forests based on 3D technologies of remote sensing’ (HUSK/1101/1.2.1/0141). The project is a two years duration project which was started on the 1st of September, 2012. There are three partners in the project; the Lead Partner (LP) is the National Forest Centre (NLC), Zvolen, the cross-border partner is the Hungarian Forest Research Institute (ERTI) and the Partner2 is the University of West Hungary (NYME). The main aim of the project is to develop innovative methods and support a common approach in forest monitoring in Danube floodplain area influenced by the hydro-power plant at Gabcikovo according to mutual Király, Brolly, Inventory and monitoring of Danube floodplain forests…

1

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

obligatory cross-border commitments of Hungary and Slovakia. Data and methods of remote sensing has been used for evaluation of floodplain forests state since 1992, but innovative remote sensing technology of ALS still has not got effective and operational data-processing methodologies. The planning of the aerial survey has been carried out at a very problematic period. There is a quite new act concerning surveying and mapping (2012/XLVI. Act: about surveying and mapping work) which was announced on 22nd of May, 2012, and came into force on the 5th day, 27/05/2012. This act says that the airborne remote sensing would be (is) regulated by order. This new regulation is the following: 399/2012. (XII. 20.) Government Decree on Regulation about Airborne Remote Sensing Permission and the Usage of Remote Sensing Data, which came into force on 1st of January, 2013 theoretically, but the important parts came into force on 1st of March, 2013. There have been several interpretation problem of this new order, and all the issues have to be addressed cross-border, so the planning and permissioning of the flight has been prolonged significantly. So finally the permission was issued on the 2nd of September, and the leaves-on aerial survey has been carried out on the 8th of September, 2013.

3

ALS DATA-PROCESSING

The data from the INMEIN project is not yet available currently, because of the above mentioned reason. We had the chance to process the data from another HU-SK project, called DUREFLOOD to start our investigations. The map of the overlapping area of the two projects can be seen on Figure 1.

Figure 1 – The overlapping area of the INMEIN (red) and DUREFLOOD (blue) Projects.

The dataset was acquired in the frame of the DUREFLOOD project on the 5th of March, 2013. The applied sensor was a Leica ALS70 airborne laser scanner. The data processing contains several steps, we split this process to the followings: 1. geometric check of the data (data pre-processing) 2. creation of digital surface models, DSM, DTM and nDSM (data-processing) 3. forest parameters extractions (data-post-processing)

3.1

Pre-processing

One of the first pre-processing steps of the airborne laser scanning (ALS) data is to calculate the point density of the flight lines. The next figure (Figure 2) shows the point density:

Király, Brolly, Inventory and monitoring of Danube floodplain forests…

2

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

Figure 2 – The point density of the overlapping DUREFLOOD flight lines

It can be seen, that the stripe-overlaps are just less, than 50%; resulting a very inhomogeneous point density from approximately 7 points/m2 on the simply covered area up to ~100 points/m2 on the double (or even triple) covered areas. All the four stripes have double boundaries, because of a new ‘split-beam technology’ feature of this scanner. There can be some differences between the strips, and it is very important to reduce them, which is quite similar to relative orientation of photogrammetry (Figure 3). Mostly because of this reason, some special strips crossing all the others, called cross-strips are recorded at the end of the flight.

Figure 3 – Relative orientation of the strips (Kager, 2003)

The differences between the stripes (relative orientation) can be seen on Figure 4. These differences are not really high, but the differences between the 2 middle stripes are bigger, than acceptable.

Király, Brolly, Inventory and monitoring of Danube floodplain forests…

3

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

Figure 4 – Height differences between the stripes

The differences between the two split-beams has also been checked, and turned out, there is a small but systematic deviations between the two channels (Figure 5).

Figure 5 – Relative orientation of the two channels inside one strip

All the above mentioned differences are not really high, but induced some waves on the Digital Surface Model (DSM) (Figure 6).

Figure 6 – The undulating shaded Digital Surface Model

Király, Brolly, Inventory and monitoring of Danube floodplain forests…

4

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

These discrepancies resulted a real need for reprocessing the stripes, with a better calibration, and relative orientation values. The newly processed stripes served as a basis for the further investigations.

3.2

DSM, DTM and nDSM extraction

The Digital Terrain Model (DTM) has been created with the progressive TIN method (Axelsson, 2000) implemented in TerraSolid software. The Digital Surface Model (DSM) has been interpolated with the matching parabolid algorithm (Király et. al., 2012). The difference between the Digital Surface Modell and Digital Terrain Modell results the so called normalised Digital Surface Model (nDSM), sometimes referred to as Canopy Height Modell (CHM), according to the following equation:

nDSM = DSM − DTM A sample from the DTM, DSM and nDSM can be seen in Figure 7.

Figure 7 – Samples from the DTM, DSM and nDSM

3.3

Forest parameters extraction

The forest parameter extraction is currently divided into two groups: The area- or plot-based methods, and the individual tree based methods. Both types of the methods are presented on a sample compartment ‘Győrzámoly 9D’ (see Figure 8)

Figure 8 – The sample compartment (Győrzámoly 9D) on the shaded nDSM. Green areas are below 0.5 m

Király, Brolly, Inventory and monitoring of Danube floodplain forests…

5

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

The reference data about this compartment has been acquired from the National Forest Service, Sopron. The data is based on forest planning in year 2006. The current data has been generated from the planning data through growth-modelling (e.g. age, diameter at breast height (DBH) and heights (H)), or directly from forest inspection, as one part of this compartment was cut in 2009. The compartment’s main tree species is Pannonian poplar (Populus x euramericana cv. ‘Pannónia’), 40% old (27 years, 41 cm DBH, 39 m H), 39% young (4 years, 3 cm DBH, 3 m H), and there is 13% Salix alba cv ‘Bédai’ (27 years, 34 cm DBH, 22 m H) and 8% Populus x canescens (27 years, 31 cm DBH, 24 m H). So the compartment is very complex, which makes it a good demonstrative example. The area based methods generally can determine forest stand type characteristics. The simplest way to estimate the height of a forest stand is to investigate the distribution of the points’ relative heights (Figure 9).

Figure 9 – The distribution of the points’ relative height in the sample compartment (Győrzámoly 9D)

There are several parameters can be extracted from the 3D point cloud. The different percentiles (e.g. 90%) are commonly used height estimation method (Næsset, 2002). The previously calculated nDSM can also be used directly for forest stand height determination. The following figure shows the frequencies diagram of the nDSM (Figure 10).

Figure 10 – The distribution of the cells in the nDSM (Győrzámoly 9D)

The nDSM can be used directly to calculate the ‘Volume’ (or the growing space) of the stand, and the average height based on this (Király, 2007). The crown closure can be determined using the nDSM, as well. As the current Hungarian Forest Law (2009/XXXVII. Act) determines the minimum average forest height in 2 meters, we can easily calculate the area of the nDSM below (red) and above (green) 2 metres (Figure 11). Király, Brolly, Inventory and monitoring of Danube floodplain forests…

6

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

Figure 11 – The classified nDSM of the compartment ‘Győrzámoly 9D’

The individual tree based approaches are presented through the tree-top determination using the inverse watershed modeling (Gougeon 1995, Király 1998). In these works, very high resolution aerial images with adequate lighting conditions created the basis. Treating the image as an inverse digital terrain model, where the brightest pixels are the lowest points, the individual tree crowns are the watersheds, which can easily be separated by well-known algorithms (Jenson and Domingue 1988). These image-based methods were adapted successfully on airborne laser scanning data, where the top of the trees were determined with the Digital Surface Model (Hyyppä et al. (1999)).

Figure 11 – The determined treetops (white points) in the compartment ‘Győrzámoly 9D’, and a detail

The extracted tree tops can be used for single tree height determination afterwards by determining the height of the nDSM at the location of these sink-points. The tree-top determination can serve as a basis for the number of tree in the compartments, as well (Figure 11).

4

RESULTS AND DISCUSSION

The distribution of the relative heights of the point (Figure 9) already has foreshadowed the need for subcompartments delineation. The average stand height for the whole compartment based on the nDSM is 6.1 m

Király, Brolly, Inventory and monitoring of Danube floodplain forests…

7

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

against the reference 21.6 m values. Additionally the classified nDSM (Figure 11) resulted a 51.8% forestcover against the 80% reference value. So the separation of the compartment was really needed. The 2 m contour-line of the nDSM has been used, to separate the old stand and the reforestation area. The ‘no data’ area has been delineated for the area covered by water, as no (or very few) points returned from here. The delineated three sub-compartments and their area can be seen on the following figure.

Name Area (m2) Old stand 40 709 Reforestation 28 665 Water 4 103 Sum 73 477 Figure 12 – The automatically separated sub-compartments of Győrzámoly 9D

The recalculated forest cover for the old sub-stand is 79.7% which is extremely close to the 80% reference. The recalculated forest cover for the reforested sub-stand is 12.1%, which is much-much lower, that the reference value (100%), but here are some administrative restrictions for the forest inspectors, and the 2 m threshold is a bit high for such a young reforestation. The 90% percentile of the points’ relative height is 31.6 m, which is still lower, that the reference data measured in 2006 (34 m). The statistics of the single tree based relative heights of the tree-tops can be seen on the following figure.

Minimum: Maximum: Mean: St. Dev.: Count: Area: n:

Old Reforesstand tation Unit 2.09 0.20 m 37.59 23.41 m 22.36 2.01 m 6.51 2.48 m 1559 1668 piece 4.071 2.867 ha 383 582 piece/ha

Figure 13 – The height statistics of the tree-tops in the sub-compartments of ‘Győrzámoly 9D’

The achieved average tree numbers/ha are far higher, than the reference values, 383 (reference is 227), and 582 (reference 100), but here again, the data of 100 doesn’t represent the real number unfortunately.

5

CONCLUSIONS

The airborne laser scanning is a very promising tool to estimate our forest stand parameters with much higher accuracy, and – which is more important – with spatially explicitly. However the reference data currently

Király, Brolly, Inventory and monitoring of Danube floodplain forests…

8

International conference Catchment processes in regional hydrology: experiments, modeling and predictions in Carpathian drainage basins Sopron, Hungary, 28 October 2013

available have to be treated with care, the forest growth models are often out of date. So these kinds of investigations can help to introduce better field surveys, and more accurate forest stand description in the near future. This sample compartment also shows the close relationship between forest gaps and very low sites. The biggest gap in the old sub-compartment is situated in the deepest pit (155.3 m HAE) in the whole compartment (see Figure 7). But the investigation on these correspondences between the forest stands and the forest sites will be carried out in another research paper later.

6

AKNOLEDGMENTS

We thank the Károly Róbert College for the processing, and specially for the re-processing the DUREFLOOD data, and the National Forestry Service for the reference data on the compartment.

7

REFERENCES

Axelsson, P. (2000) DEM Generation from Laser Scanner Data Using Adaptive TIN Models. International Archives of Photogrammetry and Remote Sensing. Vol. 33, Part B4., pp 110-117. Hyyppä, J., and Inkinen, M., (1999) Detecting and estimating attributes for single trees using laser scanner. The Photogrammetric Journal of Finland. 16 (2): 27-43. Jenson, S. K., and Domingue, J. O., (1988) Extracting Topographic Structure from Digital Elevation Data for Geographic Information System Analysis. In: Photogrammetric Engineering and Remote Sensing. 54 (11): 1593-1600 Kager, H. (2003) Simultaneous Georeferencing of Aerial Laser Scanner Strips. Österreichische Zeitschrift für Vermessung und Geoinformation. 91, 4, 235 - 242 Király, G., (1998) Multiscale images in forestry. In: ISPRS International Archieves of Photogrammetry and remote sensing. XXXII, 7. ISPRS Comission VII Symposium on „Resource and Environmental monitoring.” Budapest, Hungary. 1-4 September, 1998. 365-369 Király Géza (2007): A távérzékelés erdészeti alkalmazása. Doktori (PhD) értekezés, Sopron, 2007., p 121. Király, G., Brolly, G., Burai, P. (2012) Tree Height and Species Estimation Methods for Airborne Laser Scanning in a Forest Reserve. In: Full Proceedings of SilviLaser 2012; 12th International Conference on LiDAR Applications for Assessing Forest Ecosystems, Vancouver, BC, Canada, 2012.09.16-19. pp 260-270. Næsset, E., (2002) Predicting forest stand characteristics with airborne scanning laser using a practical twostage procedure and field data. Remote Sensing of Environment 80, pp.88-99.

Király, Brolly, Inventory and monitoring of Danube floodplain forests…

9