Home Articles Monitoring trends in tree crown condition using remote sensing data

Monitoring trends in tree crown condition using remote sensing data

Introduction
Quantifying rates of forest cover change is important for improved carbon accounting and climate change modeling, management of forestry and agricultural resources, and biodiversity monitoring (Hansen 2010). Monitoring forest change is also important for global climate because of both deforestation emissions and altered land-atmosphere exchanges of energy, water and carbon. For any mountainous country, trees are important resources and monitoring individual tree crown detection and delineation has become steadily important for watershed management. The role of remote sensing in watershed management is concentered on the supply of data and information for the planning operation of the infrastructure (Meijerink 1988). In view of this, monitoring crown condition trend in Lorpa watershed is considered important for watershed management and process of implementing afforestation practices.

Traditional methods for assessing forest structure like field inventories and aerial photo interpretation are intrinsically limited in providing spatially continuous information over a large area (Tiede 2012). The field study on forest area is laborious and expensive. The field investigation on wide area of mountainous country is nearly impossible. Utilizing remote sensing data reduces the amount of field sampling and hence information gathering becomes more cost-effective (Wang, Le 2004). Remote sensing has been a valuable source of information over the course of past few decades in mapping and monitoring forest. It provides a cost-effective tool to help forest managers better understand forest characteristics, such as forest area, locations, and species, even down to the level of characterizing individual trees (Ke 2009).Modern forest management requires that forest resources be efficiently managed, not only for timber production, but also for such purposes as maintaining biodiversity and meeting wildlife, environmental, and recreational needs (Wang 2004). Recently, very high resolution satellites have created both new opportunities and challenges in remote sensing and earth observation information extraction. Remotely sensed, high spatial resolution images have great potential in assisting vegetation in watershed level with the wide use of object based approaches in remote sensing image analysis, individual tree crown delineation. Across the environmental, urban, surveying, planning and forestry sectors automated feature extraction and classification is the main use of OBIA in remote sensing (Blaschke 2010; lang 2008). Remote sensing based algorithms help in identifying separable tree crowns using high spatial resolution images. Most of these algorithms are developed for the low altitude or plan land areas tree crown delineation (Pinz, 1989; Gougeon, 1995; Pollock, 1996; Brandtberg& Walter, 1998; Larsen &Rudemo, 1998; Culvenor, 2002; Olofsson, 2002; Pouliot et al., 2002). Nepal has been conducting forest resources survey using field survey. With the help of visual interpretation of aerial photographs and field survey, the forest resources survey office conducted the first forest inventory during the period 1963–1967. Using aerial photography from 1953–1958 and 1963–1964 forest, crop, grass, urban, water, badly eroded and barrenland cover was identified (Acharya 2011). The present study aims to express methods to use object based image analysis by the potential of high resolution data to automatically delineate individual tree crowns and followed by change in crown shape.

Material and Methods

Study area
The study area, Lorpa watershed, is situated mid-northern part of Jumla district, Nepal (Figure 1). The total area of study area is 13 km2.


Figure 1. Location of study area

Data used
Very high spatial resolution (VHSR) satellite image from QuickBird satellite of Nov 18, 2006 and IKONOS-2 of Oct 24, 2011 was used to generate tree crown. Sub-meter resolution QuickBird image has Panchromatic at 0.62m spatial resolution and Multispectral spectral Range 450-545 nm (blue), 466-620 nm (green), 590-710 nm (red), 715-918 nm (near infrared) at 2.62m spatial resolution. IKONOS image has Panchromatic at 0.82m spatial resolution and Multispectral spectral range 445–516 nm (blue), 506–595 nm (green), 632–698 nm (red), 757–853 nm (near infrared) 4m Spatial Resolution.

Tree crown delineation
First of all, the IKONOS and QuickBird images were orthorectified with the cubic convolution method into UTM, Zone 44 based on SRTM DEM and RPC files using ERDAS Image 2011. Since higher resolution panchromatic layer is single band and multispectral spectral band has the lower resolutions, pan sharpening process of merging high-resolution panchromatic and lower resolution multispectral imagery was used to produce high resolution of rectified multispectral imagery. After that using convolution 3×3 low-pass filters were used to reduce local variation, remove noise, enhance tree features and improve the quality of the analyzed satellite images. eCognition object based analysis (OBIA) was used for the study toward delineation tree crown. Initially, resolution merged inserted in eCognition and replacing each pixel by the average of the square area of the matrix centered on the pixel convolution filter was done using gauss blur for the blue band. Next step was converting image pixel level to object level according to their attributes, such as shape, color and relative position by typically segmentation steps. Presently eCognition developer provides eight type’s of different approaches to segmentation, ranging from very simple algorithms, such as chessboard and quad tree-based segmentation, to highly sophisticated methods such as multi resolution segmentation and contrast filter segmentation. Present study used ‘‘multi resolution segmentation’’ algorithm which consecutively merges pixels or existing image objects that essentially identifies single image objects of one pixel in size and merges them with their neighbors, based on relative homogeneity criteria (Blaschkeet al.2001). Multi resolution segmentations are those groups of similar pixel values which merge the homogeneous areas into larger objects and heterogeneous areas in smaller ones (Baatz et al. 2006).

In the analysis stage, the Normalized Difference Vegetation Index (NDVI) image was created using customised features applying the formula: NDVI = (RED – IR)/(RED + IR ). The vegetation indices like the NDVI is a standardised index allowing generating an image displaying greenness (relative biomass). Index values can range from -1.0 to 1.0, normally an area containing a healthy vegetation canopy will tend to positive values (say 0.3 to 0.8) while clouds and snow fields will be characterised by negative values of this index. Also in general within the image object average highest NDVI pixels value represents tree crown top points. In the following stage, from the recognised detected local maxima or tree top location image objects enlarged defined minimum NDVI value to merge all direct neighboring image objects according to the parameters using grow region (Figure 2). Lastly, tree crown object border smoothed by the pixel-based binary morphology operations, removed unwanted smaller image objects and exported tree crown further analysis.


Figure 2: a-satellite image, b- image with local maxima/tree top and c-distance from the tree top

Forest fragmentation quantification
For forest fragmentation, large and contiguous forests were divided into smaller patches due to natural processes and development activities. To calculate the forest fragmentation (Riiters et al. 2000) two variables that are derived from neighborhood-relations between forest- and non-forest pixels were identified. First, delineated tree crown merged and defined as forest area for fragmentation exploration. Landscape Fragmentation Tool of ArcGIS was used to study forest fragmentation and edge effects. Vogt et al. 2007 developed the Landscape Fragmentation Tool which is based on a procedure to map forest fragmentation – a model to quantify forest fragmentation from raster forest cover maps and the procedures developed are believed to be more consistent and reliable than previous methods for mapping fragmentation at the landscape level. Forested areas were classified into 4 main categories of increasing disturbance core, perforated, edge and patch based on a key metric called edge width (Vogt et al. 2007). The edge width indicates the distance within which other land covers can degrade the forest. Based on available national research, an edge width of 100 meters was used core, perforated, edge and patch.

Results and discussion
The tree crown map of 2006 and 2011 are given in Figure 3. In 2006, number of tree crown was 41689 and in 2011 it declined by 47121. Although forest cover area declined between 2006 and 2011, new plantation took place in those areas. The change from in high tree crown count of 125-150 to 75-100 in the higher elevations depicts degree of impact on forest cover. In the south eastern parts of the watershed, certain degree of improvement in the tree crown count has been observed. This kind of high resolution based spatial and temporal information on tree vegetation will be of immense use in watershed conservation efforts and as reliable indicator for watershed monitoring. The loss/increase of a few number of trees does not necessarily result in to change/shift in crown density category. The present study did not attempt to delineate canopy density to a finer number of classes considering the degree of field calibration required. However, taking advantage of very high resolution satellite used in the current study, we have attempted to delineate tree crowns. The tree crown map provides information on occurrence of trees both in forest and non-forest categories. The number of tree crown present over one hectare are counted, after delineating them using digital image analysis methods. It may also be noted that a given ‘Tree Crown’ delineated may represent more than one tree. Therefore, the tree crown map produced strongly provides information on the occurrence of trees and direction of change in tree count. The map thus provides additional information on changes on tree vegetation apart from deforestation and changes in broad crown density categories.


Figure 3: Tree crown top and number of crowns per hectare

Based on crown covered statistic of 2006, forested lands dominated with 634 ha and in 2011 significant forest area declined by 494 ha contributing to 37% of the watershed. Majority of forests is present over upper elevation ranges and far away from the existing settlement. Considering the degree of deforestation found in the watershed, an attempt has been made to assess how spatially the deforestation affected the forest condition by forest fragmentation analysis. The analysis of spatial distribution of forest and non-forest areas has shown that only 47 ha are covered by core forests. The loss of these core forests by 2011 signifies the accessibility and vulnerability of forests for both anthropogenic and natural change factors (Table 2). In addition the forest patches which do not have any forest cover close by 100 meter distance also have increased from 54 ha to 121 ha during 2006-2011. The reduction of area under perforated and edge condition and increase in forest patches reveals the edge effects of deforestation.


Table 1: Changes tree crown between 2006 and 2011

 


Table 2: Forestfragmentation between 2006 and 2011

Conclusions
This paper aims to demonstrate the potentiality of remote sensing technology as a cost effective alternative to field based assessment approach in delineating individual tree crown of Lorpa watershed between 2006 and 2011. GIS analysis has taken into consideration for forest fragmentation identification. This study considered the possibility of using IKONOS and QuickBird images to estimate tree crown in watershed level. Temporal tree crown analysis has shown that in 2006 tree crown was 41689 and in 2011 declined by 47121.

Acknowledgements I acknowledged the Asian Development Bank (ADB) for supporting the initiative by funding the High Mountain Agribusiness and Livelihood Project (HIMALI). Thanks are also offered to Govinda Joshi and Madhav Dhakal for extensive field information for the results validation. My gratitude goes especially Mr. Basanta Shrestha, Regional Program Manager Dr. MSR Murthy Theme Leader, Geospatial Solutions, Dr. Rajan Kotru Team Leader, InFEWS, Ecosystem Services and MR Birendra Bajracharya, Programme Coordinator – Regional Database Initiative for the encouragement and support extended to bring out this report. Cordial thanks goes to our project partners Local Initiatives for Biodiversity, Research and Development (LI-BIRD) for a good and pleasant cooperation Local Initiatives for Biodiversity, Research and Development (LI-BIRD). Special thanks go to Hammad Gilani for the valuable support during the tree crown analysis.

Reference:

  • Hansen, Matthew C., Stephen V. Stehman, and Peter V. Potapov. “Quantification of global gross forest cover loss.” Proceedings of the National Academy of Sciences 107(19), 8650-8655.(2010).
  • Lehmann, Eric A., Jeremy F. Wallace, Peter A. Caccetta, Suzanne L. Furby, and Katherine Zdunic. “Forest cover trends from time series Landsat data for the Australian continent.” International Journal of Applied Earth Observation and Geoinformation (2012).
  • Tiede, Dirk, Stefan Lang, and Bernhard Maier. “Transferability of a tree-crown delineation approach using region-specific segmentation.” In Proceedings of XIII Brazilian Remote Sensing Symposium, edited by José Carlos NevesEpiphanio, LênioSoaresGalvão, and Leila Maria Garcia Fonseca, pp. 21-26. (2007).
  • Ke, Yinghai, and Lindi J. Quackenbush. “Individual tree crown detection and delineation from high spatial resolution imagery using active contour and hill-climbing methods.” In Proceedings of 2009 ASPRS Annual Conference, American Society of Photogrammetry and Remote Sensing, pp. 9-13. 2009.
  • Wang, Le, Peng Gong, and Gregory S. Biging. “Individual tree-crown delineation and treetop detection in high-spatial-resolution aerial imagery.”Photogrammetric Engineering and Remote Sensing, 70(3), 351-358. (2004).
  • Acharya, K. P., R. B. Dangi, and M. Acharya. “Understanding forest degradation in Nepal.” Unasylva 62, no. 2 (2011): 238.
  • Blaschke, T. ( 2010) Object based image analysis for remote sing. ISPRS journal of Photogrammetry and remote sensing, 65, 3-16
  • Lang, S. (2008) Object-based image analysis for remote sensing application: modeling reality- dealing with complexity analysis for remote sensing.
  • Blaschke, T., Hay, G. J. Object-oriented image analysis and scale-space: theory and methods for modeling and evaluating multiscale landscape structure. International Archives of Photogrammetry and Remote Sensing, 2001, 34(4), 22-29.
  • Baatz, M.; Arini, N.; Schäpe, A.; Binnig, G.; Linssen, B. Objectoriented image analysis for high content screening: Detailed quantification of cells and sub cellular structures with the Cellenger software. Cytometry Part A 2006, 69(7), 652-658.