Logo 2016 Dragon 3 Final Results Symposium

Conference Agenda

Overview and details of the sessions and sub-session of this conference. Please select a date or session to show only sub-sessions at that day or location. Please select a single sub-session for detailed view (with abstracts and downloads if available).

 
Session Overview
Session
E2: ID.10666 Forest DRAGON 3
Time:
Tuesday, 05/Jul/2016:
9:00am - 10:00am

Session Chair: Christiana Cornelia Schmullius
Session Chair: Yong Pang
Workshop: Forest Mapping & Retrievals
Location: Sun Moon Room -1, 5.5 Floor, Junyi Dynasty Hotel

Show help for 'Increase or decrease the abstract text size'
Presentations
Oral presentation

Forest DRAGON-3: Decadal trends of Northeastern Forests in China from Earth Observation Synergy

Christiana Cornelia Schmullius1, J. Balling1, P. Schratz1, C. Thiel1, M. Santoro2, U. Wegmuller2, Zengyuan Li3, P. Yong3

1Friedrich-Schiller-University Jena, Germany; 2Gamma Remote Sensing, Gümlingen, Switzerland; 3Chinese Academy of Forestry, P.R. China;

Forest DRAGON 3 has investigated the synergy of Earth Observation products to derive information on decadal trends of forest in Northeast China. 20 years of spaceborne, airborne and in situ data are available for an unprecedented analysis of forest ecosystems in a region that experienced substantial changes in terms of forest cover since 1990.

Previous and new forest GSV products have been investigated to derive information on trends in vegetation cover and carbon storage of forests in Northeast China. The launch of the Sentinel satellites has allowed the development of new forest mapping algorithms based on synergy of radar and optical measurements. The Chinese partner institutions have focussed on algorithms adaptation to Eastern Russia and Continental Southeast Asia, and to do comparative studies on forest changes with the neighbouring China forests. For this purpose, new algorithms for forest mapping in tropical forests and change detection in comparative sites have been developed.

Results include large area growing-stock volume (GSV) mapping for NE and SW China and change monitoring in the Guangxi region. Growing-stock volume and land cover products and their change products were used for plausibility check, but large discrepancies were discovered which lead to new research questions. The GSV products for 2005 and 2010 with a geometric resolution of 1 km were derived using the BIOMASAR algorithm, which is independent from in-situ reference data, utilizing ENVISAT ASAR data. Biomass modelling has been undertaken involving multisource data and forest type classification involving the new generation of Sentinel-1 and -2 data and a new approach to retrieve tree geometry from TLS-data to be included in radar modelling has been developed.


Oral presentation

Forest Aboveground Biomass Estimation and Change Analysis in Greater Mekong Subregion and Russian Siberia

Yong Pang1, Zengyuan Li1, Guoqing Sun2, Zhiyu Zhang2, Christiane Schmullius3, Erxue Chen1, Xianlin Qin1, Bingxiang Tan1, Qingwang Liu1, Ying Guo1, Lina Bai1, Shiming Li1, Xin Tian1

1Chinese Academy of Forestry, China, People's Republic of; 2Chinese Academy of Sciences; 3Friedrich-Schiller-University;

Forests play a vital role in sustainable development and provide a range of economic, social and environmental benefits, including essential ecosystem services such as climate change mitigation and adaptation. We summarized works in forest coverage in the Greater Mekong Subregion (GMS), and forest aboveground biomass estimation in GMS and Russian Siberia (RuS). Both regions are rich in forest resources. These mapping and estimation works were based on multiple sources remote sensing data and some field measurements. The forests of GMS are undergoing rapid changes due to human activities. The forests of RuS are mainly suffering changes due to natural disturbances like fire. The overall changes and detail spatial patterns are analyzed.

Pang-Forest Aboveground Biomass Estimation and Change Analysis_Cn_version.pdf

Poster

Coupling Lidar Plots Data and Geometric-optical Approach to Estimate Canopy Cover for Greater Khingan Forest Mapping and Management

Chengyan Gu, Xin Tian, Zengyuan Li, Min Yan, Chunmei Li, Zongtao Han, Wenwu Fan

Research Institute of Forest Resource Information Techniques, Chinese Academy of Forestry, Dongxiaofu No.1, Xiangshan Rd, Beijing, 100091,P.R. China;

Achieving precision estimation of forest canopy cover (CC), which is for understanding and evaluating the impact of human activity and climate change on forest ecosystems has very important significance. Forest CC usually relates to the ratio of the vertical projection area within the forest canopy and the forest area and its spatial distribution is a key index for forest productivity and decomposition rates. Compared to field mensuration, LiDAR provides a relatively cost-efficient solution to estimate CC; however, wall-to-wall airborne LiDAR surveys over large areas remain expensive. In this study, we elaborate a large-area, fine-scale (30 m) mapping solution to estimate CC across Greater Khingan forests by integrating LiDAR data, and Landsat imagery, requiring a geometric-optical modeling technique. First, 1 m CHM derived from the LiDAR data was used to guide the selection of appropriate training samples by verifying the structural differences between trees and low vegetation. Second, spectral mixture analysis (SMA) was used to extract the image endmembers and generate fraction images. Third, LiDAR data were used to calibrate the inverted geometric-optical model by adjusting the model's three key fractional inputs: sunlit crown, sunlit background, and shade fraction, based upon the SMA derived images. Results show that the pixels CC of white birch as the dominant tree species showed N (0.57, 0.08) in spatial distribution and that of larch as the dominant tree species showed N(0.6,0.1).

Gu-Coupling Lidar Plots Data and Geometric-optical Approach_Cn_version.pdf

Poster

Plausibility Check Of The BIOMASAR GSV Change Map Of 2005 - 2010 Using Existing Remote Sensing Products In NE China

Patrick Johann Schratz1, Johannes Balling1, Christiane Schmullius1, Maurizio Santoro2, Yong Pang3, Xianlin Qin3

1Friedrich-Schiller-University, Jena, Germany; 2Gamma Remote Sensing, Gümligen, Switzerland; 3Research Institute of Forest Resource and Information Technology - Chinese Academy of Forestry, Beijing, China;

In times of risen awareness and knowledge regarding global climate change the quantification and validation of complex carbon storage in biomass, as a major climatic driver, moved in the center of interdisciplinary research. To tackle this issue, GSV biomass maps were produced within ESA Forest-DRAGON 2 for NE China. The GSV products for 2005 and 2010 with a geometric resolution of 1 km were derived using the BIOMASAR algorithm, which is independent from in-situ reference data, utilizing ENVISAT ASAR data. GSV refers to the stem/bore volume of living trees for all living species, including bark and excluding branches and stumps. The ongoing Forest DRAGON 3 project is focusing on the validation of the created BIOMASAR GSV maps.

To verify the BIOMASAR GSV estimations, various remote sensing products were compared with the BIOMASAR GSV change map of 2005 to 2010. By only taking GSV decreases smaller than 20 [m³/ha] and GSV increases greater than 20 [m³/ha] as “real” GSV changes into account, variations of GSV values caused by the BIOMASAR algorithm respectively the methodology were avoided.

All pixels of the BIOMASAR GSV change map that showed decreasing GSV values and fire activity based on MODIS burned area product from 2005 to 2010 and Landsat based burned area product of the DRAGON - Forest Fires & Emissions project from 2005 to 2010 were marked as plausible decreases of GSV. By using remote sensing based vegetation products, land cover products and the CCI MERIS time series it was also possible to detect plausible GSV decreases and increases. As vegetation products MODIS VCF from 2005 to 2010 and MODIS NDVI for the same time period were integrated. By subtracting aggregated values for the years 2005 – 2007 and 2008 – 2010 trends for MODIS VCF and MODIS NDVI were calculated. These trends were then compared with the BIOMASAR GSV change map and marked as plausible in cases of accordance. The land cover products LC CCI of 2005 and 2010 were reclassified into "forest" and "non-forest" classes regarding a reclassification key. Only pixels that stayed "forest" and pixels that changed from "non-forest" to the class "forest" were marked as plausible increases of GSV based on the LC CCI change map. Moreover pixels that stayed "non-forest" and pixels that changed from "forest" to "non-forest" from 2005 to 2010 within the LC products were marked as plausible GSV decreases. By calculating trends of the CCI MERIS time series from 2005 to 2010 and compare them with the BIOMASAR GSV change map, plausible GSV changes were selected.

All marked pixels of plausible GSV losses or gains explained by the mentioned remote sensing products were used to create a map of agreement. The results show that the majority of the BIOMASAR GSV changes are plausible based on the information of existing remote sensing products.

This work was undertaken during the Young Scientist Exchange between ESA and NRSCC at Chinese Academy for Forestry (CAF) in China, to contribute to the tasks of Phase 1 and 2 of the ESA-NRSCC/CAF Project "Forest-DRAGON 3”.

Schratz-Plausibility Check Of The BIOMASAR GSV Change Map Of 2005_Cn_version.pdf

Poster

Pure And Mixed Pixel Analysis For The BIOMASAR Growing Stock Volume Maps 2005 And 2010 In NE China

Johannes Balling1, Patrick Johann Schratz1, John Truckenbrodt1, Christiane Schmullius1, Maurizio Santoro2, Yong Pang3

1Friedrich-Schiller-University Jena, Germany; 2Gamma Remote Sensing, Gümligen, Switzerland; 3Research Institute of Forest Resource and Information Technology - Chinese Academy of Forestry, Beijing, China;

In the last decades biomass estimation utilizing remote sensing data has moved into the center of attention of interdisciplinary research for the quantification and validation of complex carbon storage. The latter is the most complex part of the carbon cycle and has therefore a huge impact in the analysis of the global climate change.

To help in this issue, the ENVISAT ASAR based BIOMASAR maps for 2005 and 2010 with a spatial resolution of 1 km give information about Growing Stock Volume (GSV) in Northeast China. GSV refers to the stem/bore volume of living trees for all living species, including bark and excluding branches and stumps. The BIOMASAR maps were produced within ESA Forest-DRAGON 2. The follow up Forest-DRAGON 3 project focuses on the validation of these products.

To help addressing this goal, a pure- and mixed pixel analysis for forest areas of the BIOMASAR maps was undertaken. To detect pure-and mixed forest pixels, the National Land Cover Dataset (NLDC) of China for the years 2000 and 2010 as well as ESA's Climate Change Initiative (CCI) Land Cover products for the years 2005 and 2010 were utilized. All further analysis steps were applied to both land cover products and compared in the end.

For the pure pixel analysis at first a reclassification of the land cover products focusing on forest classes was applied. Afterwards continuous forest pixels according to the land cover products for both years 2005 and 2010 were selected. To ensure purity, the Pixel-Purity-Index (PPI) was calculated on all selected pixels for both years using Landsat 5 Surface Reflectance images and followed by a rejection of all pixels not being 100 percent pure. Furthermore, the NDVI was calculated and applied with the condition of being >= 0.2 to be free of cloud cover problems in the PPI calculation. By this, it was also ensured that pure non-forest areas (due to possible misclassification of the LC products) are not falsely included into further analysis steps.

Mixed pixel analysis was performed with the same datasets as used for the pure pixel investigation. Mixed forest classes were set up in 25 percent intervals and are based on the percentage share of non-forest classes, e.g. “0-25% shrub & 75-100% forest”. For both land cover datasets (NLCD & CCI) an a priori reclassification was undertaken featuring the three main classes shrub, crop and grassland, ending up with 12 mixed-forest classes in total (3 types á 4 classes). Statistics for each class were calculated over the whole study area on the base of the 1km BIOMASAR grid.

First results of the pure pixel analysis on a small subset reveal some interesting areas with very low GSV values for both years. However, it has to be further investigated whether this error relies on the BIOMASAR maps or the land cover products.

The mixed pixel analysis showed similar reasonable results for the mixed classes based on grassland and crop with a decreasing GSV value when featuring a higher percentage of non-forest elements. However, the mixed-forest classes based on shrub showed unexpected behaviors with a non-consistent trend for an increase of non-forest area. For NLCD, no mixed-forest class based on shrubs could be taken out as the number of detected pixels was too small.

A further investigation of the undertaken analysis is needed with a special focus on the detected unreasonable results of both pure-and mixed pixel analysis.

This work was undertaken during the Young Scientist Exchange between ESA and NRSCC at Chinese Academy for Forestry (CAF) in China, to contribute to the tasks of Phase 1 and 2 of the ESA-NRSCC/CAF Project "Forest-DRAGON 3”.

Balling-Pure And Mixed Pixel Analysis For The BIOMASAR Growing Stock Volume Maps 2005 And 2010_Cn_version.pdf

Poster

Forest investigation from combined ALS and TLS methods.

Zhichao Wang1, Christiane Schmullius1, Jussi Baade2

1Department of Earth Observation, Friedrich-Schiller-University, Germany; 2Department of Physical Geography ,Friedrich-Schiller-University, Germany;

Since laser scanning technology are increasingly used in forest investigation, aerial laser scanning (ALS) and terrestrial laser scanning(TLS) take two major parts for large area and spot level researches. The common method for biomass investigation using ALS data is to extract a nDSM grid as well as CHM from point cloud. Then this grid can be fitted with different empirical models. Here come out a consequent issue that how to validate results from the ALS processing. Many peers’ works followed the exactly way the same as inventory as ground truth data for calibrating ALS processing which biomass value were derived from a certain combination of tree height(H), diameter at breast height(DBH). Regardless the advantages of this approach such as maintaining stabile standard to obtain biomass for a long time period, if we simply compare to the point density (usually 10 -1000 per sq. meter depending on types of sensor vehicle) of ALS data with the point density for measurement of H, DBH which require theoretically less than 4 points per sample spot corresponding to ALS data, we could find a huge data loss and it’s obviously improper to use such as kind of ground truth retrieved from few points to calibrate result retrieved form higher magnitude on points. From the viewpoint of magnitude, TLS methods are naturally compatible to provide ground truth for ALS biomass. However, the scales of object for ALS and TLS are clearly different, we found that TLS are typically used for single tree modelling in related researches which

requires certain circumstance of field champion such as low planting density, equal distribution of scan positions, full cover for target trees, leaf-off situation and so on. Thus point clouds with high quality can be provide for single modelling algorithm like quantitative structure modelling(QSM). Comparing to the ideal measurement condition for single tree modelling, we took TLS scanning covering the same area as ALS coverage and got a point cloud with poor quality for single tree modelling. How to compensate those poor quality point cloud is first focus in this study.

Our research site is in the forest near Stadtroda, Germany(N50°50’23”,E11°40’57”), 11 TLS scan positions were set up and covered roughly 30000sq.meter’s forest. For the consideration of full cover for each trees within certain area. A smaller core area 4700sq.m were clipped from whole dataset. 195 trees were detected in this point cloud with 3 types. 31 trees which locate in the boundary area with missing part of stem or canopy were throw out of dataset. 32 trees canopy are mixed with adjacent trees and the rest 132 trees can be well segmented to individual point cloud. Those 164 trees were reconstructed using QSM to get 3D model. As what was stated as a first focus in this research, the quality of point cloud is the significant drawback for single tree modelling. After a serious processing which included de-noising, classifying and clustering, usually for each tree 20%-30% points from the pick were filtering out of data set. On another hand, we could only model a tree form bottom until the part at average 70%-80% height. This is the significant difference between this research and those studying specific to single tree. In order to compensate those data loss, allometric methods had to be used to regress the missing part of each tree. There are 2 types of structure missing, on stem and on branches. For stem missing, we used two methods to predict the missing volume. The first one is to calculate using regression based on stem curve and the second one is to consider that diameter of stem in the upper parts of the tree shrinks themselves very fast, more close to the tree pick, more less difference with morphological characteristics for branches, therefore we could sort stem loss into branched loss. For branches loss, we presented a new method based on the hypothesis that a certain branch can hold up a certain crown volume. Thus we can compensate the branch loss with reconstructed braches plus ratio between whole crown volume and the reconstruct crown volume. With the help of serious compensation, we cloud finally determine branch volume, stem volume and together as wood biomass.

With known ground truth from TLS measurement, linking ALS and TLS approaches can be second focus in this research. We have two psychical points clouds. One is taken from airplane with average above ground height 820m, another is taken form TLS. For the TLS data, besides for single tree reconstruction, we treated it as an ultra-high quality ALS data. We did full process the same as what we did for ALS data. the purpose of this part studying is to mix up the boundary between ALS and TLS measurement and investigate the potential to apply some TLS method into ALS data because ALS is the convicted approach both in large area investigation and even in plot level with the fast developing of UAV in the future.

This work contributes to the tasks of Phase 1 and 2 of the ESA-NRSCC/CAF Project "Forest-DRAGON 3”.

Wang-Forest investigation from combined ALS and TLS methods_Cn_version.pdf
Wang-Forest investigation from combined ALS and TLS methods_ppt_present.pdf


 
Contact and Legal Notice · Contact Address:
Conference: 2016 Dragon 3 Final Results Symposium
Conference Software - ConfTool Pro 2.6.94
© 2001 - 2016 by H. Weinreich, Hamburg, Germany