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.10676 Forest Modeling
Time:
Tuesday, 05/Jul/2016:
3:00pm - 4:00pm

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 type and above ground biomass estimation based on Sentinel-2A and WorldView-2 data – evaluation of predictor and data suitability

Andreas Fritz1, Fabian Enßle1, Xiaoli Zhang2, Barbara Koch1

1Albert-Ludwigs University, Chair of Remote Sensing and Landscape Information Systems (FeLis), Germany; 2Beijing Forestry University, Institute of Forestry, P.R. China;

The European Space Agency successfully launched the Sentinel-2A satellite on 23 June 2015. With its radiometric and spectral properties it offers a broad range of applications. One of the prospected opportunities of these sensor characteristics is an advancement in forest mapping. Especially the SWIR channels at a spatial resolution of 20m are of importance for forest ecosystem monitoring. Even though, radiometric resolution of the Sentinel-2 sensor is eminent, to assess stand specific characteristics WorldView-2 imagery with its very high geometric resolution is incorporated in the analyses in addition. The present study analyses the two earth observation sensors regarding their capability of modelling above ground biomass and forest density. Our research is carried out at two demonstration sites. One is located in south-western Germany (region Karlsruhe) and the other is located in southern China in Jiangle County (Province Fujian). With this two geographical significant different sites, we can draw conclusion regarding the performance in both, moderate and sub-tropical climate and vegetation conditions. Due to the new availability of Sentinel-2A-data we used late spring data, which we believe shows an acceptable state of vegetation in middle Europe. A set of spectral and spatial predictors are computed from both, Senitel-2A and WorldView-2 data. WordView2 data with a geometric resolution of 0.5m in the pan-channel and 2m for the multi-spectral bands is used to calculate textural predictors with several window sizes. Window sizes in the range of 3*3 pixels to 21*21 pixels are computed to cover the full range of the canopy sizes of mature forest stands. Textural predictors of first and second order (grey-level-co-occurrence matrix) are calculated and then used within a feature selection algorithm. Additionally common spectral predictors from WorldView-2 and Sentinel-2 data such as all relevant bands, band ratios and NDVI are integrated in the analyses. To examine the most important predictors, a predictor selection algorithm is applied to the data and the entire predictor set of more than 80 predictors is used to find most important ones. Out of the original set only the 10 most important predictors are then further analysed. With a minimum subset of three predictors, the classification and biomass estimation will be performed. With a stepwise increase of the amount of predictors, the influence on the classification and regression by the number of integrated predictors has been examined. The analyses further clarifies what the optimal number of predictors with respect to accuracy and computational effort is. Predictor selection is done with the Boruta package in R (Kursa & Rudnicki 2010), whereas classification and regression is computed with random forest and support vector regression. Prior the classification and regression a tuning of parameters is done by a repetitive model selection (100 runs), based on the .632 bootstrapping. Both are implemented in the caret R package. To account for the variability in the data set 100 independent runs are performed. Within each run 80 percent of the data is used for training and the 20 percent are used for an independent validation. With the subset of original predictors mapping of tree types and of above ground biomass is performed. Furthermore, the analyses is carried out on the two data sources separately to draw conclusion regarding their individual performance. The study demonstrates the applicability of the random forest algorithm and of support vector regression/classification on novel earth observation data.

Fritz-Forest type and above ground biomass estimation based_Cn_version.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