Esther Merlo Snchez 1 Miguel Pieiro Garca 1 Oscar Santaclara Estvez - - PowerPoint PPT Presentation
Remote sensing technologies as proxy forecast of wood technological properties. Preliminary results Esther Merlo Snchez 1 Miguel Pieiro Garca 1 Oscar Santaclara Estvez 1 Mara Julia Yage Ballester 2 1 MADERA plus; 2 GMV
Remote sensing technologies as proxy forecast of wood technological properties. Preliminary results
Esther Merlo Sánchez 1 Miguel Piñeiro García 1 Oscar Santaclara Estévez 1 María Julia Yagüe Ballester 2 1 MADERA plus; 2 GMV
MADERA PLUS CALIDAD FORESTAL S.L.
Tecnologycal wood properties prediction by indirect methods
Evolving . . .
2012 IEBT NDT in standing trees Wood properties prediction model by NDT in standing trees with field data in different species 2014 2015 2019 Support wood industry decision with information from wood fiber quality
From Knowledge to aplication
European Union’s Horizon 2020
Grant agreement No 776045. The 36-month long project, coordinated by GMV, aims to develop a pre- commercial service and platform for forest stakeholders to integrate Earth Observation into daily decision-making processes and operations.
Partners:
Mysustainableforest project
Service wood characterisation:
Context
Wood fiber atributes are linked to product potential and performance and is fundamental to optimise the use (i.e., pulp yield, strength and stiffness of lumber) Basic wood density is a wood quality characteristic very important for paper pulp industry: when it increases, the raw material needs are lower and the pulp yields are bigger. Wood density plays a role in biomass and carbon storage estimation Wood density is correlated with cavitation resistance and participate indirectly in water transport. (Drought risk detection?)
Context
the forest industry: Strategic business industry decisions before clear-cuts, Optimize forest-wood chain cost-effectiveness, More efficient planning and sustainable management.
Why remote sensing methods ?
(Satellite images, LiDAR, Clima, Phisiography……)
surface data.
aspects related to vigor can be eventually detected in images collected with remote sensing satellites in the near infrared and medium infrared bands (chlorophyll , nutrient state. ..)
can be determine by LiDAR data, also climatic data during growing season are correlated with early wood/latewood proportion, and other aspects related to physiography data (slope, altitude,..) also affect wood properties.
Why Eucalyptus globulus?
Around 400.000 ha eucalyptus plantation (globulus and nitens) 4,5 millones M3/year
Why basic wood density?
Basic wood density from 350 to 650 kg/m3 that influence its commercial value and industrial processing
Objectives
Aid the segmentation and selection forest masses (trees) and later management plans, based on basic wood properties, in
exploitation of resources in woodlands
globulus based on EO derived parameters (satellite) and in combination with forest attributes, climatic and physiographic data.
product that allows predicting the basic density of specific masses of Eucalyptus globulus.
Matherial and Methods
A total of 25 circular field plots (r=14,1m) were randomly sampled across the distribution area of Eucalyptus globulus in the Galician coastal ecoregion Forest growth variables : (all trees by plot) diameter at breast height (DBH), Height, crown
Fiber wood quality traits (10 trees by plot): dynamic modulus of elasticity using the ST300 (Fibre-Gen, NZ) and Basic wood density by partial wood core sample extracted at breast height (1.3m) Multitemporal climatic data during the growing season Temperature: average monthly temperature (during last 10 years). Precipitation: average monthly precipitation (during last 10 years). The data were determined from May to September using an extensive set of meteorological station data Satellite data collection Values of satellite bands and indexes (for Sentinel1, Sentinel 2 and Landsat 8, from May to September, in the period 2013-2018. These parameters are provided by GMV (satellite data). Satellites images comes from Copernicus and in-situ data from project field survey activities Said periodic series of satellite images were duly processed by GMV.
GMV image processing chain is built in a modular way. GMV´s end-to-end processing chains have been designed to perform massive data processing, using parallel computing techniques and advances cloud architectures.
Satellite image processing
Variable Dataset Age at sampling (years) 7-27 years Diameter at breast height (cm) 10−38 Height (m) 10−42 Slenderness (m m-1) 37−99 Mean average annual air temp (oC) 11−14 Total annual rainfall (mm) 981−2315 Basic wood density (Kg/m3) 346-610 Dynamic modulus of elasticity (GPa) 952−33.19
The sampled plots
Modelisation
WOOD MODELs
WOOD RANKINGS
LiDAR Data:.
Climatic Data: Physiographic Data (DEM): . Satellite Data: Inventory Data: Growth Two different mathematical models 1) The stepwise regression methodology, which is a statistical procedure with which we can find the parameters which best fit using the Akaike information criterion (AIC) (Posada e Buckley, 2004)1 (the smaller the AIC, the fitting the better) 2) Partial least squares Regression (PLS), which is a methodology very useful in cases like this, when the number of predictors is bigger than the number of observations . Evaluation of the model performance : two statistical indices are used: The adjusted coefficient of determination (R2), which reflects the part of the total variance that is explained by the model. It is expressed as follows: 𝑆2 = 1 − 𝑜 − 1 𝑧𝑗 − 𝑧 𝑗 2
𝑜 𝑗=1
𝑜 − 𝑞 𝑧𝑗 − 𝑧 2
𝑜 𝑗=1
The Root Mean Square Error (RMSE) which analyses the precision of the estimates. The expression is: 𝑆𝑁𝑇𝐹 = 𝑧𝑗 − 𝑧 𝑗 2
𝑜 𝐽=1
𝑜 − 𝑞
Modelisation
Model for basic wood density prediction was developed using PLS regression with multitemporal satellite images obtained during the growth period in the last two years and climatic data with R2=0,77 and RMSE=26,6 Kg/m3 The coefficient of determination (R2)reflects that 77% of the total variance is explained by the model
Basic Wood Density prediction Model
First basic wood density prediction model in Eucalyptus Stands by remote data
To validate!
www.mysustainableforest.com
The best models were obtained by PLS regression model. Using multitemporal series of satellite images of different Sentinel II bands , with the multitemporal average monthly rainfall and temperature data of the end of the vegetative period, we was
Dynamic Modulus of Elasticity with R2 = 0.66 and RMSE = 2, 2 GPa
Good preliminary results also for MOE
Conclusions
EO data ( Satellite) and Non-EO data (climate and physiography) are correlated with the field basic wood density values and dynamic modulus of elasticity. The preliminary models developed obtain a reliable approach to the dynamic modulus of elasticity and basic wood density (at stand level) to provide the wood industry and forest owners a product that allows predicting fiber wood quality of specific masses of Eucalyptus globulus. MySustainableForest provides an advanced and flexible new service to the End-Users ground-breaking objectives to improve their operations. The synergies from different types of satellites systems (EUMETSAT meteorological ones, Copernicus Sentinels missions including both
innovation at all links of the added value chain.
Financial support for this research was provided through European Commisión MySustainableForest proyect H2020 Grant No. H2020-EO-2017 Proposal number: 776045. The collaboration of ENCE enterprise for allowing us to sample sites is also gratefully acknowledged.
Acknowledgments
Thank you! maderaplus@maderaplus.es