Object-based random forest modelling of aboveground forest biomass outperforms a pixel-based approach in a heterogeneous and mountain tropical environment

Silveira, Eduarda Martiniano de Oliveira; Silva, Sérgio Henrique Godinho; Júnior, Fausto Weimar Acerbi; Carvalho, Mônica Canaan; Carvalho, Luís Marcelo Tavares de; Scolforo, José Roberto Soares; Wulder, Michael A.

doi:10.1016/j.jag.2019.02.004

Cited by 72 publications

(46 citation statements)

References 91 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…This can arguably be due to the relatively higher saturation of SRS derivatives (i.e., vegetation indices and SAR backscatters) in forests with high vegetation density and complex structure, such as in case of tropical forests, compared to, for instance, temperate and boreal forests. The use of texture measures [77], object-based image analysis [78], and radiative transfer models [79] showed improved estimations and provide detailed insight into the structural assessment of tropical forests. A common recommendation from most studies is that SRS-based forest structure assessments will benefit from data acquired in dry seasons.…”

Section: Discussionmentioning

confidence: 99%

Linking Terrestrial LiDAR Scanner and Conventional Forest Structure Measurements with Multi-Modal Satellite Data

Mulatu

Decuyper

Brede

et al. 2019

Forests

View full text Add to dashboard Cite

Obtaining information on vertical forest structure requires detailed data acquisition and analysis which is often performed at a plot level. With the growing availability of multi-modal satellite remote sensing (SRS) datasets, their usability towards forest structure estimation is increasing. We assessed the relationship of PlanetScope-, Sentinel-2-, and Landsat-7-derived vegetation indices (VIs), as well as ALOS-2 PALSAR-2- and Sentinel-1-derived backscatter intensities with a terrestrial laser scanner (TLS) and conventionally measured forest structure parameters acquired from 25 field plots in a tropical montane cloud forest in Kafa, Ethiopia. Results showed that canopy gap-related forest structure parameters had their highest correlation (|r| = 0.4 − 0.48) with optical sensor-derived VIs, while vegetation volume-related parameters were mainly correlated with red-edge- and short-wave infrared band-derived VIs (i.e., inverted red-edge chlorophyll index (IRECI), normalized difference moisture index), and synthetic aperture radar (SAR) backscatters (|r| = −0.57 − 0.49). Using stepwise multi-linear regression with the Akaike information criterion as evaluation parameter, we found that the fusion of different SRS-derived variables can improve the estimation of field-measured structural parameters. The combination of Sentinel-2 VIs and SAR backscatters was dominant in most of the predictive models, while IRECI was found to be the most common predictor for field-measured variables. The statistically significant regression models were able to estimate cumulative plant area volume density with an R2 of 0.58 and with the lowest relative root mean square error (RRMSE) value (0.23). Mean gap and number of gaps were also significantly estimated, but with higher RRMSE (R2 = 0.52, RRMSE = 1.4, R2 = 0.68, and RRMSE = 0.58, respectively). The models showed poor performance in predicting tree density and number of tree species (R2 = 0.28, RRMSE = 0.41, and R2 = 0.21, RRMSE = 0.39, respectively). This exploratory study demonstrated that SRS variables are sensitive to retrieve structural differences of tropical forests and have the potential to be used to upscale biodiversity relevant field-based forest structure estimates.

show abstract

Section: Discussionmentioning

confidence: 99%

Linking Terrestrial LiDAR Scanner and Conventional Forest Structure Measurements with Multi-Modal Satellite Data

Mulatu

Decuyper

Brede

et al. 2019

Forests

View full text Add to dashboard Cite

show abstract

“…The model produces the OOB error and the variable importance to assess the prediction accuracy and indicate the contribution of each variable. The RFR model is a well-known method that has been widely used in forest AGB estimations in previous studies [22,75].…”

Section: Random Forest Regression (Rfr)mentioning

confidence: 99%

Comparison of Machine Learning Methods for Estimating Mangrove Above-Ground Biomass Using Multiple Source Remote Sensing Data in the Red River Delta Biosphere Reserve, Vietnam

et al. 2020

View full text Add to dashboard Cite

This study proposes a hybrid intelligence approach based on an extreme gradient boosting regression and genetic algorithm, namely, the XGBR-GA model, incorporating Sentinel-2, Sentinel-1, and ALOS-2 PALSAR-2 data to estimate the mangrove above-ground biomass (AGB), including small and shrub mangrove patches in the Red River Delta biosphere reserve across the northern coast of Vietnam. We used the novel extreme gradient boosting decision tree (XGBR) technique together with genetic algorithm (GA) optimization for feature selection to construct and verify a mangrove AGB model using data from a field survey of 105 sampling plots conducted in November and December of 2018 and incorporated the dual polarimetric (HH and HV) data of the ALOS-2 PALSAR-2 L-band and the Sentinel-2 multispectral data combined with Sentinel-1 (C-band VV and VH) data. We employed the root-mean-square error (RMSE) and coefficient of determination (R 2 ) to evaluate the performance of the proposed model. The capability of the XGBR-GA model was assessed via a comparison with other machine-learning (ML) techniques, i.e., the CatBoost regression (CBR), gradient boosted regression tree (GBRT), support vector regression (SVR), and random forest regression (RFR) models. The XGBR-GA model yielded a promising result (R 2 = 0.683, RMSE = 25.08 Mg·ha −1 ) and outperformed the four other ML models. The XGBR-GA model retrieved a mangrove AGB ranging from 17 Mg·ha −1 to 142 Mg·ha −1 (with an average of 72.47 Mg·ha −1 ). Therefore, multisource optical and synthetic aperture radar (SAR) combined with the XGBR-GA model can be used to estimate the mangrove AGB in North Vietnam. The effectiveness of the proposed method needs to be further tested and compared to other mangrove ecosystems in the tropics.

show abstract

“…According to the results listed in Table 1, the medium class has more likelihood of confusion with other classes, especially the high class. A suggested future line of research is to apply the Random Forest algorithm [41] to classify the reference data with the aim of minimizing the confusion between classes.…”

Section: Discussionmentioning

confidence: 99%

“…This information was georeferenced and used to generate training sites based on the previous segmentation of the multispectral reflectance. OBIA was applied by [41] for modelling aboveground forest biomass.…”

Section: Reference Data Mapmentioning

confidence: 99%

Evaluation of Fire Severity Indices Based on Pre- and Post-Fire Multispectral Imagery Sensed from UAV

et al. 2019

View full text Add to dashboard Cite

Fire severity is a key factor for management of post-fire vegetation regeneration strategies because it quantifies the impact of fire, describing the amount of damage. Several indices have been developed for estimation of fire severity based on terrestrial observation by satellite imagery. In order to avoid the implicit limitations of this kind of data, this work employed an Unmanned Aerial Vehicle (UAV) carrying a high-resolution multispectral sensor including green, red, near-infrared, and red edge bands. Flights were carried out pre- and post-controlled fire in a Mediterranean forest. The products obtained from the UAV-photogrammetric projects based on the Structure from Motion (SfM) algorithm were a Digital Surface Model (DSM) and multispectral images orthorectified in both periods and co-registered in the same absolute coordinate system to find the temporal differences (d) between pre- and post-fire values of the Excess Green Index (EGI), Normalized Difference Vegetation Index (NDVI), and Normalized Difference Red Edge (NDRE) index. The differences of indices (dEGI, dNDVI, and dNDRE) were reclassified into fire severity classes, which were compared with the reference data identified through the in situ fire damage location and Artificial Neural Network classification. Applying an error matrix analysis to the three difference of indices, the overall Kappa accuracies of the severity maps were 0.411, 0.563, and 0.211 and the Cramer’s Value statistics were 0.411, 0.582, and 0.269 for dEGI, dNDVI, and dNDRE, respectively. The chi-square test, used to compare the average of each severity class, determined that there were no significant differences between the three severity maps, with a 95% confidence level. It was concluded that dNDVI was the index that best estimated the fire severity according to the UAV flight conditions and sensor specifications.

show abstract

Object-based random forest modelling of aboveground forest biomass outperforms a pixel-based approach in a heterogeneous and mountain tropical environment

Cited by 72 publications

References 91 publications

Linking Terrestrial LiDAR Scanner and Conventional Forest Structure Measurements with Multi-Modal Satellite Data

Linking Terrestrial LiDAR Scanner and Conventional Forest Structure Measurements with Multi-Modal Satellite Data

Comparison of Machine Learning Methods for Estimating Mangrove Above-Ground Biomass Using Multiple Source Remote Sensing Data in the Red River Delta Biosphere Reserve, Vietnam

Evaluation of Fire Severity Indices Based on Pre- and Post-Fire Multispectral Imagery Sensed from UAV

Contact Info

Product

Resources

About