Fusion of Five Satellite-Derived Products Using Extremely Randomized Trees to Estimate Terrestrial Latent Heat Flux over Europe

Shang, Ke; Yao, Yunjun; Li, Yufu; Yang, Junming; Jia, Kun; Zhang, Xiaolin; Chen, Xiao‐Wei; Bei, Xiangyi; Guo, Xihong

doi:10.3390/rs12040687

Cited by 30 publications

(24 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Since it does not require the determination of optimal discretization thresholds, the ERT algorithm has an advantage in terms of computing time. To date, the ERT algorithm was utilized to predict terrestrial latent heat fluxes [31], air quality [51], and streamflow models [52], but its performance in estimating forest AGB has not yet been explored.…”

Section: Tree-based Modelsmentioning

confidence: 99%

“…Although these above studies have examined or compared the performances of machine learning algorithms for estimating AGB, they were mainly limited to several commonly used algorithms. Some algorithms, such as the extremely randomized trees (ERT) and CatBoost methods, have attracted much attention in the machine learning community and have outperformed commonly used algorithms when applied in other fields [31,32], but their performances for AGB estimation have not been fully investigated. Moreover, recent studies mainly focused on the overall accuracy of estimated AGB, and the problems of overestimation and underestimation when estimating AGB with an inversion algorithm were largely ignored [21].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

An Evaluation of Eight Machine Learning Regression Algorithms for Forest Aboveground Biomass Estimation from Multiple Satellite Data Products

Zhang

Liang

et al. 2020

Remote Sensing

View full text Add to dashboard Cite

This study provided a comprehensive evaluation of eight machine learning regression algorithms for forest aboveground biomass (AGB) estimation from satellite data based on leaf area index, canopy height, net primary production, and tree cover data, as well as climatic and topographical data. Some of these algorithms have not been commonly used for forest AGB estimation such as the extremely randomized trees, stochastic gradient boosting, and categorical boosting (CatBoost) regression. For each algorithm, its hyperparameters were optimized using grid search with cross-validation, and the optimal AGB model was developed using the training dataset (80%) and AGB was predicted on the test dataset (20%). Performance metrics, feature importance as well as overestimation and underestimation were considered as indicators for evaluating the performance of an algorithm. To reduce the impacts of the random training-test data split and sampling method on the performance, the above procedures were repeated 50 times for each algorithm under the random sampling, the stratified sampling, and separate modeling scenarios. The results showed that five tree-based ensemble algorithms performed better than the three nonensemble algorithms (multivariate adaptive regression splines, support vector regression, and multilayer perceptron), and the CatBoost algorithm outperformed the other algorithms for AGB estimation. Compared with the random sampling scenario, the stratified sampling scenario and separate modeling did not significantly improve the AGB estimates, but modeling AGB for each forest type separately provided stable results in terms of the contributions of the predictor variables to the AGB estimates. All the algorithms showed forest AGB were underestimated when the AGB values were larger than 210 Mg/ha and overestimated when the AGB values were less than 120 Mg/ha. This study highlighted the capability of ensemble algorithms to improve AGB estimates and the necessity of improving AGB estimates for high and low AGB levels in future studies.

show abstract

Section: Tree-based Modelsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

An Evaluation of Eight Machine Learning Regression Algorithms for Forest Aboveground Biomass Estimation from Multiple Satellite Data Products

Zhang

Liang

et al. 2020

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…ANN , RF , BR and RANSAC were used to estimate ocean LHF at a spatial resolution of 0.25°. Some of these four ML methods were successfully used to estimate a terrestrial LE fusion product, such as ANN and RF [ 49 , 67 , 68 ]. According to validation against buoy observations ( Figure 4 ), our results illustrate that ensemble ocean LHF from ML methods performed much better than the estimated LHF from four individual ocean LHF products ( MERRA-2 , JOFURO-3 , ERA-I and GSSTF -3).…”

Section: Discussionmentioning

confidence: 99%

“…Fan et al [ 48 ] developed four tree-based ensemble models ( RF , M5Tree , GBDT and XGBoost ) to estimate daily ET using limited meteorological data; the developed XGBoost and GBDT models have accurate predictions, strong model stability and low calculation cost. Shang et al [ 49 ] applied four ML methods (Extremely Randomized Trees ( ETR ), Gradient Boosting Regression Tree ( GBRT ), Random Forest ( RF ) and Gaussian Process Regression ( GPR )) to improve terrestrial LE estimations over Europe based on five individual terrestrial LE product; the validation results illustrate that the LE estimation using ETR method increased R 2 and decreased RMSE . Even though the ML methods have been widely used to estimate terrestrial biophysical variables, there is a lack of experiments on dataset fusion to improve ocean LHF estimates by combining multiple LHF products.…”

Section: Introductionmentioning

confidence: 99%

ANN-Based Estimation of Low-Latitude Monthly Ocean Latent Heat Flux by Ensemble Satellite and Reanalysis Products

Chen

Yao

Li³

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

Ocean latent heat flux (LHF) is an essential variable for air–sea interactions, which establishes the link between energy balance, water and carbon cycle. The low-latitude ocean is the main heat source of the global ocean and has a great influence on global climate change and energy transmission. Thus, an accuracy estimation of high-resolution ocean LHF over low-latitude area is vital to the understanding of energy and water cycle, and it remains a challenge. To reduce the uncertainties of individual LHF products over low-latitude areas, four machine learning (ML) methods (Artificial Neutral Network (ANN), Random forest (RF), Bayesian Ridge regression and Random Sample Consensus (RANSAC) regression) were applied to estimate low-latitude monthly ocean LHF by using two satellite products (JOFURO-3 and GSSTF-3) and two reanalysis products (MERRA-2 and ERA-I). We validated the estimated ocean LHF using 115 widely distributed buoy sites from three buoy site arrays (TAO, PIRATA and RAMA). The validation results demonstrate that the performance of LHF estimations derived from the ML methods (including ANN, RF, BR and RANSAC) were significantly better than individual LHF products, indicated by R2 increasing by 3.7–46.4%. Among them, the LHF estimation using the ANN method increased the R2 of the four-individual ocean LHF products (ranging from 0.56 to 0.79) to 0.88 and decreased the RMSE (ranging from 19.1 to 37.5) to 11 W m-2. Compared to three other ML methods (RF, BR and RANSAC), ANN method exhibited the best performance according to the validation results. The results of relative uncertainty analysis using the triangle cornered hat (TCH) method show that the ensemble LHF product using ML methods has lower relative uncertainty than individual LHF product in most area. The ANN was employed to implement the mapping of annual average ocean LHF over low-latitude at a spatial resolution of 0.25° during 2003–2007. The ocean LHF fusion products estimated from ANN methods were 10–30 W m−2 lower than those of the four original ocean products (MERRA-2, JOFURO-3, ERA-I and GSSTF-3) and were more similar to observations.

show abstract

“…ERT is a top-down method that is very similar to the random forest approach, but is different from the latter in two points: first, it does not adopt bootstrap sampling with a replacement strategy but directly uses the original training samples to reduce the deviation; second, the bifurcation value is completely random, which can achieve the bifurcation of a decision tree. The result is smaller and more stable than that of the random forest [ 30 ]. To apply this technique, we used library – “sklearn.…”

Section: Experimental Materials and Methodsmentioning

confidence: 99%

Comparison of new hyperspectral index and machine learning models for prediction of winter wheat leaf water content

Zhang

Xiong

et al. 2021

Plant Methods

View full text Add to dashboard Cite

Background The leaf water content estimation model is established by hyperspectral technology, which is crucial and provides technical reference for precision irrigation. Methods In this study, two consecutive years of field experiments (different irrigation times and seven wheat varieties) in 2018–2020 were performed to obtain the canopy spectra reflectance and leaf water content (LWC) data. The characteristic bands related to LWC were extracted from correlation coefficient method (CA) and x-Loading weight method (x-Lw). Five modeling methods, spectral index and four other methods (Partial Least-Squares Regression (PLSR), Random Forest Regression (RFR), Extreme Random Trees (ERT), and K-Nearest Neighbor (KNN)) based characteristic bands, were employed to construct LWC estimation models. Results The results showed that the canopy spectral reflectance increased with the increase of irrigation times, especially in the near-infrared band (750–1350 nm). The prediction accuracy of the newly developed differential spectral index DVI (R1185, R1307) was higher than that of the existing spectral index, with R2 of 0.85 and R2 of 0.78 for the calibration and validation, respectively. Due to a large amount of hyperspectral data, the correlation coefficient method (CA) and x-Loading weight (x-Lw) were used to select the water characteristic bands (100 and 28 characteristic bands, respectively) from the full spectrum. We found that the accuracy of the model based on the characteristic bands was not significantly lower than that of the full spectrum-based models. Among these models, the ERT- x-Lw model performed the best (R2 and RMSE of 0.88 and 1.46; 0.84 and 1.62 for the calibration and validation, respectively). In addition, the accuracy of the LWC estimation model constructed by ERT-x-Lw was higher than that of DVI (R1185, R1307). Conclusion The two models based on ERT-x-Lw and DVI (R1185, R1307) can effectively predict wheat leaf water content. The results provide a technical reference and a basis for crop water monitoring and diagnosis under similar production conditions.

show abstract

Fusion of Five Satellite-Derived Products Using Extremely Randomized Trees to Estimate Terrestrial Latent Heat Flux over Europe

Cited by 30 publications

References 64 publications

An Evaluation of Eight Machine Learning Regression Algorithms for Forest Aboveground Biomass Estimation from Multiple Satellite Data Products

An Evaluation of Eight Machine Learning Regression Algorithms for Forest Aboveground Biomass Estimation from Multiple Satellite Data Products

ANN-Based Estimation of Low-Latitude Monthly Ocean Latent Heat Flux by Ensemble Satellite and Reanalysis Products

Comparison of new hyperspectral index and machine learning models for prediction of winter wheat leaf water content

Contact Info

Product

Resources

About