Estimating the Growing Stem Volume of Chinese Pine and Larch Plantations based on Fused Optical Data Using an Improved Variable Screening Method and Stacking Algorithm

Spatial differentiation of carbon emissions from energy consumption based on machine learning algorithm: A case study during 2015-2020 in Shaanxi, China

Carbon emissions resulting from energy consumption have become a pressing issue for governments worldwide. Accurate estimation of carbon emissions using satellite remote sensing data has become a crucial research problem. Previous studies relied on statistical regression models that failed to capture the complex nonlinear relationships between carbon emissions and characteristic variables. In this study, we propose a machine learning algorithm for carbon emissions, a Bayesian optimized XGboost regression model, using multi-year energy carbon emission data and nighttime lights (NTL) remote sensing data from Shaanxi Province, China. Our results demonstrate that the XGboost algorithm outperforms linear regression and four other machine learning models, with an R2 of 0.906 and RMSE of 5.687. We observe an annual increase in carbon emissions, with high-emission counties primarily concentrated in northern and central Shaanxi Province, displaying a shift from discrete, sporadic points to contiguous, extended spatial distribution. Spatial autocorrelation clustering reveals predominantly high-high and low-low clustering patterns, with economically developed counties showing high-emission clustering and economically relatively backward counties displaying low-emission clustering. Our findings show that the use of NTL data and the XGboost algorithm can estimate and predict carbon emissions more accurately and provide a complementary reference for satellite remote sensing image data to serve carbon emission monitoring and assessment. This research provides an important theoretical basis for formulating practical carbon emission reduction policies and contributes to the development of techniques for accurate carbon emission estimation using remote sensing data.

Stacking and ridge regression-based spectral ensemble preprocessing method and its application in near-infrared spectral analysis

Spectral preprocessing techniques can, to a certain extent, eliminate irrelevant information, such as current noise and stray light from spectral data, thereby enhancing the performance of prediction models. However, current preprocessing techniques mostly attempt to find the best single preprocessing method or their combination, overlooking the complementary information among different preprocessing methods. These preprocessing techniques fail to maximize the utilization of useful information in spectral data and restrict the performance of prediction models. This study proposed a spectral ensemble preprocessing method based on the rapidly developing ensemble learning methods in recent years and the ridge regression (RR) model, named stacking preprocessing ridge regression (SPRR), to address the aforementioned issues. Different from conventional ensemble learning methods, the proposed SPRR method applied multiple different preprocessing techniques to the original spectral data, generating multiple preprocessed datasets. These datasets were then individually inputted into RR base models for training. Ultimately, RR still served as the meta-model, integrating the output results of each RR base model through stacking. This approach not only produced diversity in base models but also achieved higher accuracy and lower computational complexity by using a single type of base model. On the apple spectral dataset collected by our team, correlation analysis showed significant complementary information among the data produced by different preprocessing techniques. This provided robust theoretical support for the proposed SPRR method. By introducing the currently popular averaging ensemble preprocessing method in a comparative experiment, the results of applying the proposed SPRR method to six datasets (apple, meat, wheat, olive oil, tablet, and corn) demonstrated that compared to the single preprocessing method and averaging ensemble preprocessing method, SPRR yielded the best accuracy and reliability for all six datasets. Furthermore, under the same conditions of the training and test datasets, the proposed SPRR method demonstrated better performance than the four commonly used ensemble preprocessing methods.

Mapping forest and site quality of planted Chinese fir forest using sentinel images

Normally, forest quality (FQ) and site quality (SQ) play an important role in evaluating actual and potential forest productivity. Traditionally, these assessment indices (FQ and SQ) are mainly based on forest parameters extracted from ground measurement (forest height, age, density, forest stem volume (FSV), and DBH), which is labor-intensive and difficult to access in certain remote forest areas. Recently, remote sensing images combined with a small number of samples were gradually applied to map forest parameters because of the various advantages of remote sensing technology, such as low cost, spatial coverage, and high efficiency. However, FQ and SQ related to forest parameters are rarely estimated using remote sensing images and machine learning models. In this study, the Sentinel images and ground samples of planted Chinese fir forest located in the ecological "green-core" area of Changzhutan urban cluster, were initially employed to explore the feasibility of mapping the FQ and SQ. And then, four types of alternative variables (backscattering coefficients (VV and VH), multi-spectral bands, vegetation indices, and texture characteristics) were extracted from Sentinel-1A and Sentinel-2A images, respectively. After selecting variables using a stepwise regression model, three machine learning models (SVR, RF, and KNN) were employed to estimate various forest parameters. Finally, the FQ of the study region was directly mapped by the weights sum of related factors extracted by the factor analysis method, and the SQ was also extracted using mapped forest height and age. The results illustrated that the accuracy of estimated forest parameters (DBH, H, and Age) was significantly higher than FSV, FCC, and Age and the largest and smallest rRMSEs were observed from FSV (0.38~0.40) and forest height (0.20~0.21), respectively. Using mapped forest parameters, it also resulted that the rRMSEs of estimated FQ and SQ were 0.19 and 0.15, respectively. Furthermore, after normalization and grading, the grades of forest quality were mainly concentrated in grades I, II, and III in the study region. Though the accuracy of mapping FQ and SQ is limited by the saturation phenomenon, it is significantly proved that using machine learning models and Sentinel images has great potential to indirectly map FQ and SQ.

Inversion of Coniferous Forest Stock Volume Based on Backscatter and InSAR Coherence Factors of Sentinel-1 Hyper-Temporal Images and Spectral Variables of Landsat 8 OLI

Forest stock volume (FSV) is a basic data source for estimating forest carbon sink. It is also a crucial parameter that reflects the quality of forest resources and forest management level. The use of remote sensing data combined with a support vector regression (SVR) algorithm has been widely used in FSV estimation. However, due to the complexity and spatial heterogeneity of the forest biological community, in the FSV high-value area with dense vegetation, the optical re-mote sensing variables tend to be saturated, and the sensitivity of synthetic aperture radar (SAR) backscattering features to the FSV is significantly reduced. These factors seriously affect the ac-curacy of the FSV estimation. In this study, Landsat 8 (L8) Operational Land Imager multispectral images and C-band Sentinel-1 (S1) hyper-temporal SAR images were used to extract three re-mote sensing feature datasets: spectral variables (L8), backscattering coefficients (S1), and inter-ferometric SAR factors (S1-InSAR). We proposed a feature selection method based on SVR (FS-SVR) and compared the FSV estimation performance of FS-SVR and stepwise regression analysis (SRA) on the aforementioned three remote sensing feature datasets. Finally, an estima-tion model of coniferous FSV was constructed using the SVR algorithm in Wangyedian Forest Farm, Inner Mongolia, China, and the spatial distribution map of coniferous FSV was predicted. The experimental results show the following: (1) The coherence amplitude and DSM data ob-tained based on S1 images contain information relat-ed to forest canopy height, and the hy-per-temporal S1 image data significantly enrich the diversity of S1-InSAR feature factors. There-fore, the S1-InSAR dataset has a better FSV response than remote sensing factors such as the S1 backscattering coefficient and L8 vegetation index, and the corresponding root mean square er-ror (RMSE) and relative RMSE (rRMSE) values reached 47.6 m3/ha and 20.9%, respectively. (2) The integrated dataset can provide full play to the synergy of the L8, S1, and S1-InSAR remote sensing data. Its RMSE and rRMSE values are 44.3 m3/ha and 19.4% respectively. (3) The proposed FS-SVR method can better select remote sensing variables suitable for FSV estimation than SRA. The average value of the rRMSE (23.17%) based on the three datasets was 13.8% lower than that of the SRA method (26.87%). This study provides new insights into forest FSV retrieval based on active and passive multisource remote sensing joint data.

Improving Leaf Area Index Retrieval Using Multi-Sensor Images and Stacking Learning in Subtropical Forests of China

The leaf area index (LAI) is a key indicator of the status of forest ecosystems that is important for understanding global carbon and water cycles as well as terrestrial surface energy balances and the impacts of climate change. Machine learning (ML) methods offer promising ways of generating spatially explicit LAI data covering large regions based on optical images. However, there have been few efforts to analyze the LAI in heterogeneous subtropical forests with complex terrain by fusing high-resolution multi-sensor data from the Sentinel-1 Synthetic Aperture Radar (SAR), Sentinel-2 Multi Spectral Instrument (MSI), and Advanced Land Observing Satellite-1 digital elevation model (DEM). Here, forest LAI mapping was performed by integrating the MSI, SAR, and DEM data using a stacking learning (SL) approach that incorporates distinct predictions from a set of optimized individual ML algorithms. The method’s performance was evaluated by comparison to field forest LAI measurements acquired in Xingguo and Gandong of subtropical China. The results showed that the addition of the SAR and DEM images using the SL model compared to the inputs of only optical images reduced the mean absolute error (MAE) and root mean square error (RMSE) by 26% and 18%, respectively, in Xingguo, and by 12% and 8%, respectively, in Gandong. Furthermore, the combination of all images had the best prediction performance. SL was found to be more robust and accurate than conventional individual ML models, while the MAE and RMSE were decreased by 71% and 64%, respectively, in Xingguo, and by 68% and 59%, respectively, in Gandong. Therefore, the SL model using the three-source data combination produced satisfied prediction accuracy with the coefficients of determination (R2), MAE, and RMSE of 0.96, 0.17, and 0.28, respectively, in Xingguo and 0.94, 0.30, and 0.47, respectively, in Gandong. This study revealed the potential of the SL algorithm for retrieving the forest LAI using multi-sensor data in areas with complex terrain.

Study on the Estimation of Forest Volume Based on Multi-Source Data

We performed a comparative analysis of the prediction accuracy of machine learning methods and ordinary Kriging (OK) hybrid methods for forest volume models based on multi-source remote sensing data combined with ground survey data. Taking Larix olgensis, Pinus koraiensis, and Pinus sylvestris plantations in Mengjiagang forest farms as the research object, based on the Chinese Academy of Forestry LiDAR, charge-coupled device, and hyperspectral (CAF-LiTCHy) integrated system, we extracted the visible vegetation index, texture features, terrain factors, and point cloud feature variables, respectively. Random forest (RF), support vector regression (SVR), and an artificial neural network (ANN) were used to estimate forest volume. In the small-scale space, the estimation of sample plot volume is influenced by the surrounding environment as well as the neighboring observed data. Based on the residuals of these three machine learning models, OK interpolation was applied to construct new hybrid forest volume estimation models called random forest Kriging (RFK), support vector machines for regression Kriging (SVRK), and artificial neural network Kriging (ANNK). The six estimation models of forest volume were tested using the leave-one-out (Loo) cross-validation method. The prediction accuracies of these six models are better, with RLoo2 values above 0.6, and the prediction accuracy values of the hybrid models are all improved to different extents. Among the six models, the RFK hybrid model had the best prediction effect, with an RLoo2 reaching 0.915. Therefore, the machine learning method based on multi-source remote sensing factors is useful for forest volume estimation; in particular, the hybrid model constructed by combining machine learning and the OK method greatly improved the accuracy of forest volume estimation, which, thus, provides a fast and effective method for the remote sensing inversion estimation of forest volume and facilitates the management of forest resources.

A Novel Method for Estimating Spatial Distribution of Forest Above-Ground Biomass Based on Multispectral Fusion Data and Ensemble Learning Algorithm

Optical remote sensing technology has been widely used in forest resources inventory. Due to the influence of satellite orbits, sensor parameters, sensor errors, and atmospheric effects, there are great differences in vegetation spectral information captured by different satellite sensor images. Spectral fusion technology can couple the advantages of different multispectral sensor images to produce new multispectral data with high spatial and spectral resolution, it has great potential for improving the spectral sensitivity of forest vegetation and alleviating the spectral saturation. However, how to quickly and effectively select the multi-spectral fusion data suitable for forest above-ground biomass (AGB) estimation is a very critical issue. This study proposes a scheme (RF-S) to comprehensively evaluate multispectral fused images and develop the appropriate model for forest AGB estimation, on the basis of random forest (RF) and the stacking ensemble algorithm. First, four classic fusion methods are used to fuse the preprocessed GaoFen-2 (GF-2) multispectral image with Sentinel-2 image to generate 12 fused Sentinel-like images. Secondly, we apply a comprehensive evaluation method to quickly select the optimal fused image for the follow-up research. Subsequently, two feature combination optimization methods are used to select feature variables from the three feature sets. Finally, the stacking ensemble algorithm based on model dynamic integration and hyperparameter automatic optimization, as well as some classic machine learners, are used to construct the forest AGB estimation model. The results show that the fused image NND_B3 (based on nearest neighbor diffusion pan sharpening method and Band3_Red) selected by the evaluation method proposed in this study has the best performance in AGB estimation. Using the stacking ensemble method and NND_B3 image, we get the highest estimation accuracy, with the adjusted R2 and relative root mean square error (RMSEr) of 0.6306 and 15.53%, respectively. The AGB estimation RMSEr of NND_B3 is 19.95% and 24.90% lower than those of GF-2 and Sentinel-2, respectively. We also found that the multi-window texture factor has better performance in the area with low AGB, and it can suppress the overestimation significantly. The AGB spatial distribution estimated using the NND_B3 image matches the field observations well, indicating that the multispectral fusion image combined with the Stacking algorithm can increase the accuracy and saturation of the AGB estimates.

Coniferous Plantations Growing Stock Volume Estimation Using Advanced Remote Sensing Algorithms and Various Fused Data

Spatial distribution prediction of growing stock volume (GSV) for supporting the sustainable management of forest ecosystems, is one of the most widespread applications of remote sensing. For this purpose, remote sensing data were used as predictor variables in combination with ground data obtained from field sample plots. However, with the increase in forest GSV values, the spectral reflectance of remote sensing imagery is often saturated or less sensitive to the GSV changes, making accurate estimation difficult. To improve this, we examined the GSV estimation performance and data saturation of four optical remote sensing image datasets (Landsat 8, Sentinel-2, ZiYuan-3, and GaoFen-2) in the subtropical region of Central South China. First, various feature variables were extracted and three optimization methods were used to select optimal feature variable combinations. Subsequently, k-nearest-neighbor (kNN), random forest regression, and categorical boosting algorithms were employed to build the GSV estimation models, and evaluate the GSV estimation accuracy and saturation. Second, Gram Schmidt (GS) and NNDiffuse pan sharpening (NND) methods were employed to fuse the optimal multispectral images and explore various image fusion schemes suitable for GSV estimation. We proposed an adaptive stacking (AdaStacking) model ensemble algorithm to further improve GSV estimation performance. The results indicated that Sentinel-2 had the highest GSV estimation accuracy exhibiting a minimum relative root mean square error of 20.06% and saturation of 434 m3/ha, followed by GaoFen-2 with a minimum relative root mean square error of 22.16% and a saturation of 409 m3/ha. Among the four fusion images, the NND-B2 image—obtained by fusing the GaoFen-2 green band and Sentinel-2 multispectral image with the NND method—had the best estimation accuracy. The estimated optimal RMSEs of NND-B2 were 24.4% and 16.5% lower than those of GaoFen-2 and Sentinel-2, respectively. Therefore, the fused image data based on GF-2 and Sentinel-2 can effectively couple the advantages of the two images and significantly improve the GSV estimation performance. Moreover, the proposed adaptive stacking model is more effective in GSV estimation than a single model. The GSV estimation saturation value of the AdaStacking model based on NND-B2 was 5.4% higher than that of the KNN-Maha model. The GSV distribution map estimated by AdaStacking model used the NND-B2 dataset corresponded accurately with the field observations. This study provides some insights into the optical image fusion scheme, feature selection, and adaptive modeling algorithm in GSV estimation for coniferous forest.

Estimating the aboveground biomass of coniferous forest in Northeast China using spectral variables, land surface temperature and soil moisture

As a crucial indicator of forest growth and quality, estimating aboveground biomass (AGB) plays a key role in monitoring the global carbon cycle and forest health assessments. Novel methods and applications in remote sensing technology can greatly reduce the investigation time and cost and therefore have the potential to efficiently estimate AGB. Random forest (RF), combined with remote sensing images, is a popular machine learning method that has been widely used for AGB estimation. However, the accuracy of the ordinary linear variable selection method in the AGB estimation of coniferous forests is challenging due to the complexity of these forest biomes. In this study, spectral variables (spectral reflectance and vegetation index), land surface temperature (LST) and soil moisture were extracted from the operational land imager (OLI) and thermal infrared sensor (TIRS) of Landsat 8, and optimized RF regressions were established to estimate the AGB of coniferous forests in the Wangyedian forest farm, Inner Mongolia, Northeast China. We applied one linear (Pearson correlation coefficient (PC)) and four nonlinear (Kendall's τ coefficient (KC), Spearman coefficient (SC), distance correlation coefficient (DC) and the importance index) indices to select variables and establish optimized RF regressions for AGB estimation. The results showed that all the nonlinear indices provided significantly lower estimation errors than the linear index, in which the minimum root mean square error (RMSE) of 40.92 Mg/ha was obtained by the importance index in the nonlinear indices. In addition, the inclusion of LST and soil moisture significantly improved AGB estimation. The RMSE of the models constructed through the five indices decreased by 12.93%, 7.31%, 8.33%, 6.28% and 10.78%, respectively, following the application of the LST variable. In particular, when LST and soil moisture were both added into the model, the RMSE decreased by 31.47%. This study demonstrates that combining the nonlinear variable selection method with optimized RF regression can improve the efficiency of AGB estimation to support regional forest resource management and monitoring.

Mapping the Growing Stem Volume of the Coniferous Plantations in North China Using Multispectral Data from Integrated GF-2 and Sentinel-2 Images and an Optimized Feature Variable Selection Method

Accurate measurement of forest growing stem volume (GSV) is important for forest resource management and ecosystem dynamics monitoring. Optical remote sensing imagery has great application prospects in forest GSV estimation on regional and global scales as it is easily accessible, has a wide coverage, and mature technology. However, their application is limited by cloud coverage, data stripes, atmospheric effects, and satellite sensor errors. Combining multi-sensor data can reduce such limitations as it increases the data availability, but also causes the multi-dimensional problem that increases the difficulty of feature selection. In this study, GaoFen-2 (GF-2) and Sentinel-2 images were integrated, and feature variables and data scenarios were derived by a proposed adaptive feature variable combination optimization (AFCO) program for estimating the GSV of coniferous plantations. The AFCO algorithm was compared to four traditional feature variable selection methods, namely, random forest (RF), stepwise random forest (SRF), fast iterative feature selection method for k-nearest neighbors (KNN-FIFS), and the feature variable screening and combination optimization procedure based on the distance correlation coefficient and k-nearest neighbors (DC-FSCK). The comparison indicated that the AFCO program not only considered the combination effect of feature variables, but also optimized the selection of the first feature variable, error threshold, and selection of the estimation model. Furthermore, we selected feature variables from three datasets (GF-2, Sentinel-2, and the integrated data) following the AFCO and four other feature selection methods and used the k-nearest neighbors (KNN) and random forest regression (RFR) to estimate the GSV of coniferous plantations in northern China. The results indicated that the integrated data improved the GSV estimation accuracy of coniferous plantations, with relative root mean square errors (RMSErs) of 15.0% and 19.6%, which were lower than those of GF-2 and Sentinel-2 data, respectively. In particular, the texture feature variables derived from GF-2 red band image have a significant impact on GSV estimation performance of the integrated dataset. For most data scenarios, the AFCO algorithm gained more accurate GSV estimates, as the RMSErs were 30.0%, 23.7%, 17.7%, and 17.5% lower than those of RF, SRF, KNN-FIFS, and DC-FSCK, respectively. The GSV distribution map obtained by the AFCO method and RFR model matched the field observations well. This study provides some insight into the application of optical images, optimization of the feature variable combination, and modeling algorithm selection for estimating the GSV of coniferous plantations.

Mapping the Forest Canopy Height in Northern China by Synergizing ICESat-2 with Sentinel-2 Using a Stacking Algorithm

The forest canopy height (FCH) plays a critical role in forest quality evaluation and resource management. The accurate and rapid estimation and mapping of the regional forest canopy height is crucial for understanding vegetation growth processes and the internal structure of the ecosystem. A stacking algorithm consisting of multiple linear regression (MLR), support vector machine (SVM), k-nearest neighbor (kNN), and random forest (RF) was used in this paper and demonstrated optimal performance in predicting the forest canopy height by synergizing Sentinel-2 images acquired from the cloud-based computation platform Google Earth Engine (GEE) with data from ICESat-2 (Ice, Cloud, and Land Elevation Satellite-2). This research was conducted to achieve continuous mapping of the canopy height of plantations in Saihanba Mechanical Forest Plantation, which is located in Chengde City, northern Hebei province, China. The results show that stacking achieved the best prediction accuracy for the forest canopy height, with an R2 of 0.77 and a root mean square error (RMSE) of 1.96 m. Compared with MLR, SVM, kNN, and RF, the RMSE obtained by stacking was reduced by 25.2%, 24.9%, 22.8%, and 18.7%, respectively. Since Sentinel-2 images and ICESat-2 data are publicly available, this opens the door for the accurate mapping of the continuous distribution of the forest canopy height globally in the future.

Estimating the Growing Stem Volume of Coniferous Plantations Based on Random Forest Using an Optimized Variable Selection Method

Forest growing stem volume (GSV) reflects the richness of forest resources as well as the quality of forest ecosystems. Remote sensing technology enables robust and efficient GSV estimation as it greatly reduces the survey time and cost while facilitating periodic monitoring. Given its red edge bands and a short revisit time period, Sentinel-2 images were selected for the GSV estimation in Wangyedian forest farm, Inner Mongolia, China. The variable combination was shown to significantly affect the accuracy of the estimation model. After extracting spectral variables, texture features, and topographic factors, a stepwise random forest (SRF) method was proposed to select variable combinations and establish random forest regressions (RFR) for GSV estimation. The linear stepwise regression (LSR), Boruta, Variable Selection Using Random Forests (VSURF), and random forest (RF) methods were then used as references for comparison with the proposed SRF for selection of predictors and GSV estimation. Combined with the observed GSV data and the Sentinel-2 images, the distributions of GSV were generated by the RFR models with the variable combinations determined by the LSR, RF, Boruta, VSURF, and SRF. The results show that the texture features of Sentinel-2’s red edge bands can significantly improve the accuracy of GSV estimation. The SRF method can effectively select the optimal variable combination, and the SRF-based model results in the highest estimation accuracy with the decreases of relative root mean square error by 16.4%, 14.4%, 16.3%, and 10.6% compared with those from the LSR-, RF-, Boruta-, and VSURF-based models, respectively. The GSV distribution generated by the SRF-based model matched that of the field observations well. The results of this study are expected to provide a reference for GSV estimation of coniferous plantations.