# Integration of Multi-Sensor Data to Estimate Plot-Level Stem Volume Using Machine Learning Algorithms–Case Study of Evergreen Conifer Planted Forests in Japan

^{1}

^{2}

^{3}

^{4}

^{5}

^{*}

## Abstract

**:**

^{2}up to 0.665 and RMSE up to 66.87 m

^{3}/ha (rRMSE = 11.95%) depending on the input variables (best result with canopy height, canopy size, canopy cover, and Sentinel-1 data), and the SVR method showed fitting R

^{2}up to 0.519 and RMSE up to 80.12 m

^{3}/ha (rRMSE = 12.67%). The RFR outperformed the SVR method, which could delineate the relationship between the variables for better model accuracy. This work has demonstrated that incorporating various remote sensing data to satellite data, especially adding finer resolution data, can provide good estimates of forest parameters at a plot level (10 by 10 m), potentially allowing advancements in precision forestry.

## 1. Introduction

_{2}through photosynthetic processes, and some global actions tend to rely on this sequestration to mitigate the impacts of global warming [3]. Reducing Emissions from Deforestation and Forest Degradation (REDD+) [4], the global goals discussed in the Paris Agreement [5], and the Sustainable Development Goals (SDGs) set by the United Nations [6] are important frameworks and goals that promote such forest functions. To commit to strategic actions and to shed light on the achievement of these goals, advances are required in the realm of precision forestry.

## 2. Study Area

## 3. Methodology

#### 3.1. PALSAR2 Mosaic Data and Sentinel-1 TOPSAR Preprocessing

^{0}is the backscattering intensity in decibel units and CF is the calibration factor: −83.0 dB [42]. The global mosaic data are already radiometric corrected and in a form of the plane perpendicular to the line of sight from sensor to the ground surface [43]. The second data set was from the Sentinel-1 Terrain Observation with Progressive Scans SAR (TOPSAR) developed by the European Space Agency (ESA). The C-band wavelength is shorter can interact more within the canopy [44], and is thus able to collect information from the top layers of the forests. The data were downloaded via the webpage, and the Level-1 Ground Range Detected (GRD) product observed in the Interferometric Wideswath (IW) mode, containing dual polarized (VV, VH) data, is used. The data were observed on 22 June 2017. The data were processed with thermal noise removal and radiometric slope correction using the 5 m digital elevation model published by the Geospatial Information Authority of Japan and Range Doppler terrain correction to convert the map into geographical coordinates. The terrain-corrected data were processed with a 3-by-3 Lee filter [45] to reduce the artifacts from the conversion. Only a subset of the processed data corresponding to the surroundings of the study site was extracted. The processing of Sentinel-1 uses SNAP (Sentinel-1 Toolbox ver. 6.0.0) developed by the ESA. Sentinel-1 was available only for VV and VH, which is due to the acquisition plans for this region.

#### 3.2. UAS Observation and Image Processing

#### 3.3. Canopy Segmentation

#### 3.4. TLS Survey

#### 3.5. Non-Destructive Biophysical Parameter Retrieval for Ground Truth

^{3}), D is DBH (cm) and H is tree height (m). The parameters a, b and c are coefficients, and for the cypress trees in this region, their values are as follows: a = −4.31101, b = 1.83546, c = 1.10655 [52]. Figure 6a,b shows the sum of the volume for each grid and the relationship between the DBH and tree height of individual trees within all grid areas. A total of 33 plots were examined, and all of the individual trees within each plot were measured. The cross and circles on Figure 6a indicate the plots further used for training (21 plots) and validation (12 plots). The total number of trees measured for all plots was n = 403. The volume data per plot was multiplied by 100 to compute the volume per unit area (m

^{3}/ha).

#### 3.6. Generating Remotely Sensed Variables

_{avg}(CA

_{avg}), canopy area

_{min}(CA

_{min}), canopy area

_{max}(CA

_{max}), CHM

_{avg}(m), CHM

_{min}, CHM

_{max}, and fraction of canopy cover (FCC) (%). First, for the canopy area, the rasterized segmentation results were used as the base information. A mask image for shadows and canopy gaps was generated using the orthoimages. The mask was applied to the segmentation images to filter out areas that did not require computation. The segmentation image was processed to calculate the average, minimum and maximum value of the canopy area (in m

^{2}) within each 10 m by 10 m grid. The second variable was the tree height, expressed as the canopy height model (CHM). The DSM generated from the SfM method was used with the DTM data computed from the TLS data. By subtracting the DTM from the DSM, we obtained the height of the forested area (CHM). Additionally, the average, minimum, and maximum values of CHM were computed for each grid. Thirdly, for the FCC, the total area of canopy coverage within the grid area was computed (as a percentage).

^{0}

_{HH}, γ

^{0}

_{HV}, γ

^{0}

_{VV}, and γ

^{0}

_{VH}backscattering intensity (dB) was prepared. Both Sentinel-1 and PALSAR have different ground resolutions of 10 m and 25 m, respectively. Therefore, to match the UAS data, both radar images were resampled to the DSMs using the cubic convolution method for further processing purposes.

#### 3.7. Machine Learning Algorithms

#### 3.7.1. Support Vector Regression (SVR)

^{2}, 2

^{3}, 2

^{4}…2

^{7}. More than 60 models were trained first, and from the preliminary result adjustment were made for the finer tuning of the model. The best model had values of ε = 0.01, cost = 64, as well as a radial basis function kernel, gamma = 0.1. Remaining plot area (12 plots) was used for the validation. The predicted values were averaged within the plot area and compared with the reference value.

#### 3.7.2. Random Forest Regression (RFR)

#### 3.8. Correlation of Each Variables and Validating the Predicted Models

^{2}), root mean square error (RMSE) and relative RMSE (rRMSE) between the predicted value and the reference values. The RFR models were further tested to compute the importance of each variables, whether or not they significantly contributed to the prediction of the model. The rfPermute R-package [67] was used, which assess a null distribution of variable’s importance against the observed distribution. Each variable was shown if it was significantly different from the null distribution, considering a 5% significance level. We would like to emphasize that the R

^{2}here is considered as how well the predicted model fits to the 1:1 line, and not the proportion of the variance in the dependent variable that is predictable from the independent variable. To avoid confusion, it will be denoted as fitting R

^{2}. Fitting R

^{2}was computed from the following equation:

^{2}> 0.6 indicating high correlation. Although low R

^{2}can still be plausible if the RMSE is low. However, if the model shows extreme errors then the calculations may show negative R

^{2}values; in such case it is assumed to be an erroneous model. In addition, to understand what each variable contributes, a forward stepwise selection method was applied, adding one variable at a time starting with the one that resulted in the lowest residual sum of squares. The behavior of each variable to the result is checked. Table 1 summarizes the combination of the variables for each model. Furthermore, a combination of the variables were additionally modeled for three cases: minimum UAS variables (Model

_{UAS}: CHM

_{avg}, CA

_{avg}and FCC), SAR only variables (Model

_{SAR}: γ

^{0}

_{HV}, γ

^{0}

_{HH}, γ

^{0}

_{VV}and γ

^{0}

_{VH}) and both combined (Model

_{UAS+SAR}).

## 4. Results

#### 4.1. Correlation Matrix for Variable Comparison

_{avg}(r = 0.66) and CHM

_{min}(r = 0.66), RSP (r = −0.63), CHM

_{max}(r=0.60), followed by γ

^{0}

_{VV}(r = 0.51), FCC (r = 0.42) and the CA

_{avg}(r = −0.39). For a brief exploration of other variables correlations, CHM

_{min}sigificantly correlated with RSP (r = −0.60), followed by FCC (r = 0.54), γ

^{0}

_{HV}(r = 0.53) and other SAR variables

_{.}The SAR variables show where γ

^{0}

_{HH}correlates to CHM

_{min}(r = −0.37), γ

^{0}

_{HV}to CHM

_{min}(r = −0.53), CHM

_{avg}(r = −0.46) and FCC (r = −0.33). γ

^{0}

_{VH}although little lower correlation, similarity is shown to γ

^{0}

_{HV.}The SAR variables were not significant in their relationship to volume, except for γ

^{0}

_{VV}(r = 0.51). γ

^{0}

_{VV}was also correlated with RSP (r = −0.54) and CHM variables (r = 0.42−0.52). All of the SAR variables were not significant for the CA variables.

#### 4.2. Prediction Power of RFR and SVR

_{RFR18}(fitting R

^{2}= 0.665, RMSE = 66.87 m

^{3}/ha, rRMSE = 11.95%), which considered CHM

_{min}, CHM

_{avg}, γ

^{0}

_{VH}, FCC, γ

^{0}

_{VV}, CA

_{min}and CA

_{avg}variables. The SVR method showed lower accuracy compared to RFR, where the best model was Model

_{SVR2}(fitting R

^{2}= 0.519, RMSE = 80.12 m

^{3}/ha, rRMSE = 12.67%), which considered CHM

_{avg}variable. Figure 10 shows the result of the analysis with Model

_{UAS}, Model

_{SAR}and Model

_{UAS+SAR}for RFR. Presentation for the SVR is omitted due to the failure of the modeling with the same variables as RFR. Model

_{UAS}(CHM

_{avg}, CA

_{avg}and FCC variables) results R

^{2}= 0.06, RMSE = 111.90 m

^{3}/ha and rRMSE = 20.95%. Model

_{SAR}results R

^{2}= 0.11, RMSE = 109.23 m

^{3}/ha and rRMSE = 17.71%. Model

_{UAS+SAR}improves overall (R

^{2}= 0.51, RMSE = 80.90 m

^{3}/ha and rRMSE = 15.52%) and it is the only useful model compared to the former two, although majority of the validation plots were still overestimated.

^{2}, RMSE, and rRMSE values of each model computed. As mentioned, a forward stepwise selection method was implemented to determine how the model behaves with the addition of each variable. For the RFR, clearly the addition of each variable enhances both the fitting R

^{2}and RMSE. The fitting R

^{2}and RMSE both showed dramatic increase in accuracy, however beyond Model

_{18}, the results started to saturate. The results also show that using only the second variable (CHM

_{avg}) can perform estimation fairly well (fitting R

^{2}= 0.173, RMSE = 105.09 m

^{3}/ha, rRMSE = 20.6%), from which we can presume that the volume is contributed more by the vertical structure (i.e., tree height). Variables related to canopy area (Model

_{6–8}) show higher errors when used alone, which is understandable since the volume is difficult to estimate only from the horizontal information. There are differences of results for each variables, though in general it can be said that multiple regression performs better than single variable regression. SVR showed a similar trend of reducing errors in RMSE with increasing variables, although not much improvement was seen for the fitting R

^{2}. The RMSE showed lower errors when all variables were combined (Model

_{SVR23}: fitting R

^{2}= 0.126, RMSE = 108.01 m

^{3}/ha, rRMSE = 19.92%), however it achieved a better result just by using the second variable (Model

_{SVR2}: fitting R

^{2}= 0.519, RMSE = 80.12 m

^{3}/ha, rRMSE = 12.67%). Some combinations, such as from Model

_{SVR15}, failed to delineate the information from each variable, resulting in low fitting R

^{2}(negative value) and high RMSE and rRMSE, but gradually improve towards the full model. SVR can have better chances using the single influential variable and needs care when selecting multiple variables.

#### 4.3. Importance and Significance of Variables (RFR)

_{RFR23}) was CHM

_{min}, shown in Figure 12 for the Gini importance measure. The variables following were CHM

_{avg}, FCC, CA

_{avg}, CA

_{min}, CA

_{max}and CHM

_{max}, which were all indicated to be significant for partitioning the data (red bars). The additional models Model

_{UAS}, Model

_{SAR}, and Model

_{UAS+SAR}are shown also. For Model

_{UAS+SAR}, the UAS variables show that the influencing parameter for the estimation is significant, while the SAR variables, although shown to have some influence, were not significant. Using both variable sets on their own (Model

_{UAS}and Model

_{SAR}) showed that all the variables were significant in partitioning the data. The ranking of the Model

_{SAR}variables changed from Model

_{RFR23}, where γ

^{0}

_{VV}influenced more than γ

^{0}

_{VH}.

## 5. Discussion

#### 5.1. Challenges of Multiple Regression Analysis with Multi-Sensor Data

^{2}= 0.665, RMSE = 66.87 m

^{3}/ha (rRMSE = 11.95%) for RFR and R

^{2}= 0.519 and RMSE = 80.12 m

^{3}/ha (rRMSE = 12.67%) for SVR. They could perform better compared to the predictive power of multilinear regression (MLR) [69,70,71]; in contrast to other works, this work focused on a wide range of volumes, from approximately 100 to 800 m

^{3}/ha. If we look at a case with a stem volume range similar to that of Adbullahi et al. [71] (up to 1049 m

^{3}/ha), this work produces a better plot level estimate (although it focuses on mixed forests and the plot size is different: 500 m

^{2}circular plots) compared to using only X-band SAR data with its best accuracy of 155 m

^{3}/ha (rRMSE = 41.90%). The author discusses the possible saturation effect of the radar signals limiting the prediction. SAR saturation effects are always in discussion, and this could be caused from the large spatial unit aggregating the variations in backscattering signals (other than the limitation of radar frequency). For example, Iizuka et al. [72] tested the NDVI response to FCC with different grid resolutions. With coarser resolution, the range of the NDVI became more limited, indicating that variations were aggregated in larger units. Similar results might be shown with SAR, which limits extracting the feature information at a limiting area with lower resolution data. It was not clear in our results whether such effects were causing limitations, nevertheless, they should be noted when SAR is used in dense forest regions. This work also shows improvements compared to the works by Shataeea et al. [73] by using airborne laser scanner (ALS) and Landsat-TM data at plot level (plot size of the grid of TM data) with RF yielding 179.39 m

^{3}/ha (rRMSE = 42.93%), expressing that by adding the third dimension (i.e., tree height) data to the variable can enhance in the estimation. This was done by adding CHM information in our work. Although the ALS can produce data at a larger spatial extent, the footprints of laser scanning density may result in coarser tree information (e.g., crown delineation), although this can be overcome when utilizing UAS information. Unfortunately, similar work with the same observational scale and the integration of a multi-sensor approach was not found. Therefore, it is difficult to directly compare our results with others. Navarro et al. [40] showed the integration of UAS and Sentinel-1 and -2 for above ground biomass (AGB) estimation of mangroves in Senegal. Although it is expressed as integration, the UAS-based measurements were used as a reference for estimation with Sentinel data, and only Sentinel-1 and -2 were used. Their SVR approach using both data sets shows slightly lower validation RMSE (RMSE = 2.35 Mg/ha) compared with using only Sentinel-1 (RMSE = 2.22 Mg/ha), but improved from using only Sentinel-2 data (RMSE = 3.74 Mg/ha). Similar to our results, the multiple regression of SVR is failing. The approach of optical and SAR is possible, but needs more investigation in higher density forests.

^{2}= 0.665 (ΔR

^{2}= 0.146)), and the RMSE was much lower than that of the SVR approach (66.87 m

^{3}/ha (ΔRMSE = 13.25 m

^{3}/ha)). The test was conducted in a small area with few sample plots (n = 33) and with a small spatial extent. It might not compensate with broad scale analysis (unless such fine resolution data is collected broadly), however the demands of quantifying at a local scale (and extreme site dependence) might be possible from applying the proposed method. Considering the cost-effectiveness, utilizing SAR information with the aid of UAS variables of tree height and diameter would be an improvement.

#### 5.2. Variable Selections

#### 5.2.1. UAS Remote Sensing Variables

#### 5.2.2. TLS Variables

#### 5.2.3. Radar Variables

^{0}

_{HH}, γ

^{0}

_{HV}, γ

^{0}

_{VV}, and γ

^{0}

_{VH}to estimate the volume, and the RFR resulting had moderate accuracy. The results indicate that it is possible to use the RFR approach for plot level analysis, even by using only SAR variables. Implementing SAR data is still a major candidate for global estimations for its advantages of weather and daylight independence. But it has been shown in this work that potentials can be seen for assessing local (plot-level) analysis by incorporating finer detailed information, such as from UASs. Incorporating the PALSAR and Sentinel-1 data can also produce a stronger prediction because of the incorporation of information from both the bottom and top layers of the forests [78]. Including the third variable for the forest structure (e.g., tree height) should improve its predictions. γ

^{0}

_{HH}was correlated with CHM

_{min}, γ

^{0}

_{HV}and γ

^{0}

_{VH}was correlated with CHM

_{min}and CHM

_{avg}. γ

^{0}

_{VV}was the only one correlating with all CHM variables. It could be noted that C-band co-polarization response is affected more by the vertical structure of objects, and thus has a similar contribution to CHM. The cross-polarized variable usually interacts with volume, however it was not correlated significantly, and only γ

^{0}

_{VV}did so. The stem volume of Chamaecyparis obtusa is influenced more by tree height, thus the γ

^{0}

_{VV}correlating to all CHM variables corresponded also to volume. The complex local topography could have affected the backscattering information, which could not be as expected from other works related to L-band estimation (especially at plot level), and moreover scale difference of data resolution might have affected characterization of the local backscattering trend (brief explanation in Section 5.3). Considering the fine spatial/temporal resolution of Sentinel-1 and its availability free to the public, it can be recommended for further investigations.

#### 5.3. Model Errors

#### 5.4. Scale Difference of Multi-Sensor Approach

#### 5.5. Data Processing of Point Clouds

#### 5.6. Beyond Precision Forestry

## 6. Conclusions

^{2}= 0.665, RMSE = 66.87 m

^{3}/ha, and rRMSE =11.95%. The SVR approach showed a good correlation when a single variable was used (fitting R

^{2}= 0.519, RMSE = 80.12 m

^{3}/ha and rRMSE =12.67%), however its accuracy was reduced when multiple variables were combined (fitting R

^{2}= 0.126, RMSE = 108.01 m

^{3}/ha and rRMSE =19.92%). TLS data reveals the precise ground truth information of stem volume and terrain data. Ultra-high resolution UAS imagery has contributed to delineating forest attributes at the individual stand level, and SAR backscattering corresponded mainly with the vertical structure of the forests. The Gini index has presented the most influential variable for RFR modeling with UAS variables (canopy height, canopy size, and canopy cover), ranked by CHM

_{min}, CHM

_{avg}, FCC, CA

_{avg}, CA

_{min}, CA

_{max}, and CHM

_{max}. Although the case study was set in a small area, the experiment showed the potential for the incorporation of conventional remote sensing data to outperform traditional methods, utilizing only LiDAR, UAS photogrammetry, or satellite data with RFR. The proposed method could provide important information on upcoming trends in precision forestry, as various global activities need strategic actions in response to REDD+, the Paris Agreements, and the SDGs. Future challenges will be considered in the development of a comprehensive model for estimating forest parameters over broader regions and potentially at a landscape scale.

## Author Contributions

## Funding

## Conflicts of Interest

## References

- Holopainen, M.; Vastaranta, M.; Hyyppä, J. Outlook for the Next Generation’s Precision Forestry in Finland. Forests
**2014**, 5, 1682–1694. [Google Scholar] [CrossRef] [Green Version] - O’Brien, M.; Bringezu, S. Assessing the Sustainability of EU Timber Consumption Trends: Comparing Consumption Scenarios with a Safe Operating Space Scenario for Global and EU Timber Supply. Land
**2017**, 6, 84. [Google Scholar] [CrossRef] [Green Version] - Iizuka, K.; Tateishi, R. Estimation of CO
_{2}Sequestration by the Forests in Japan by Discriminating Precise Tree Age Category using Remote Sensing Techniques. Remote Sens.**2015**, 7, 15082–15113. [Google Scholar] [CrossRef] [Green Version] - Di Lallo, G.; Mundhenk, P.; Zamora López, S.E.; Marchetti, M.; Köhl, M. REDD+: Quick Assessment of Deforestation Risk Based on Available Data. Forests
**2017**, 8, 29. [Google Scholar] [CrossRef] - Climate Focus. Forests and Land Use in the Paris Agreement. The Paris Agreement Summary. 2015. Available online: http://www.climatefocus.com/publications/cop21-paris-2015-climate-focus-overall-summary-and-client-briefs (accessed on 20 May 2020).
- Bastos Lima, M.G.; Kissinger, G.; Visseren-Hamakers, I.J.; Braña-Varela, J.; Gupta, A. The Sustainable Development Goals and REDD+: Assessing institutional interactions and the pursuit of synergies. Int. Environ. Agreem.
**2017**, 17, 589–606. [Google Scholar] [CrossRef] [Green Version] - Němec, P. Comparison of modern forest inventory method with the common method for management of tropical rainforest in the Peruvian Amazon. J. Trop. For. Sci.
**2015**, 27, 80–91. [Google Scholar] - Fazakas, Z.; Nilsson, M.; Olsson, H. Regional forest biomass and wood volume estimation using satellite data and ancillary data. Agric. For. Meteorol.
**1999**, 98–99, 417–425. [Google Scholar] [CrossRef] - Sandberg, G.; Ulander, L.M.H.; Fransson, J.E.S.; Holmgren, J.; Le Toan, T. L- and P-band backscatter intensity for biomass retrieval in hemiboreal forest. Remote Sens. Environ.
**2011**, 115, 2874–2886. [Google Scholar] [CrossRef] - Naidoo, L.; Mathieu, R.; Main, R.; Kleynhans, W.; Wessels, K.; Asner, G.; Leblon, B. Savannah woody structure modelling and mapping using multi-frequency (X-, C- and L-band) Synthetic Aperture Radar data. ISPRS J. Photogramm. Remote Sens.
**2015**, 105, 234–250. [Google Scholar] [CrossRef] [Green Version] - Dash, J.P.; Watt, M.S.; Bhandari, S.; Watt, P. Characterising forest structure using combinations of airborne laser scanning data, RapidEye satellite imagery and environmental variables. For. Int. J. For. Res.
**2016**, 89, 159–169. [Google Scholar] [CrossRef] [Green Version] - Barrett, F.; McRoberts, R.E.; Tomppo, E.; Cienciala, E.; Waser, L.T. A questionnaire-based review of the operational use of remotely sensed data by national forest inventories. Remote Sens. Environ.
**2016**, 174, 279–289. [Google Scholar] [CrossRef] - Saarela, S.; Grafström, A.; Ståhl, G.; Kangas, A.; Holopainen, M.; Tuominen, S.; Nordkvist, K.; Hyyppä, J. Model-assisted estimation of growing stock volume using different combinations of LiDAR and Landsat data as auxiliary information. Remote Sens. Environ.
**2015**, 158, 431–440. [Google Scholar] [CrossRef] - Goodbody, T.R.H.; Coops, N.C.; Marshall, P.L.; Tompalski, P.; Crawford, P. Unmanned aerial systems for precision forest inventory purposes: A review and case study. For. Chron.
**2017**, 93, 71–81. [Google Scholar] [CrossRef] [Green Version] - Sanga-Ngoie, K.; Iizuka, K.; Kobayashi, S. Estimating CO
_{2}Sequestration by Forests in Oita Prefecture, Japan, by Combining LANDSAT ETM+ and ALOS Satellite Remote Sensing Data. Remote Sens.**2012**, 4, 3544–3570. [Google Scholar] [CrossRef] [Green Version] - Vaglio Laurin, G.; Pirotti, F.; Callegari, M.; Chen, Q.; Cuozzo, G.; Lingua, E.; Notarnicola, C.; Papale, D. Potential of ALOS2 and NDVI to Estimate Forest Above-Ground Biomass, and Comparison with Lidar-Derived Estimates. Remote Sens.
**2017**, 9, 18. [Google Scholar] [CrossRef] [Green Version] - Badreldin, N.; Sanchez-Azofeifa, A. Estimating Forest Biomass Dynamics by Integrating Multi-Temporal Landsat Satellite Images with Ground and Airborne LiDAR Data in the Coal Valley Mine, Alberta, Canada. Remote Sens.
**2015**, 7, 2832–2849. [Google Scholar] [CrossRef] [Green Version] - Richards, J.A. Remote Sensing with Imaging Radar; Springer: New York, NY, USA, 2009. [Google Scholar]
- Dobson, M.C.; Ulaby, F.T.; Le Toan, T.; Beaudoin, A.; Kasischke, E.S.; Christensen, N. Dependence of Radar Backscatter on Coniferous Forest Biomass. IEEE Trans. Geosci. Remote Sens.
**1992**, 30, 412–415. [Google Scholar] [CrossRef] - Santoro, M.; Fransson, J.E.S.; Eriksson, L.E.B.; Magnusson, M.; Ulander, L.M.H.; Olsson, H. Signatures of ALOS PALSAR L-Band Backscatter in Swedish Forest. IEEE Trans. Geosci. Remote Sens.
**2009**, 47, 4001–4019. [Google Scholar] [CrossRef] [Green Version] - Lucas, R.M.; Armston, J.; Fairfax, R.; Fensham, R.; Accad, A.; Carreiras, J.; Kelley, J.; Bunting, P.; Clewley, D.; Bray, S.; et al. An Evaluation of the ALOS PALSAR L-Band Backscatter―Above Ground Biomass Relationship Queensland, Australia: Impacts of Surface Moisture Condition and Vegetation Structure. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.
**2010**, 3, 576–593. [Google Scholar] [CrossRef] - Motohka, T.; Shimada, M.; ISoguchi, O.; Ishihara, M.I.; Suzuki, S.N. Relationships between PALSAR Backscattering Data and Forest Above Ground Biomass in Japan. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium 2011, Vancouver, BC, Canada, 24–29 July 2011; pp. 3518–3521. [Google Scholar]
- Kobayashi, S.; Widyorini, R.; Kawai, S.; Omura, Y.; Sanga-Ngoie, K.; Supriadi, B. Backscattering Characteristics of L-Band Polarimetric and Optical Satellite Imagery over Planted Acacia Forests in Sumatra, Indonesia. J. Appl. Remote Sens.
**2012**, 6, 063519–063525. [Google Scholar] - Iizuka, K.; Tateishi, R. Simple Relationship Analysis between L-Band Backscattering Intensity and the Stand Characteristics of Sugi (Cryptomeria japonica) and Hinoki (Chamaecyparis obtusa) Trees. Adv. Remote Sens.
**2014**, 3, 219–234. [Google Scholar] [CrossRef] [Green Version] - Srinivasan, S.; Popescu, S.C.; Eriksson, M.; Sheridan, R.D.; Ku, N.-W. Terrestrial Laser Scanning as an Effective Tool to Retrieve Tree Level Height, Crown Width, and Stem Diameter. Remote Sens.
**2015**, 7, 1877–1896. [Google Scholar] [CrossRef] [Green Version] - Hayakawa, Y.S.; Kusumoto, S.; Matta, N. Application of terrestrial laser scanning for detection of ground surface deformation in small mud volcano (Murono, Japan). Earth Planets Space
**2016**, 68, 114. [Google Scholar] [CrossRef] [Green Version] - Momo Takoudjou, S.; Ploton, P.; Sonké, B.; Hackenberg, J.; Griffon, S.; De Coligny, F.; Kamdem, N.G.; Libalah, M.; Mofack, G.I.; Le Moguédec, G.; et al. Using terrestrial laser scanning data to estimate large tropical trees biomass and calibrate allometric models: A comparison with traditional destructive approach. Methods Ecol. Evol.
**2018**, 9, 905–916. [Google Scholar] - Flynn, K.F.; Chapra, S.C. Remote Sensing of Submerged Aquatic Vegetation in a Shallow Non-Turbid River Using an Unmanned Aerial Vehicle. Remote Sens.
**2014**, 6, 12815–12836. [Google Scholar] [CrossRef] [Green Version] - Getzin, S.; Nuske, R.S.; Wiegand, K. Using Unmanned Aerial Vehicles (UAV) to Quantify Spatial Gap Patterns in Forests. Remote Sens.
**2014**, 6, 6988–7004. [Google Scholar] [CrossRef] [Green Version] - Luna, I.; Lobo, A. Mapping Crop Planting Quality in Sugarcane from UAV Imagery: A Pilot Study in Nicaragua. Remote Sens.
**2016**, 8, 500. [Google Scholar] [CrossRef] [Green Version] - Ota, T.; Ogawa, M.; Shimizu, K.; Kajisa, T.; Mizoue, N.; Yoshida, S.; Takao, G.; Hirata, Y.; Furuya, N.; Sano, T.; et al. Aboveground Biomass Estimation Using Structure from Motion Approach with Aerial Photographs in a Seasonal Tropical Forest. Forests
**2015**, 6, 3882–3898. [Google Scholar] [CrossRef] [Green Version] - Jucker, T.; Caspersen, J.; Chave, J.; Antin, C.; Barbier, N.; Bongers, F.; Dalponte, M.; van Ewijk, K.Y.; Forrester, D.I.; Haeni, M.; et al. Allometric equations for integrating remote sensing imagery into forest monitoring programmes. Glob. Chang. Biol.
**2017**, 23, 177–190. [Google Scholar] [CrossRef] - Panagiotidis, D.; Abdollahnejad, A.; Surový, P.; Chiteculo, V. Determining tree height and crown diameter from high-resolution UAV imagery. Int. J. Remote Sens.
**2017**, 38, 2392–2410. [Google Scholar] [CrossRef] - Iizuka, K.; Yonehara, T.; Itoh, M.; Kosugi, Y. Estimating Tree Height and Diameter at Breast Height (DBH) from Digital Surface Models and Orthophotos Obtained with an Unmanned Aerial System for a Japanese Cypress (Chamaecyparis obtusa). For. Remote Sens.
**2018**, 10, 13. [Google Scholar] [CrossRef] [Green Version] - Jaakkola, A.; Hyyppä, J.; Yu, X.; Kukko, A.; Kaartinen, H.; Liang, X.; Hyyppä, H.; Wang, Y. Autonomous Collection of Forest Field Reference—The Outlook and a First Step with UAV Laser Scanning. Remote Sens.
**2017**, 9, 785. [Google Scholar] [CrossRef] [Green Version] - Schlund, M.; Davidson, M.W.J. Aboveground Forest Biomass Estimation Combining L- and P-Band SAR Acquisitions. Remote Sens.
**2018**, 10, 1151. [Google Scholar] [CrossRef] [Green Version] - Shao, Z.; Zhang, L. Estimating Forest Aboveground Biomass by Combining Optical and SAR Data: A Case Study in Genhe, Inner Mongolia, China. Sensors
**2016**, 16, 834. [Google Scholar] [CrossRef] [Green Version] - Cutler, M.; Boyd, D.; Foody, G.; Vetrivel, A. Estimating tropical forest biomass with a combination of SAR image texture and Landsat TM data: An assessment of predictions between regions. ISPRS J. Photogramm. Remote Sens.
**2012**, 70, 66–77. [Google Scholar] [CrossRef] [Green Version] - Karlson, M.; Ostwald, M.; Reese, H.; Sanou, J.; Tankoano, B.; Mattsson, E. Mapping Tree Canopy Cover and Aboveground Biomass in Sudano-Sahelian Woodlands Using Landsat 8 and Random Forest. Remote Sens.
**2015**, 7, 10017–10041. [Google Scholar] [CrossRef] [Green Version] - Navarro, J.A.; Algeet, N.; Fernández-Landa, A.; Esteban, J.; Rodríguez-Noriega, P.; Guillén-Climent, M.L. Integration of UAV, Sentinel-1, and Sentinel-2 Data for Mangrove Plantation Aboveground Biomass Monitoring in Senegal. Remote Sens.
**2019**, 11, 77. [Google Scholar] [CrossRef] [Green Version] - Kosugi, Y.; Takanashi, S.; Ueyama, M.; Ohkubo, S.; Tanaka, H.; Matsumoto, K.; Yoshifuji, N.; Ataka, M.; Sakabe, A. Determination of the gas exchange phenology in an evergreen coniferous forest from 7 years of eddy covariance flux data using an extended big-leaf analysis. Ecol. Res.
**2013**, 28, 373–385. [Google Scholar] [CrossRef] - Japan Aerospace Exploration Agency (JAXA). PALSAR Calibration Factor Updated. Available online: http://www.eorc.jaxa.jp/en/about/distribution/info/alos/20090109en_3.html (accessed on 12 June 2018).
- Small, D. Flattening gamma: Radiometric terrain correction for SAR imagery. IEEE Trans. Geosci. Remote Sens.
**2011**, 49, 3081–3093. [Google Scholar] [CrossRef] - Omar, H.; Misman, M.A.; Kassim, A.R. Synergetic of PALSAR-2 and Sentinel-1A SAR Polarimetry for Retrieving Aboveground Biomass in Dipterocarp Forest of Malaysia. Appl. Sci.
**2017**, 7, 675. [Google Scholar] [CrossRef] [Green Version] - Lee, J.S. Speckle suppression and analysis for synthetic aperture radar images. Opt. Eng.
**1986**, 25, 636–643. [Google Scholar] [CrossRef] - Mlambo, R.; Woodhouse, I.H.; Gerard, F.; Anderson, K. Structure from Motion (SfM) Photogrammetry with Drone Data: A Low Cost Method for Monitoring Greenhouse Gas Emissions from Forests in Developing Countries. Forests
**2017**, 8, 68. [Google Scholar] [CrossRef] [Green Version] - Girardeau-Montaut, D. CloudCompare. Available online: http://www.cloudcompare.org/ (accessed on 13 June 2018).
- Conrad, O.; Bechtel, B.; Bock, M.; Dietrich, H.; Fischer, E.; Gerlitz, L.; Wehberg, J.; Wichmann, V.; Böhner, J. System for Automated Geoscientific Analyses (SAGA) v. 2.1.4. Geosci. Model Dev.
**2015**, 8, 1991–2007. [Google Scholar] [CrossRef] [Green Version] - Trimble Navigation Limited (2012) Datasheet Trimble TX5 Scanner. Available online: http://www.trimble.com/globalTRL.asp?nav=Collection-91149 (accessed on 2 July 2018).
- Yamamoto, W. Forest inventory of Japanese red pine for stem volume and diameter at breast height (あかまつノ単木幹材積表並胸高形数表). Bull. For. Exp.
**1918**, 16, 147–164. (In Japanese) [Google Scholar] - Schumacher, F.X.; Hall, F.D.S. Logarithmic expression of timber-tree volume. J. Agric. Res.
**1933**, 47, 719–734. [Google Scholar] - Hosoda, K.; Mitsuda, Y.; Iehara, T. Differences between the present stem volume tables and the values of the volume equations, and their correction. Jpn. Soc. For. Plan.
**2010**, 44, 23–39. (In Japanese) [Google Scholar] - Zhang, W.; Qi, J.; Wan, P.; Wang, H.; Xie, D.; Wang, X.; Yan, G. An Easy-to-Use Airborne LiDAR Data Filtering Method Based on Cloth Simulation. Remote Sens.
**2016**, 8, 501. [Google Scholar] [CrossRef] - Adams, H.R.; Barnard, H.R.; Loomis, A.K. Topography alters tree growth–climate relationships in a semi-arid forested catchment. Ecosphere
**2014**, 5, 148. [Google Scholar] [CrossRef] - MacMillan, R.A.; Pettapiece, W.W.; Nolan, S.C.; Goddard, T.W. A generic procedure for automatically segmenting landforms into landform elements using DEMs, heuristic rules and fuzzy logic. Fuzzy Sets Syst.
**2000**, 113, 81–109. [Google Scholar] [CrossRef] - Breiman, L. Random Forests. Mach. Learn.
**2001**, 45, 5–32. [Google Scholar] [CrossRef] [Green Version] - Boser, B.E.; Guyon, I.M.; Vapnik, V.N. A Training Algorithm for Optimal Margin Classifiers. In Proceedings of the 5th Annual Workshop on Computational Learning Theory (COLT’92), Pittsburgh, PA, USA, 27–29 July 1992; pp. 144–152. [Google Scholar]
- Shao, Y.; Lunetta, R.S. Comparison of support vector machine, neural network, and CART algorithms for the land-cover classification using limited training data points. ISPRS J. Photogramm. Remote Sens.
**2012**, 70, 78–87. [Google Scholar] [CrossRef] - Negri, R.G.; Dutra, L.V.; Sant’Anna, S.J.S. An innovative support vector machine based method for contextual image classification. ISPRS J. Photogramm. Remote Sens.
**2014**, 87, 241–248. [Google Scholar] [CrossRef] - Wu, J.; Yao, W.; Choi, S.; Park, T.; Myneni, R.B. A Comparative Study of Predicting DBH and Stem Volume of Individual Trees in a Temperate Forest Using Airborne Waveform LiDAR. IEEE Geosci. Remote Sens. Lett.
**2015**, 12, 2267–2271. [Google Scholar] [CrossRef] - Marabel, M.; Alvarez-Taboada, F. Spectroscopic Determination of Aboveground Biomass in Grasslands Using Spectral Transformations, Support Vector Machine and Partial Least Squares Regression. Sensors
**2013**, 13, 10027–10051. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Gao, Y.; Lu, D.; Li, G.; Wang, G.; Chen, Q.; Liu, L.; Li, D. Comparative Analysis of Modeling Algorithms for Forest Aboveground Biomass Estimation in a Subtropical Region. Remote Sens.
**2018**, 10, 627. [Google Scholar] [CrossRef] [Green Version] - Liaw, A.; Wiener, M. Classification and Regression by randomForest. R News
**2002**, 2, 18–22. [Google Scholar] - Meyer, D.; Dimitriadou, E.; Hornik, K.; Weingessel, A.; Leisch, F. e1071: Misc Functions of the Department of Statistics, Probability Theory Group (Formerly: E1071), TU Wien. R Package Version 1.7-3. 2019. Available online: https://CRAN.R-project.org/package=e1071 (accessed on 20 May 2020).
- Mounce, S.R.; Mounce, R.B.; Boxall, J.B. Novelty detection for time series data analysis in water distribution systems using support vector machines. J. Hydroinform.
**2011**, 13, 672–686. [Google Scholar] [CrossRef] - García, M.; Riaño, D.; Chuvieco, E.; Salas, J.F.; Danson, M. Multispectral and LiDAR data fusion for fuel type mapping using Support Vector Machine and decision rules. Remote Sens. Environ.
**2011**, 115, 1369–1379. [Google Scholar] [CrossRef] - Archer, E. rfPermute: Estimate Permutation p-Values for Random Forest Importance Metrics. R Package Version 2.1.81. 2020. Available online: https://CRAN.R-project.org/package=rfPermute (accessed on 20 May 2020).
- Alexander, D.L.J.; Tropsha, A.; Winkler, D.A. Beware of R
^{2}: Simplee, unambiguous assessment of the prediction accuracy of QSAR and QSPR models. J. Chem. Inf. Model.**2015**, 55, 1316–1322. [Google Scholar] [CrossRef] [Green Version] - Lindberg, E.; Hollaus, M. Comparison of Methods for Estimation of Stem Volume, Stem Number and Basal Area from Airborne Laser Scanning Data in a Hemi-Boreal Forest. Remote Sens.
**2012**, 4, 1004–1023. [Google Scholar] [CrossRef] [Green Version] - He, Q.; Chen, E.; An, R.; Li, Y. Above-Ground Biomass and Biomass Components Estimation Using LiDAR Data in a Coniferous Forest. Forests
**2013**, 4, 984–1002. [Google Scholar] [CrossRef] [Green Version] - Abdullahi, S.; Kugler, F.; Pretzsch, H. Prediction of stem volume in complex temperate forest stands using TanDEM-X SAR data. Remote Sens. Environ.
**2016**, 174, 197–211. [Google Scholar] [CrossRef] - Iizuka, K.; Kato, T.; Silsigia, S.; Soufiningrum, A.Y.; Kozan, O. Estimating and Examining the Sensitivity of Different Vegetation Indices to Fractions of Vegetation Cover at Different Scaling Grids for Early Stage Acacia Plantation Forests Using a Fixed-Wing UAS. Remote Sens.
**2019**, 11, 1816. [Google Scholar] [CrossRef] [Green Version] - Shataeea, S.; Weinaker, H.; Babanejad, M. Plot-level Forest Volume Estimation Using Airborne Laser Scanner and TM Data, Comparison of Boosting and Random Forest Tree Regression Algorithms. Procedia Environ. Sci.
**2011**, 7, 68–73. [Google Scholar] [CrossRef] [Green Version] - Sumida, A.; Miyaura, T.; Toori, H. Relationships of tree height and diameter at breast height revisited: Analyses of stem growth using 20-year data of an even-aged Chamaecyparis obtusa stand. Tree Physiol.
**2013**, 33, 106–118. [Google Scholar] [CrossRef] [PubMed] - Nagakura, J.; Shigenaga, H.; Akama, A.; Takahashi, M. Growth and transpiration of Japanese cedar (Cryptomeria japonica) and Hinoki cypress (Chamaecyparis obtusa) seedlings in response to soil water content. Tree Physiol.
**2004**, 23, 1203–1208. [Google Scholar] [CrossRef] - Kobayashi, S.; Omura, Y.; Sanga-Ngoie, K.; Widyorini, R.; Kawai, S.; Supriadi, B.; Yamaguchi, Y. Characteristics of Decomposition Powers of L-Band Multi-Polarimetric SAR in Assessing Tree Growth of Industrial Plantation Forests in the Tropics. Remote Sens.
**2012**, 4, 3058–3077. [Google Scholar] [CrossRef] [Green Version] - Varghese, A.O.; Suryavanshi, A.; Joshi, A.K. Analysis of different polarimetric target decomposition methods in forest density classification using C band SAR data. Int. J. Remote Sens.
**2016**, 37, 694–709. [Google Scholar] [CrossRef] - Sivasankar, T.; Lone, J.M.; Sarma, K.K.; Qadir, A.; Raju, P.L.N. The potential of multi-frequency multipolarized ALOS-2/PALSAR-2 and Sentinel-1 SAR data for aboveground forest biomass estimation. Int. J. Eng. Technol.
**2018**, 10, 797–802. [Google Scholar] [CrossRef] [Green Version] - Jin, S.; Su, Y.; Gao, S.; Hu, T.; Liu, J.; Guo, Q. The Transferability of Random Forest in Canopy Height Estimation from Multi-Source Remote Sensing Data. Remote Sens.
**2018**, 10, 1183. [Google Scholar] [CrossRef] [Green Version] - Wu, H.; Li, Z.-L. Scale Issues in Remote Sensing: A Review on Analysis, Processing and Modeling. Sensors
**2009**, 9, 1768–1793. [Google Scholar] [CrossRef] [PubMed] - Jelinski, D.E.; Wu, J. The modifiable areal unit problem and implications for landscape ecology. Landsc. Ecol.
**1996**, 11, 129–140. [Google Scholar] [CrossRef] - Ulander, L.M.H.; Smith, G.; Eriksson, L.; Folkesson, K.; Fransson, J.E.S.; Gustavsson, A.; Hallberg, B.; Joyce, S.; Magnusson, M.; Olsson, H.; et al. Mapping of wind-thrown forests in southern Sweden using space- and airborne SAR. In Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS), Seoul, Korea, 25–29 July 2005; pp. 3619–3622. [Google Scholar]
- Carrer, M.; Castagneri, D.; Popa, I.; Pividori, M.; Lingua, E. Tree spatial patterns and stand attributes in temperate forests: The importance of plot size, sampling design, and null model. For. Ecol. Manag.
**2018**, 407, 125–134. [Google Scholar] [CrossRef] - Saarinen, N.; Kankare, V.; Vastaranta, M.; Luoma, V.; Pyörälä, J.; Tanhuanpää, T.; Liang, X.; Kaartinen, H.; Kukko, A.; Jaakkola, A.; et al. Feasibility of Terrestrial laser scanning for collecting stem volume information from single trees. ISPRS J. Photogramm. Remote Sens.
**2017**, 123, 140–158. [Google Scholar] [CrossRef] - Roşca, S.; Suomalainen, J.; Bartholomeus, H.; Herold, M. Comparing terrestrial laser scanning and unmanned aerial vehicle structure from motion to assess top of canopy structure in tropical forests. Interface Focus
**2018**, 8, 20170038. [Google Scholar] [CrossRef] [PubMed] - Tian, J.; Dai, T.; Li, H.; Liao, C.; Teng, W.; Hu, Q.; Ma, W.; Xu, Y. A Novel Tree Height Extraction Approach for Individual Trees by Combining TLS and UAV Image-Based Point Cloud Integration. Forests
**2019**, 10, 537. [Google Scholar] [CrossRef] [Green Version] - Forestry Agency, Japan. State of Japan’s Forests and Forest Management. 2019. Available online: https://www.maff.go.jp/e/policies/forestry/attach/pdf/index-8.pdf (accessed on 16 May 2020).

**Figure 1.**Characteristics of the Kiryu Experimental Watershed. Moderately dense trees are seen at the site with heterogeneous microtopography. The middle image shows the flux tower at the site from the surface, and the right image shows the flux tower and the surrounding landscape as seen from the unmanned aerial system (UAS) (Photo by author: 19 December, 2017). Detail of geographical location is shown by Iizuka et al. [34].

**Figure 2.**Flowchart of the methodology. Terrestrial laser scanning (TLS) was used for ground truthing and generation of the digital terrain model (DTM). A UAS was utilized for collecting aerial images, which were then processed to obtain a digital surface model (DSM) and orthoimages. Phased Array type L-band Synthetic Aperture Radar-2 (PALSAR-2) and Sentinel-1 data were preprocessed and used as independent variables. All variables were incorporated in the end with different regression analysis for the test.

**Figure 3.**(

**a**) Orthoimage of the study site, and (

**b**) DSM generated from the aerial imageries and the structure from motion (SfM) method. 345 images taken from the flight, the SfM method was successfully applied to the site.

**Figure 4.**(

**a**) Full dense point cloud of the study site and (

**b**) smoothed point cloud processed with the moving least squares (MLS) method. (

**c**) The polygons generated from the smoothed point cloud using the watershed segmentation, and (

**d**) the final canopy shape from masking the canopy gaps and shadows. Random colors represent the canopy shapes, while black is the background (gap and shadows).

**Figure 5.**(

**a**) Merged TLS scans of the study site, and (

**b**) example of the extracted sample area with the 10 by 10 m grid size. Each individual tree was measured to compute tree height and diameter at breast height (DBH) within the grids.

**Figure 6.**(

**a**) Total stem volume per grid (m

^{3}/ha) and (

**b**) the tree height–DBH relationship curve for the trees within the sample grids (n = 403). Shadow and canopy gaps are masked out in the raster image. The crosses and circles on each grids indicates the plot area used for training (21 plots) and validation (12 plots), respectively.

**Figure 7.**Dense point cloud of the merged UAS SfM points and TLS points of the study site (

**a**) before the CSF filter and (

**b**) after the filtering. The CSF successfully excluded the top surface objects (e.g., trees, the tower, etc.) and left only the surface terrain.

**Figure 8.**Correlation matrix with significant levels between stem volume and other 12 variables. Diagonal matrices show the histogram of the data distribution. The left half is the bivariate scatter plot with fitted line. The right half shows the Pearson correlation and the significance level associated with a symbol: p-values < 0.001 (***), < 0.01 (**), < 0.05 (*).

**Figure 9.**Model validation results from the regression analysis. Comparison of the reference and predicted stem volumes at each validation plots (indicated in Figure 6a) for (

**a**) random forest regression (RFR) and (

**b**) support vector regression (SVR). The figure shows the best result from the modelling. RFR = Model

_{RFR18}; SVR = Model

_{SVR2}. The solid black line indicates the 1:1 line. The dotted line is the linear trend line between the predicted and reference volume just for the visual reference, and it does not represent the R

^{2}shown.

**Figure 10.**Results from the RFR analysis. Comparison of the reference and predicted stem volumes for additional models using, (

**a**) minimum UAS variables (CHM

_{avg}, CA

_{avg}, FCC) (

**b**) SAR variables, and (

**c**) UAS+SAR variables. The solid black line indicates the 1:1 line. The dotted line is the linear trend line between the predicted and reference volume just for the visual reference, and it does not represent the R

^{2}shown.

**Figure 11.**Summary of the results for all regression analyses tested. Each result for the correlation of determination (R

^{2}), root mean square error (RMSE), and relative RMSE (rRMSE) is shown for (

**a**) RFR and (

**b**) SVR. Only positive R

^{2}is drawn and the models and without R

^{2}plots indicates all negative fitting R

^{2}, hence erroneous models.

**Figure 12.**Importance of each variables explained as the mean increase in node purity (i.e., decrease in node impurity: Gini Index) for (

**a**) Model

_{RFR23}, (

**b**) Model

_{UAS}, (

**c**) Model

_{SAR}, and (

**d**) Model

_{UAS+SAR}. The upper shows more influential variables used in the random forest model for estimating stem volume. Red color bars indicates the parameter that significantly partitioned the data.

**Table 1.**Summary of the model combinations and its variables used. Model

_{1-12}is in the order from the lowest sum of residual which was checked through linear model.

Model | Variables | RFR Model | SVR Model |
---|---|---|---|

Model_{1} | CHM_{min} | Model_{RFR1} | Model_{SVR1} |

Model_{2} | CHM_{avg} | Model_{RFR2} | Model_{SVR2} |

Model_{3} | γ^{0}_{VH} | Model_{RFR3} | Model_{SVR3} |

Model_{4} | FCC | Model_{RFR4} | Model_{SVR4} |

Model_{5} | γ^{0}_{VV} | Model_{RFR5} | Model_{SVR5} |

Model_{6} | CA_{min} | Model_{RFR6} | Model_{SVR6} |

Model_{7} | CA_{avg} | Model_{RFR7} | Model_{SVR7} |

Model_{8} | CA_{max} | Model_{RFR8} | Model_{SVR8} |

Model_{9} | CHM_{max} | Model_{RFR9} | Model_{SVR9} |

Model_{10} | γ^{0}_{HV} | Model_{RFR10} | Model_{SVR10} |

Model_{11} | γ^{0}_{HH} | Model_{RFR11} | Model_{SVR11} |

Model_{12} | RSP | Model_{RFR12} | Model_{SVR12} |

Model_{13} | Model_{1+2} | Model_{RFR13} | Model_{SVR13} |

Model_{14} | Model_{1+2+3} | Model_{RFR14} | Model_{SVR14} |

Model_{15} | Model_{1+2+3+4} | Model_{RFR15} | Model_{SVR15} |

Model_{16} | Model_{1+2+3+4+5} | Model_{RFR16} | Model_{SVR16} |

Model_{17} | Model_{1+2+3+4+5+6} | Model_{RFR17} | Model_{SVR17} |

Model_{18} | Model_{1+2+3+4+5+6+7} | Model_{RFR18} | Model_{SVR18} |

Model_{19} | Model_{1+2+3+4+5+6+7+8} | Model_{RFR19} | Model_{SVR19} |

Model_{20} | Model_{1+2+3+4+5+6+7+8+9} | Model_{RFR20} | Model_{SVR20} |

Model_{21} | Model_{1+2+3+4+5+6+7+8+9+10} | Model_{RFR21} | Model_{SVR21} |

Model_{22} | Model_{1+2+3+4+5+6+7+8+9+10+11} | Model_{RFR22} | Model_{SVR22} |

Model_{23} | Model_{1+2+3+4+5+6+7+8+9+10+11+12} | Model_{RFR23} | Model_{SVR23} |

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Iizuka, K.; Hayakawa, Y.S.; Ogura, T.; Nakata, Y.; Kosugi, Y.; Yonehara, T.
Integration of Multi-Sensor Data to Estimate Plot-Level Stem Volume Using Machine Learning Algorithms–Case Study of Evergreen Conifer Planted Forests in Japan. *Remote Sens.* **2020**, *12*, 1649.
https://doi.org/10.3390/rs12101649

**AMA Style**

Iizuka K, Hayakawa YS, Ogura T, Nakata Y, Kosugi Y, Yonehara T.
Integration of Multi-Sensor Data to Estimate Plot-Level Stem Volume Using Machine Learning Algorithms–Case Study of Evergreen Conifer Planted Forests in Japan. *Remote Sensing*. 2020; 12(10):1649.
https://doi.org/10.3390/rs12101649

**Chicago/Turabian Style**

Iizuka, Kotaro, Yuichi S. Hayakawa, Takuro Ogura, Yasutaka Nakata, Yoshiko Kosugi, and Taichiro Yonehara.
2020. "Integration of Multi-Sensor Data to Estimate Plot-Level Stem Volume Using Machine Learning Algorithms–Case Study of Evergreen Conifer Planted Forests in Japan" *Remote Sensing* 12, no. 10: 1649.
https://doi.org/10.3390/rs12101649