Estimating Forest Stock Volume in Hunan Province, China, by Integrating In Situ Plot Data, Sentinel-2 Images, and Linear and Machine Learning Regression Models

Abstract

The forest stock volume (FSV) is one of the key indicators in forestry resource assessments on local, regional, and national scales. To date, scaling up in situ plot-scale measurements across landscapes is still a great challenge in the estimation of FSVs. In this study, Sentinel-2 imagery, the Google Earth Engine (GEE) cloud computing platform, three base station joint differential positioning technology (TBSJDPT), and three algorithms were used to build an FSV model for forests located in Hunan Province, southern China. The GEE cloud computing platform was used to extract the imagery variables from the Sentinel-2 imagery pixels. The TBSJDPT was put forward and used to provide high-precision positions of the sample plot data. The random forests (RF), support vector regression (SVR), and multiple linear regression (MLR) algorithms were used to estimate the FSV. For each pixel, 24 variables were extracted from the Sentinel-2 images taken in 2017 and 2018. The RF model performed the best in both the training phase (i.e., R 2 = 0.91, RMSE = 35.13 m 3 ha −1 , n = 321) and in the test phase (i.e., R 2 = 0.58, RMSE = 65.03 m 3 ha −1 , and n = 138). This model was followed by the SVR model (R 2 = 0.54, RMSE = 65.60 m 3 ha −1 , n = 321 in training; R 2 = 0.54, RMSE = 66.00 m 3 ha −1 , n = 138 in testing), which was slightly better than the MLR model (R 2 = 0.38, RMSE = 75.74 m 3 ha −1 , and n = 321 in training; R 2 = 0.49, RMSE = 70.22 m 3 ha −1 , and n = 138 in testing) in both the training phase and test phase. The best predictive band was Red-Edge 1 (B5), which performed well both in the machine learning methods and in the MLR method. The Blue band (B2), Green band (B3), Red band (B4), SWIR2 band (B12), and vegetation indices (TCW, NDVI_B5, and TCB) were used in the machine learning models, and only one vegetation index (MSI) was used in the MLR model. We mapped the FSV distribution in Hunan Province (3.50 × 10 8 m 3 ) based on the RF model; it reached a total accuracy of 63.87% compared with the official forest report in 2017 (5.48 × 10 8 m 3 ). The results from this study will help develop and improve satellite-based methods to estimate FSVs on local, regional and national scales.The forest stock volume (FSV, m 3 ha −1 ) is the sum of the stem volumes of all the living trees per unit area, and is one of key forest variables for forest resources management and assessments on local, region and country scales [1]. FSVs have a strong relationship with the aboveground biomass (AGB) and carbon stocks [2]. To understand the spatial distribution of carbon in forests and to derive predictions for monitoring carbon stock trends, the FSV must be quantified [3]. Traditionally, the FSV is estimated by sampling several plots, which involves substantial manpower, materials, and financial resources [4]. With the development of remote sensing technology, particularly the Landsat series since 1972, satellite imagery has played an important role in forest inventory. Various studies have been performed to estimate the forest variables ...

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

“…However, we found that the estimates were associated with uncertainties. The uncertainty of GSV estimates often results from atmospheric effects, sensor effects, sample plot GSV measurement errors, feature variable selection, and estimation model [ 27 , 59 ]. At present, the atmospheric effects and sensor errors cannot be completely eliminated [ 57 , 59 ].…”

“…These images were acquired during the field investigation time on 22 September, 2017 ( Table 2 ). The official Sen2cor module version 2.5.5 was used to transform the Level-1C product into the Level-2A product [ 27 ]. The Level-2A product is the bottom-of-atmosphere corrected reflectance after radiometric calibration and atmospheric correction.…”

“…Currently, Sentinel-2 is the only source of freely available optical data that exceeds three red-edge bands. Sentinel-2 data can, therefore, provide a more cost-effective potential for accurately mapping GSV of forest plantations [ 27 ]. Compared with MODIS and Landsat, the Sentinel-2 data has a higher spatial resolution, which can obtain more accurate vegetation information.…”

“…In addition, Sentinel-2 data characterizes higher temporal resolution (revisit period of five days using 2A and 2B). The time-series data provided by the satellite system allow us to obtain high-quality records of seasonal forest changes for forest resources monitoring [ 27 ].…”

“…Pearson correlation coefficient combined with linear stepwise regression (LSR) is commonly used for selecting feature variables, which are linearly related [ 30 ]. However, due to the complexity of forest ecosystems, the linear relationship may limit the estimation accuracy of models [ 27 ]. The importance assessment of random forest (RF) provides a metric for judging the contributions of feature variables to increasing model prediction accuracy, which can offer an evaluation of non-linear relationships between feature variables and GSV.…”