Estimation of Forest Residual Biomass for Bioelectricity Utilization towards Carbon Neutrality Based on Sentinel-2A Multi-Temporal Images: A Case Study of Aizu Region of Fukushima, Japan

Abstract

Forest biomass is expected to remain a key part of the national energy portfolio mix, yet residual forest biomass is currently underused. This study aimed to estimate the potential availability of waste woody biomass in the Aizu region and its energy potential for local bioelectricity generation as a sustainable strategy. The results showed that the available quantity of forest residual biomass for energy production was 191,065 tons, with an average of 1.385 t/ha in 2018, of which 72% (146,976 tons) was from logging residue for commercial purposes, and 28% (44,089 tons) was from thinning operations for forest management purposes. Forests within the biomass–collection radius of a local woody power plant can provide 45,925 tons of residual biomass, supplying bioelectricity at 1.6 times the plant’s capacity, which could avoid the amount of 65,246 tons of CO₂ emission per year by replacing coal-fired power generation. The results highlight the bioelectricity potential and carbon-neutral capacity of residual biomass. This encourages government initiatives or policy inclinations to sustainably boost the production of bioenergy derived from residual biomass.

1. Introduction

Energy represents a cornerstone in social development and economic growth by providing fundamental modern services for fulfilling basic social needs, driving economic growth, and fueling human development [1]. The longstanding global dependence on fossil fuels has led to pressing concerns such as air pollution, greenhouse gas emissions, and the formation of a “carbon bubble” [2,3,4]. As the world shifts away from fossil fuels, renewable energy (particularly woody biomass) emerges as a promising alternative after wind, solar, and geothermal technologies’ market penetration [3,4,5,6]. Unlike the variable outputs of solar and wind power due to their fluctuation with weather and season [7], biomass offers consistent energy production throughout the year, positioning it as a reliable base load power source. This stability, combined with its ability to balance the intermittency of other renewables, underscores the potential of biomass in creating a diversified, sustainable energy system [8,9].

In Japan, 88% of the country’s 2019 total primary energy supply relied on fossil fuels, and they heavily depend on imports to meet energy demands [10,11]. As coal is the most carbon-intensive fossil fuel and accounts for 31.6% of Japan’s total power generation [12], transitioning to renewables is not just environmentally beneficial but also crucial for Japan’s 2050 carbon neutrality goal. To accelerate this transition, policies like the feed-in tariff policy have been implemented, promoting coal-fired projects integrating renewable energy since 2012 [5,13]. A total of 307 power generation plants have adopted, or are planning to adopt, biomass co-firing energy resources to phase out coal in the power sector.

A total of 67% of Japan’s total land area is covered by forests, and 41% of the country’s forests are mature post-WWII-planted forests (primarily Japanese cedar (44%) and Japanese cypress (25%)) featuring fast growth and construction suitability [14]. The majority of forest resources have been utilized for papermaking, building, and furniture industries. Residues from lumber (processed wood ready for housing construction or furniture making) preparation or thinning are the main woody fuel resource for power generation [15]. Current wood consumption for electricity is low, accounting for 0.058% of the total wood supply [16]. Challenges like labor shortages, high management and collection costs, and competition with other sectors have hindered optimal utilization of forest residues and often led to the abandonment of residues in the thinning site or landing field [14,17,18,19,20,21,22]. Other reasons that have hindered the expansion of woody biomass for biomass energy include the high initial investment for facilities, fuelwood procurement, transportation costs, lower energy conversion efficiency (25–30%), source competition with other sectors, and a general decrease across the entire domestic forestry industry affected by the import of cheaper foreign timber products (70% of Japan’s total wood supply and 30.7% of fuel wood are imported) [5,16,23,24]. Life-cycle studies on energy consumption and greenhouse gas emissions indicate that substituting fossil fuels with forest residues and fast-growing woods may promote carbon neutrality [25,26]. Tapping into the potential of forest residues could enhance Japan’s domestic energy supplies and reduce fossil fuel reliance, making it imperative to assess their quantity, distribution, and bioelectricity conversion potential. Such data would serve as an important reference for sustainable forest management practices and appropriate strategies to promote renewable energy and revitalize rural society [27,28].

Japan has a long tradition of research into forest biomass and its associated energy utilization, including the estimation of availability, energy supply potential, cost estimation, and technical/economic feasibility [7,18,20,29,30,31,32]. There are studies focusing on forest residues, however they remain in the minority. Moreover, many of them estimated logging residue availability based on the existing forest register book (GIS format) which can be updated every several years. The most used GIS data provided by the New Energy and Industrial Technology Development Organization (2011) features a coarse resolution (1 km²) and the estimations rely on decade-old data, limiting their applicability to smaller-scale regions [33]. Thus, up-to-date forest information and a fine-resolution forest map that enables the classification of detailed tree types from advanced remote sensing images are essential for reflecting field conditions, estimating forest biomass, and conducting economic analysis to achieve more sustainable and efficient forest resource management [34,35,36].

The evolution of satellite remote sensing technology has revolutionized the methods used to observe the land surface. Sentinel-2 stands out for its free-of-charge, fine spatial and temporal resolution, as well as the availability of a red-edge band, achieving higher accuracies compared to images from other medium-resolution satellites [8,25]. Sentinel-2 multispectral satellite products were provided by a joint effort of the European Space Agency (ESA) and the European Union (EU) [37]. Sentinel-2 features 13 multispectral bands and provides fine-resolution (10–60 m) imagery for land cover/use monitoring [38]. Furthermore, GIS methods have become powerful tools for analyzing, visualizing, and interpreting spatial data. The combination of remote sensing, GIS, and ground-based measurements enables a comprehensive and accurate approach for monitoring and evaluating land surfaces, and it has become prevalent in estimating above-ground biomass (AGB) [39,40,41,42,43,44,45,46,47,48].

This study aimed to estimate the availability and energy potential of forest residual biomass derived from thinning and logging activities for electricity generation in the Aizu region of Fukushima for local bioelectricity generation. To do this, residual biomass was estimated based on local forest management practices and a high-resolution forest map extracted from multi-temporal Sentinel-2A images [49]. Road density data was used to filter the availability of residuals for transportation, and the corresponding potential for bioelectricity supply was determined. As there are hundreds of woody biomass power plants that are either operational or planned to be in operation in Japan, we believe that this paper will serve as a valuable reference for informing decision makers as well as those planning or operating woody biomass plants by assessing the potential of residual biomass energy, practicing sustainable forest management (health and harvest), providing guidance to evaluate coal displacement opportunities [27,50].

2. Materials and Methods

2.1. Study Area

Aizu region is located in western Fukushima Prefecture (Figure 1, 36°54′N to 37°50′N and 139°10′E to 140°17′E). The area covers nearly 542,069 ha, which accounts for 39% of Fukushima Prefecture. Administratively, it consists of two cities, eleven towns, and four villages in order of administrative ranking. The average temperature of the warmest month (August) exceeds 20 °C, and the coldest month (January) is below −2.03 °C. The annual precipitation is nearly 1715.6 mm. The average wind speed is 1.4 m/s [51].

Forests covered nearly 415,666 ha in 2010, accounting for 44% of the total forest area of Fukushima Prefecture [52]. The Aizu region has the highest number, accounting for 33% of Fukushima’s total whole tree volume [52]. In 2013, the Aizu region accounted for 14% of the prefecture’s total felling and bucking, and its forest industry contributed 0.08% of the prefecture’s total gross domestic product [53].

Figure 1. Location of the Fukushima Prefecture (left), and SRTM DEM map (U.S. Geological Survey, 2013 [54]) of Aizu region (right).

Figure 1. Location of the Fukushima Prefecture (left), and SRTM DEM map (U.S. Geological Survey, 2013 [54]) of Aizu region (right).

The forest is divided into state (40.1% of its total area) and private (59.9% of its total area) forests [52]. In the region’s private forest, sawtooth oak (Quercus acutissima) and jolcham oak (Quercus serrata) are the major broad-leaved species, whereas Japanese cedar (Cryptomeria japonica) and Hinoki cypress (Chamaecyparis obtusa) are the major needle-leaved species [52]. Natural forests are dominated by broad-leaved species such as the Japanese beech (Fagus crenata) and the Japanese mizunara oak (Quercus crispula Blume), whereas Japanese cedar (Cryptomeria japonica) and Japanese larch (Larix kaempferi) are the major needle-leaved species dominating the artificial forest.

2.2. Data Preprocessing and Classification

Sentinel-2A products were used to generate a forest map of the Aizu region. Fifty-one images of Sentinel-2A were acquired from 2018 to 2020 through the online cloud platform Google Earth Engine (GEE) (https://earthengine.google.org, accessed on 23 February 2021). The Sentinel-2A data is already applied with atmospheric correction. The overall methodology used for processing Sentinel-2A images in this study is shown in Figure 2. Initially, within the GEE code editor (https://code.earthengine.google.com/, accessed on), cloud cover filter was utilized and set for a cloud coverage of less than 10% to remove the clouds. A median composition method (by taking the median of each cloud-free pixel available) was performed to composite cloud-free images and cover the entire study area. This method enables the minimization of potential distortions from seasonal or short-term environmental variations [55]. The periods during which the images were acquired were categorized into three seasons: (1) spring (April–May), when broad-leaf trees start to green up; (2) summer (June–August), when forested areas reach peak greenness; (3) autumn (October–November), when some leaves start to gradually change colors. Among the 51 scenes acquired for image composition, 10 scenes were from spring, 34 from summer, and 7 from autumn. As a result, three seasonal composited images were generated.

Based on the seasonal composited images, three indexes including normalized difference water index (NDWI), soil-adjusted vegetation index (SAVI), and normalized difference vegetation index (NDVI) [56,57,58] were derived for each seasonal composition to facilitate the land cover classification. These indices have been proven to be robust and reliable indicators for discriminating water bodies from vegetation and soil, as well as for tracking changes in vegetation over time [59,60,61,62,63]. As a result, nine index images (three seasonal images × three index images) were generated.

To export the seasonal composited images and their corresponding derived index images, Google Drive was used as an export data receiver. However, due to the size restriction of exportation to Google Drive, the images were exported by dividing each image into three parts of the study area and were stitched again in ENVI 5.6 software (Exelis Visual Information Solutions, Boulder, CO, USA).

Sentinel-2 mission provides multi-spectral (12 bands) at a spatial resolution of 10–60 m. For this study, the 10 m bands (B2, B3, B4, and B8) and 20 m bands (B5, B6, B7, B8, B11, and B12) were used. The bands 1 and 10 were removed as redundant information of coastal aerosol and cirrus-type cloud radiation. To fully utilize the multispectral information provided by the Sentinel-2A satellite, we resampled the bands with a native resolution of 20 m (B5, B6, B7, B8, B11, and B12) down to a finer 10 m resolution in ENVI 5.6 software. The resampled bands were mosaiced with 10 m bands (B2, B3, B4, and B8) to create an image at a resolution of 10 m. Furthermore, the resampled seasonal composition images were ultimately combined with index (NDWI, SAVI, and NDVI) images to form a singular image for land cover/land use classification. Figure 2 shows the flowchart of processing Sentinel-2A images.

Field visits and ancillary datasets were utilized to collect comprehensive landscape information, including land cover/use types and tree species, which facilitated the differentiation of diverse landscapes and improvement of satellite image interpretation. Ground truth data for Sentinel-2A image classification, primarily consisting of homogeneous tree species in the Aizu region, were collected during field visits along major roads and forest trails. A total of 126 sites were visited and the typical land cover and forest types, as well as their GPS locations, were recorded using a MAP64SJZ GPS receiver (Garmin, Olathe, KS, USA) and digitalized. Additionally, the high-quality images and geotagged photos available in Google Earth Pro™ (https://www.google.com/earth/, accessed on 20 March 2021) were used to derive reference information through visual image interpretation. The land cover classes observed were recorded with geo-referenced information, and these records served as an auxiliary reference dataset for selecting regions of interest (ROI). Further supplementary data sources included 10 plots (3 were NLF and 7 were BLF) from the National Forest Inventory (http://www.rinya.maff.go.jp/j/keikaku/tayouseichousa/, accessed on 10 April 2021) for the study area and the forest register of Fukushima Prefecture provided by Forestry Promotion Division of Fukushima, both available in GIS file format. These resources provided crucial tree information (e.g., species, age, and ownership) aiding in tree species identification for satellite image interpretation. Based on our field experience and supplementary data, we selected the sample points (pixels) showing similar features through visual interpretation of images. A total of 23,136 sample points (pixels) of Sentinel-2A images were selected as our ROI. Subsequently, 70% and 30% of these pixels were randomly assigned as training and testing samples following the methodology of Breiman and Spector et al. [64].

Classification categories were defined based on the land use and land cover map products (https://www.eorc.jaxa.jp/ALOS/en/dataset/lulc_e.htm, accessed on 15 March 2021) provided by the Japan Aerospace Exploration Agency (JAXA) and visual interpretation (

Table 1. Land cover and land use classes with descriptions for the Aizu region.

Major Category	Sub-Category	Definition
Forest	Broad-leaf forest (BLF)	Forest dominated by deciduous broad-leaf trees, including Quercus (Quercus acutissima and Quercus serrata), Fagus (Fagus crenata and Fagus japonica), and Castanea crenata
	Needle-leaf forest (NLF)	Forest dominated by coniferous trees, including Cupressaceae (Cryptomeria japonica and Chamaecyparis obtusa) and Pinaceae (Larix kaempferi and Pinus densiflora)
Non-forest	Water body	Rivers, lakes, reservoirs, and swimming pools
	Cropland	Cultivated land, fallow land, or uncultivated land after harvest
	Grassland	Land dominated by grass-like features and low shrubs
	Bare land	Land dominated by mining fields, quarries, riverbanks, rocky mountainous areas, or unpaved playgrounds
	Built-up area	Land dominated by buildings and paved surfaces

).

Random forest (RF) classifier was used for image classification. High classification accuracies have been reported for the land cover/use classification of Sentinel-2 data using RF compared with other classifiers such as the maximum likelihood classifier (MLC), support vector machine (SVM), classification tree (CT), and k-nearest neighbor (k-NN) [65,66,67,68]. RF is an ensemble method based on classification and regression trees that can be constructed in parallel without strong dependencies among individual learners [69,70,71]. The decision trees are created based on variables including the object attributes (independent variables) and their visually identified label (dependent variable) of the training set (ROIs) [72]. For each decision tree node, a random subset of the training set is assessed and used to develop other decision trees. By default, two-thirds of the training set (so-called in-bag samples with a total number of 494 samples of all classes) were used for decision tree development, whereas the remaining one-third of the training set (so-called out-of-bag samples with a total number of 277 samples of all classes) were used for assessing the prediction performance of the RF [73]. By aggregating the predictions of all individual decision trees, the final class of a certain land cover type was determined by the prediction with the highest majority vote [72,73,74]. We defined the number of decision trees as 10,000 to build the RF. The classification result was assessed using a confusion matrix [75], the overall classification accuracy, kappa coefficient, producer’s accuracy, and user’s accuracy [76]. Non-forest classes were removed from the resulting classification map so that only forest classes remained.

2.3. Estimation and Mapping of Residual Woody Biomass

2.3.1. Trunk Volume Estimation

As a key parameter in forest ecosystems, trunk volume (TV) (m³/ha) is closely correlated with the aboveground biomass of forest ecosystems and allows for the prediction of all related biomass components (stems, branches, foliage, roots, and understory) for various research purposes.

In this study, we utilized the BGC-ES sub-biomass model developed by Ooba et al. [77] to estimate the TV. The sub-biomass model can estimate the TV based on tree age and tree type information by using the growth function and population function that considers factors like water availability, light, soil nutrients, and forest management. The process began with the estimation of average tree height (hm) as a function of forest age (t), followed by the determination of population number (N), and culminated in the calculation of TV (V).

The average tree height was calculated using a growth function hf (t) as follows:

h = sc h_max hf (t),

(1)

where h_max is the maximum tree height (m), sc is the site coefficient, and hf (t) is a growth function that is defined by the Mitscherlich [78] (Equation (2)) or Gompertz equations [79] (Equation (3)) as follows:

hf (t) = 1 − h_b exp(−h_ct),

(2)

hf (t) = exp(−h_b exp(−h_ct)),

(3)

where h_b and h_c are species-specific parameters that characterize the growth rate and the shape of the growth curve, respectively.

The population number in a forest unit is estimated using the theoretical 3/2-law [80]. The maximum volume of the trunk of a plant V_max (m³ ha⁻¹) depends on the maximum population number N_max (ha⁻¹):

$$V_{\max }=k{N}_{max}^{-\alpha },$$

(4)

where k and $\alpha$ are parameters (

Table 2. Parameter values used in equations.

		NLF	BLF
Growth	Unit
hmax	m	29.94	18.93
hb		1	1.515
hc		0.021	0.0181
ch		0.972	0.9059
Equation type *		M	G
Population
α		0.953	0.4998
k	104 m3 ha−1	96.06	1.538
k *	104 m3 ha−1	347.2	3.995
$p_a$	m−3 ha	0.00021	0.07004
$p_b$		−1.175	−1.071
$p_c$	m−3 ha	2665.12	20,850
$p_d$		−2.322	−3.214

*M: Mitscherlich (Equation (2)); G: Gompertz (Equation (3)).

). $\alpha$ is an empirical constant typically valued at 1/2. Equation (4) is represented as a line with a slope of −0.5 in a logarithmic coordinated N–V plane.

Subsequently, the population number (N) was estimated using the natural mortality curve, which takes into account the intraspecific competition:

$$1=\frac{V}{{k^{*}N}_0^{-\alpha }}+\frac{N}{N_0},$$

(5)

where k^∗ is a scaling parameter, $N_0$ represents the initial population number (ha⁻¹), and α is an empirical constant typically valued at 1/2, reflecting the 3/2-law of self-thinning.

Finally, trunk volume (V) was calculated by inverting the relationship between volume and population number, integrating the average tree height (hm):

$$V^{-1}=p_a{h}_m^{p_b}+\frac{p_C{h}_m^{p_d}}{N},$$

(6)

where $p_a$, $p_a$, $p_c$ and $p_d$, were sourced from the National Forestry Agency of Japan [81]. (Table 2), adjusted for the region and tree type. By inputting the age data into these equations, we estimated the TV (V) for individual trees.

To calculate the annual total TV per hectare (G_v, m³ ha⁻¹), TV (V) was multiplied by the population number of trees (ha⁻¹):

G_v = N·V,

(7)

It is worth mentioning that the sub-biomass model was developed under relatively simple assumptions (such as an increase in the tree diameter at breast height changes with competition and an increase in the averaged upper-layer tree height mainly depending on the forest age) about the forest ecosystem. Furthermore, the coefficients used for these functions vary with region and tree type. Forest inventory and forest report review are necessary when determining the coefficients. More detailed investigations are necessary for more specific woody biomass estimations.

2.3.2. Trunk Volume Data Preparation

The tree age and tree type information (forest registrator book) of our study area was obtained in GIS file format provided by the Forestry Promotion Division of Fukushima.

As a result of Section 2.3.1, two maps of TV (G_v, m³ ha⁻¹) for NLF and BLF were generated. Then, through the raster calculation in ArcGIS 10.8 software (https://www.esrij.com, accessed on 1 December 2023), the forest map obtained in Section 2.2 was linked geographically with the two TV maps of NLF and BLF by using the information of tree type (NLF or BLF). Thus, an integrated TV (G_v, m³ ha⁻¹) map of 10 m resolution was created.

A flowchart summarizing the preparation of the TV map is shown in Figure 3 (yellow color background).

2.3.3. Estimation of Forest Residual Biomass

Forest residual biomass (waste woody biomass) is comprised of residues resulting from two kinds of anthropogenic practices (thinning and logging). Estimations of forest residual biomass were performed separately for the two different practices (thinning and logging) (Figure 3).

Thinning activity, according to the local forestry management plan [53], was performed mainly on certain ages of young needle-leaf trees to improve the health and growth of a forest. Therefore, the estimation of thinned residual biomass was focused on NLF of age classes 3–6 and 8 (one age class is equal to 5 years). BLF was excluded from this step, as there were no thinning plans for BLF. The thinning rate was 35% for trees of all thinning ages combined [53]. The entire portion of each thinned NLF was assumed to be used for bioelectricity generation. The available thinned NLF volume was calculated by multiplying the annual total NLF TV (G_v m³ ha⁻¹) by a thinning rate of 35%.

For logging activity, according to the local forestry management plan [53], cutting ages for logging were performed mainly on NLF trees aged at least 45 years (age class 9 or greater) and BLF trees aged at least 40 years (age class 8 or greater). Unlike in the case of thinning, the logging residue including branches, leaves, stumps, roots, tops, and bark that had been stripped from the harvested raw wood and was assumed to be used for bioenergy generation. In the Aizu region, logging residues are typically found at felling sites in the forest or wood piles alongside the road [53].

The amount of forest logging residue was estimated by using the method developed by Japan’s Forestry and Forest Products Research Institute (Equation (8)):

$${\mathrm{Q}}_{\mathrm{residue}}={\mathrm{G}}_{\mathrm{v}}\times {\mathrm{f}}_{\mathrm{residue}}\times (1-{\mathrm{W}}_{\mathrm{d}})$$

(8)

where

Q_residue represents the quantity of dry wood logging residue (t ha⁻¹);

G_v represents the annual logging of trunk volume per unit (m³ ha⁻¹);

f_residue represents the logging residue generation coefficient (0.07 for NLF and 0.091 for BLF) [82];

W_d represents the 15% air-dry water content of logged woods.

Because the utilization of potential woody biomass is largely restricted by transportation infrastructures for the collection of woody biomasses and its transportation out of the forest, we needed to consider the road network (density of the forest road and the distance from the road required for wood collection) in forests. Therefore, the quantities of transportable (available for energy production) thinned woods and logging residues in each pixel in the raster data were further estimated by integrating the road network information, including the density of the forest road network and the distance from the road at which point wood is collectible using Equation (9). It is worth mentioning that this method has been commonly used in Japan and is interpretable across multiple disciplines and by policymakers [14,33,82]. However, the use of a fixed road density across the study area may not accurately capture the variations in road network densities, potentially leading to overestimations or underestimations of transportable biomass in different sub-regions such as remote mountainous areas (with less road density) and lower areas (with higher road density).

$${\mathrm{V}}_{\mathrm{transport}}={\mathrm{G}}_{\mathrm{V} \mathrm{thinned}/\mathrm{residual}}\times {\mathrm{D}}_{\mathrm{road}}\times {\mathrm{R}}_{\mathrm{collect}}\times {10}^{-4}$$

(9)

where

V_transport represents the annual volume of transportable residual woody biomass (m³);

G_{V thinned/residual} represents the annual volume of thinned NLF or logging residues of NLF or BLF (m³ ha⁻¹);

D_road represents the density of the forest road network (6.29 m/ha);

R_collect represents the collection distance for forest residues, which is assumed to be 25 m from the forest road to the mountain slope and 25 m to the valley slope, equalling a total of 50 m from the roads [83];

10⁻⁴ is the scale factor (ha m⁻²).

The mass of the transportable (available for bioenergy production) forest residual biomass (Q_transport) in tonnes of dry matter per hectare per year (t dry-weight (ha⁻¹) in each pixel was derived from the following formula:

$${\mathrm{Q}}_{\mathrm{transport}}={\mathrm{D}}_{\mathrm{bulk}}\times {\mathrm{V}}_{\mathrm{transport}}$$

(10)

where

D_bulk represents the bulk density of NLF (0.314 t m⁻³) or BLF (0.573 t m⁻³) [84].

A flowchart summarizing the process of the overall estimation is shown in Figure 2 (blue color background).

2.3.4. Available Bioelectricity Potential

The available potential of woody bioelectricity represents the available quantity of biomass that can be technically and economically harvested and used for energy purposes [85]. In Japan, wood fuels used for electricity generation are generally made from residuals from the sawmill process, the manufacture of wood products and construction, and harvesting/thinning residuals in the forest. The available bioelectricity potential from combustion was evaluated using Equation (11), developed by Tatebayashi et al. [86].

$${\mathrm{E}}_{\mathrm{biosolid}}={\mathrm{Q}}_{\mathrm{transport}}\times {\left(1-\mathrm{W}\mathrm{p}\right)}^{-1}\times \mathrm{YR}\times \mathrm{LHV}$$

(11)

where

E_biosolid represents the woody biomass energy (Gigajoule ha⁻¹, hereafter referred to as GJ ha⁻¹);

Q_transport represents the dry wood matter of transportable biomass (t ha⁻¹);

Wp represents the water content percentage, which was set at 10% for air-dried processed wood residues, and (1 − Wp)⁻¹ represents the composition ratio of wood solids;

YR represents the yield (the ratio of the output to the total input of materials in a process), and 20% was set as the output loss of wood materials during the processing;

LHV represents the lower heating value of 16 GJ t⁻¹.

To make the unit of TV map pixel (10 × 10 m²) consistent with the units of the quantity of transportable (collectible) residual biomass and the residual biomass energy (Q_transport and E_biosolid, respectively), we converted the units of Q_transport and E_biosolid from t ha⁻¹ and GJ ha⁻¹ to t 100 m⁻² and GJ·100 m⁻² by multiplying them by a scale factor of 0.01. Each pixel then represented the biomass quantity/volume of a certain type of tree.

At last, the overall quantity of residual biomass over the study area was calculated by multiplying the biomass quantity/volume values with the statistically counted pixel numbers that share the same type of tree and age. Thus, the overall quantity of transportable woody biomass and the energy feedstock potential of the whole study area was obtained.

However, the amount of woody biomass that can be collected and used for power generation is restricted by various factors such as terrain features, the road network for collecting and transporting wood materials, and the collection range from the power plant to the chip processing sites. We selected the Green Power Aizu Power Plant as the target of this analysis as it is the first local woody biomass power plant and is supported by MAFF, aiming to recycle local waste woody biomass into electricity generation. The sources of fuelwood are thinned and logging residues. The electricity generated mainly serves the Aizu region’s largest city of Wakamatsu, catering to power producers and suppliers of industrial electricity. The plant’s electricity generation capacity (generating-end output) is about 5700 kW, and the total annual electricity generation capacity is 167,443.2 GJ. Operating 24 h a day for 340 days a year, the plant requires an annual feedstock of 36,000 t of dried chips.

Therefore, to more accurately estimate the realistic potential of accessible woody biomass for the power plant, we used data from the forest plan of the Aizu Region [53] to set and apply several filters to the local forest register data. The filters and the reasons for their application were as follows:

(1)

Because the national average woody material collection radius is 50 km [87], which is also the operating radius of the towable wood chippers that are adopted by most of the small scale power plants, the area within 50 km was set as the maximum biomass-collection buffer radius to the Green Energy Aizu Power Plant.

(2)

Under the distance reachable by lumber-collecting machinery, the distance from the center of each unit of each forest land unit (forest compartment) to the nearest road was set at no more than 500 m.

(3)

The slope of land was set at no more than 35° due to difficulties in logging/thinning, collecting, and transporting operations. Moreover, many areas with slopes greater than 35° are classified as protected land to prevent natural disasters such as landslides and soil erosion.

(4)

Forests in which cutting was forbidden were not included, as local government has strict land-use rules to protect national parks or lands that are at a high risk of natural disasters.

Based on the estimated bioelectricity potential within a 50 km radius of the power plant, we estimated the potential reduction in CO₂ emissions by replacing coal-fired power generation with biomass-fired power generation. This estimation was derived from the difference in CO₂ emissions between the biomass-fired generation system (0.0618 tons CO₂/MWh) and the coal-fired generation system (0.96 tons CO₂/MWh) in Japan [88].

3. Results

3.1. Forest Cover Mapping

A total of 84% of the study area was classified as forest (Figure 4). Almost 32% of the forest was BLF, distributed below an elevation of 1400 m across the study area. Most of these BLFs are secondary woods planted after clean cutting for woodfires and mainly distributed around villages. A total of 52% of the forest was NLF, most of which had been planted along streams, rivers, urban areas, and villages below 1000 m elevation.

Table 3. Accuracy of forest classification for the Aizu region.

Class	NLF	BLF
NLF (needle-leaf forest)	1752	62
BLF (broad-leaf forest)	37	1321
Producer accuracy (%)	97.93	92.57
User accuracy (%)	95.84	96.42
Overall accuracy (%)	92.32%
Kappa coefficient	0.9086

gives an accuracy assessment of land cover classification results. The class-specific producer’s accuracies were 97.93% and 92.57% for NLF and BLF, respectively. An overall accuracy of 92.32% and a Kappa index of 0.9086 indicate a satisfactory accuracy, which underscores the robustness of the RF classifier in incorporating the multi-temporal dimension of the dataset into the analysis and evaluating the different features which are critical for accurate classification. In previous research, multitemporal data and RF classification showed satisfying forest mapping and facilitated further above-ground biomass estimation [89,90,91,92].

However, class confusion between NLF and BLF did occur. Some tree species have different patterns of phenological variation in their leaves and flowering over seasons (e.g., BLF leaves do not always turn yellow or red in autumn, and the leaves of certain pine tree species (such as Metasequoia) turn reddish in autumn and winter). This could have caused confusion between NLF and BLF, as the spectral reflection of plants was one of the indicators used in our classification. Some of the pixels of Sentinel-2A contained mixed forest stands (e.g., broad-leaf and needle-leaf, or mixed-age) and were classified into homogeneous forest types (NLF/BLF). Furthermore, due to the limited data set that contained forest information about our study area, we only used two general forest types (NLF and BLF) for the biomass estimation. As the biomass varies among different tree species, there might be some discrepancies between our estimation with the actual forest biomass. Combining active remote sensing data such as LiDAR/SAR that can be used to extract forest structural attributes including tree height and classifying tree species can enhance the precision of detailed forest mapping and biomass estimation.

3.2. Spatial Analysis of Forest Residual Biomass

3.2.1. Analysis of Thinned Residue Biomass

For thinned residue biomass, we mapped the spatial distribution of transportable thinned residue biomass (Figure 5a) and bioelectricity potential (Figure 5b).

There was a uniform distribution pattern of both low and high biomass quantity across the study area. According to elevation information provided by the forest register, the residue biomass left by thinning was distributed mainly at elevations of 180 m to 900 m along the rivers over the study area, except in populated areas such as Wakamatsu city.

The TV (Gv in table) of NLF (ages 15, 20, 25, 30, and 40) ranged from 89.34 to 374.54 m³ ha⁻¹ year⁻¹. The volume of residues (Gv-residue) extracted from thinned NLF trees ranged from 31.27 to 131.09 m³ ha⁻¹ year⁻¹. The corresponding mass of residues (Q_residue) ranged from 9.82 to 41.46 t dry-weight ha⁻¹ year⁻¹ with a mean value of 24.08 t dry-weight ha⁻¹ year⁻¹. Of this residue mass, the transportable amount of residues (Q_transport) varied from 0.31 to 1.29 t dry-weight ha⁻¹ year⁻¹, available for delivery to a power plant for energy production. Consequently, the bioelectricity potential (E_biosolid) of this transportable mass of residues ranged from 4.1 to 18.10 GJ ha⁻¹ with a mean value of 10.77 GJ ha⁻¹ (Table 3).

3.2.2. Analysis of Logging Residue Biomass

The distributions of logging residue biomass and corresponding bioelectricity production potential of NLF (Figure 6a,b) and BLF (Figure 6c,d) were mapped separately.

As in the case of thinning residues, the logging residues also showed a general uniform distribution pattern over the study area. However, some high biomass values (dark blue in Figure 6a) were distributed in the south-western part of our study area, which is a remote mountainous area with relatively higher elevation than other parts of the study area and has less residential area.

The TV of NLF (aged at least 45 years) ranged from 89.34 to 374.54 m³ ha⁻¹ year⁻¹. The logging residue mass ranged from 9.17 to 14.26 t dry-weight ha⁻¹ with a mean value of 12.79 t ha⁻¹, while the mean value of mass available for energy production (transportable amount) reduced to 0.29–0.45 t dry-weight ha⁻¹ with a mean value of 0.4 t ha⁻¹. The corresponding bioelectricity potential ranged from 4.1 to 6.38 GJ ha⁻¹ with a mean value of 5.72 GJ ha⁻¹ (Figure 7).The TV of BLF (aged at least 40 years) did not vary markedly (range, 128.5 to 138.58 t dry-weight ha⁻¹) because the growing volume tends to be stable for mature BLF. Compared with NLF, their mean values of logging residues were close, but the minimum value of BLF was greater than NLF while the maximum value of BLF was smaller than NLF. The corresponding bioelectricity potential ranged from 3 to 3.23 GJ—which is lower than NLF (Figure 7).

3.2.3. Estimation of Potential Bioelectricity

By using the pixel-based values of estimated bioelectricity for different tree age classes of the residues of thinned NLF and the logging residues of BLF and NLF, we calculated the total amount of potential bioelectricity for each tree age class by multiplying the pixel values of bioelectricity by the number of pixels that shared the same values. The sum value of the calculated total potential bioelectricity was then regarded as the overall potential bioelectricity of the entire study area. All the transportable amounts of forest residues can provide 744,964 tons of woody biomass. The relative potential bioelectricity of 2018 was approximately 2,717,366 GJ (627,046 GJ for thinned NLF, 1,031,436 GJ for BLF logging residues, and 1,058,884 GJ for NLF logging residues), which is 16.2 times the plant’s power-generating capacity. NLF logging residues contributed to 39% of the Aizu region’s total bioelectricity potential, BLF logging residues contributed to 38%, and thinned NLF contributed to 23%. According to the Wood Demand Chart of 2020 [16], the coefficients for converting wood to chips are 2.2 for NLF and 1.7 for BLF. After conversion of the collectable dry-matter biomass quantity to chip production, the total dried-chip production potential was approximately 246,052 t (63,823 t for thinned NLF; 74,451 t for BLF logging residues; and 107,778 t for NLF logging residues). The potential chip production is more than 6.8 times the annual chip demand of the Green Energy Aizu Power Plant which requires an annual feedstock of 36,000 t of dried chips.

Within the extracted 50 km radius of the Green Energy Aizu Power Plant, approximately 70% of the extracted forest was NLF and 30% was BLF. The forest residues within the extracted area can provide 45,925 tons of woody biomass. The total dried-chip production potential was approximately 22,960 t, which can meet the 64% of the annual chip demand of the power plant. The spatial distribution of potential chip production indicated the values range from 0.0005 to 0.645 t/ha with a majority of 0.233–0.645 t/ha (Figure 8a). The estimated total energy potential was 261,507 GJ—approximately 1.6 times the annual electricity-generation capacity of the power plant. The spatial distribution of potential bioelectricity indicated that most forests within the area have the potential to provide 3.2–6.3 GJ/ha of bioelectricity (as represented by the blue color in Figure 8b). The potential bioelectricity would also supply 2% of the total population-based energy consumption of Wakamatsu city in 2015 (102.43 GJ/person for a population of 124,100 people) [93,94]. Additionally, the substitution of coal fuel with this residual biomass is estimated to reduce CO₂ emissions by approximately 65,246 tons/year. The reduction of CO₂ emissions from the entire study area’s residual biomass-fired power generation was estimated to be 677,983 tons/year, which corresponds to 1.84% of the Fukushima Prefecture’s CO₂ emissions as of 2019 (36,876,452 tons/year) [95].

4. Discussion

This study used growth and population functions from a sub-biomass model of BGC-ES developed by Ooba [29]. BGC-ES is advantageous for estimating forest resource dynamics as it incorporates factors such as water, light, soil nutrients, and forest management into the simulation. It also facilitates area-wide estimations easily, particularly when using satellite imagery-based land cover classifications as the data source. However, since it provides estimations of average size and growth of trees, more detailed investigations are necessary for more specific woody biomass projects. In the present study, it was observed that although NLF for thinning has a lower TV (mean value: 219.1 m³ ha⁻¹) compared to NLF for logging (mean values: 581.99 m³ ha⁻¹) due to the younger age of the thinned trees, it yields more residues per unit (mean value: 76.68 m³ ha⁻¹), as the whole tree was cut down and considered as a biomass resource in thinning activities (Figure 9). In logging activities, only the stripped parts of the tree, excluding the merchantable trunk, are considered as residual biomass. Among the logging residues, the BLF has a lower TV and fewer residues compared to NLF. This is because most of the trees in BLF areas are mature, and their trunk volumes are relatively stable. NLF and BLF logging residues together account for 77% of the total residual biomass in the study area (Figure 9), indicating that they are the primary contributors to the bioenergy production potential.

In the present study, the estimations for transportable biomass were based on the existing forest road network and ground slope. As a result, the calculated values for both biomass and bioelectricity potential were significantly lower than potential utilization of the abundant forest resources in the study area. One limitation of our study is the adoption of a fixed road density and thinning/logging intensity to represent the entire study area, which may not capture the variations in road network densities in individual sub-compartments of forest. Estimating availability based on road density in each sub-compartment would yield a more accurate biomass availability estimation and facilitate the investigation of optimized strategies adapted to different availabilities and logistics.

According to Yoshioka et al. [22], the lowest costs for hauling, transporting, and chipping residues range from USD 150/t to USD 227/t of dry biomass. Consequently, the total cost for transporting residual biomass in our study area was calculated to range from USD 28,659,750 to USD 43,371,755. Compared to processing residues alongside forest roads or at plants, processing them directly at thinning/logging sites using advanced machinery is preferable for cost minimization. This is particularly challenging given the current state of the Japanese forestry industry, which suffers from low-capacity forestry machines and a lack of well-trained operators [96]. The shortage of well-trained operators is a multifaceted issue, driven by an aging workforce, insufficient training and education, and the slow adoption of technology in the forestry industry [24]. To achieve more efficient harvesting, processing, and management of forest resources, thereby minimizing procurement costs and increasing overall productivity, it is essential to have better machines and well-trained operators skilled in the use of advanced forestry equipment and implementing sustainable forest management practices [22,97]. Government initiatives that provide skill-building opportunities in forestry and revitalizing the forestry workforce is necessary [5]. Furthermore, subsidies aimed at expanding current road density in woodland areas and supporting the development of efficient forestry techniques are crucial for improving this situation. Such measures would attract more forestry businesses and facilitate the conversion of forest waste into energy.

Additionally, power generation based on residual biomass has the potential to reduce CO₂ emissions by replacing coal-fired power generation. To promote renewable energy use and reduce CO₂ emissions, Japan has implemented policies and measures such as the feed-in tariff (FIT) system and a carbon tax. However, the level of carbon tax in Japan is lower than in many other countries. Therefore, the introduction of more robust measures such as large-scale carbon pricing and emissions trading systems, is essential to further incentivize the shift towards renewable energy sources and achieve more substantial reductions in CO₂ emissions.

Removing waste residual biomass has potential ecological impacts on soil nutrients, biodiversity, and carbon cycling. Its removal could lead to soil nutrient depletion, reduced habitat complexity, and altered forest dynamics [98]. Careful management is essential to balance biomass removal with ecological sustainability.

5. Conclusions

The availability and potential of forest biomass and its energy utilization from logging residues and thinned trees were discussed with the combination of BGC-ES sub-biomass model, remote sensing, and GIS tools. Firstly, a distribution map of forest biomass map was completed by combing Sentinel-2A-derived forest maps and tree trunk volume modeling. Then, the transportable availability of residual biomass was estimated taking into consideration thinning/logging intensities, road density, and wood collectable range. Finally, the corresponding bioelectricity generation potential and its CO₂ reduction effects was discussed.

As a result, NLF logging residue (the annual available amount was 74,453 t/y (dry weight)) proved to be the primary biomass contributor to the bioelectricity potential (2,717,366 GJ), followed by broad-leaved forests (72,523 t/y (dry weight)) and thinned trees (44,089 t/y (dry weight)). The forests located within a 50 km radius of this power plant can supply biomass (45,925 t/y (dry weight)) that equates to 1.6 times the facility’s capacity, also fulfilling 2% of the energy demands of the region’s central city. By substituting a coal-fired power generation system with a biomass-fired system, a reduction of CO₂ emissions of 677,983 tons/year can be achieved from the study area’s residual biomass, which can potentially be contributed to a reduction of 1.84% of the Fukushima Prefecture’s yearly CO₂ emissions. The results highlight the bioelectricity potential and carbon-neutral capacity of residual biomass. This encourages government initiatives and policy inclinations to sustainably boost the production of bioenergy derived from residual biomass.

A significant challenge in harnessing this potential is the existing forest road network as well as forestry machinery, which poses constraints on the forwarding and transportation of usable residual wood biomass. This underscores the critical need to enhance the forest road infrastructure and calls for both technical and financial support for the forestry sector. Future studies could apply economic feasibility analyses to assess the collection and transportation of residual forest biomass for supplying sustainable and low-carbon bioenergy feedstocks.