Assessment of above ground biomass and soil organic carbon in the forests of Nepal under climate change scenario

1. Introduction

Forest ecosystems are the largest carbon reservoirs storing ∼2 billion tons of CO₂ per year (UNDESA and UNFFS, 2021). The 2006 Intergovernmental Panel on Climate Change (IPCC) guidelines for the national greenhouse gas inventories indicate three major carbon pools (biomass, dead organic matter, and soil) in the forest ecosystem (Eggleston et al., 2006; IPCC, 2006). Most of the forest carbon is found in soil organic matter (45%) followed by living biomass (44%) i.e., above-ground tree biomass (AGTB) and root biomass and remaining in dead organic matter, i.e., in dead wood and litter (FAO, 2020).

Several climatic and edaphic factors influence forest carbon storage (Hofhansl et al., 2020). AGTB is influenced by altitude (Powell et al., 2010; Van der Laan et al., 2014; Rajput et al., 2017), temperature and precipitation (Yan et al., 2015), water availability, soil nitrogen content, and tree cover (Requena Suarez et al., 2021). Similarly, soil organic carbon (SOC) is affected by the amount of above-ground litter fall and root turnover (Andivia et al., 2016), temperature and precipitation (Sun et al., 2019), soil conditions and vegetation (Reyna-Bowen et al., 2019), species diversity (Gamfeldt et al., 2013), soil properties and moisture (Hounkpatin et al., 2018), altitude (Zinn et al., 2018), slope aspect, and soil depths (Zhu et al., 2017).

Climate change is contributing to global warming due to the steady increase in temperature since the 1960s (NOAA, 2023). It is projected to increase the severity of impacts in both the natural and human systems (IPCC, 2023). Climate change, rising temperature particularly, in the future has shown to have a negative effect on AGTB (Larjavaara et al., 2021; Li Y. et al., 2022) and SOC (Kirschbaum, 2000; Zhao et al., 2021) while a positive effect of the rising temperature on AGTB and SOC has also been studied under different climate change scenarios (Fu et al., 2017; Azian et al., 2022). The carbon sink of the forest is sensitive to CO₂ emission change resulting from increasing temperature, hydrological changes, and forest dynamics (Hubau et al., 2020).

Efficient estimation of above ground biomass and soil organic carbon is crucial for the study of carbon dynamics in forest ecosystems. Different assessment methods for the estimation of AGTB and SOC have been carried out. The 2006 IPCC guidelines have provisioned simple to robust method for the estimation of above and below carbon in Tier 1, Tier 2 and Tier 3 categories (IPCC, 2006). Design-based estimation (using ground-based sample plots) is one of the most used approaches for estimating AGTB and SOC (DFRS, 2014, 2015a,b; DFRS/FRA, 2014). Though it provides the precise evaluation of changes (stand structure, tree attributes) due to small standard error (Schadauer and Gabler, 2007), it is time-consuming, less cost- effective and difficult to implement in poorly accessible forest areas (Köhl et al., 2011; Kandel, 2013). Alternatively, a regression model (model-based estimation) has been used for the estimation of AGTB and SOC (Tian et al., 2014; Mohd Zaki et al., 2016; Pokhre, 2018; Li et al., 2019; Malla et al., 2022) that allows more flexibility to provide estimates outside the sample plots (Ståhl et al., 2016). Thus, model based estimation (regression model) is cost-effective and also able to estimate target variables of poorly accessible areas.

Recently, several studies have used a machine learning method such as random forest model (RFM) and gradient boosting (GB) for the prediction of AGTB and SOC (Powell et al., 2010; John et al., 2020; Lee et al., 2020; Li et al., 2020; López-Serrano et al., 2020; Nguyen and Kappas, 2020; Vorster et al., 2020). The RFM model uses machine learning algorithms for classification and regression based on decision trees (Jin et al., 2020). It is appropriate for large datasets with large numbers of variables, non-linear responses, both continuous and categorical variables and is less affected by the multicollinearity problem (Lu et al., 2016). Several studies found RFM superior to the regression model in terms of lowering mean squared error (Hounkpatin et al., 2018; Zhu et al., 2020; Xie et al., 2021), handling non-linear relations (Pahlavan Rad et al., 2014; Hengl et al., 2015), and indifference of assumptions of having probability distribution (normality) and no multicollinearity among independent variables (Lu et al., 2016; López-Serrano et al., 2016). Moreover, RFM does not require several numbers of sample plots, as in the case of design-based estimation, thus it is cost-effective. It can also estimate the target variable of the poorly accessible area in the presence of readily available independent variables (i.e., temperature, precipitation, slope, altitude, etc.).

Previous studies have used spectral values of satellite images as an independent variable to predict a response variable such as AGTB and SOC in the past period (Powell et al., 2010; Vicharnakorn et al., 2014; Angelopoulou et al., 2019; López-Serrano et al., 2020; Zhu et al., 2020; Kumar et al., 2022). However, the response of AGTB and SOC against change in climatic variables (temperature and precipitation) in the future has been lacking in the national scenario in Nepal. The influence of temperature and precipitation on the quantity of AGTB and SOC (Mehta et al., 2014; Bennett et al., 2020; Saimun et al., 2021) helps estimate these target variables in future climate change scenarios. Therefore, this study aims to answer the questions (1) Which are the variables (topographic, forest variables and climatic) significant to influence AGTB and SOC? (2) Are these variables likely to contribute to the amount of AGTB and SOC under the climate change scenario? The study covered all the forest covers of Nepal using forest resource assessment data. A RFM was used to better examine the influence of climatic, topographic and forest variables on the amount of AGTB and SOC. The research will improve our understanding of how climate change affects AGTB and SOC in the forests.

2. Materials and methods

2.1. Study area

For this study, we selected Nepal (Map 1) as a study site due to its varied site conditions. In Nepal, hilly region occupies a higher chunk of the land (∼86% of the total land area) while lowland (less than 300 m altitude) occupies only 14%. Wide altitudinal variations (<300–8,848 m), resulting in diverse climatic conditions, have produced different physiographic zones, i.e., Terai and Siwalik (lowlands), Mid-hills, High mountains and High Himal (LRMP, 1986), which influence the composition of flora and fauna (HMGN/MFSC, 2002). Stainton (1972) classified 35 forest types in Nepal that were further broadly categorized into 10 major groups based on the altitudinal range (HMGN/MFSC, 2002).

The climate of Nepal varies seasonally. For the last 30 years (1991–2020), the average monthly temperature ranges from ∼ 5°C in January to ∼18°C in July, whereas average rainfall ranges from ∼20 mm in November to ∼340 mm in July (ADB and WB, 2021). Nepal is likely to experience a higher rate of warming in two future periods (2016–2045 and 2036–2065) compared to the reference period, i.e., 1981 to 2010 (GoN/MoFE, 2021) and spatiotemporal changes in precipitation over the period from 1981 to 2010 (Karki et al., 2017). Diverse current and future climatic conditions within comparatively small areas (Dawadi, 2017) make Nepal an ideal place to study the effects of climate change on forests.

2.2. Data collection

The primary data used in this study were obtained from the third national forest inventory (NFI), which was carried out during 2010–2014. The NFI adopted a two-phase systematic sampling design, composed of 450 clusters containing 1,553 Permanent Sample Plots (PSPs)-after excluding inaccessible PSPs - in the real ground (See Figure 1). Data were collected only from the accessible PSPs (slope up to 100 % or 45°). On the sample plots tree related attributes such as diameter at breast height (DBH) and tree height were recorded for the analysis of growing stock, above ground tree biomass and carbon. The third NFI is the first assessment in Nepal that collected soil samples to analyze the SOC of the forests. Four soil pits were established in a cardinal direction in each PSP to collect soil samples. At each cardinal direction, soil pits of appropriate size were dug within the 2 m * 2 m area size at a 21 m distance from the PSP center. In each soil pit, soil samples were collected from three different horizons (1–10 cm, 10–20 cm, and 20–30 cm) up to the depth of 30 cm and were mixed together resulting in 3 soil samples representing three different soil horizons in each PSP (DFRS/FRA, 2014).

Besides forest inventory data, the study used 19 bioclimatic variables representing historic data (near current) representing average figures for the years 1970–2000 at 30 arc sec (∼1 km²) resolution (Fick and Hijmans, 2017). The study also used future climate data from the WorldClim data set¹ at 30 arc sec (∼1 km²) resolution. representing Couple Modeled Inter-comparison Project Phase 6 (CMIP6) based on shared socio-economic pathways (SSP2 4.5) scenario from 2041 to 2060 (i.e., 2050 on average) with resulting global warming of 1.6 –2.5°C (IPCC, 2021). We used this scenario in the study because it is an intermediate scenario among five prescribed by Intergovernmental Panel on Climate Change (IPCC) and is based on the current level of CO₂ emission until the middle of the century.

2.3. Soil organic carbon analysis

Altogether 1,049 PSPs out of 1,553 PSPs were used for SOC analysis. Data from 504 PSPs were removed for one or more of the factors: inappropriateness of the site condition e.g., presence of rock or boulder instead of soil, and missing data for important variables such as aspect, distance to settlement, etc. The Black wet combustion method (Walkley and Black, 1934) was applied in the Nepalese Department of Forest Research and Survey (DFRS) soil laboratory to analyze the SOC content. In addition, dry combustion and LECO CHN Analyzer were used in the Metla Soil Laboratory, Finland, to assure the quality of the laboratory test.

2.4. Above ground tree biomass analysis

Above-ground tree biomass was also estimated from the same PSPs used for SOC analysis. DBH of the tree greater than 5 cm was recorded from the PSPs. The stem volume of the tree was calculated using the equation given by Sharma and Pukkala (1990a).

where,

ln = Natural logarithm to the base 2.71828,

d = DBH in cm.

h = Total tree height in m.

a, b and c are parameters of the volume equation (Annex 1).

To get stem volume in a cubic meter, the model estimation must be divided by 1,000. According to Sharma and Pukkala (1990b), the air-dried wood densities of the tree species range from 352 kg/m³ for Trewia nudiflora L. to 960 kg/m³ for Acacia catechu (L.F.) wild.

In order to estimate AGTB, firstly stem biomass was calculated using following equation.

where,

Volume = Stem volume (m³).

Density = Air-dried wood density (kg/m³).

Branch biomass and foliage biomass of the trees were calculated using branch-to-stem and foliage-to-stem ratios, respectively based on tree species and three classes of the size of the stem (small = < 28 cm, medium = 28–53 cm and large = > 53 cm) at diameter at breast height (Sharma and Pukkala, 1990a). Finally, above ground tree biomass (AGTB) of each tree in the PSPs was calculated by using an equation (3). The individual tree biomass (Kg/m³) within PSP was calculated and it was further converted into ton/ha using the plot expansion factor.

2.5. Partition of data set

In order to have independent data sets for model development and model testing, the data were partitioned into two sets A total of 737 (70%) samples were used as training data and 312 (30%) were used as test data. The partitioning of the data was done by using the createDataPartition function in the “caret” package (Kuhn, 2008), which splits data randomly into two different sub-sets with different proportions.

2.6. Variables selection

Altogether 36 variables were identified for modeling purposes (Table 1). Out of these 36 variables, we conducted variable selection based on the importance of the variables in the model. To select the important variables, the function VSURF from the R package “VSURF” (Genuer et al., 2010) was used. This package selects important predictor variables for the model by step-wise analysis i.e., threshold, interpretation and prediction. Finally, the selected predictor variables were applied in the model development.

2.7. Estimation of SOC and AGTB using random forest model

Estimation of the SOC and AGTB was conducted (including all predictor variables and only important predictor variables) using a random forest model (RFM) by a function randomForest under the “randomForest” package in R software (version 4.2.1). RFM is a machine learning tool using bootstrap aggregating to develop models with an improved prediction (Jin et al., 2020). It is based on two parameters i.e., Number of predictor variables (Mtree) and the number of decision trees (Ntree). The random selection of predictor variables and the records in the data set to generate one decision tree helps to achieve higher accuracy in subsequent iterations. In this way, the RFM function generates many decision trees and averages to give an estimation for the response variable. Averaging a large number of decision trees helps to increase accuracy. Moreover, RFM generates IncNodePurity which is a total decrease in node impurities when splitting the predictor variables. An increase in the IncNodePurity value of the predictor variables indicates the higher importance of the variables. Furthermore, the partial dependence plot was plotted using the partialPlot function under the “randomForest” package in the R program. The plot shows the marginal effects of predictor variables on the response variable in the model (Friedman, 2001). It is generally used to evaluate whether the relationship between the predictor and response variable is linear, non-linear, or more complex.

2.8. Model validation

Observed data (test data) was plotted against predicted data (model output) to see their relationship for visual interpretation. Moreover, RMSE, RMSE% and R² value was calculated to determine the efficiency of the model developed using the rmse function (“ModelMetrics” package), rmse_per function (“forestmangr” package) and summary function in the R program. The RMSE and RMSE% were calculated as follows.

Where,

$\widehat{\text{y}}$i = the predicted SOC or AGTB on the i^th plot,

yi = the observed SOC or AGTB on the i^th plot,

${\overline{\text{y}}}_{\text{i}}$ = the average value of SOC or AGTB.

n = Number of samples.

3. Results

3.1. Variables used in the model

Altogether 35 independent variables were used for the prediction of SOC or AGTB in the study. Of which, nine variables were selected for the prediction of SOC (Bio1, Bio4, Bio7, Bio8, Bio10, Bio12, Forest type, Distance to settlement and Crown cover) and four variables for the prediction of AGTB (Basal area, Altitude, Bio5 and Bio14).

3.2. Variables importance in the model

The selected 9 and 4 Predictor variables for estimating SOC and AGTB, respectively showed different importance values in the models. The predictor variable “Bio8” was found to be the most important variable for the prediction of SOC followed by Bio1, Bio10, Forest type, Bio7, Bio4, Distance to settlement, Bio12 and Crown cover (Figure 1A) whereas Basal area showed its importance highest for the prediction of AGTB followed by Altitude, Bio5, and Bio14 (Figure 1B).

3.3. SOC and AGTB estimation

The random forest model was run in two ways. Firstly, all 35 predictor variables (RFM1 and RFM3) were used in the model (RMF1 and RMF3) for the estimation of SOC and AGTB. Secondly, only predictor variables with high-importance values were used in the model (RFM2 and RFM4) for the same estimation (Table 2). The root mean square error (RMSE), RMSE% and coefficient of determination (R²) were found similar for using all 35 predictor variables and using only 9 predictor variables for the estimation of SOC. On the other hand, the performance of the model for the estimation of AGTB was found slightly better while using 35 predictor variables compared to 4 predictor variables (Table 2).

3.4. Relation between number of decision trees and error in the model

The number of decision trees (or “trees”) in the Random forest model represents the number of sub-samples selected randomly from the original data set. Increasing the number of decision trees helps to reduce the error in the model. The error was sharply reduced when the number of sub-samples selected from the sample population increased from 1 to 100 and slowed down afterward in both the SOC (Figure 2A) and AGTB (Figure 2B) models.

3.5. Accuracy assessment

Model performance varied in the estimation of SOC (RFM2) and AGTB (RFM4) using test data. RMSE% was found lower in the estimation of SOC as compared to the estimation of AGTB (Table 3).

Moreover, the degree of fitness of the model calculated from the predicted value against the observed value for the estimation of SOC was found to be strong i.e., R² = 0.759 and the relation was found significant (p < 0.05) (Figure 3A). A similar degree of fitness was also found in the case of AGTB estimation i.e., R² = 0.762 and (p < 0.05) (Figure 3B).

3.6. Partial dependence plots (Response plots)

Partial dependence plots for each important predictor variable were plotted for both SOC (RFM2) and AGTB (RFM4) models. Our study found that the response variable SOC responded positively with Crown cover, Distance to settlement and Bio12, and responded negatively with Bio1, Bio7, Bio8 and Bio10, whereas it responded both ways (non-linear relation) with Bio4.

An increase in distance to settlement from the forests up to 8,000 m contributed to the increase in SOC, while for longer distances no effect on SOC was found. Similarly, an increase in crown cover and Bio12 also contributed to the increase in SOC. Furthermore, Bio1, Bio8 and Bio10 did not contribute to SOC up to the temperature of 12, 17, and 19°C, respectively. However, the increase in temperature after those limits contributed to a decrease in SOC. In contrast, Bio4 contributed to a decrease in SOC up to 500 mm and afterward, it contributed to an increase in SOC. Lastly, The comparison of forest types revealed that 1, 11, and 17 contributed more to SOC than the other forest types (Figure 4).

Above-ground tree biomass responded differently with the four predicted variables (Basal area, Altitude, Bio5 and Bio14). Basal area and Bio5 showed a positive relation with AGTB, while Bio14 and Altitude showed both positive and negative (Figure 5). Basal area up to 80 m²/ha of the forests increased AGTB, and then the amount of AGTB stayed more or less stable, while an increase in Bio5 further increased AGTB. In contrast, altitude and Bio14 decreased AGTB up to 2,000 m and 7 mm, respectively, and after those limits, these variables increased AGTB.

3.7. Amount of soil organic carbon (SOC) and above ground tree biomass (AGTB) using climate change scenario (CMIP6, SSP2 4.5 for 2050)

The CMIP, SSP2 4.5 scenario showed an effect of climate change on SOC and AGTB, assuming other predictors to be the same. An average SOC stock of 63.6 tons/ha was found in the near current period, while it would decrease to 61.15 tons/ha in the future scenario. Unlikely, an average AGTB would increase to 210.57 tons/ha in the future scenario compared to the near current period (204.51 ton/ha). Our result shows that the amount of SOC would likely decrease by 3.85% while AGTB would likely increase by 2.96% in the future climate change scenario (Table 4).

The SOC and AGTB were plotted over the individual PSP. The blue lines in both figures represent SOC/ATGB in the near current period (1970–2000) whereas red lines represent them in the future scenario (2040–2060). The blue line has exceeded the red line indicating decreasing trend of SOC in the future scenario (Figure 6A). But, for the amount of AGTB, a red line has exceeded the blue line indicating the trend of AGTB in the future (Figure 6B).

4. Discussion

4.1. Performance of the random forest models

A random forest model has been used in this study to estimate SOC and AGTB in the current and future climate change scenario. The RFM has been popular and considered to produce better accuracy than the multiple linear regression (Powell et al., 2010; Hounkpatin et al., 2018). The multiple linear regression approach is though popular, it does not well capture the complex relationships between the forest variables; and soil-landscape relationships subject to non-linear dynamics (Grimm et al., 2008; Chen et al., 2012). The coefficient of determination (R² value) produced by our model for the estimation of AGTB is found strong, i.e., 0.74, which is higher than or similar to the other previous studies that used different predictor variables to predict AGTB using RFM (Powell et al., 2010; López-Serrano et al., 2020; Nguyen and Kappas, 2020; Li Z. et al., 2022). Similarly, the RMSE percent of the AGTB model in our study is slightly higher than the results reported by Musthafa and Singh (2022), Wai et al. (2022) and slightly lower than result of Zhu et al. (2020). These studies completely used other predictors (Image pixel value, age, crown density etc.) compared to our studies (especially temperature and precipitation). Moreover, R² and RMSE% of the model for the estimation of SOC is smaller and higher, respectively than other studies (Hounkpatin et al., 2018; Lee et al., 2020). The possible reason could be the use of different independent variables in those studies than our study.

If we compare the estimated quantity of SOC and AGTB of the Random forest model with the forest resource assessment result (DFRS, 2015c) based on design based estimation, the quantity is found similar. The estimated average of SOC (63.6 ton/ha) in this study is 4.9% lower than the forest resource assessment result (66.88 ton/ha) whereas the average of AGTB (204.51 ton/ha) is 5.14% higher than the forest inventory result (i.e., 194.51 ton/ha). Though number of samples used in the model is lower than the samples used in design based approach, the Random forest model seems to be capable to produce better accuracy.

4.2. Factors influencing above ground tree biomass (AGTB)

Based on the previous studies, altitude, stand characteristics (tree age, density), slope, aspect, temperature and precipitation affect the AGTB (Powell et al., 2010; Van der Laan et al., 2014; Yan et al., 2015; Zhang et al., 2016; Rajput et al., 2017; Shen et al., 2018). Similar to the other studies (Wang et al., 2017; Bennett et al., 2020; Larjavaara et al., 2021), our study reports the effect of climate attributes on AGTB, particularly due to the maximum temperature of the warmest month (Bio5) and precipitation of the driest month (Bio14).

The RFM used in this study helps understand AGTB as functions of predictors such as altitude and climatic variables. Previous studies also used RFM to estimate AGTB, but were confined to a few predictor variables such as image pixel value, canopy height, topography, vegetation indices, and texture feature (Li Z. et al., 2022; Musthafa and Singh, 2022; Wai et al., 2022).

Our model shows an increase of AGTB under future climate change scenarios, a finding that is consistent with the results reported by Day et al. (2008), Saeed et al. (2019), Wang et al. (2019). Temperature is the most determining climatic factor that helps in accumulation of tree biomass particularly in the growth season (Devi et al., 2020). Similarly, an increase in precipitation in the driest months (Bio14) helps increase AGTB by lengthening the growing season that supports plant growth (Vaganov et al., 1999). Our results show a positive effect of Bio14 and warmer in the summer (similar to Bio5) with AGTB is consistent with the study conducted by Lewis et al. (2013), Devi et al. (2020), Noguchi et al. (2022). Unlike the forests in Nepal, rising temperature is likely to decrease above-ground biomass in the old-growth tropical forests (Larjavaara et al., 2021).

4.3. Factors influencing soil organic carbon (SOC)

Nine predictor variables, including topographic variables, climatic variables, forest types, distance to settlement and crown cover, are important to influence SOC distribution. Previous studies also report similar influencing variables for SOC, topography (altitude, slope and aspect), above-ground biomass, basal area, canopy cover, climate and forest types (Kara et al., 2008; Song et al., 2012; Mohammad and Rasel, 2013; Liu et al., 2016; Bangroo et al., 2017; Chaturvedi and Sun, 2018; Jakšić et al., 2021; Shapkota and Kafle, 2021). Apart from other variables, distance to settlement has also an effect on SOC. Our result shows that an increase in distance to settlement- which is likely to reduce human disturbances- results increase in SOC stock (Figure 4). SOC distribution is likely to be more in the area with less human disturbance (Mehta et al., 2008; Eshaghi Rad et al., 2018). Human disturbance such as logging and tree harvest result in a decrease in soil carbon and organic matter (Latty et al., 2004; Moreno et al., 2007).

Our study shows the mean temperature of the wettest quarter (Bio8) as a major predictor variable to estimate SOC in particular. In general, climatic variables are dominating other variables for the prediction of SOC. Similar to our study, previous studies have reported the effect of climate (temperature and precipitation) on SOC (Chen et al., 2015; Alani et al., 2017; Sun et al., 2019; Odebiri et al., 2020; Fang et al., 2022). But, other studies also found altitude as a major variable for SOC prediction (Dieleman et al., 2013; Odebiri et al., 2020). This is also true because altitude though does not directly influence SOC but is an indicator of various climatic functions that govern different vegetation and soil formation processes (Hanawalt and Wittaker, 1976). Thus, altitude can be used as a proxy of climatic variables (Malla et al., 2022).

Furthermore, our model shows a decrease in SOC amount in the future climate change scenario which is similar to the finding reported by Dimobe et al. (2018). Owing to global warming, surface temperature will continue to increase, at least, until 2050 under all emission scenarios (IPCC, 2021). The result shows an increase in temperature (in the future scenario) leads to a decrease in SOC amount, which is supported by other studies (Liu et al., 2021; Zhao et al., 2021). The possible reason could be an increase of soil microbial decomposition due to higher temperature resulting less SOC amount (Dong et al., 2021; Song et al., 2021). Similarly, the negative association of precipitation (in the future scenario) with SOC in our result is similar to the result reported by Alani et al. (2017). The higher amount of precipitation possibly causes to leach dissolved organic carbon of the soil resulting less SOC accumulation.

4.4. Implications of the study

4.4.1. Model implications

Our model shows the effect of climatic variables, topographic variables, forest variables, and distance to settlements on the amount of SOC and AGTB. Particularly, climatic variables (temperature and precipitation) have a direct relation with the formation process of SOC and AGTB. Mean annual precipitation is a driver of the amount of SOC and AGTB (Mehta et al., 2014). Precipitation influences soil moisture and hydrological processes (Heisler and Weltzin, 2006) which is an important factor in SOC cycling (Aanderud et al., 2010) and affects AGTB through functional traits (Cheng et al., 2021). Similarly, temperature also affects the amount of SOC (Zinn et al., 2018; Zhang et al., 2021) and the amount of AGTB (Poudel et al., 2011; Larjavaara et al., 2021). An increase in temperature helps soil microbial decomposition resulting in higher carbon emission or lower SOC accumulation (Dong et al., 2021; Song et al., 2021) whereas warming temperature enhances tree growth resulting in an increase in AGTB (Way and Oren, 2010).

However, most of the previous studies were focused on forest inventory data accompanied by satellite imageries to estimate AGTB and SOC of the latest period (Angelopoulou et al., 2019; López-Serrano et al., 2020). But for the future prediction of AGTB and SOC under climate change scenario, projected bioclimatic variables are necessary as input variables to produce a precise result. These projected bioclimatic variables have been widely used in species distribution modeling, and habitat suitability under different climate change scenarios (Fyllas et al., 2022; Khan et al., 2022; Shrestha et al., 2022) however, the use of these variables have been very limited for SOC prediction (Liu et al., 2021; Zhao et al., 2021).

Inclusion of Bio2 and Bio6 bioclimatic variables with inventory data helps estimate AGTB and SOC, respectively in a better way. Readily available bioclimatic variables not only improve the performance of the model but also reduce the cost of the model. Combining bioclimatic variables with other variables for the prediction of SOC and AGTB can be a viable option to understand the present scenario.

Moreover, using easily available projected bioclimatic variables under different climate change scenarios see text footnote 1 has benefited us in getting a better understanding the trend of SOC and AGTB in the future. Thus, our model shows an advantage over previous model to assess AGTB and SOC in the future climate change scenario using freely available climatic data.

4.4.2. Implications to Nepal

The forest policy of Nepal emphasizes managing forest resources largely through community participation. Almost half of the total forests have been managed under the broad regime of community-based forest management (Ghimire and Lamichhane, 2020). After the involvement of local people in forest resource management, Nepal has received positive changes in the forest condition. The forest cover of Nepal has been in an increasing trend reported by different assessments, i.e., 29% (DFRS, 1999), 40.36% (DFRS, 2015c), 41.69% (FRTC, 2022). Despite these facts, our model shows the amount of SOC is likely to be decreased in the future, whereas there will be a slight gain in the AGTB. In order to increase SOC in the future, the result highlights the need of management intervention to reduce forest degradation and deforestation through sustainable forest management in all the forests of Nepal to deal with climate change impact.

5. Conclusion

Climatic variables (temperature and precipitation) show an effect on the amount of SOC and AGTB in the future climate change scenario. However, the effect of climate on the SOC and AGTB is opposite (positive with AGTB while negative with SOC). Therefore, management intervention through sustainable forest management is crucial in all forest types to maintain SOC level in the future climate change scenario.

Our study proposed an approach for estimating the AGTB and SOC of Nepal using forest inventory data combined with world climate data (bioclimatic variables). Integrating readily available bioclimatic variables along with other predictor variables helps estimate SOC and AGTB in the near current and future scenario, leading to a better understanding of AGTB and SOC dynamics.

Data availability statement

The datasets presented in this article are not readily available because data sets are available from Forest Research and Training Center, Kathmandu, Nepal upon the request of the researchers, students or institutions. Requests to access the datasets should be directed to Forest Research and Training Center, info@frtc.gov.np.

Author contributions

RM contributes on data acquisition, data analysis and drafting manuscript. PN and MK contribute from draft stage to the final stage of the manuscripts. All authors discussed and revised the manuscript and read and approved the final manuscript.

Funding

This research was partially funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy–EXC 2037 ‘CLICCS–Climate, Climatic Change, and Society’–Project Number: 390683824, contribution to the Center for Earth System Research and Sustainability (CEN) of Universität Hamburg.

Acknowledgments

We thank FRTC, Kathmandu for the provision of data, Sudiksha Joshi, Ph.D (USA) for proofreading, and the reviewers for their constructive comments and suggestions.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

1. ^ www.worldclim.org

References

ANNEX