Abstract
Forest aboveground biomass (AGB) measurement is a direct estimator of the live carbon stock of that forest region. Increasing emission and concentration of CO2 is a global threat as it is a major cause of today’s global warming. The forest AGB is a live carbon sequester that plays a major role by absorbing atmospheric CO2. There are field-based measurement methods of AGB, but the main disadvantage is that they are primarily destructive. Several authors have undertaken AGB estimation using different remote sensing data types, but they are mostly not cost-effective for extensive study areas. We have created a cost-effective algorithm for AGB estimation using multispectral (MSS) data. In this study, Indian Remote-Sensing Satellite-P6 (IRS P6) Linear Imaging Self-Scanning Sensor-4 (LISS-IV) MSS data have been used for the analysis. The research has tried to estimate the AGB of different types of forests existing in the study area by using various vegetation indices and the gray-level co-occurrence matrix (GLCM) and created a hybrid methodology combining the vegetation indices and GLCM. Among all vegetation indices, the simple ratio (SR) highly correlates with AGB of pure deciduous and coniferous forests. In a mixed forest region, due to a mixture of two canopy stands, there is a mixture of foliage angle and optical scattering distribution. Therefore, modified simple ratio (MSR) becomes dominant in mixed forest AGB estimation. Previously there was no study to justify this GLCM texture parameter selection. In this study, we have justified the parameter selection of GLCM texture statistics. This parameter selection will help researchers choose the proper GLCM texture parameter for their study. Integration of GLCM textures with vegetation indices enhances the AGB model strength for all forest regions. The deciduous forest map gives validation R2 of 0.89 with an RMSE of 1.93 ton/pixel. The validation R2 of the Coniferous Forest map is 0.83 with an RMSE of 1.35 ton/pixel. There is a comparatively identifiable improvement in mixed forest with validation R2 of 0.96 and RMSE of 0.25 ton/pixel. This study shows AGB storage of deciduous forest has a maximum share over other forest region of Kalimpong forest.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Introduction
Greenhouse gases are a significant contributor to global warming. Among all other greenhouse gases, CO2 contributes the most to global warming. The forest AGB is a live sequester of emitted CO2. The assessment of forest aboveground biomass (AGB) is an essential part of national development planning as it incorporates the productivity of an ecosystem, carbon budget, and etc. (Parresol 1999; Zianis & Mencuccini 2004; Zheng et al. 2004; Hall et al. 2006). In addition to the economic aspect, it dramatically impacts global climatic variables.
Field-based AGB measurements were the standard methods for AGB estimation of a forest area. The field-based methods were mostly destructive and not practicable in a mountainous terrain where most of the area is inaccessible. Remote sensing has been mainly used to estimate forest AGB as it is more economical and less time-consuming to measure the AGB of a forest than field-based estimation. Remote sensing methods are the only way to assess the AGB of forests in hilly terrain with inaccessible tracts of land parcels.
Researchers have studied different remote sensing approaches. These approaches are majorly divided into two parts: (1) optical or passive remote sensing and (2) active remote sensing approach. In the active remote sensing-based approach, researchers have mainly focused on the synthetic-aperture radar (SAR)-based approach for monitoring AGB. Among all SAR data, only L band data have penetration capability through the surface canopy layer and then get scattered back by the trunk and main branches (Blomberg et al. 2018). The L band data to be used should have to be cross-polarized (i.e., HV or VH) as cross-polarized data can give the volumetric backscatter (Xiang et al. 2016) from the tree trunks and branches that have a reasonable correlation with the AGB of that forest (Luckman et al. 1997; Kurvonen et al. 1999; Sun et al. 2002; Günlü and Ercanlı, 2020). Although the L band cross-polarized data have a reasonable correlation with forest AGB, temporal measurements of AGB at a specified period are very costly and uneconomical. The researchers have studied optical data to estimate forest AGB to make the estimation more economical. Multispectral and hyperspectral optical data have been used to estimate forest AGB by different researchers all over the globe. Although hyperspectral data demonstrates some AGB estimation successes, the data suffers from the problem of band redundancy. The application of hyperspectral data is significantly less in AGB estimation because of its minimal availability (Hyperspectral data are mainly airborne and captured in small areas) (Lu et al. 2016).
Multispectral (MSS) data are the most used data for forest AGB assessment among all other remote sensing data due to the availability of its various spatial, spectral, radiometric, and temporal resolutions. There are various MSS data available like Landsat-5 TM (Roy & Ravan 1996; Wylie et al. 2002; Foody et al. 2003; Phua & Saito 2003; Lu 2005; Lu et al. 2005 (a); Du et al. 2012; Singh & Das 2014; Günlü et al. 2014; Barrachina et al. 2015; Das & Singh 2016), Landsat-7 ETM + (Zheng et al. 2004; Avitabile et al. 2012), Landsat-8 OLI (Ali et al. 2018; Li et al. 2018), Sentinel-2 (Askar et al. 2018; Ali et al. 2018; Pandit et al. 2018; Keleş et al. 2021), and LISS-3 (Kumar et al. 2013; Mayamanikandan et al. 2017; Nandy et al. 2017), Aster (Fuchs et al. 2009). Due to free availability and good spectral resolution, Landsat series and Sentinel 2 are the most commonly used MSS data for forest AGB estimation. Although the spatial resolution of Sentinel 2 data does not meet the accuracy required in the estimation of forest AGB, researchers are also using the high-resolution MSS data like IKONOS (Thenkabail et al. 2004; Kayitakire et al. 2006), Quickbird (Fuchs et al. 2009; Sousa et al. 2015), Worldview (Obeyed et al. 2018), GeoEye (Mareya et al. 2018), and RapidEye (Gascón et al. 2019) for estimation of forest AGB. Due to high-cost involvement and lack of availability of IKONOS data, it is challenging to identify the AGB of a forest where regular monitoring is required at a specific interval of time.
Previously, a few studies on AGB estimation have been done by some researchers using Linear Imaging Self-Scanning Sensor-4 (LISS-IV). Madugundu et al. (2008) used LISS-IV data to estimate AGB by leaf area index (LAI) determination and got an R2 value of 0.63 between the estimated and field-observed AGB of Haliyal and Yellapur Forest Divisions, Western Ghats of Karnataka, India. On the other hand, Pargal et al. (2017) studied the AGB of different forest types of Yellapur Forest Division, Uttara Kannada District, Western Ghats of Karnataka, India, with LISS-IV. He used the vegetation index, NDVI, for his analysis. He got R2 = 0.82 for his AGB model. Attri and Kushwaha (2018) have used LISS-IV data on Barkot Forest Range, Dehradun, India. For identification of AGB using NDVI, he got R2 = 0.71 for his AGB model.
Very few studies apply gray-level co-occurrence matrix (GLCM) texture parameters on AGB estimation. Lu and Batistella (2005) and Lu (2005) have used eight textural parameters of Landsat-5 TM data to identify AGB and got maximum R2 = 0.68 and 0.71, respectively. Kayitakire et al. (2006) have used GLCM of IKONOS data. He got an R2 value of 0.82. This work indicates that high-resolution GLCM textures have a high correlation with forest AGB. For AGB estimation, some researchers have integrated both vegetation indices with the GLCM texture. Lu (2005) has used the integrated model with Landsat-5 TM data and got R2 = 0.77. Fuchs et al. (2009) have used coarse resolution ASTER data and high-resolution Quickbird data and got R2 = 0.63 and 0.69, respectively. Avitabile et al. (2012) had used Landsat-7 ETM + data and got R2 = 0.81. Nandy et al. (2017) had used LISS-3 data and got R2 = 0.746. Gascón et al. (2019) had used RapidEye data and got R2 = 0.69 for their AGB models.
The forest region of Kalimpong has dense forest cover. There is no study available on the estimation of AGB of Kalimpong forest. Due to a gradual increase in human habitation, deforestation is a significant concern for these forests. In addition to this, monitoring of AGB is one of the essential measures to identify forest health. Kalimpong is hilly terrain, with most places inaccessible for collecting physical measurements of forest AGB due to stiff slopes. Not only stiff slopes but several reserve forests, protected forests, and Indian army-occupied forest regions are not permitted entry due to government rules. Therefore, physical identification and forest inventory-based sample collection are challenging in the Kalimpong forest region. For delineating these problems of the study area, this research attempted to identify a cost-effective method for AGB estimation, which can be used by the authorities for the measurement of AGB periodically. In this work, MSS data that is IRS P6 LISS-IV, which is a meager cost high-resolution data, have been used to create a cost-effective, accurate methodology for estimating AGB. There are few studies on AGB estimation using vegetation indices of high-resolution LISS-4 data. However, no study is available on the relationship of GLCM-based texture parameters of LISS-4 bands with AGB of the forest. There is no study on the impact of forest vegetation indices and textures of spectral response with AGB of different forest classes present in Himalayan Forest regions. An attempt has also been made to identify whether the integration of texture and vegetation indices influences the improvement of AGB measurement.
This study has been made to identify the models using LISS-4 generated vegetation indices and GLCM-based texture parameters with AGB of different forest types (coniferous, deciduous, and mixed) of the Kalimpong district. The best fit models have been correlated to come out with models using vegetation indices and the GLCM parameters to increase the accuracy of assessment of AGB of various types of forest in the study area.
Materials and methods
Study area
The Kalimpong district of West Bengal, India is a part of the north-eastern Himalayan region. It lies between 27° 11′ 44″ N to 26° 51′ 40″ N latitude and 88° 23′ 16″ E and 88° 53′ 00″ E longitude. The areal extent of the Kalimpong district is 1095.18 km2 (Fig. 1). The mean monthly temperature of this area lies between 30 and 9 °C. The annual average rainfall is 2200 mm. The forests under the Kalimpong district mostly fall under the Kalimpong Forest Division of West Bengal Forest Development Corporation (WBFDC), excluding the area under Neora Valley National Park that had been handed over to Wild Life Wing Forest Directorate. The elevation of the study area ranges from 150 to 3700 m. The upper altitude region consists of evergreen alpine coniferous forest, and the lower altitude is covered by temperate deciduous forest. Being a hilly location, most of the forested area is inaccessible, and the accessible places also pose difficulty in collecting the field data. Additionally, shadows of the hills cause many problems in using satellite data in the study area. Only minimal data are available for reliably estimating the existing forest biomass in the Kalimpong district of West Bengal. Deforestation due to the increasing pressure of the growing population and frequent landslides on many forested slopes are affecting the biomass stock in that region, so estimating actual biomass present in that region is necessary to monitor the forests. The detailed methodology flowchart is shown in Fig. 2.
Study area
Methodology flowchart
Field inventory data collection and AGB estimation
A total of 59 random sample plots were collected from the different forests of Kalimpong in places that are accessible. There were 18 coniferous forest plots, 26 deciduous forest plots, and 15 mixed forest plots. The field plots were established using purposive sampling (Nesha et al., 2020) due to the constraints of accessibility in the presence of steep slopes and also administrative permissions. Picea rubens and Juniperus virginiana were the major species found in coniferous forests. In the deciduous forest, the primary species were Tectona grandis, Garuga pinnata, Toona ciliate, Holarrhena pubescens, Albizia procera, Shorea robusta, Alnus nepalensis, Terminalia myriocarpa, Quercus pachyphylla, Bucklandia populnea, Alnus nepalensis, Ficus cunia, Schima wallichii, Michelia champaca, and etc. The diameter at breast height (DBH), tree height, wood density, and plot area were collected from the field. The details of field inventoried data of the sample areas in coniferous, deciduous, and mixed forests are shown in Table 1.
The field plot distribution has been shown on the LISS-IV MSS data (Fig. 1). The field estimation of AGB has been calculated from this field-collected inventory data.
The AGB was calculated using the volumetric conversion method (Brown & Lugo 1992). AGB density (t/ha) = F
where VOB = volume over bark; WD = volume-weighted average wood density (tons/m3 or g/cm3); and BEF = biomass expansion factor (ratio of oven-dry AGB of trees to oven-dry biomass of inventoried volume). (Brown, 1997).
Volume over bark (VOB) has been calculated using the DBH value and the height. Using VOB per hector and volume-weighted average wood density, the biomass of the inventoried volume has been calculated. Biomass expansion factor (BEF) has been calculated using the biomass of the inventoried volume. Volume expansion factor (VEF) has been calculated using the VOB30 (i.e., this VOB includes the DBH of trees having a minimum diameter greater than 30 cm) value. VOB10 (i.e., this VOB includes the DBH of trees having a minimum diameter greater than 10 cm) has been calculated using the volume expansion factor and VOB30. Finally, AGB has been calculated using VOB10, biomass expansion factor, and volume-weighted average wood density. The AGB in tons/ha of all the 59 plots measured in the field has been calculated using this methodology. Finally, the AGB thus calculated was divided into three classes: coniferous, deciduous, and mixed type of forest for further analysis. It has been found that among 59 plots, there are 18 coniferous forest plots, 26 deciduous forest plots, and 15 mixed forest plots available. We have divided these data into 70% training sample plots (i.e., those sample plots have been used to correlate and model making) and 30% test sample plots (i.e., those sample plots have been used to validate the model).
where V1, V2,…. Vn = volume of species 1, 2,.. to the nth species and Vt = total volume WD1, WD2,….. Wdn = wood density of species 1, 2,…… to the nth species. (Brown, 1997)
where BEF = biomass expansion factor (dimensionless) (Brown, 1997);
where Wcrown = tree crown dry weight (kg), composed of foliage, thick and thin branches; Wbole = tree bole dry weight (kg) (i.e., trunk weight) (Brown, 1997).
where BV = biomass of inventoried volume in t/ha, calculated as the product of VOB/ha (m3/ha) and wood density (t/m3) (Brown, 1997).
LISS-IV data accusation
Two cloud-free scenes of IRS P6 LISS-IV were acquired for Kalimpong district from NRSC, Hyderabad, India. Those images were geometrically and atmospherically corrected. LISS-IV data have a swath of 70 km. It consists of three spectral bands: B2 (green (0.52–0.59 mm)), B3 (red (0.62–0.68 mm)), and B4 (NIR (0.76–0.86 mm)). The details of those LISS-IV scenes are given in Table 2. The landuse landcover, forest class map, and vegetation indices of Kalimpong forest have been calculated from this data with the help of the ERDAS Imagine software.
Preparation of forest classification map
The landuse and landcover map has been prepared using a supervised classification based on the field-observed training data points with a maximum likelihood algorithm. The study area has been classified into eight classes: forest, agriculture, waterbody, settlement, barren land, open scrub, tea garden, and sand over the Kalimpong district. Processing of images has been done using a supervised classification based on collected training sets. Among all classes, the forest areas have the maximum coverage of about 817.01 km2 (about 74.57% of the total Kalimpong district). Other than forest, the agricultural land has coverage of about 89.91 km2 (8.21%); settlement has coverage of about 89.58 km2 (8.17%), open scrub has coverage of about 44.77 km2 (4.08%), waterbody has coverage of about 24.13 km2 (2.20%), barren land has coverage of about 20.38 km2 (1.85%), tea garden has coverage of about 9.07 km2 (0.83%), and the sand deposit has coverage of about 0.73 km2 (0.066%). The landuse landcover map has been validated using field-collected 191 test datasets. An accuracy of 87.96% and an overall Kappa 0.81 have been achieved for this landuse landcover map (Fig. 3 (a)). The confusion of landuse and landcover distribution is shown in Table 3.
a Landuse landcover map. b Forest map.
The forest class map (Fig. 3(b)) has been prepared by extracting the landuse classified forest area from the LISS 4 MSS data using a supervised classification of the extracted LISS 4 MSS data. The distribution of forest cover is described in Table 4. The forest map has been validated using field-collected 150 test datasets. An accuracy of 89.33% and an overall Kappa 0.80 have been achieved for the forest map. The confusion of landuse and landcover distribution is shown in Table 5.
Estimation of vegetation indices
From the spectral response curve of the vegetation region, it is identified that the blue and red reflectance is significantly less in the visible spectrum than green. However, there is a sudden increment in reflectance from vegetation beyond the visible range in the infrared region. Red reflectance is sensitive to chlorophyll content, and the near-infrared reflectance is sensitive to the mesophyll structure of leaves. The higher the difference between the red and near-infrared reflectance, the higher the green vegetation present in that pixel. Using this spectral phenomenon of vegetation, researchers have developed several vegetation indices to relate the biophysical parameters of vegetation, like leaf area index, percentage vegetation cover, a fraction of absorbed photo-synthetically active radiation (fAPAR), photosynthetic capacity, and carbon dioxide fluxes, and also, they identified a relationship between the forest biomass. In our study, six vegetation indices generated from the high-resolution LISS-IV data have been correlated to measure the AGB of Kalimpong forest. The selected vegetation indices are as given in Table 6.
Relationship between AGB of different forests with vegetation indices
The whole field-collected dataset is divided into three major parts according to the forest classification, i.e., deciduous, coniferous, and mixed. Furthermore, the dataset is divided into training (70%) and testing (30%) datasets to establish the model and validate that model. The different vegetation indices are compared to identify the correlation (Pearson correlation) with AGB for each forest class individually. The AGB density has been converted into AGB of each pixel. Those per pixel AGB have been correlated with vegetation index of that pixel generated from LISS-IV data. Plot-wise vegetation index distribution of coniferous, deciduous, and mixed forest AGB are shown in Appendix Tables 22, 23 and 24 respectively. The correlations of vegetation indices with AGB are shown in Table 7.
Among all vegetation indices, simple ratio (SR) has the maximum correlation in coniferous (r = 0.81) and deciduous (r = 0.75) forest. In mixed forest, modified simple ratio (MSR) has maximum correlation (r = 0.84). The relationship between AGB of coniferous, deciduous, and mixed forests is shown in Fig. 4 a, b, and c, respectively.
Relationship between AGB of coniferous (a), deciduous (b), and mixed (c) with vegetation indices
Estimation of GLCM (gray-level co-occurrence matrix)-based texture parameters
The GLCM-based texture parameters show different combinations of a pixel’s gray-level occurrence in an image scene by relating with its neighborhood pixel’s gray value. This study generated ten second-order statistics from the GLCM of 3 bands of LISS-4 data (i.e., 30 GLCM texture parameters). Those texture maps have been correlated with the AGB of three different types of forests of Kalimpong. The details of those GLCM second-order statistics are given below (Tables 8 and 9).
The description of notations of the equations given in Table 8 is as follows:
P(i,j) = (i,j)th entry in a normalized gray-tone spatial dependence matrix. (i,j) stands for the number of times gray tones i and j have been neighbors; µ and σ are the mean and standard deviation respectively (Haralick et al. 1973).
Relationship between AGB of different forests with GLCM parameters
The details of training data of coniferous, deciduous, and mixed forest AGB with the all GLCM texture features are in Appendix, Tables 25, 26, 27, 28, 29, 30, 31, 32 and 33. The correlation of AGB with GLCM parameters is given in Table 10. It has been identified that (Table 9) among all GLCM textures, angular second moment of the green band (ASM_GREEN), homogeneity of the red band (HOM_RED), and entropy of infrared band (ENT_IR) have a maximum correlation with AGB of coniferous forest. In the deciduous forest, the entropy of green (ENT_GREEN), red (ENT_RED), and infrared (ENT_IR) has the highest correlation with AGB. Similarly, the contrast of green (CON_GREEN), infrared (CON_IR), and entropy of red (ENT_RED) have the highest correlation with mixed forest AGB.
The relationships between the highest correlated GLCM texture features with AGB of coniferous forest (Fig. 4a, b, c), deciduous forest (Fig. 4e, f, g), and mixed forest (Fig. 5h, i, j) have been chosen for establishing GLCM-based multi-linear regression (MLR) models for estimation of AGB of coniferous forest (Table 11), deciduous forest (Table 12), and mixed forest (Table 13).
Relationship between AGB of coniferous forest (a-c), deciduous forest (d-f), and mixed forest (g-i) with GLCM texture features
Model developed by combining vegetation indices with the GLCM texture parameters
Results
In vegetation index-based models, SR generated models were used for the coniferous and deciduous forests to generate an AGB distribution map of both forests. Similarly, the mixed forest AGB distribution map was generated using the MSR generated model. Those maps have been validated using 30% of the test datasets. The validation plots are shown in Fig. 6a, b, and c for coniferous, deciduous, and mixed forest, respectively.
Validation plot between observed and predicted AGB of coniferous (a), deciduous (b), and mixed (c) forest using vegetation indices
Among all GLCM-based MLR models, model 7 has the highest R2 with AGB of coniferous forest (Table 11). Also, deciduous forest model 7 has the highest R2 (Table 12), and mixed forest model 7 has the highest R2 (Table 13). Those GLCM-based parameters have been used to generate AGB distribution maps. Those maps have been validated using 30% of the test datasets. The validation plots are shown in Fig. 7a, b, and c for the coniferous, deciduous, and mixed forest, respectively.
Validation plot between observed and predicted AGB of coniferous (a), deciduous (b), and mixed (c) forest using GLCM-based models
The model 7 of the combined model of all forests has the highest R2 (Tables 14, 15, 16) with AGB. That shows the importance of all the chosen parameters. These combined models have been used to generate AGB maps of each forest. Those maps have been validated using 30% of the test datasets. The validation plots are shown in Fig. 8a, b, and c for the coniferous, deciduous, and mixed forest, respectively.
Validation plot between observed and predicted AGB of coniferous (a), deciduous (b), and mixed (c) forest using combined modeling
The detailed model statistics generated from vegetation indices, GLCM, and combined model are discussed in Table 17.
Discussions
The study area is hilly terrain with 74.57% of the forest where most places are inaccessible. There is an urgent need to identify the biomass content of the district. This biomass measurement is for regulatory measures to control the degradation of forests and maintain the forest’s health. Keeping this objective in view, this work envisages creating a methodology that will be economical for periodically measuring biomass of the district. There are many options available today for biomass measurement by remote sensing methods, but in this work, LISS-4 data was selected to keep the investigation cost as low as possible.
A few studies were available on the applicability of LISS-4 data as an AGB estimator to date. Madugundu et al. (2008) used LISS-IV generated NDVI to estimate LAI as an identifier of AGB of Haliyal and Yellapur Forest Divisions, Western Ghats of Karnataka, India. However, Madugundu et al. (2008) did not directly relate to forest AGB and vegetation index (NDVI). Madugundu et al.’s (2008) study was only based on Haliyal and Yellapur Forest’s deciduous forest of Western Ghats of Karnataka, India. Pargal et al. (2017), on the other hand, used LISS-IV to investigate the AGB of different forest types in the Yellapur Forest Division, Uttara Kannada District, Western Ghats of Karnataka, India. He used the vegetation index, only NDVI, for his analysis. Pargal et al.’s (2017) AGB model achieved R2 = 0.82. However, Pargal et al. (2017) cannot generate different AGB models for different forest classes. Attri and Kushwaha (2018) have used LISS-IV data on Barkot Forest Range, Dehradun, India. Attri and Kushwaha (2018) used only NDVI as a vegetation index to identify AGB and got R2 = 0.71 for his AGB model. Bindu et al. (2020) used kg/pixel-based AGB estimation using LISS-4 generated NDVI. Bindu et al. (2020) achieved an R2 of 0.71 for his NDVI-based AGB model. However, no studies have used all LISS-4 generated vegetation indices for their AGB modeling. No study has generated an individual AGB model for different forest classes using LISS-4. To date, no study also used LISS-4 generated GLCM-based textures to model forest AGB.
This study correlated high-resolution LISS-4 MSS generated six vegetation indices with AGB. It has been identified that the pure coniferous (r2 = 0.81) and deciduous forest (r2 = 0.75) AGB are strongly correlated with SR (Table 7). Due to mixed patches of coniferous and deciduous stands in mixed forest regions, the response of SR is comparatively weaker than the nonlinear vegetation index MSR (Chen 1996). We have found that MSR has a comparatively strong correlation with mixed forest AGB (r2 = 0.84). Although SR has a strong correlation with pure forest regions, the vegetation index-based model standard error (Table 16) shows that the ability of SR-based model to estimate coniferous forest AGB (SE = 1.08 ton/pixel) is comparatively better than AGB of deciduous forest (SE = 2.17 ton/pixel) due to the presence of varying tree species and so varying spectral responses in deciduous forest. Due to different optical and geometrical surfaces of mixed forest canopies, MSR is a good estimator of mixed forest AGB with SE = 0.81 ton/pixel. These model generated AGB maps have been validated with the field-collected test data sets. The validation of maps also has a strong coefficient of determination (R2) with field-observed AGB and map generated AGB of deciduous (R2 = 0.84), coniferous (R2 = 0.87), and mixed forest (R2 = 0.82). However, the RMSE of validation is relatively higher in the deciduous forest (2.29 ton/pixel) compared to coniferous (1.38 ton/pixel) and mixed forest (0.54 ton/pixel).
Spectral responses play more essential roles in biomass estimation than textural images when the forest stand structure is relatively simple, but textural images are more important than spectral responses in complex forest stand structures (Lu 2005). Our study has generated 10 GLCM-based texture parameters of 3 different spectral bands of LISS-4 MSS data. These GLCM texture parameters have been correlated with AGB to identify the effect of forest canopy complexity responses among the neighboring pixels. In this study, we have discussed the reasons and justifications for GLCM properties’ choice (Table 9). The adjusted R2 of GLCM models for deciduous (0.51) and mixed forest (0.55) are comparatively better (Table 15) than coniferous forest (0.44). GLCM texture has a better response in complex forest structures with varying tree species. The coniferous forest has fewer tree species than deciduous and mixed forests. The coniferous GLCM model is weaker than the deciduous and mixed forest. Due to higher complexity in tree species variation, the GLCM model of the mixed forest has a higher response than the deciduous forest. These model generated maps of each forest have been validated with the test data. The validation shows that RMSE has reduced in each forest in GLCM models compared to vegetation indices (Table 17). It is seen that the validation R2 of GLCM models is poor compared to the vegetation index model.
An attempt has been made to combine the models generated by vegetation indices and GLCM-based texture parameters to increase the accuracy of AGB measurement. It has been identified that the combined models have an improvement over individual models (Table 17) in all forest classes. Therefore, the combined models have been chosen to estimate the AGB of each forest class. After generating coniferous, deciduous, and mixed forest AGB maps, the AGB maps have been merged to generate the AGB distribution map of the Kalimpong forest region (Fig. 9). The deciduous forest map shows a validation R2 of 0.89 with an RMSE of 1.93 ton/pixel. Coniferous forest map validation R2 is 0.83 with an RMSE of 1.35 ton/pixel. There is a comparatively identifiable improvement in mixed forest with validation R2 of 0.96 and RMSE of 0.25 ton/pixel. ANOVA report of coniferous, deciduous, and mixed forest is shown in Tables 18, 19, and 20. The equations used to generate the AGB distribution map are given below. The detailed AGB report generated from the AGB distribution map is in Table 21.
-
Coniferous forest: (r2 = 0.83)
AGB = (0.820297320671545*0.7524*EXP(0.5584*SR)) + 0.330865636425123 * (1.6716 * ASM_GREEN − 6.8163) + 0.524162541451282 *(10.194 * HOM_RED − 0.3108) − 0.0173250138195351 * (6.8828 * EXP(− 3.322*ENT_IR)) − 2.9838956546503.
-
Deciduous forest: (r2 = 0.74)
AGB = 0.901864914911025*0.2615*EXP(0.6789*SR) − 0.31626023959724 * 0.0014 * EXP(1.257 * ENT_GREEN) + 0.984231244376142 * 0.014* EXP(0.902* ENT_RED) + 0.116195076761434 * 0.0037 * EXP(0.9324*ENT_IR) − 1.2694170570619.
-
Mixed forest: (r2 = 0.92)
AGB of Kalimpong forest from combined model
AGB = (0.648609247582477*2.4128*MSR^2.8338) + (0.376613117313149*(0.0016*CON_GREEN + 1.556)) + (0.12037360444411*(1.6644*ENT_RED − 6.8199)) + (0.396131933575747 *(0.7872 * EXP(0.001 *CON_IR))) − 1.34826615320316.
Conclusion
The work envisaged a cost-effective methodology for identifying the AGB of a study area in the Himalayan region. Most of the area in the study area is inaccessible due to rugged terrain and is covered mainly by forest. Due to poor per capita income in the study area, there is much pilferage of forest inventory. To maintain the health of the forest and for regulatory measurement of the forest, it was decided to use LISS-4 data for this work. There are various options available today in identifying AGB using the remote sensing approach, but using low-cost data will reduce the total cost of the analysis for periodic measurement of AGB.
This study suggested that LISS-4 MSS data can estimate the high-resolution AGB distribution of the Himalayan Forest region with adequate accuracy. Various options available for AGB estimation using optical remote sensing data were attempted in work. In this study, six vegetation indices have been used for AGB estimation of different forests of Kalimpong forest regions. Among them, SR gives the highest correlation with AGB of pure deciduous and coniferous forests. In mixed forest regions, due to a mixture of two canopy stands, there is a mixture of foliage angle and optical scattering distribution. Therefore, the nonlinear vegetation index of SR (i.e., MSR) becomes dominant in AGB estimation in mixed forest. It was found that GLCM-based texture parameters of LISS-4 bands have the ability of AGB estimation. Attempts were made to identify the AGB using the GLCM parameters. The results obtained depicted varying accuracy wherein some categories of forest GLCM parameters showed better results, whereas in some types the vegetation indices had better accuracy. An attempt was made to integrate GLCM textures with vegetation indices to identify whether a better accuracy could be obtained in the AGB estimation of the study area. The results obtained have enhanced the AGB model strength for all forest regions, so the integrated model using vegetation indices and the GLCM parameters were selected for the calculation of AGB of the study area.
This study shows AGB storage of deciduous forest has a maximum share over other forest regions of Kalimpong forest. Not only in pure deciduous and coniferous regions, this study has developed an adequately accurate model for mixed forest regions also, where there is a mixture of different canopy cover. The study has also concluded an adequately accurate model with more cost-effectiveness than L band microwave data for AGB modeling.
This model can be used to estimate accurate AGB of different forest regions. Integration of microwave data with LISS-4 can improve the accuracy of AGB monitoring in the future. This model can be beneficial for use in carbon budgeting.
References
Ali A, Ullah S, Bushra S, Ahmad N, Ali A, Khan M (2018) Quantifying forest carbon stocks by integrating satellite images and forest inventory data. Austrian J for Sci 135(2):93–118
Askar NN, Phairuang W, Wicaksono P, Sayektiningsih T (2018) Estimating aboveground biomass on private forest using Sentinel-2 Imagery. J Sens 2018:11. https://doi.org/10.1155/2018/6745629
Attri P, Kushwaha S (2018) Estimation of biomass and carbon pool in Barkot Forest Range, UK using geospatial tools. ISPRS Ann Photogramm Remote Sens Spatial Inf Sci 4(5):121–128. https://doi.org/10.5194/isprs-annals-IV-5-121-2018
Avitabile V, Baccini A, Friedl M, Schmullius C (2012) Capabilities and limitations of Landsat and land cover data for aboveground woody biomass estimation of Uganda. Remote Sens Environ 117:366–380
Barrachina M, Cristóbal J, Tulla A (2015) Estimating above-ground biomass on mountain meadows and pastures through remote sensing. Int J Appl Earth Obs Geoinf 38:184–192
Bindu G, Rajan P, Jishnu ES, Joseph KA (2020) Carbon stock assessment of mangroves using remote sensing and geographic information system. Egypt J Remote Sens 23(1):1–9
Blomberg E, Ferro-Famil L, Ulander L (2018) Forest biomass retrieval from L-band SAR using tomographic ground backscatter removal. IEEE Geosci Remote Sens Lett 15(7):1030–1034
Brown S (1997) Estimating biomass and biomass change of tropical forests: a primer. A Forest Resources Assessment publication, Rome
Brown S, Lugo A (1992) Above ground biomass estimates for tropical moist forests of the Brazilian Amazon. Interciencia Interciencia 17(1):8–18
Chen J (1996) Evaluation of vegetation indices and a modified simple ratio for boreal applications. Can J Remote Sens 22(3):229–242
Chen J, Chilar J (1996) Retrieving leaf area index of boreal conifer forests using Landsat TM images. Remote Sens Environ 55(2):153–162
Das S, Singh T (2016) Forest type, diversity and biomass estimation in tropical forests of Western Ghat of Maharashtra using geospatial techniques. Small-Scale for 15:517–532
Du H, Zhou G, Ge H, Fan W, Xu X, Fan W, Shi Y (2012) Satellite-based carbon stock estimation for bamboo forest with a non-linear partial least square regression technique. Int J Remote Sens 33(6):1917–1933
Dube T, Mutanga O, Shoko C, Adelabu S, Bangira T (2016) Remote sensing of aboveground forest biomass: a review. Trop Ecol 57(2):125–132
Foody G, Boyd D, Cutler M (2003) Predictive relations of tropical forest biomass from Landsat TM data and their transferability between regions. Remote Sens Environ 85(4):463–474
Fuchs H, Magdon P, Kleinn C, Flessa H (2009) Estimating aboveground carbon in a catchment of the Siberian forest tundra: combining satellite imagery and field inventory. Remote Sens Environ 113(3):518–531
Gascón L, Ceccherini G, Haro F, Avitabile V, Eva H (2019) The potential of high resolution (5 m) RapidEye optical data to estimate above ground biomass at the national level over Tanzania. Forests 10(2):107
Günlü A, Ercanli I (2020) Artificial neural network models by ALOS PALSAR data for aboveground stand carbon predictions of pure beech stands: a case study from northern of Turkey. Geocarto in 35(1):17–28
Günlü A, Ercanli I, Başkent E, Çakır G (2014) Estimating aboveground biomass using Landsat TM imagery: a case study of Anatolian Crimean pine forests in Turkey. Ann for Res 57(2):289–298
Hall R, Skakun R, Arsenault E, Case B (2006) Modeling forest stand structure attributes using Landsat ETM+ data: application to mapping of aboveground biomass and stand volume. For Ecol Manage 225(1–3):378–390
Haralick R, Shanmugam K, Dinstein I (1973) Textural features for image classification. IEEE Trans Syst Man Cybern Syst SMC 3(6):610–621
Huete A (1988) A soil-adjusted vegetation index (SAVI). Remote Sens Environ 25(3):295–309
Jackson R, Slater P, Pinter P (1983) Discrimination of growth and water stress in wheat by various vegetation indices through clear and turbid atmospheres. Remote Sens Environ 13(3):187–208
Kayitakire F, Hamel C, Defourny P (2006) Retrieving forest structure variables based on image texture analysis and IKONOS-2 imagery. Remote Sens Environ 102(3–4):390–401
Keleş S, Günlü A, Ercanli I (2021) Estimating aboveground stand carbon by combining Sentinel-1 and Sentinel-2 satellite data: a case study from Turkey. Forest Resources Resilience and Conflicts 117–126. https://doi.org/10.1016/B978-0-12-822931-6.00008-3
Kumar P, Sharma L, Pandey P, Sinha S, Nathawat M (2013) Geospatial strategy for tropical forest-wildlife reserve biomass estimation. IEEE J Sel Top Appl Earth Obs Remote Sens 6(2):917–923
Kurvonen L, Pulliainen J, Hallikainen M (1999) Retrieval of biomass in boreal forests from multitemporal ERS-1 and JERS-1 SAR images. IEEE Trans Geosci Remote Sens 37(1):198–205
Li B, Wang W, Bai L, Chen N, Wang W (2018) Estimation of aboveground vegetation biomass based on Landsat-8 OLI satellite images in the Guanzhong Basin. China Int J Remote Sens 40(10):3927–3947
Lu D (2005) Aboveground biomass estimation using Landsat TM data in the Brazilian Amazon. I Int J Remote Sens 26:2509–2525
Lu D, Batistella M (2005) Exploring TM image texture and its relationships with biomass estimation in Rondônia. Brazilian Amazon Acta Amazonica 35(2):249–257
Lu D, Batistella M, Moran E (2005) (a)) Satellite estimation of aboveground biomass and impacts of forest stand structure. Photogramm Eng Remote Sens 71(8):967–974
Lu D, Chen Q, Wang G, Liu L, Li G, Moran E (2016) A survey of remote sensing-based aboveground biomass estimation methods in forest ecosystems. Int J Digital Earth 9(1):63–105
Luckman A, Baker J, Kuplich T, Yanasse C, Frery A (1997) A study of the relationship between radar backscatter and regenerating forest biomass for space borne SAR instrument. Remote Sens Environ 60(1):1–13
Madugundu R, Nizalapur V, Jha C (2008) Estimation of LAI and above-ground biomass in deciduous forests: Western Ghats of Karnataka, India. Int J Appl Earth Obs Geoinf 10(2):211–219
Mareya H, Tagwireyi P, Ndaimani H, Gara T, Gwenzi D (2018) Estimating tree crown area and aboveground biomass in Miombo Woodlands from high-resolution RGB-only imagery. IEEE J Sel Top Appl Earth Obs Remote Sens 11(3):868–875
Mayamanikandan T, Jha C, Das I, Amminedu E, Reddy C (2017) Forest biomass estimation in tropical deciduous forests of western ghats using remote sensing data and GIS. In: 3rd International Conference on Environmental Management. Centre for Environment, JNTU, Hyderabad. https://www.researchgate.net/publication/329012038_Forest_Biomass_Estimation_in_Tropical_Deciduous_Forests_of_Western_Ghats_using_Remote_sensing_data_and_GIS
Nandy S, Singh R, Ghosh S, Watham T, Singh Kushwaha S, Kumar A, Dadhwal V (2017) Neural network-based modelling for forest. Carbon Manage. https://doi.org/10.1080/17583004.2017.1357402
Nesha M, Hussin Y, Leeuwen L, Sulistioadi Y (2020) Modeling and mapping aboveground biomass of the restored mangroves using ALOS-2 PALSAR-2 in East Kalimantan, Indonesia. Int J Appl Earth Observ Geoinfo 91:102158
Obeyed M, Mustafa Y, Akrawee Z (2018) Estimating and Mapping Aboveground Biomass of Natural Quercus Aegilops Using WorldView-3 Imagery. International Conference on Advanced Science and Engineering (ICOASE), Kurdistan Region, Iraq. https://ieeexplore.ieee.org/document/8548859
Pandit S, Tsuyuki S, Dube T (2018) Estimating above-ground biomass in sub-tropical buffer zone community forests, Nepal, using Sentinel 2 data. Remote Sens 10(4):601
Pargal S, Fararoda R, Rajashekar G, Balachandran N, Réjou-Méchain M, Barbier N, Jha CS, Pélissier R, Dadhwal VK, Couteron P (2017) Inverting aboveground biomass-canopy texture relationships in a landscape of forest mosaic in the western Ghats of India using very high resolution Cartosat imagery. Remote Sens 9:228. https://doi.org/10.3390/rs9030228
Parresol B (1999) Assessing tree and stand biomass: a review with examples and critical comparisons. For Sci 45(4):573–593
Pearson R, Miller L (1972) Remote mapping of standing crop biomass for estimation of the productivity of the shortgrass prairie. In: Proceedings of the 8th International Symposium on Remote Sensing of the Environment. Colorado State University, Pawnee National Grasslands, Colorado. https://eurekamag.com/research/000/179/000179997.php
Perry C, Lautenschlager L (1984) Functional equivalence of spectral vegetation indices. Remote Sens Environ 14(1–3):169–182
Phua M, Saito H (2003) Estimation of biomass of a mountainous tropical forest using Landsat TM data. Can J Remote Sens 29(4):429–440
Roujean J, Breon F (1995) Estimating PAR Absorbed by Vegetation from Bidirectional Reflectance Measurements. Remote Sens Environ 51:375–384. https://doi.org/10.1016/0034-4257(94)00114-3)
Rouse J, Haas R, Schell J, Deering D, Harlan J (1974) Monitoring the Vernal Advancement and Retrogradation (Greenwave Effect) of Natural Vegetation. NASA/GSFC Type III Final Report. Greenbelt, MD: NASA/ GSFC. https://ntrs.nasa.gov/citations/19740022555
Roy P, Ravan S (1996) Biomass estimation using satellite remote sensing data—an investigation on possible approaches for natural forest. J Biosci 21(4):535–561
Singh T, Das S (2014) Predictive analysis for vegetation biomass assessment in Western Ghat Region (WG) using geospatial techniques. J Indian Soc Remote Sens 42(3):549–557
Sousa A, Gonçalves A, Mesquita P, da Silva J (2015) Biomass estimation with high resolution satellite images: a case study of Quercus rotundifolia. ISPRS J Photogramm Remote Sens 101:69–79
Sun G, Ranson K, Kharuk V (2002) Radiometric slope correction for forest biomass estimation from SAR data in the western Sayani Mountains. Siberia Remote Sens Environ 79(2–3):279–287
Thenkabail P, Stucky N, Griscom B, Ashton M, Diels J, van der Meer B, Enclona E (2004) Biomass estimations and carbon stock calculations in the oil palm plantations of African derived savannas using IKONOS data. Int J Remote Sens 25(23):5447–5472
Wylie B, Meyer D, Tieszen L, Mannel S (2002) Satellite mapping of surface biophysical parameters at the biome scale over the North American grasslands: a case study. Remote Sens Environ 79(2–3):266–278
Xiang D, Ban Y, Su Y (2016) The cross-scattering component of polarimetric SAR in urban areas and its application to model-based scattering decomposition. Int J Remote Sens 37(16):3729–3752
Zheng D, Rademacher J, Chen J, Crow T, Bresee M, Moine J, Ryua S (2004) Estimating aboveground biomass using Landsat 7 ETM+ data across a managed landscape in northern Wisconsin, USA. Remote Sens Environ 93(3):402–411
Zhou J, Yan Guo R, Sun M et al (2017) The Effects of GLCM parameters on LAI estimation using texture values from Quickbird Satellite Imagery. Sci Rep 7:7366. https://doi.org/10.1038/s41598-017-07951-w
Zianis D, Mencuccini M (2004) On simplifying allometric analyses of forest biomass. For Ecol Manage 187(2–3):311–332
Acknowledgements
The researchers are thankful to the Department of Science and Technology, New Delhi for providing the opportunity to work in this area.
Funding
The researchers received funding for this project from the Department of Science and Technology, New Delhi.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Additional information
Responsible Editor: Biswajeet Pradhan
Appendix
Appendix
Rights and permissions
About this article
Cite this article
Ghosal, K., Das Bhattacharya, S. & Paul, P.K. Estimation of aboveground forest biomass in Himalayan region of West Bengal, India using IRS P6 LISS-IV data. Arab J Geosci 15, 601 (2022). https://doi.org/10.1007/s12517-022-09898-3
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s12517-022-09898-3