Comparison of Airborne Laser Scanning (3D) with Sentinel-2 (2D) for Estimating Above Ground Biomass/Carbon Stock in a Subtropical Chir Pine Forest of Balakot, Pakistan

Forest ecosystems act as a sink of atmospheric carbon dioxide in the form of biomass, and plays one of the crucial role for carbon sequestration and in regulating the global carbon cycle. Few studies based on ground sample plots were conducted for estimating forest biomass/carbon stock across Pakistan. This study comparing the rst time the potential of three dimensional (3D) airborne laser scanning (ALS) with two dimensional (2D) Sentinel-2 to estimate above-ground biomass/carbon stock (AGB/C) in a Subtropical Chir Pine forest of Balakot, Pakistan. We derived height and density metrics from the ALS canopy height model (CHM), and different metrics from Sentinel-2 images, and were regressed with eld measured AGB/C at sample plots locations. We found R 2 = 0.86 with RMSE% = 25.70, and R 2 = 0.62 with RMSE% = 43.92 for ALS and for Sentinel-2 respectively with ground measured AGB/C at sample plots locations. Our study demonstrated that 3D ALS technology has greater potential and is the most accurate option as compared to 2D Sentinel-2 for regular planning and monitoring of AGB/C in the context of the national forest inventory of Pakistan. Our study will be useful for the accomplishment of the REDD+ in measuring, reporting, and verication of forest resources, and future sustainable utilization of forest, safeguarding the livelihoods of forest-dependent people, and reducing pressure on forest ecosystems.


Introduction
Regular and updated information about vegetation biomass assessment is crucial for quantifying carbon stocks of different forest types, their storing potential, and energy source for future protection, conservation, and reporting. Many countries have committed themselves to regular reporting of carbon stock under the background of climate change initiatives such as the Kyoto protocol which is based on biomass inventories (Fassnacht et al. 2021;Patenaude et al. 2005). Hence, a developing country like Pakistan also needs to make a strategy for regularly assessing their vegetation biomass potential at a national level or at least in smaller spatial units. Very few studies have been reported based on ground sample plots for estimating forest biomass/carbon stock across Pakistan (Ahmad et al. 2018a;Ahmad et al. 2018b;Ahmad et al. 2014;Ahmad and Nizami 2015;Amir et al. ;Mannan et al. 2019;Uddin et al. 2019). However, in the case of ground sample inventories, the accuracy can be limited to the sample plots, and cannot cover spatial variation of large geographical forest area. Combining remote sensing datasets with a eld-based sample forest inventory can be a promising outcome for the wall-to-wall mapping of large geographical forest areas (Ullah et al. 2017a;Ullah et al. 2019).
Among remote sensing datasets, airborne laser scanning (ALS) technology has revolutionized forest mapping, and are using constantly for operational forest management in many Nordic countries (Maltamo and Packalen 2014;Naesset 2007;Naesset 2014;Ullah et al. 2017a). Over the last decade, many worthiest studies have been reported by estimating vegetation biomass in different forest regions across the globe using ALS technology. (Domingo et al. 2018;Ene et al. 2018;Ene et al. 2016;Gobakken et al. 2012;Hansen et al. 2015;Maltamo et al. 2016). Unfortunately, not a single study was reported yet in Pakistan by assessing the potential usage of ALS technology for estimating forest biomass and other structural parameters. Thanks to the 477 Pakistan Army Survey Group for introducing this wonderful technology which is now available in Pakistan for different applications and has greater potential in forest science. Also, none of the ALS based studies speci cally focused on characterizing vegetation biomass of subtropical Chir pine forest. To our knowledge, this will be the rst study in Pakistan by testing the capability of ALS for the assessment of above-ground vegetation biomass/carbon (AGB/C) stock in the subtropical Chir Pine forest of Balakot, Pakistan.
Besides the availability of 3D technology such as ALS for research, and smaller spatial units to forest communities in Pakistan, the open-source 2D spaceborne sensors has also greater potential for regular planning and monitoring of natural resources. Although, some remote sensing studies have already been reported by using the space-borne multispectral 2D data in Pakistan (Butt et al. 2015;Iqbal and Khan 2014;Qamer et al. 2016;Qasim et al. 2011;Raziq et al. 2016;Saleemi 2015;Shehzad et al. 2014;Ullah et al. 2016;Ullah et al. 2017c), but all of them related to spatial-temporal forest cover and others land-use change assessment. Hence, we are also evaluating the open-source 2D Sentinel-2 sensor for the assessment and mapping of AGB/C to explore viable options to forest administration for regular and updated information about the forest resources of Pakistan.
More speci cally, our research aim is to compare the potential of ALS (3D) technology with Sentinel-2 (2D) for estimating above AGB/C stock in a Subtropical Chir Pine Forest of Balakot, Pakistan. The output of this study will be used as baseline information to the forest administration of Pakistan to think about the possible introduction of remote sensing technology for regular planning and monitoring of their natural resources.

Study area, and eld inventory
The study was conducted at Balakot, district Mansehra of Khyber Pakhtunkhwa province of Pakistan. It is extending from latitude 34°33'21.32"N, and longitude 73°19'58.03"E to latitude 34°32'58.34"N and longitude 73°20'41.50"E on the right side of River Kunhar. Its elevation ranges from 1031 meters to 1472 m above sea level. The rainfall takes place mostly from February-March (late winter) and June-August (the monsoon season). The forest type of the study site dominated by Pinus roxburghii (Chir pine) with a few associated species that include Poplus nigra (Poplar), Ailanthus altissima (tree of heaven), Salix tetrasperma (willow tree), Ficus caricia (Fig), and Broussonetia papyrifera (Paper mulberry). Figure 1 depicts the geographical location of the study site.
Field data were collected from 24 August 2019 to 05 September 2019 where tree measurements were done from randomly selected 30 sample plots each having 17.84 m of radii. The center coordinates of sample plots were recorded with Stonex X5 GPS which has sub-metric accuracy. In every selected plot area, all trees having a width of greater than 7 cm were selected and their diameter at breast height point (DBH) and heights was measured. Finally, above-ground biomass per KG was derived for each tree using the following allometric equation developed by Ali et al. (2020).
Chir Pine biomass (KG) = 0.0224 × (D^2× (H)^0.9767) Eq-1 Where H represents the height of the tree, and D stands for Diameter at the level of breast height point. The single measured tree biomass per sample plots was summed up and nally converted into tons/hectare by using the expansion factor. The carbon stock is usually regarded as about half of the dry biomass. Hence, the total biomass per sample plots/KG was converted by multiplying the 0.50 value (Walker et al. 2011). Table 1 depicts the summary of the measured biomass, and carbon stock of the ground inventory sample plots. Remote sensing data and its processing The details of the ALS data used in our study are shown in Table 1. The 3D ALS point clouds were used to generate the Digital Surface Model (DSM) and Digital Terrain Model (DTM) with 1 m spatial resolution. The DSM generation was accomplished through the selection of the maximum option which uses the highest points within the 1 m spatial resolution pixel as a ltering and natural neighbor as an interpolation. Similarly, for the DTM generation, the minimum option was selected in which the lowest point is used within a resolution of 1 m pixel as a ltering and natural neighbor as an interpolation. Finally, the Canopy Height Model (CHM) of 1 m was generated by subtracting the corresponding DTM from DSM.

Extraction of metrics from ALS CHM
The most commonly used derived metrics from ALS CHM in forest inventory are the height, and its percentiles (Naesset 2004). A total of 15 height metrics extracted from the ALS CHMs using a 12 m radius circle corresponds to the size of ground sample plots. For instance, the metrics extracted were the minimum (h min ), mean (h mean ), (h max ), range (h range ), and the height percentiles i.e., hp99, hp95, hp90, hp80 . . ., hp10. We also calculated the coe cient of variation (h CV ), standard deviation (h STD ), and range from the CHMs by taking into consideration the variation and heterogeneity of ALS CHM. The metrics, as mentioned above, derived from the vertical distribution of CHMs. We also calculated the canopy cover density (cd), and canopy volume (h sum ), for the horizontal distribution of the canopy structure. The cd is calculated by dividing the number of pixels with heights above 2 m by the total number of pixels within an area of 12 m radius circular sample plots. Besides, 10 types of other forest cover density metrics such as cd1, cd2, and cd3 . . . . cd10 were extracted at sample plots location. More detail of the abovementioned metrics can be found in our previously published papers ( Modeling for predicting carbon stock Before tting multiple regression models, the multicollinearity problem was resolved between the extracted explanatory variables of the ALS CHM and Sentinel-2. The problem of multicollinearity between the explanatory variables (> 0.7 correlation) was eliminated by using the variance in ation factors. The variation in ation factor for each of the variables was calculated and the variables were sequentially dropped, which were found to be above a pre-speci ed threshold value (>2) according to a suggestion made by Zuur et al. (2010). We used the CAR package of the R-statistics for calculating the variance in ation factor (Fox and Weisberg 2018). After solving the problem of multicollinearity between the explanatory variables, a multiple linear regression model was tted between the height variables/metrics obtained from the ALS CHMs, and Sentinel-2 as explanatory variables, and the biomass/carbon stock (tons/ha) as the response variables. Subsequently, a stepwise method was applied to remove the explanatory variables by minimizing the Akaike information criterion. For reporting RMSE and RMSE%, leave-one-out cross-validation was used in which each observation was regarded as a testing set whereas the remaining data were used as a training set. The caret package in R-statistics software was used to analyze and compute all the calculations (Kuhn 2008). After the selection of the nal most model, the nal metrics selected were generated on a rasterized cell size corresponds to the ground sample plots using a wall-to-wall mapping approach. Figure 2 depicts the ow chart of the research methodology adopted in our study.

Results
The nal most explanatory variable selected for the prediction of biomass/carbon stock using ALS CHM were the h range , h mean , and h STD (Table 4). Similarly, the nal most explanatory variable selected for the prediction of biomass/carbon stock using Sentinel-2 were the NDVI, and NLI (Table 4). We found R 2 =0.86, and R 2 = 0.62 for the ALS CHM, and Sentinel-2 respectively with the ground measured biomass/carbon stock. More detail can be found in Table 4. Our results showed that ALS based CHM showed higher accuracy with RMSE% = 25.70 as compared to Sentinel-2 with RMSE% = 43.92 (Table 5).  Figure 3 showed scatter plots between the observed versus predicted biomass/carbon stock (tons/ha -1 ) for both ALS and Sentinel-2.
The nal predicted AGB/C maps for both ALS CHM, and Sentinel-2 are shown in Figure 4.
The summary statistics of predicted maps between the predicted and observed AGB/C are shown in Figure 5. Our results also showed here that ALS CHM is more closed to eld based observation as compared to Sentinel-2 in terms of comparison of mean AGB/C (ton/ha).

Discussion
3D ALS has revolutionized forest mapping, and many developed countries adopted this technology for operational forest management. However, the potential usage of this technology is never explored for different forestry applications in developing countries like Pakistan. That's the reason that forest administration in Pakistan is unaware of the potential usage of ALS data for different forestry applications. We assess the potential usage of a 3D ALS active airborne sensor for the rst time in Pakistan to estimate AGB/C in a Subtropical Chir Pine Forest of Balakot, Pakistan. For ALS based CHM, we obtain R 2 = 0.85 with RMSE% = 25.70 again ground AGB/C (ton/ha). Our obtained biomass prediction results are in agreement with other comparable earlier studies in the context of estimating forest AGB/C across the different forested regions (Baccini and Asner 2013; Bright et al. 2012;García et al. 2010;Naesset 2011;Stephens et al. 2007). It can be noted that we didn't nd any previous study about Pakistan, and Subtropical Chir Pine forest for estimating forest AGB/C (T/ha).
Besides the availability of the ALS data for only research and smaller spatial scales in Pakistan, we also tested the 2D open-source multispectral Sentinel-2 data to explore the other possible viable option to the forest community for regular reporting and monitoring of the forest resources. For 2D Sentinel-2 multispectral data, we obtained R 2 = 0.62 with RMSE = 43.92 for estimating AGB/C in a Subtropical Chir Pine Forest of Dir Kohistan. Our obtained results in term of RMSE % values of Sentinel-2 imagery for estimating AGB/C seems to be slightly lower but can be par as compared to previous results obtained by Nuthammachot et al. (2020), Nuthammachot et al. (2018), Pandit et al. (2018), andVafaei et al. (2018) which can due to different forest types, altitudinal gradients, etc.
Summarizing the ndings of our results 3D ALS sensor showed high accuracy and thus has greater potential as compared to 2D multispectral Sentinel-2 imagery for estimating AGB/C in a Subtropical Chir Pine forest of Balakot, Pakistan. Further, the prediction models derived from both ALS CHM and Sentinel-2 images ful ll the normality assumption since they are highly signi cant and their probability is greater than 0.05. The out of the results can be successfully employed for the regular assessment and mapping of AGB/C of the Subtropical Chir Pine forest at both the local and regional scale studies.

Conclusions/outlook
Despite well-proven tested ALS technology in developed countries, the forest community in Pakistan is unaware of its potential usage in the eld of forest science. The present study attempt to motivate the forest communities towards the potential usage and possible introduction of remote sensing technology such as 3D ALS data for different application in the eld forest sciences across Pakistan. We demonstrated successfully the usage and potential of 3D ALS sensors for estimation forest AGB/C of Subtropical Chir Pine forest for the rst time in the history of Pakistan. However, it can be tested in the future as well for the other essential forest types of Pakistan. Besides the availability of the costly option such as ALS data, or which can only be affordable for research and smaller spatial scales in Pakistan, we also tested the 2D open-source multispectral Sentinel-2 data to open other viable ways to forest community for regular reporting and monitoring of the forest resources. The nding of the results shares useful information and methodology regarding the uses of 3D ALS technology, and 2D multispectral Sentinel-2 images in combination with eld data for provincial, national forest department, environments, climatologist, and REDD + policymakers in Pakistan.

Author contribution
Dr. Sami Ullah is the lead author and was involved in the overall project designing, remote sensing and eld data collection, analysis, and writing etc. Mr. Tahir Sayed arrange the ALS data, and it's preprocessing. Dr. Muhammad Sha que help in writing up the manuscript. Mr. Muhammad Saad help in allometric equation and processing of eld data. Mr. Adnan help in eld data collection, Sentinel-2 images preprocessing steps, and writing. partnership of Shaheed Benzair Bhutto University, Center of GIS and Space Applications for providing remote sensing processing facilities, nancing eld data collection, and expertise etc. The geographical location of the study test site, the yellow circular depicts the location of ground inventory sample plots. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.   Final predicted biomass/carbon stock maps (Tons/ha-1) i.e., (A) ALS biomass (B) ALS Carbons Stock (C) Sentinel-2 Biomass, and (D) Sentinel-2 Carbon Stock. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.