Biomass estimation models for Acacia saligna trees in restored landscapes

Acacia saligna, originating from Australia, is a naturalized multipurpose tree species widely grown to restore degraded lands of Africa. The contribution of A. saligna in biomass restoration can be quantified using a precise estimation of tree biomass carbon. This study developed species-specific allometric models and evaluated the spatial variation of tree biomass across restored areas in exclosures and open grazing landscapes. These models could play a considerable role in the monitoring of carbon dynamics across A. saligna planation dominated areas. We harvested, excavated, and weighed twenty-one sample trees representing different size classes to develop allometric models for the estimation of aboveground (AGB), belowground (BGB) and total tree (TB) biomass. The average dry-to-fresh mass ratio and the root-to-shoot ratio was 0.47 (±0.13) and 0.28 (±0.14), respectively. Tree biomass significantly correlated with diameter at breast height (r = 0.93; P < 0.001), diameter at stump height (r = 0.88, P < 0.001) and tree height (r = 0.56, P < 0.05). Our best biomass estimation models explained 86%, 82% and 87% of variations in AGB, BGB, and TB, respectively. Models using DSH and DSH & H explained 70%–78% of the variation in AGB, BGB, and TB. Estimated C-stock showed a significant relationship with stem density (R 2 = 0.91, P < 0.01). Estimated TB varied between 1.5–18 Mg ha−1 on grazed land and exclosures. Estimated C-stocks in the exclosure exceeded the estimated C-stock in the open grazing land by ∼60%. This implies that with proper management practices and enrichment planting A. saligna significantly contributes to increasing carbon accumulation on degraded landscapes, playing a key role in climate change mitigation efforts while improving land productivity.


Introduction
Acacia saligna is a native tree species to Western Australia [1].It is a fast-growing and drought-tolerant nitrogenfixing tree species in resource poor environments [2,3].The drought adaptive traits of the species have led to its current distribution across degraded and drought prone regions all over the tropics outside its original distribution area.As a member of the Fabaceae family, it has the capability to fix nitrogen, therefore being an important species for improvement of soil properties.Due to its soil stabilization capacity, it is useful for the rehabilitation of degraded landscapes, and thereby animal feed/fodder and a source of fuel wood [4][5][6].Due to these properties, A. saligna has a great potential to be used as an agroforestry tree, while contributing for climate change mitigation through carbon sequestration [7].It is identified as one of the three priority multipurpose tree species [8] and is found on more than 300, 000 ha as a plantation tree species globally in arid and semi-arid regions [6].In Ethiopia, A. saligna was first introduced in the 1970s and established as a trials in Tigray region [9], aiming to rehabilitation of massive degraded landscapes.Since 2000th, it is planted extensively [10].Despite of its vast distribution, the species does not show any character of invasiveness in Ethiopia [11] Because of its many positive effects A. saligna has been widely used to restore degraded landscapes and for climate mitigation efforts in Ethiopia and globally [11].
The plantations of fast-growing tree species, like A. saligna play a key role in re-vegetating degraded landscapes [8].In this regard, understanding the carbon storage dynamics of plantations and restored landscapes is important for climate change adaptation and mitigation-oriented program implementation and impact monitoring [12][13][14].More importantly, addressing climate change issues requires an understanding of the tree growth performance and a reliable quantification of forest carbon sequestration and their spatiotemporal dynamics.Despite this, the knowledge gap concerning the species' potential for emission reduction through its biomass accumulation is lacking for most tree species including A. saligna.
Destructive harvesting to build biomass estimation models is rarely conducted in the tropics [12,15].Study showed that grouping all species together and using generalized allometric relationships that are stratified by broad forest types or ecological zones has been highly effective in the tropics [16].However, there are very few existing species-specific allometric equations represent only 1% of tree species that are dominantly grown in Sub-Saharan Africa (SSA) [12,15].This further constrain the development of generalized allometric model for the SSA region [15].The lack of good number of species-specific allometric models limits the development of forest inventories for commercial biomass, and carbon estimation in the SSA [15].This problem is not only relevant for scientific purposes, but also obtaining financial rewards for emission reductions from protected and restored forest areas is currently of great interest to many [15,17].Generally, species-specific allometric equations are more preferred because tree species may differ greatly in their morphological characteristic which determine the nature of biomass estimation models [18][19][20].
However, in the absence of site-and species-specific allometric models, generalized allometric equations have been used to estimate forest carbon dynamics [21].Moreover , incautious utilization of generalized equations often leads to large errors in biomass estimates, thus it requires the ability to control the uncertainties and biases involved in using inaccurate biomass equations [22].This uncertainties might be linked to variations in site-and species-specific growth influencing factors, including rainfall variation, topography, soil conditions, species competition and composition [18][19][20].Aboveground tree components are used to derive species-specific models in addition to estimation of errors due to model selection.Thus, the belowground biomass component was usually estimated using root-to-shoot ratio (RSR) values, which varies across stands structure, tree age and size, soil conditions, and rainfall amount [18,23,24].For instance, many tree species adapted to dry conditions allocate more biomass to the roots compared to species adapted to humid regions [24].Hence, estimation of belowground biomass from aboveground biomass using root-to-shoot ratio (RSR) potentially increases uncertainties in the estimation of total biomass [24].Furthermore, diameter at breast height (DBH) and total tree height have often been used as standard predictors of biomass estimation [19,21,25,26] which limits biomass estimation where trees are felled, and logs are taken away illegally.This is mainly because of the common method of estimating deforestation through estimating DBH by developing equations relating DBH-DSH (DSH -stumps left behind) from sample trees which are measured for both DBH and DSH [19,27].The estimated DBH is used to estimate removed biomass by means of allometric equations.Through these sequence of procedures, it is clear that there would be an accumulation of propagated errors from the estimation of DBH to the estimation of biomass [19,[28][29][30].Thus, the most dependable way of reducing the magnitude of propagated errors is to estimate biomass directly from DSH models.Such models are also important to monitor forest degradation and carbon removal rates due to illegal logging and natural disasters from the remaining stump.Therefore , it is recommended to apply species-and site-specific biomass estimation models while converting dendrometric measurement to carbon accumulation and dynamics [12,22].More importantly, the IPCC, under Tier-2 or Tier-3 approaches, indicated that national greenhouse gas inventories must report estimates based on country-specific data [31].Thus, developing site-and species-specific biomass estimation models have both local, regional or/and global significance, because it is required to produce a reliable national greenhouse gas estimates; to develop regional and/or global scale generalized biomass estimation model and for local validation of biomass estimates using generalized biomass estimation models [17,21].Therefore, the present study was conducted to develop (i) species-and site-specific biomass estimation models for A. saligna with different predictor variables for use of flexibility; (ii) to identify the best allometric equation for the restored landscape and to compare it with previously developed species-specific and generalized models, and (iii) to estimate carbon accumulations dynamics in exclosures and adjacent open grazing land supported by A. saligna plantations in northern Ethiopia.

Study area
The study area was in Maybrazyu kebele/tabia (the smallest administrative unit), Tahtay Maychew district, central Tigray zone, Ethiopia (figure 1(a), (b)).Geographically, it is located at 13°52′ and 14°19′ N and 38°29′ and 38°42′ E, with sparse forest cover (figure 1(b)).The study landscape encompasses exclosures (henceforth ExSASP) and adjacent open grazing land (henceforth OGL) where both land use supported by A. saligna plantation.The total area of the study village was about 85 ha where the specific study exclosure and adjacent grazing land shares about 39% (33 ha) of the study village (figure SI1).Enrichment planting started in 2004.The average altitude of the study landscape is about 2133 m above sea level, with slope angles ranging from 11%-38%.Based on high-resolution gridded datasets (CRU TS V4.03 -from 1901-2018), the mean annual rainfall and maximum temperature are 568.4(±146.4)mm and 21.8 °C, respectively [32] (figure 1(c), (d)).The district belongs to the arid and semi-arid agro-ecological zones, with erratic rail fall and rugged terrain [33].

Vegetation data collection
We conducted a reconnaissance survey based on the information from the district Office of Agriculture and Natural Resources to select a landscape encompassing both the exclosure and adjacent open grazing land enriched by A. saligna plantation.Sample ExSASP and OGL supported by A. saligna enrichment plantation were used to conduct the vegetation survey.The ExSASP and OGL had similar environmental conditions before applying restoration practices.We applied systematic transect sampling techniques for the vegetation data collection.The first plot was randomly placed followed by systematic placements of the other plots within an equal interval in each transect.According to Mekuria and Aynekulu [34], the first plots should be laid at least 30 m away from the edge to avoid edge effects in exclosures and grazing lands.In each of ExSASP and OGL, we laid three transects established perpendicular to the main slope of the terrain considering topographic variation within the study site.Then, three sample plots of 20 m x 20 m (400m 2 ) at 100 m and 150 m distances between plots and transects, respectively were established [12,35].In total, we established eighteen sample plots across ExSASP and OGL.Finally, stem dendrometric parameters such as diameter at breast height (DBH, 1.3 m from the ground), diameter at stump height (DSH 0.3 m from the ground), and height (H) were recorded for each tree species encountered in each plot.We measured stem diameter was measured using a diameter tape, a graduated stick to measure tree height and a hypsometer for taller trees.In addition, we recorded tree density and calculated basal area (BA) for each plot.Elders living nearby to the study area identified the local names of recorded woody species.Then, the scientific name of each species was recorded using expert knowledge and published materials [36].

Destructive sampling and determination of total tree biomass
We stratified the vegetation inventory data into five diameter classes to understand tree size distribution of the study population (Supplementary information (SI), figure SI2).Then, proportional to size-class distribution, 21 sample trees were randomly identified, measured for DBH, DSH, H and harvested for aboveground biomass (AGB) estimation [12,37].We separated the harvested tree components into stem, branch, and leaf + twig.The total fresh biomass of each component was weighed on the site using a spring balance (±0.01 kg) [12].The dry matter content of the sampled trees was determined using representative sub-samples that were collected randomly from each component of the tree.We weighed, sealed with a plastic bag the sub-samples in the field before they were transported to Shire Soil Research Center (SSRC) for dry weight determination.
For belowground biomass (BGB) estimation, the root components of the harvested trees (n = 21) were excavated manually using hand tools and then carefully removed from the soil by hand [24,38,39].We collected coarse roots ( 2 mm) from the stump to the root tip.The fine roots (<2 mm) were also collected as much as possible [40].Soil attached to all collected root components was removed using a brush and water [21,38,39].We measured the fresh weight of the root components in the field using a spring balance (precision of ±0.01 kg) to take representative sub-samples, sealed in plastic bags, and transported to SSRC to determine the root dry mass.Finally, all sub-samples (stem, branch, twig+ leaf, and root) were oven-dried at a temperature of 105 °C until they reached constant weight [12,24,38].Digital sensitive balance with a precision of ± 0.01 g was used to weight the total ovendried samples and the fresh-to-oven-dry weight ratios were calculated.The dry weight ratios were used to convert the total fresh weights of sample trees measured in the field into total oven-dry weights [12].The carbon content in the aboveground (AGB) and below ground (BGB) biomass was estimated by multiplying each values by the default IPCC carbon fraction value of 0.47 [41].We summed AGB and BGB to calculate the total tree biomass (TB) and carbon accumulation.Finally, tree biomass predictive models for all components of the tree were developed using oven dry biomass and measured dendrometric parameters (DBH, DSH, H).

Species-specific allometric equation development
Tree biomass estimation models were developed using linear and non-linear regression equations based on either stem diameter (DBH), stump diameter (DSH) and total tree height (H) alone or in a combination of DBH and H; DSH and H, as an independent variables (table 1) [12,37,[42][43][44].It is important to note that, using more than one predictor at a time may introduce potential problems of collinearity, resulting in poor precision in the estimates of the corresponding regression coefficients [22].A multicollinearity test using Variance Inflation Factors (VIF = 1/(1−r 2 ), where VIF value exceeding 10 indicates a significant collinearity impact on the prediction of the parameters and thus, using the two parameters should be discouraged [12].Finally, separate predictive AGB, BGB and TB models were developed.Model performance was evaluated using the Dynamic Curve fitting function using the SigmaPlot statistical software.We developed the graphics using SigmaPlot statistical software.https://systatsoftware.com/sigmaplot-software/.[45].Thus, we further calculated Percent Relative Standard Error (PRSE) of the coefficients and Weighted Akaike information criterion (AICiw) were also calculated to check the reliability of model parameter estimates [22,46].Moreover, collinearity could influence a model with the largest R 2 or smallest RMSE and AIC, resulting in parameter estimates with large errors (PRSE > 25%).Thus, a model becomes unreliable when PRSE is greater than 25% for one or more parameters [22,46].A cross-validation test was conducted to validate and select the best fitting equation for each biomass component (AGB, BGB, and TB).In principle, an independent dataset should be used to validate the best fitting allometric equations; however, such data for A. saligna in the research locations were not available.Thus, an allometric model was cross-validated following a splitting sample approach [12,47].Following this approach, the collected tree samples were randomly divided into two data set, i.e. 16 samples for model calibration and 5 sample trees for model validation [12,47].We portioned the data set using the size class distribution of harvested sample trees.The goodness-of-fit statistics and the coefficients of the 'training' data set equations were used to compare with those derived using the full data set.Moreover, we compared the estimated and measured biomasses for the five 'test' trees.Finally, we used the full dataset to build the final biomass estimation models for each tree component (AGB, BGB, and TB).

Comparison of biomass estimation models
We compared the performance of our best models with previously published biomass estimation models (AGB, BGB, TB) developed based on DBH, DSH, and H from Ethiopia and elsewhere in the tropics.Unfortunately, we could not find allometric models developed for the study species in Ethiopia.Hence, we selected widely used allometric models in the tropical regions including Ethiopia for comparison (Supplementary table SI1).Finally, we used our best model to convert forest inventory data into tree biomass.

Tree biomass and carbon stock estimations
We converted vegetation inventory data to tree biomass carbon stock using our best models for aboveground, belowground, and total tree biomass of the study area.In addition, species-specific root-to-shoot ratio values were compared using modelled BGB with the estimated belowground biomass.The carbon content in the tree biomass was estimated by multiplying the values by the default IPCC carbon fraction value of 0.47 [41].The summation of the aboveground and belowground carbon stock was employed to calculate the biomass carbon stock of trees.Then, we converted the plot level carbon pools to hectare (Mg C ha −1 ).Furthermore, to determine the weight of carbon dioxide sequestered per hectare (Mg CO 2-equivalent ha −1 ), we multiplied the weight of biomass carbon (Mg C ha −1 ) by 3.67 [14].

Statistical analysis
Dendrometric parameters and measured carbon stock were correlated using Pearson correlation tests to identify which tree dendrometric variables were most strongly correlated with measured tree biomass and carbon stock at different components of trees (root, leaf + twig, stem, branch).The differences among tree aboveground and belowground components in carbon accumulation of the studied land use types were assessed using one-way analysis of variance and the significance of the differences was tested using the least significant difference test (LSD) with P < 0.05 [12].

Model development and cross-validation test
As a preliminary step to model calibration, the degree of collinearity among predictors (DBH, DSH, H) was analyzed and VIF values were found in the range between 1.5 and 9.0, indicating multicollinearity effects in the estimation of biomass.Thus, we avoided using a parameters which have showed collinearity from model development [46].Thus, for each biomass component (i.e., AGB, BGB, TB), nine model forms including DBH, DSH, and H were evaluated to develop a predictive model to estimate AGB, BGB, and TB of A. saligna from tree dendrometric properties (tables 3, 4).Based on model performance test results, AGBE1 and TBE1 (tables 3,4, SI3-5), and BGBE1 (tables 3, SI3-5) ranked as the best models using DBH (tables 3, 4), whereas the best performing model using DSH as predictive parameters was BGBE5 (table 4).Our best models passed all rigorous  verification and cross-validation statistical tests (SI, tables SI2-4) and the regression residuals for both AGB, BGB and TB were normally distributed (figure SI4).The best selected models, AGBE1, BGBE1, and TBE1 explained more than 82% (using DBH) and 75% (using DSH) of the variance in the biomass estimates (tables 3, 4).The RMSE from the cross-validation test statistics is close to the standard error of estimate; thus, our best models are robust to capture the variations in biomass estimation (tables 3, 4, SI3-5).Moreover, the parameter estimates for coefficients a and b for the selected models using the training dataset showed negligible differences from those parameter estimates using the full dataset (SI, tables SI3-5).In addition, the selected equations are reliable (tables 3, 4) as the standard errors of the coefficients (a), (b) of our best models were not inflated  (i.e., PRSE < 30%), with PBIAS < 5%.More importantly, the biomass of the cross-validation dataset estimated by model formed training dataset, showed negligible differences compared to estimated biomass applying models developed using the full dataset (figure 4).Thus, the selected model can estimate biomass from tree morphological characteristics of A. saligna trees

Comparison of aboveground biomass equations with previously published equations
We plotted the AGB, BGB, TB estimated with our models (AGBE1, BGBE5, TBE1) and with previously published equations (figure 5, table SI6) against the measured tree biomass components in figure 5. Our best models provided the lowest average relative error (PBIAS % = < 5%) compared to the error produced using previously published equations (table SI6).The magnitudes of error (PBIAS %) produced using other published models were considerable higher than produced using our models (table SI6).These models significantly overestimated the total tree biomass of A. saligna plants (table SI6).Generally, model comparison results showed that estimated biomass using site-and species-specific allometric models are more accurate than estimates using generalized models.

Tree density, estimated tree biomass and carbon stocks
In total, we inventoried 1040 trees and shrubs that belong to fifteen plant species across eighteen sample plots from both ExSASP and OGL plots.The dominant tree species was A. saligna, which comprised about 72%, and 21% of all individuals in ExSASP and ODL, respectively (figure 6).Stem density and estimated biomass varied between land uses (table SI8).Estimated total biomass (TB) per tree ranged from 1.5-16.3Kg/tree (mean = 4.8 ± 2.3, Kg/tree) and 0.9-47.5 Kg/tree (mean = 5.0 ± 3.6) in ExSASP and OGL, respectively (table SI7).Estimated biomass and C-stocks significantly varied between ExSASP and OGL (P < 0.01) (table SI8).Estimated tree biomass in exclosures and grazing land ranged from 3.6-18.0Mg ha −1 and 1.5-7.5 Mg ha −1 , respectively (table SI8).At a landscape level, estimated total tree biomass (AGB + BGB) varied between 1.5-18 Mg ha −1 , with the overall mean of 7.0 ±3.6 Mg ha −1 .The most important species in terms of C-stocks were A. saligna in exclosures (69%) and Euclea schimperi (34%) in the open grazing land.Moreover, A. saligna also contributes 22% of C-stock in grazing lands (figure 6).Estimated C-stock showed a significantly positive correlation with stem density (p < 0.001) (figure SI5).

Discussion
Dendrometric relationships, dry to fresh mass ratio and root to shoot ratio in A. saligna.The significant relationships between DBH, DSH and tree biomass indicate that stem or stump diameter are one of the main predictors of tree biomass.High correlations between these parameters were also reported in other studies that concluded DBH/DSH is the main determining factor for tree biomass accumulation [12,21,42,48].
The dry-to-fresh mass ratio was higher for stems (0.61 ±0.15) followed by branches (0.45 ±0.2), root (0.433 ±0.19) and twig + leaves (0.36 ±0.15).A weak relationship was found between RSR and tree size (figure SI2), indicating that RSR is independent of tree size [49].The mean RSR was 0.28, indicating that belowground biomass constitutes ∼28% of the total of A. saligna.Our RSR was slightly higher compared to other studies from Ethiopia and elsewhere.For instance, RSR of 20% was reported for Acacia abyssinica from northern Ethiopia [50], 22% for subtropical forests [44], 24% for tropical dry forests [51].On the other hand, our finding is comparable with the global average RSR of 30% for tropical/sub-tropical dry forests [51].However, higher RSR values were also reported for Juniperus procera (42%) and Acacia abyssinica (54%) from northern Ethiopia [52].Such variations indicate that the root-shoot ratio might vary depending on tree species, tree size, tree age, agroecology, soil conditions, and water stress level [18,23,24].Thus, it is important to be cautious when using RSR values to estimate the belowground biomass, particularly when there is no available species-specific RSR values for the study tree species [24].The RSR value generated from estimated BGB, and AGB was 25%, which is the same as RSR calculated from measured AGB and BGB, indicating a reliable estimate of belowground biomass using both site-and species-specific RSR and root biomass estimation models within tolerable error range.

Species-specific allometric models and their performance
We developed biomass estimation models for A. saligna which constitutes 72% of the tree population in the study area.The predictive performance of tested models using the full data set ranged from 11%-87% for AGB, 0.47%-0.84%for BGB and 0.21-0.87for TB (tables 3, 4).The variation in predictive power of the tested models might be due to differences in the type and number of variables used in the models their allometric differences [12,22].
Our best selected models explained more than 84% and 75% of the variance in the biomass estimates while using DBH and DSH, respectively (tables 3, 4).This shows that DBH could provide the best estimation of tree biomass and carbon accumulated in A. saligna trees in the study area.More importantly, the parameter estimates for the coefficients a, and b for best selected models (AGBE1, BGBE5, & TBE1) using the training data set showed negligible differences from those parameter estimates using the full dataset (tables SI3-5).The PRSE analysis further confirmed that parameter estimates were stable and reliable for selected models (PRSE < 30) (tables 4) [12,22].Moreover, estimated biomass of the cross-validation dataset using model developed by the training and full dataset showed negligible differences (figure 4).The deviations (PBIAS%) of estimated averages from observed average biomass were less than a percent (PBIAS < 5%) using the training data set and the full data set equations of our best models (table SI3-5), indicating that our best models are reliable and can be used to convert inventory data to tree biomass across the study area.
In summary, our findings further imply that using DBH or DSH alone as a predictor provides a reliable estimate of AGB, BGB, and TB of A. saligna and co-occurring tree species.In agreement with our results, several studies indicated that DBH or DSH is a good predictor of tree biomass for many tropical tree species [12,42,44,48,[53][54][55].Although the DBH/DSH alone models provided a reliable estimate of BGB, using DBH/ DSH with H slightly improved BGB biomass estimations (tables 3, 4).Similarly, studies have also recommended using tree heigh with DBH to increase model performance and reduce uncertainty [12,21,56].It is widely accepted that developing and using simple predictive models is important for several practical reasons, like interpretability in biologically meaningful ways [22].Therefore, our best models, AGBE1, BGBE5, and TBE1 could provide a reliable estimate of the aboveground, belowground, and total tree biomass of A. saligna and other co-occurring tree species.

Model comparison and importance of site-specific allometric equation
Estimated tree biomass (AGB, BGB) using our best models are close to measured values (table SI6).However, estimates using other published models (Listed in table SI1) showed large variability of predictive power (table SI6).Moreover, there was a significant (P < 0.001) difference between biomass estimated using our best models and that estimated by other models (table SI6).Our best model produced the lowest estimation error (PBIAS < 5%) compared to other published models, which showed PBIAS% ranging from −1328%-49% (table SI6).Except for the studies of Aneseyee et al [57] and Zhao et al [44], all compared allometric models over-predicted AGB, BGB and TB of the study tree species (table SI6).Our result further confirmed that site-and species-specific tree biomass equations are more robust and reliable to convert forest assessment data to tree biomass and C-stock [12,42,48,53,55,[57][58][59].This further highlight that increasing the number of species-and site specific allometric models could also contribute to the reliability of information on the national emission levels as well as for the development of generalized model for SSA region.

Biomass and biomass carbon accumulation potential of a. saligna dominated landscapes
The estimated total tree biomass and C-stocks of the study landscape may have an error, as we applied the same species-specific model to estimate the biomass of all other co-exiting trees species contributed about 28% of the population.The estimated biomass and C-stocks were linearly related to tree size and showed a significant correlation with DBH and DSH (P < 0.001).Estimated tree biomass showed that aboveground biomass constitutes about 78% of the estimated total biomass accumulation in ExSASP and OGL.On the other hand, belowground components of the tree grown in the study location stored ∼22% of the estimated tree biomass.Our result implies that reliable information on tree biomass accumulated in all components of the tree (AGB, BGB) is important for proper valuation and monitoring of restoration impacts in improving land productivity and reducing climate change impact through carbon sequestration.The total estimated AGB (C-stock) in ExSASP was significantly different (P< 0.001) and exceeded C-stocks accumulated in OGL by 60% (Table SI8).This indicates that exclosures supported by enrichment planting can contribute to a rapid rehabilitation of degraded landscapes and contribute to reducing climate change impacts in a reasonable period (figure 7).The estimated AGB in the study site is comparable with the biomass accumulation reported for exclosures and grazing lands from northern Ethiopia (e.g.[12,52].However, it was much lower than estimated biomass for exclosure and grazing lands in some parts of Tigray region [35].In general, our result indicates that exclosure sites supported through enrichment planting (A.saligna plantation) could significantly contribute to climate change mitigation through sequestration of atmospheric carbon dioxide in different tree components (figure 7).Hence, enrichment planting can be valuable for carbon credits, thereby improving the resilience of rural communities through earning incentives and support for environmental protection.Thus, is a need to promote A. saligna plantation should be promoted across degraded landscapes, to fight both climate change and land degradation impacts while improving soil fertility (figure 7).

Conclusions
This study presents the first site-and species-specific allometric equations for A. saligna species from restored landscapes in Tigray, Ethiopia where this finding could serve as basic information for other parts of Ethiopia and other countries with similar agroecologies and landscape conditions.Increasing the number of databases for specific models will increase the ability to precisely estimate biomass and carbon from different biomes.Our best carbon estimation models explained 86%, 84% and 87% of the variation measured in AGB, BGB, and TB, respectively.This shows that our species-specific models are robust and dependable to estimate biomass carbon accumulation in different components of the tree.Moreover, areas with similar site characters, tree species and tree-size distribution to the study area can use this model.Moreover, our biomass estimation model is relevant to monitor the temporal and spatial changes in tree biomass carbon across restoring landscapes and supported by A. saligna plantations.Estimated carbon stock in ExSASP indicates that A. saligna plantation across exclosures could play a significant role in restoring degraded landscapes and mitigating climate change through sequestering atmospheric CO 2 .Moreover, the reported tree biomass carbon values are important baseline information and might help to monitor carbon dynamics across exclosures supported by A. saligna plantation in Ethiopia and beyond.

Figure 1 .
Figure 1.Study area showing the Land Use Land Cover (LULC) of the study district and climatic characteristics with mean annual and monthly rainfall and Temperature (source from (CRU TS V4.03 -from 1901-2018).On figure 1, Eth_Region (represents Ethiopian regional boundary), AVgTm represents Average temperature, MMMT represents mean monthly maximum and MMMiT designates Minimum temperature, and MMR represents mean Monthly rainfall.

Figure 2 .
Figure 2. Below ground biomass of harvested trees as a function of aboveground biomass.

Figure 3 .
Figure 3. Above ground biomass (AGB), below ground biomass (BGB), and total biomass (TB) of harvested trees as a function of diameter at breast height (DBH) and diameter at stump height (DBH, DSH), and total tree height (H).

Figure 4 .
Figure 4. Relationships between estimated and measured total aboveground biomass of the cross-validations 'test' dataset.The solid rectangles (black shade) are the biomass estimates using the full dataset equation, the crosses are the estimates calculated using the training data set.The diagonal 1:1 line represents the perfect fit between the measured and estimated values.In figures, diameter at breast height (DBH), diameter at stump height (DSH), height (H), indicates parameter used to develop the model.AGB, BGB & TB refers to Above ground biomass, below ground biomass, and total biomass, respectively.

Figure 5 .
Figure 5. Graphic comparison of observed (black dash line) and predicted (smooth line) values using our best models.

Figure 6 .
Figure 6.Species frequency and contribution to total tree biomass in the study landscape.

Figure 7 .
Figure 7. Photographs depict Acacia saligna plantations established in degraded landscapes and their performance in resources poor landscape.Red arrows show A. saligna trees.

Table 2 .
Summary of tree dendrometric properties and oven dry biomass of harvested trees.

Table 3 .
Equations and goodness-of-fit performance statistics for estimating biomass (kg /plant, based on DBH, H) of A. saligna grown in exclosures and in open grazing land.Bold coefficient values are significant at P < 0.001.Bold PRSE values > 30 indicate unreliable parameter estimates.

Table 4 .
Equations and goodness-of-fit performance statistics for estimating biomass (kg /plant based on DSH, H) of A. saligna grown in exclosures and in open grazing land.Bold coefficient values are significant at P < 0.001.Bold PRSE values indicate > 30 unreliable parameter estimates.