Modelling tree canopy cover and evaluating the driving factors based on remotely sensed data and machine learning

Quantifying urban tree cover is important to ensure sustainable urban ecosystem. This study calculates urban percent tree cover (PTC) for Bursa city, Turkey from Sentinel-2 data and evaluates the driving factors of PTC using an Artificial Neural Network-Multi Layer Perception (ANN-MLP) approach. For the PTC calculation, a Regression Tree (RT) analysis was performed using several vegetation indices (NDVI, LAI, fCOVER, MSAVI2, and MCARI) to improve accuracy. Socio-economic, topographic, and biophysical variables were incorporated into the ANN-MLP approach to evaluate the factors that drive urban PTC. A PTC prediction map was generated with an accuracy of 0.95 and a coefficient of determination of 0.87. The ANN-MLP training process yielded a correlation coefficient value of 0.71 and an R-square of 0.82 was achieved between the predicted ANN-MLP and observed tree cover maps. A priority tree cover map was generated considering statistical relationships between the factors and the ANN-MLP prediction map in addition to visual interpretations at the urban scale. Results demonstrate that, unlike other urban forms, PTC has a statistically negative relationship with the gross dwelling density (R2 =0.31). Topographic variables including slope and DEM were positively correlated with PTC with the R2 value of 0.80 and 0.72 respectively. The integration of remote sensing data with vegetation indices and driving factors yielded accurate prediction for identifying and evaluating the variability in the urban PTC.     

The impact of pruning and mortality on urban tree canopy volume

Urban trees provide a wide range of ecosystem services for city residents, with tall, mature trees with wide crowns generally regarded as preferable. The tree biomass which is responsible for shading, pollution removal, rain runoff retention etc. gets periodically reduced by the municipal tree management practice of pruning. This is a necessary activity, which reduces the risk of infrastructure damage and falling branches, but many estimates of ecosystem service provision in cities do not consider its impact explicitly. Tree mortality is also higher in cities, preventing trees from attaining and remaining at large sizes. This study used extensive field measurements of tree structure to estimate the impact of pruning on 8 tree species in two Italian cities: Taranto and Florence. Crown widths were reduced by 1.6 m on average, however there is large variation between species variation with branches more often being removed for thinning crowns resulting in larger gap fractions, which increased by 15% on average. No significant differences were observed for crown widths or gap fraction between trees pruned 3 and 4 years previously, suggesting that tree crowns structurally recover from pruning after 3 years. A deterministic model revealed that current urban forest pruning rates (every 6 years) and mortality (1%) may create a situation in which a city dominated by the species studied benefits from 93.5% of the maximum ecosystem services possible. This work will allow more nuanced estimates of urban forest services to be calculated.     

Random point sampling to detect gain and loss in tree canopy cover in response to urban densification

Urban tree canopy cover (UTC) is a simple, and common, measure of urban forest resource. Urban infill development is likely to lead to losses in UTC under private tenure, at a time when local governments are setting ambitious targets to increase UTC overall. Simple, statistically rigorous methods are required to benchmark and track change in UTC, whilst identifying which land-use types or tenures experience change.We estimated UTC in six Melbourne suburbs in 2010 and 2015 by randomly sampling 2000 points across public land, public streetscapes and private land. We were able to detect a net change in UTC of <2% over five years to a 95% level of confidence. A significant net decrease in UTC (−2.4%) was only detected in one of the six suburbs. Two suburbs had a net increase in UTC by +2.7% over five years. On private land, there was often areas of UTC loss, but this was generally offset by canopy gain in other areas of the private realm as well as in streetscapes and public land. Losses in UTC on private land were mainly due to tree removal, with or without subsequent construction works.This study describes a simple, but statistically rigorous, method to quantify UTC change and the drivers of change in different land-use types and tenure. Despite studying two suburbs will high rates of infill development, only one suburb showed evidence of net UTC decrease. The ‘dynamic equilibrium’ in UTC, whereby canopy losses area approximately offset by concurrent canopy gain, means that ambitious targets being set by local governments to increase UTC may be difficult to achieve without changes in tree protection and infill development policy and planning.     

New morphological features for urban tree species identification using LiDAR point clouds

Urban tree species identification is the basis for studying the urban-environment coordination mechanism at the species level. Although the gradual maturity of remote sensing data and related research including light detection and ranging (LiDAR) provides a good foundation for the realization of this technology, multiple reasons such as cost, data openness, study scope limitations, and weakness of traditional morphological features make such data still challenging to apply to subtropical urban trees with heterogeneous canopy structures and high biodiversity. To address the problem, we developed two large-scale LiDAR morphological features in this research by, 1) modifying the rotate image method based on the axisymmetric structure to make it easier to use, and 2) developing an innovative adaptive ellipsoid method to extract the canopy features of the non-axisymmetric structure effectively. We evaluated the ability of these two morphological features to describe 12 common subtropical urban tree (SUT) species in Hong Kong growing in urban parks and streets, obtaining an accuracy of 88%. And the advantages of the proposed method are demonstrated by comparison with existing LiDAR morphological features and mean decrease accuracy (MDA) analysis. Our results illustrated that the rotate image feature based on the axisymmetric structure did not perform as well as the adaptive ellipsoid feature based on the non-axisymmetric structure in SUT, and the combined application of these two new morphological features got further accuracy improvement. The method proposed in this study had significant advantages in terms of accuracy, the number of species included, and generalisation capability compared to existing studies on the identification of subtropical urban trees.     

Exploring the characteristics of successful volunteer-led urban forest tree committees in Massachusetts

Citizen engagement through urban forest tree committee volunteer service may aid in providing essential experience, ideas, and skills that support municipal tree management. Using semi-structured, research interviews with tree committee (TC) representatives from across the Commonwealth of Massachusetts, this study addresses current knowledge gaps concerning the general composition, processes, and relationships of volunteer-led urban forest tree committees. Our findings indicate that TC representatives are typically motivated, passionate volunteers who generally desire to work cooperatively with the many associations, organizations, and agencies that comprise the local socio-political landscape. Our findings also indicate it is important that TC representatives make a sustained, concerted effort to work collaboratively with their local tree warden to advance the care of their community’s urban trees. Furthermore, it is also essential that municipal managers and decision-makers attempt to provide TC volunteers with appropriate training opportunities, resources, as well as demonstrate appreciation, to further encourage and solidify volunteer-engagement in urban forestry at the local level.     

Structural response of black locust (Robinia pseudoacacia L.) and small-leaved lime (Tilia cordata Mill.) to varying urban environments analyzed by terrestrial laser scanning: Implications for ecological functions and services

Quantitative measurements of structure and morphology of urban trees are hardly exhausted so far, especially in regard to variations caused by altering urban environments. However, structure and functions of trees are heavily interwoven. In fact, knowledge about structural attributes is essential for a better understanding of urban ecosystem functions and services. In order to scrutinize spatially explicit and detailed structural attributes under varying urban environments, we acquired terrestrial laser scans and applied the according methodological approaches to the common urban tree species black locust (Robinia pseudoacacia L.) and small-leaved lime (Tilia cordata Mill.). We analyzed 52 small-leaved limes and 41 black locust trees within the city of Munich (Germany). Species as well as growing location had a significant effect on the height-diameter relation. We also found greater crown volumes for small-leaved lime. Black locust however, displayed more crown projection area and likely more shade efficient crown shapes at similar volumes. Stem inclination of black locust was found to be higher in parks than in street canyons with town squares lying in between. Furthermore, black locust displayed strong crown asymmetry in park areas, likely caused by competition with neighbors. The angles of main branches did not differ significantly between both species nor between the growing location. Branch angles, branch bending, the length of the branches as well as species and growing location had a significant effect on vertical crown center position, i.e. general crown shape. Surface complexity of lime is lower than of black locust, with its lowest manifestation in parks. Fractal-like crown surface structures, increasing surface roughness and complexity, were found to be more pronounced for black locust than for small-leaved lime. Thereby, black locust featured the highest crown surface complexity in parks, the lowest in street canyons. The results suggest that studies on spatially explicit tree structures may contribute to more target oriented tree plantings and thus, more effective exploitation of ecosystem services and benefits.     

Estimating aboveground biomass of urban forest trees with dual-source UAV acquired point clouds

Urban forest is a crucial part of urban ecological environment. The accurate estimation of its tree aboveground biomass (AGB) is of significant value to evaluate urban ecological functions and estimate urban forest carbon storage. It has a high accuracy to estimate the forest AGB with field measured canopy structure parameters, but unsuitable for large-scale operations. Limited by low spatial resolution or spectral saturation, the estimated forest AGBs based on various satellite remotely sensed data have relatively low accuracies. In contrast, Unmanned Aerial Vehicle (UAV) remote sensing provides a promising way to accurately estimate the tree AGB of fragmented urban forest. In this study, taking an artificial urban forest in Ma'anxi Wetland Park in Chongqing City, China as an example, we used UAVs equipped with a digital camera and a LiDAR to acquire two point cloud data. One was produced from overlapping images using Structure from Motion (SfM) photogrammetry, and the other was resolved from laser scanned raw data. The dual point clouds were combined to extract individual tree height (H) and canopy radius (R_c), which were then input to the newly established allometric equation with tree H and R_c as predictor variables to obtain the AGBs of all dawn redwood trees in study area. In accuracy assessment, the coefficient of determination (R²) and Root Mean Square Error (RMSE) of extracted H were 0.9341 and 0.59 m; the R² and RMSE of extracted R_c were 0.9006 and 0.28 m; the R² and RMSE of estimated AGB were 0.9452 and 17.59 kg. These results proved the feasibility and effectiveness of applying dual-source UAV point cloud data and the new allometric equation on H and R_c to accurate AGB estimation of urban forest trees.     

Shade factors for 149 taxa of in-leaf urban trees in the USA

Shade factors, defined as the percentage of sky covered by foliage and branches within the perimeter of individual tree crowns, have been used to model the effects of trees on air pollutant uptake, building energy use and rainfall interception. For the past 30 years the primary source of shade factors was a database containing values from 47 species. In most cases, values were obtained from measurements on a single tree in one location. To expand this database 11,024 shade factors were obtained for 149 urban tree species through a photometric process applied to the predominant species in 17 U.S. cities. Two digital images were taken of each tree, crowns were isolated, silhouette area defined and shade factors calculated as the ratio of shaded (i.e., foliage and woody material) pixels to total pixels within the crown silhouette area. The highly nonlinear relationship between both age and diameter at breast height (DBH), and shade factor was captured using generalized additive mixed models.We found that shade factors increased with age until trees reached about 20 years or 30 cm DBH. Using a single shade factor from a mature tree for a young tree can overestimate actual crown density. Also, in many cases, shade factors were found to vary considerably for the same species growing in different climate zones. We provide a set of tables that contain the necessary values to compute shade factors from DBH or age with species and climate effects accounted for. This new information expands the scope of urban species with measured shade factors and allows researchers and urban foresters to more accurately predict their values across time and space.     

聊城市主要行道树国槐胸径与树高关系研究

采用随机抽样的方法,对聊城市主要行道树国槐的树高与胸径进行调查分析。结果表明:胸径与树高存在显著的相关性,由相关指数(R2=0.9910)可知H=a+bD+cD2多项式回归模型拟合最好,系统误差Se=0.5156%,总体精度极高,且各径阶系统误差均在±5%以内,在聊城地区可以用胸径估测平均树高,对于森林资源清查具有十分重要的意义。     

Modelling the spread of tree pests and pathogens in urban forests

Understanding the potential dynamics of tree pests and pathogens is a vital component for creating resilient urban treescapes. Epidemiologically relevant features include differences in environmental stress and tree management between street and garden trees, and variation in the potential for human-mediated spread due to intensity of human activity, traffic flow and buildings. We extend a standard spatially explicit raster-based model for pest and pathogen spread by dividing the urban tree population into roadside trees and park/garden trees. We also distinguish between naturally-driven radial spread of pests and pathogens and human-mediated linear spread along roads. The model behaviour is explored using landscape data for tree locations in an exemplar UK town. Two main sources of landscape data were available: commercially collated aerial data, which have high coverage but no information on species; and, an urban tree inventory, with low, non-random, coverage but with some species data. The data were insufficient to impute a species-specific host landscape accurately; however, by combining the two data sources, and applying either random or Matérn cluster point process driven selection of a subset of all trees, we create two sets of potential host landscapes. We find that combining the two mechanisms of dispersal has a non-additive effect, with the enhanced linear dispersal enabling new foci of infection to be established more rapidly than with radial dispersal alone; and clustering of trees by species slows down the expansion of epidemics when compared with random distribution of tree species within known host locations.     

The overlooked carbon loss due to decayed wood in urban trees

Decayed wood is a common issue in urban trees that deteriorates tree vitality over time, yet its effect on biomass yield therefore stored carbon has been overlooked. We mapped the occurrence and calculated the extent of decayed wood in standing Ulmus procera, Platanus × acerifolia and Corymbia maculata trees. The main stem of 43 trees was measured every metre from the ground to the top by two skilled arborists. All trees were micro-drilled in two to four axes at three points along the stem (0.3 m, 1.3 m, 2.3 m), and at the tree’s live crown. A total of 300 drilling profiles were assessed for decay. Simple linear regression analysis tested the correlation of decayed wood (cm²) against a vitality index and stem DBH. Decay was more frequent and extensive in U. procera, than P. acerifolia and least in C. maculata. Decay was found to be distributed in three different ways in the three different genera. For U. procera, decay did appear to be distributed as a column from the base to the live crown; whereas, decay was distributed as a cone-shape in P. acerifolia and was less likely to be located beyond 2.3 m. In C. maculata decay was distributed as pockets of variable shape and size. The vitality index showed a weak but not significant correlation with the proportion of decayed wood for P. acerifolia and C. maculata but not for U. procera. However, in U. procera, a strong and significant relationship was found between DBH and stem volume loss (R² = 0.8006, P = 0.0046, n = 15). The actual volume loss ranged from 0.17 to 0.75 m³, equivalent to 5%–25% of the stem volume. The carbon loss due to decayed wood for all species ranged between 69–110 kg per tree. Based on model’s calculation, the stem volume of U. procera trees with DBH ≥ 40 cm needs to be discounted by a factor of 13% due to decayed wood regardless of the vitality index. Decayed wood reduces significantly the tree’s standing volume and needs to be considered to better assess the carbon storage potential of urban forests.     

Redeveloping the urban forest: The effect of redevelopment and property-scale variables on tree removal and retention

The effects of urbanization on urban forest canopy cover has received significant consideration at broad scales, but little research has explored redevelopment-related influences on individual tree removal at a property scale. This study explores the effect of residential property redevelopment on individual trees in Christchurch, New Zealand. The study monitored 6966 trees on 450 residential properties between 2011 and 2015/16. Of the 450 properties, 321 underwent complete redevelopment during that time, while 129 were not redeveloped. The percentage of trees removed on redeveloped and non-redeveloped properties differed markedly, being 44% and 13.5%, respectively. A classification tree (CT) analysis was used to examine the effects of different combinations of 27 explanatory variables, describing land cover, spatial relationships, economic, and resident and household variables, on tree removal or retention on the properties. The best model included land cover, spatial, and economic variables (accuracy = 73.4%). The CT of the corresponding model shows that trees were most likely to be removed if they were within 1.4 m of a redeveloped building on a property with a capital value less than $1,060,000 NZ. The strongest predictor of tree retention was that the property was not redeveloped. The model predicted that trees were over three times as likely to be removed from a redeveloped property relative to a property that was not redeveloped. None of the seven resident and household variables were selected by the CT as important explanatory variables for tree removal or retention. These results provide insights into the factors that influence tree removal during redevelopment on residential properties, and highlight the need for effective tree protection during redevelopment.     

Survival is not enough: The effects of microclimate on the growth and health of three common urban tree species in San Francisco,California

Urban forest managers must balance social, economic, and ecological goals through tree species selection and planting location. Ornamental trees are often popular in tree planting programs for their aesthetic benefits, but studies find that they have lower survivability and growth compared to larger shade trees. To maximize ecosystem services within these aesthetic preferences, it is important to select species carefully based on their ability to grow in each particular climate. However, little locality-specific and species-specific data exist on urban trees in many regions. This study examines the growth, survival, and vigor of three common ornamental street trees in San Francisco’s three different microclimate zones after over 16 years since planting. While we found over 70% survival for all three species throughout the city, there were significant differences in health and vigor among microclimates for each species, likely due to differences in drought-tolerance. While Arbutus had the greatest proportion of healthy trees in the Fog Belt and Sun Belt zones, Prunus cerasifera had the greatest proportion in the Sun Belt, and Prunus serrulata had the greatest proportions in the Transition and the Sun Belt zones. This species-specific and climate-specific information will better equip urban foresters to target both planting and tree-care of these popular species appropriately to maximize the benefits provided by these street trees while still maintaining a diverse canopy. Finally, we argue that simple survival calculations can mask more complex differences in the health and ability of different urban tree species to provide ecosystem services.     

Estimation of urban tree canopy cover using random point sampling and remote sensing methods

Trees play an important role in urban areas by improving air quality, mitigating urban heat islands, reducing stormwater runoff and providing biodiversity habitat. Accurate and up-to-date estimation of urban tree canopy cover (UTC) is a basic need for the management of green spaces in cities, providing a metric from which variation can be understood, change monitored and areas prioritised. Random point sampling methods, such as i-Tree canopy, provide a cheap and quick estimation of UTC for a large area. Remote sensing methods using airborne Light Detection And Ranging (LiDAR) and multi-spectral images produce accurate UTC maps, although greater processing time and technical skills are required. In this paper, random point sampling and remote sensing methods are used to estimate UTC in Williamstown, a suburb of Melbourne, Australia. High resolution multi-spectral satellite images fused with LiDAR data with pixel-level accuracy are employed to produce the UTC map. The UTC is also estimated by categorising random points (a) automatically using the LiDAR derived UTC map and (b) manually using Google Maps and i-Tree canopy software. There was a minimum 1% difference between UTC estimated from the map derived from remotely sensed data and only 1000 random points automatically categorised by that same map, indicating the level of error associated with a random sampling approach. The difference between UTC estimated by remote sensing and manually categorised random point sampling varied in range of 4.5% using a confidence level of 95%. As monitoring of urban forest canopy becomes an increasing priority, the uncertainties associated with different UTC estimates should be considered when tracking change or comparing different areas using different methods.     

Optimization of tree locations to reduce human heat stress in an urban park

Trees provide cooling benefits through shading and evapotranspiration; they are regarded as an important measure in heat-resilient urban planning and policies. Knowing where to plant trees for maximum cooling benefits, given practical and resource constraints, remains a challenge in both practice and research. Literature in the field of tree modeling and location optimization is limited, either by the incompleteness in accounting for tree shading, evapotranspiration, and the modifying effect of wind, or by the slow-running speed of the Computational Fluid Dynamics model, making them less applicable in practice. This paper describes a novel method to search for the optimal locations for trees to maximize their cooling benefits in an urban environment. A rapid simulation model was applied to assess on-site heat stress under the influences of trees, which was evaluated using field measurements conducted under hot, temperate, and cool weather conditions in an urban park in Hong Kong. It was then linked to a genetic algorithm in search of a near-optimal tree layout. The proposed method was tested in the same park, and it can automatically identify locations to plant new trees to minimize heat stress, subject to practical constraints such as avoiding existing buildings and utilities. It can also identify the optimal locations to rearrange the existing 55 trees, hypothetically, which can cool the park by up to 0.3 ℃ in on-site average equivalent temperature compared with the worse scenario. Trees can cool the most if they are concentrated on the leeward side of the park, rather than spread evenly. The proposed method runs significantly faster than existing approaches, and it can inform research and landscape design practices concerning park cooling as a goal.     

The influence of sidewalk replacement on urban street tree growth

Interactions between tree roots and sidewalks can result in damage to sidewalks and when sidewalk damage is repaired adjacent tree roots are often severed. The objective of this study was to quantify the growth response of urban trees in restricted planting spaces pre- and post-sidewalk construction. The research included four trees species commonly planted along streets in Minneapolis and Saint Paul, Minnesota, USA. Species included were: Acer platanoides, Celtis occidentalis, Gleditsia triacanthos, and Tilia spp. Two street tree populations were sampled: trees adjacent to replaced sidewalk panels (<1.75 m) and trees on streets with sidewalk construction that were greater than 3 m from replaced sidewalk sections. In total, increment core samples from 292 trees were analyzed. Annual rings from each tree were measured and converted to basal area increment (BAI) for analysis. Comparisons of BAI were conducted between the two sample populations to assess differences in tree growth patterns. Pre- and post-sidewalk construction BAI was also evaluated to determine the influence of construction on growth trajectory. Growth response was quantified using resistance, resilience, and recovery indices. Species were found to differ in their response to construction disturbance. Planting space width was also found to influence post-construction growth. Tilia spp. had the highest resilience and fastest overall growth recovery post-sidewalk construction and A. platanoides exhibited the lowest resistance, resilience, and recovery post-sidewalk construction.     

Estimating aboveground carbon stocks of urban trees by synergizing ICESat-2 LiDAR with GF-2 data

Accurately mapping carbon stocks of urban trees is necessary for urban managers to design strategies to mitigate climate change. However, the aboveground carbon stocks of urban trees are usually underestimated by passive remote sensing data because of the signal saturation problem. The research is the first attempt to develop a framework to map aboveground carbon density of trees in urban areas by synergizing Ice, Cloud and Land Elevation Satellite-2 (ICESat-2) LiDAR data with Gaofen-2 (GF-2) imagery. The framework consists of three key steps. First, we used a support vector machine classifier to classify GF-2 images and extracted urban tree regions. Second, we estimated the tree carbon density of ICESat-2 strips by developing a ICESat-2 photon feature-based aboveground carbon density estimation model. Third, we mapped the carbon density of urban trees by developing a synergistic model between ICESat-2 and GF-2 data based on an object-oriented method. We tested the approach for the areas within the fifth ring road of Beijing, China. The results showed that the 50th percentile height (PH50) of nighttime photons was a good predictor for estimating carbon density of urban trees, with a R² of 0.69 and a Root Mean Square Error (RMSE) of 2.81 kg C m⁻². Using the spectral features generated by GF-2 imagery, we could further extrapolate the carbon density estimated by ICESat-2 strip data to a full coverage of accurate mapping carbon density by urban trees, resulting in a R² of 0.64 and a RMSE of 2.32 kg C m⁻². The carbon stocks within the fifth ring road of Beijing were 8.28 × 10⁸ kg in total, with the mean carbon density of 3.52 kg C m⁻². Such estimations were larger than that of previous study using passive remote sensing data only, suggesting the integration of spaceborne LiDAR and spectral data could greatly reduce the underestimation of carbon stocks of urban trees. Our approach can more accurately estimate carbon stocks of urban trees and has the potential to be applicable in other cities.     

Perception and preference of trees: A psychological contribution to tree species selection in urban areas

Trees can enhance human mental and physical well-being in urban environments. However, the tree benefits in urban planning are insufficiently recognised, and there is little knowledge on the tree characteristics that are relevant to humans and how they are evaluated. This paper presents perceptual tree parameters and their relation to human preferences. In study 1, participants sorted 24 tree images by perceived similarity. Hierarchical cluster analysis and multidimensional scaling (MDS) revealed the distinction between conifers and deciduous trees, crown shape, the two-dimensional crown size to trunk height ratio and the crown density as important to humans. In study 2, participants rated the trees based on their preferences. Multiple linear regression analyses showed that a high two-dimensional crown size to trunk height ratio and a high crown density predicted deciduous tree preferences. These findings are discussed in light of the savannah hypothesis and the Gestalt grouping principle of closure. In the task of tree selection and placement for urban areas, the identified perceptual tree parameters may allow for achieving a coherent overall picture with a simultaneous increase of tree species richness. Thus, urban landscape planning can apply the presented findings for increasing ecosystem health and residential satisfaction.     

Assessing the species diversity and vulnerability of urban tree populations in the London borough of Westminster

The level of tree species diversity in urban tree populations can have serious implications for urban forest resilience and has a direct impact on ecosystem functioning at the local level. Few studies have measured the relationship between tree species diversity and vulnerability in UK urban forests. This study analysed the species composition, species diversity and the vulnerability to pests and diseases of 10,149 public trees in the London borough of Westminster across three land use types: housing, parks, and highways. Tree species diversity was significantly different across these land use types (Shannon’s diversity index (H) was 2.47 for housing sites, 1.63 for parks and 0.83 for highways) and we found that higher diversity appears to result in reduced vulnerability, evidenced by negative correlations between tree species diversity and susceptibility to pests and diseases. A stronger negative correlation was found between vulnerability and species richness than between vulnerability and Gini-Simpson’s diversity index. Our study reinforces the role of biodiversity indices in establishing and monitoring baseline levels of UK urban tree diversity. Our findings may inform future tree planting projects, help to ensure that development does not negatively affect urban tree diversity and inform proactive strategies for urban trees to contribute to wider biodiversity conservation.     

The image recognition of urban greening tree species based on deep learning and CAMP-MKNet model

The information of urban tree species resources is of vital significance to the planning and design of urban green spaces. Tree organs, such as the bark are used as the primary features of identifying tree species. However, traditional tree identification methods need to consume a lot of manpower and time costs. In addition, the application of machine image recognition technology to tree species recognition still has problems such as heavy data preprocessing workload, small number of tree species images, uneven distribution of categories, and low recognition accuracy. In order to promote the intelligent management of urban forestry and solve the above problems, it is necessary to establish an automatic image recognition model for urban greening tree species. We captured bark images of 21 urban afforestation tree species in their natural environment and constructed a dataset that was divided into a train set, validation set, and test set in the ratio of 7:1:2. Combining Channel Attention Module (CAM) with algorithms such as Spatial Pyramid Pooling (SPP) and Mixed Depthwise Dilated Convolutional Kernels. The core algorithm is Mixed Convolution Kernel (MK), and a CAMP-MKNet Convolutional Neural Network (CNN) is constructed as a bark image classification model for urban greening tree species. The overall accuracy of the generic models ranged from 41.06% to 82.03%, whereas the overall accuracy of the experimental CAMP-MKNet model was 84.25%, with lower prediction cost. Our study shows that the CAMP-MKNet CNN model with better prediction performance and computational cost and can provide crucial insights and technical support for developing automated urban tree species image recognition systems.