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.     

Multi-feature machine learning model for automatic segmentation of green fractional vegetation cover for high-throughput field phenotyping

Background

Accurately segmenting vegetation from the background within digital images is both a fundamental and a challenging task in phenotyping. The performance of traditional methods is satisfactory in homogeneous environments, however, performance decreases when applied to images acquired in dynamic field environments.

Results

In this paper, a multi-feature learning method is proposed to quantify vegetation growth in outdoor field conditions. The introduced technique is compared with the state-of the-art and other learning methods on digital images. All methods are compared and evaluated with different environmental conditions and the following criteria: (1) comparison with ground-truth images, (2) variation along a day with changes in ambient illumination, (3) comparison with manual measurements and (4) an estimation of performance along the full life cycle of a wheat canopy.

Conclusion

The method described is capable of coping with the environmental challenges faced in field conditions, with high levels of adaptiveness and without the need for adjusting a threshold for each digital image. The proposed method is also an ideal candidate to process a time series of phenotypic information throughout the crop growth acquired in the field. Moreover, the introduced method has an advantage that it is not limited to growth measurements only but can be applied on other applications such as identifying weeds, diseases, stress, etc.

     

Modeling and mapping forest diversity in the boreal forest of interior Alaska

Context

Patterns of forest diversity are less well known in the boreal forest of interior Alaska than in most ecosystems of North America. Proactive forest planning requires spatially accurate information about forest diversity. Modeling is a cost-efficient way of predicting key forest diversity measures as a function of human and environmental factors.

Objectives

Investigate and predict the patterns and processes in tree species and tree size-class diversity within the boreal forest of Alaska for a first mapped quantitative baseline.

Methods

For the boreal forest of Alaska, USA, we employed Random Forest Analysis (machine learning) and the Boruta algorithm in R to predict tree species and tree size-class diversity for the entire region using a combination of forest inventory data and a suite of 30 predictors from public open-access data archives that included climatic, distance, and topographic variables. We developed prediction maps in a GIS for the current levels (Year 2012) of tree size-class and species diversity.

Results

The method employed here yielded good accuracy for the huge Alaskan landscape despite the exclusion of spectral reflectance data. It’s the first quantified GIS prediction baseline. The results indicate that the geographic pattern of tree species diversity differs from the pattern of tree size-class diversity across this forest type.

Conclusions

The results suggest that human factors combined with topographical factors had a large impact on predicting the patterns of diversity in the boreal forest of interior Alaska.

     

Using machine learning to identify urban forest crown bounding boxes (CBB): Exploring a new method to develop urban forest policy

Collecting and managing individual tree data is a critical activity for green sustainability strategies. Local governments are able to easily collect detailed public tree inventories, however data on trees located on private land are much more challenging and costly to collect. This means that new regulations to limit the removal of trees on private land go untested prior to their implementation, or fail to pass regulatory review processes. Without knowledge of the location of trees or the range of their different sizes, Local Government Authorities (LGAs) are unable to predict where a new policy to prohibit the removal of trees of a certain size is likely to have the greatest effect, where enforcement should be concentrated, or to convince government, the development sector and local communities of the need for action to preserve trees.The aim of this study was to explore the potential of a supervised machine learning algorithm as a cost-efficient method to understand tree sizes and locations on private land and to discuss how this information could be used for sustainable urban greening. We conclude by discussing some of the affordances of this approach to better target native vegetation protection and protect large trees; and report on the precision and recall of the detection of the urban forest.     

Quantification of carbon sequestration by urban forest using Landsat 8 OLI and machine learning algorithms in Jodhpur,India

Urban forests play a significant role in carbon cycling. Quantification of Aboveground Biomass (AGB) is critical to understand the role of urban forests in carbon sequestration. In the present study, Machine learning (ML) based regression algorithms (SVM, RF, kNN and XGBoost) have been taken into account for spatial mapping of AGB and carbon for the urban forests of Jodhpur city, Rajasthan, India, with the aid of field-based data and their correlations with spectra and textural variables derived from Landsat 8 OLI data. A total of 198 variables were retrieved from the satellite image, including bands, Vegetation Indices (VIs), linearly transformed variables, and Grey Level Co-occurrence textures (GLCM) taken as independent input variables further reduced to 29 variables using Boruta feature selection method. All the models have been compared where with RF algorithm, R² = 0.83, RMSE = 16.22 t/ha and MAE = 11.86 t/ha. For kNN algorithm R² = 0.77, RMSE = 28.04 t/ha and MAE = 24.24 t/ha and SVM where R² = 0.73, RMSE = 89.21 t/ha and MAE = 74.22 t/ha and the best prediction accuracy has been noted with XGBoost algorithm (R² = 0.89, RMSE = 14.08 t/ha and MAE = 13.66 t/ha) with predicted AGB as 0.51−153.76 t/ha. The study indicates that ML-based regression algorithms have great potential over other linear and multiple regression techniques for spatial mapping of AGB and carbon of urban forests for arid regions.     

The condition for the healing of dural defects in vitro and the influential factors

AIM:To study the condition for the healing process of dural defects in vitro and its influential factors. METHODS:Rabbit dura mater was cultured in 24-well plates that had been coated with collagen, laminin and polylysine to observe the influence of extracellular matrix on it. Dural cells were subcultured on slides for immunocytochemistry. Dural healing was observed by scanning electronic microscope. Different growth factors were added to the culture medium respectively to analyze the influential factors on it. RESULTS:Only the dura pieces cultured on collagen coated wells showed migration of cells into the central defect. Dural cells stained strongly positive with antibodies against vimentin and negative with Ⅷ factor. The dural pieces with bFGF added showed earlier migration of dural cells from defect margin after a period of 3-4 days and the dural defect healing occurred after 7-8 days. CONCLUSION:Collagen is an essential matrix for the dural healing. Cellular migration from the defect margin is an important echanism in the process of dural healing and bFGF accelerates dural healing.     

Biophysical and socioeconomic factors explaining the extent of forest cover on private ownerships in a Midwestern (USA) agrarian landscape

As landscape fragmentation continues to escalate, it is imperative that we improve our understanding of the factors that contribute to the creation and retention of forest on privately-owned land to most effectively design and implement conservation policy. This article presents the percentages of variation in the proportion of forest on private ownerships across an agriculturally-dominated landscape in north-central Indiana, USA that can be explained by biophysical characteristics, landowner (socioeconomic) attributes, and private landowner assistance programs. While biophysical characteristics of the land accounted for the majority of variation explained (17.35%, p < 0.0001, n = 194), attitudinal and demographic attributes of the landowners contributed significantly to explaining additional variation (7.97%, p<0.0001), and overlapped with biophysical characteristics to explain another 17.31%. Program familiarity and enrollment did not explain a significant amount of the variation independent of either biophysical or landowner attributes. Private landowner assistance programs should broaden their objectives and increase incentives to appeal to the variety of landowners who possess the decision-making authority for most of the land in the region and the nation as a whole.     

An investigation of uncertainty in field habitat mapping and the implications for detecting land cover change

Land cover data for landscape ecological studies are frequently obtained by field survey. In the United Kingdom, temporally separated field surveys have been used to identify the locations and magnitudes of recent changes in land cover. However, such map data contain errors which may seriously hinder the identification of land cover change and the extent and locations of rare landscape features. This paper investigates the extent of the differences between two sets of maps derived from field surveys within the Northumberland National Park in 1991 and 1992. The method used in each survey was the Phase 1 approach of the Nature Conservancy Council of Great Britain. Differences between maps were greatest for the land cover types with the smallest areas. Overall spatial correspondence between maps was found to be only 44.4%. A maximum of 14.4% of the total area surveyed was found to have undergone genuine land cover change. The remaining discrepancies, equivalent to 41.2% of the total survey area, were attributed primarily to differences of land cover interpretation between surveyors (classification error). Differences in boundary locations (positional error) were also noted, but were found to be a relatively minor source of error. The implications for the detection of land cover change and habitat mapping are discussed.     

Simulating forest cover change in the northeastern U.S.: decreasing forest area and increasing fragmentation

Landscape Ecology - Understanding how the Northern Forest landscape has changed and is likely to change, both in terms of forest extent and forest configuration, has important implications for...     

Landscape forest cover and edge effects on songbird nest predation vary by nest predator

Rates of nest predation for birds vary between and within species across multiple spatial scales, but we have a poor understanding of which predators drive such patterns. We video-monitored nests and identified predators at 120 nests of the Acadian Flycatcher (Empidonax virescens) and the Indigo Bunting (Passerina cyanea) at eight study sites in Missouri and Illinois, USA, during 2007–2010. We used an information-theoretic approach to evaluate hypotheses concerning factors affecting predator-specific and overall rates of predation at landscape, edge, and nest-site scales. We found support for effects of landscape forest cover and distance to habitat edge. Predation by Brown-headed Cowbirds (Molothrus ater) increased, and predation by rodents decreased as landscape forest cover decreased. Predation by raptors, rodents, and snakes increased as the distance to forest edges decreased, but the effect was modest and conditional upon the top-ranked model. Despite the predator-specific patterns we detected, there was no support for these effects on overall rates of predation. The interactions between breeding birds, nest predators, and the landscapes in which they reside are scale-dependent and context-specific, and may be resistant to broad conceptual management recommendations.     

Forest proximity rather than local forest cover affects bee diversity and coffee pollination services

Context

As agricultural demands for land continues to expand, strategies are urgently needed to balance agricultural production with biodiversity conservation and ecosystem service provision in agricultural landscapes.

Objectives

We used a factorial landscape design to assess the relative contributions of forest proximity and local forest cover to bee diversity and the provision of coffee pollination services.

Methods

We quantified bee diversity and fruit set in 24 sun-grown coffee fields in Southeast Region of Brazil that were selected following a factorial sampling design to test the independent effects of local forest cover (in a radius of 400 m) and proximity to forest fragments. To assess the impact of landscape simplification, we also evaluated local coffee cover.

Results

Bee richness and abundance were higher in the proximity of forest fragments, but only bee abundance decreased when the coffee cover dominated the surrounding landscapes. Coffee fruit set was 16% higher overall with bee visitations compared with bee exclusion and increased to 20% when coffee bushes were near forest fragments, and the coffee cover was low. Surprisingly, local forest cover did not affect the bee community or coffee fruit set.

Conclusion

Our results provide clear evidence that the proximity of coffee crops to forest fragments can affect the abundance and richness of bees visiting the coffee flowers and thereby facilitate the provision of pollination services. The positive association between forest proximity and fruit set reinforces the importance of natural vegetation in enhancing bee diversity and, therefore, in the provision of pollination services. The negative effect of coffee cover on fruit set at the local scale suggests that the service demand can surpass the capacity of pollinators to provide it. These effects were independent of the local forest cover, although all studied landscapes had more than 20% remaining forest cover (within a 2 km radius), which is considered the extinction threshold for Atlantic Forest species. Interspersion of forest fragments and coffee plantations in regions with more than 20% of forest cover left could thus be a useful landscape management target for facilitating pollinator flows to coffee crops and thus for increasing coffee yields.

     

Recent forest cover type transitions and landscape structural changes in northeast Minnesota,USA

Landsat TM satellite data covering an approximate 5-year interval (1990–1995) were used to quantify spatial pattern and transition rates between forest ecological states for a 2.76 million ha region in northeast Minnesota. Changes in forest cover were stratified by Ecological Subsection, management status, and by ownership categories using a 1995 digital ownership layer. Approximately 4.2% of the 1990 mature forested area was converted to early successional types by 1995. Of this 4.2%, private lands accounted for 33%, federal lands 31%, county lands 20% and state lands 16%. Notable conversion percentages by cover type category were spruce-fir (−5.3%), aspen-birch(−4.7%), jack pine (−4.6%) and black spruce(−3.0%). Transition rates were also adjusted to fit ten-year time intervals. Shannon-Weaver Eveness and edge density of cover types increased over the study period as relative contagion and interior forest area decreased. These trends suggest both smaller patches and a more even distribution of cover types. Area of upland conifers, lowland conifers and lowland hardwoods decreased while the area of mature upland hardwoods increased in most patch size classes except the > 500 ha class which showed a substantial decrease in area. The area of early successional types increased in most patch size classes. Non-industrial private forestland had the lowest proportion of interior forest of all ownership categories -decreasing by 13.5% in five years. Smaller average cut-unit size sand uncoordinated forest management is the likely cause since cutting rates between private and public forestland were similar. This revised version was published online in July 2006 with corrections to the Cover Date.     

Post-disturbance recovery of forest cover and tree height differ with management in Central Europe

Context

Recovery from disturbances is a prominent measure of forest ecosystem resilience, with swift recovery indicating resilient systems. The forest ecosystems of Central Europe have recently been affected by unprecedented levels of natural disturbance, yet our understanding of their ability to recover from disturbances is still limited.

Objectives

We here integrated satellite and airborne Lidar data to (i) quantify multi-decadal post-disturbance recovery of two indicators of forest structure, and (ii) compare the recovery trajectories of forest structure among managed and un-managed forests.

Methods

We developed satellite-based models predicting Lidar-derived estimates of tree cover and stand height at 30 m grain across a 3100 km² landscape in the Bohemian Forest Ecosystem (Central Europe). We summarized the percentage of disturbed area that recovered to >?40% tree cover and >?5 m stand height and quantified the variability in both indicators over a 30-year period. The analyses were stratified by three management regimes (managed, protected, strictly protected) and two forest types (beech-dominated, spruce-dominated).

Results

We found that on average 84% of the disturbed area met our recovery threshold 30 years post-disturbance. The rate of recovery was slower in un-managed compared to managed forests. Variability in tree cover was more persistent over time in un-managed forests, while managed forests strongly converged after a few decades post-disturbance.

Conclusion

We conclude that current management facilitates the recovery of forest structure in Central European forest ecosystems. However, our results underline that forests recovered well from disturbances also in the absence of human intervention. Our analysis highlights the high resilience of Central European forest ecosystems to recent disturbances.

     

Occupancy pattern of a long-horned beetle in a variegated forest landscape: linkages between tree quality and forest cover across spatial scales

Context

Interactions between landscape-scale processes and fine-grained habitat heterogeneity are usually invoked to explain species occupancy in fragmented landscapes. In variegated landscapes, however, organisms face continuous variation in micro-habitat features, which makes necessary to consider ecologically meaningful estimates of habitat quality at different spatial scales.

Objectives

We evaluated the spatial scales at which forest cover and tree quality make the greatest contribution to the occupancy of the long-horned beetle Microplophorus magellanicus (Coleoptera: Cerambycidae) in a variegated forest landscape.

Methods

We used averaged data of tree quality (as derived from remote sensing estimates of the decay stage of single trees) and spatially independent pheromone-baited traps to model the occurrence probability as a function of multiple cross-scale combinations between forest cover and tree quality (with scales ranging between 50 and 400 m).

Results

Model support and performance increased monotonically with the increasing scale at which tree quality was measured. Forest cover was not significant, and did not exhibit scale-specific effects on the occurrence probability of M. magellanicus. The interactive effect between tree quality and forest cover was stronger than the independent (additive) effects of tree quality and particularly forest cover. Significant interactions included tree quality measured at spatial scales ≥200 m, but cross-scale interactions occurred only in four of the seven best-supported models.

Conclusions

M. magellanicus respond to the high-quality trees available in the landscape rather than to the amount of forest per se. Conservation of viable metapopulations of M. magellanicus should consider the quality of trees at spatial scales >200 m.

     

Airborne multispectral imagery and deep learning for biosecurity surveillance of invasive forest pests in urban landscapes

Urban and peri-urban trees in major cities provide a gateway for exotic pests and diseases (hereafter “pests”) to establish and spread into new countries. Consequently, they can be used as sentinels for early detection of exotic pests that could threaten commercial, environmental and amenity forests. Biosecurity surveillance for exotic forest pests relies on monitoring of host trees — or sentinel trees — around high-risk sites, such as airports and seaports. There are few publicly available spatial databases of urban street and park trees, so locating and mapping host trees is conducted via ground surveys. This is time-consuming and resource-intensive, and generally does not provide complete coverage. Advances in remote sensing technologies and machine learning provide an opportunity for semi-automation of tree species mapping to assist in biosecurity surveillance. In this study, we obtained high resolution (≥12 cm), 10-band, multispectral imagery using the ArborCam™ system mounted to a fixed-wing aircraft over Sydney, Australia. We mapped 630 Pinus trees and 439 Platanus trees on-foot, validating their exact location on the airborne imagery using an in-field mapping app. Using a machine learning, convolutional neural network workflow, we were able to classify the two target genera with a high level of accuracy in a complex urban landscape. Overall accuracy was 92.1% for Pinus and 95.2% for Platanus, precision (user’s accuracy) ranged from 61.3% to 77.6%, sensitivity (producer’s accuracy) ranged from 92.7% to 95.2%, and F1-score ranged from 74.6% to 84.4%. Our study validates the potential for using multispectral imagery and machine learning to increase efficiencies in tree biosecurity surveillance. We encourage biosecurity agencies to consider greater use of this technology.     

华冠苹果的芳香物质特征及其影响因素

为探明华冠苹果产生异香的原因,采用气相色谱法(GC)对不同采收时期和贮藏方式下的华冠果实在贮藏期间果实中芳香物质含量变化进行了分析。结果表明,华冠果实中醇类物质明显比同期的金冠、富士含量高,随贮藏时间的延长,果实内醇类含量逐步提高、后期稍有下降;果实在贮藏期间随丁醇含量的增加而逐步呈现出特殊的异香味。延迟采收和冷藏处理有利于减少华冠果实贮藏期内丁醇积累,可用以控制和降低华冠在贮藏期间果实异香味的产生。     

Drivers of recent forest cover change in southern South America are linked to climate and CO2

Landscape Ecology - Widespread changes in forest structure&nbsp;and distribution have been documented in northern Patagonia over the past century. We employed LPJ-GUESS, a dynamic global...     

Urban forest structure and land cover composition effects on land surface temperature in a semi-arid suburban area

Using multispectral imagery and LiDAR data, we developed a high-resolution land cover dataset for a semi-arid, Colorado (USA) suburb. These data were used to evaluate patterns of land cover composition and vertical structure in relation to land use and age of development. Landsat 5 TM thermal band data for six separate dates were used to compare land surface temperature (LST) in urbanized and remnant shortgrass steppe reference areas. We used 2010 census blocks to extract LST and various explanatory variables for use in Random Forest models evaluating the relative importance of land cover composition, LiDAR-derived vertical structure variables, and the Normalized Difference Vegetation Index (NDVI) on LST patterns.We found that land cover, vertical structure, and LST varied between areas with different land use and neighborhood age. Older neighborhoods supported significantly higher tree cover and mean tree height, but differences in LST were inconsistent between Landsat image dates. NDVI had the highest variable importance in Random Forests models, followed by tree height and the mean height difference between trees and buildings. Models incorporating NDVI, vertical structure, and land cover had the highest predictive accuracy but did not perform significantly better than models using just vertical structure and NDVI. Developed areas were cooler on average than shortgrass steppe reference areas, likely due to the influence of supplemental irrigation in urbanized areas. Patterns of LST were spatially variable, highlighting the complex ways land cover composition and vertical structure can affect urban temperature.     

Spatial scaling of core and dominant forest cover in the Upper Mississippi and Illinois River floodplains, USA

Different organisms respond to spatial structure in different terms and across different spatial scales. As a consequence, efforts to reverse habitat loss and fragmentation through strategic habitat restoration ought to account for the different habitat density and scale requirements of various taxonomic groups. Here, we estimated the local density of floodplain forest surrounding each of ~20 million 10-m forested pixels of the Upper Mississippi and Illinois River floodplains by using moving windows of multiple sizes (1?C100 ha). We further identified forest pixels that met two local density thresholds: ??core?? forest pixels were nested in a 100% (unfragmented) forested window and ??dominant?? forest pixels were those nested in a >60% forested window. Finally, we fit two scaling functions to declines in the proportion of forest cover meeting these criteria with increasing window length for 107 management-relevant focal areas: a power function (i.e. self-similar, fractal-like scaling) and an exponential decay function (fractal dimension depends on scale). The exponential decay function consistently explained more variation in changes to the proportion of forest meeting both the ??core?? and ??dominant?? criteria with increasing window length than did the power function, suggesting that elevation, soil type, hydrology, and human land use constrain these forest types to a limited range of scales. To examine these scales, we transformed the decay constants to measures of the distance at which the probability of forest meeting the ??core?? and ??dominant?? criteria was cut in half (S _1/2, m). S _1/2 for core forest was typically between ~55 and ~95 m depending on location along the river, indicating that core forest cover is restricted to extremely fine scales. In contrast, half of all dominant forest cover was lost at scales that were typically between ~525 and 750 m, but S _1/2 was as long as 1,800 m. S _1/2 is a simple measure that (1) condenses information derived from multi-scale analyses, (2) allows for comparisons of the amount of forest habitat available to species with different habitat density and scale requirements, and (3) can be used as an index of the spatial continuity of habitat types that do not scale fractally.