Measuring loblolly pine crowns with drone imagery through deep learningOA

In modeling forest stand growth and yield, crown width, a measure for stand density, is among the parameters that allows for estimating stand timber volumes. However,accurately measuring tree crown size in the field, in particular for mature trees, is challenging. This study demonstrated a novel method of applying machine learning algorithms to aerial imagery acquired by an unmanned aerial vehicle(UAV) to identify tree crowns and their widths in two loblolly pine plantations in eastern Texas, US...     

Logistic regression models for human-caused wildfire risk estimation: analysing the effect of the spatial accuracy in fire occurrence data

About 90% of the wildland fires occurred in Southern Europe are caused by human activities. In spite of these figures, the human factor hardly ever appears in the definition of operational fire risk systems due to the difficulty of characterising it. This paper describes two spatially explicit models that predict the probability of fire occurrence due to human causes for their integration into a comprehensive fire risk–mapping methodology. A logistic regression technique at 1 × 1 km grid resolution has been used to obtain these models in the region of Madrid, a highly populated area in the centre of Spain. Socio-economic data were used as predictive variables to spatially represent anthropogenic factors related to fire risk. Historical fire occurrence from 2000 to 2005 was used as the response variable. In order to analyse the effects of the spatial accuracy of the response variable on the model performance (significant variables and classification accuracy), two different models were defined. In the first model, fire ignition points (x, y coordinates) were used as response variable. This model was compared with another one (Kernel model) where the response variable was the density of ignition points and was obtained through a kernel density interpolation technique from fire ignition points randomly located within a 10 × 10 km grid, which is the standard spatial reference unit established by the Spanish Ministry of Environment, Rural and Marine Affairs to report fire location in the national official statistics. Validation of both models was accomplished using an independent set of fire ignition points (years 2006–2007). For the validation, we used the area under the curve (AUC) obtained by a receiver-operating system. The first model performs slightly better with a value of AUC of 0.70 as opposed to 0.67 for the Kernel model. Wildland–urban interface was selected by both models with high relative importance.     

基于Landsat8 OLI遥感影像的森林蓄积量估测模型研究

本次试验以湖南省湘潭县为研究区,提取Landsat8 OLI影像数据的56个遥感因子作为候选因子,结合皮尔逊相关系数和主成分分析两种方法对变量进行降维,构建多元线性回归模型(MLR)、误差反向传播神经网络(BP-ANN)、K最近邻模型(KNN)和随机森林模型(RF)进行蓄积量反演,并采用决定系数(R2)、均方根误差(R...     

基于气象因子深度学习的森林火灾预测方法

森林火灾一旦发生将对生态系统造成严重的破坏,间接导致气候的变化和极端天气频发。对森林火灾的发生进行准确预测可提前采取有效的防控措施,具有重要意义。传统林火预测模型多为数学方法和浅层神经网络,当数据量增大时易出现建模困难以及预测精度降低等问题。深度学习模型在处理大量非线性数据上具有一定的优势,其模型具有多层网络结构,通过训练大量数据可提取出具有代表性的特征值,发现数据间的隐含关系,达到准确分类预测的目的。因此,本研究提出一种基于深度学习的林火预测方法,将深度信念网络(deep belief network,DBN)作为预测模型,气象因子作为输入数据,以解决传统林火预测模型在面对大量数据时预测效果不佳的问题;同时结合过采样SMOTE(synthetic minority oversampling technique)算法,平衡林火数据集和增加训练数据量,提升了森林火灾的预测准确度。结果表明,在面对更大的数据量时,该模型预测精度明显优于其他传统林火预测模型,证明了将深度学习应用在林火预测的优越性。该研究可为深度学习在林业领域的应用提供参考。     

A simple stem taper model with mixed effects for boreal black spruce

We simplified Kozak’s taper model by setting the inflection point at 1.3 m (dbh) without losing accuracy and precision. The simplification was required to facilitate the estimation of the covariance parameters when using a mixed-effects method. This method was necessary to take into account the correlation among multiple diameter measurements on an individual stem. The simple stem taper model was fitted to an extended data set collected across the province of Quebec, Canada. Comparison of the predicted stem taper and the derived stem volume with those obtained using existing models showed a comparable predictive power for the simple model. Including a prediction of the tree random effects based on supplementary diameter measurements of the bole improves the predictive ability of the model around the extra diameter observation. This model offers welcome simplicity as a means of predicting tree taper at coarse resolution for planning tree harvesting.     

A tool for identifying potential Eucalyptus nitens seed orchard sites based on climate and topography

Shy seed production in orchards of Eucalyptus nitens is a major barrier to the deployment of genetic gain in South African plantations. A machine learning method was used to identify optimal sites for the establishment of E. nitens seed orchards within the plantation forestry landscape of the summer rainfall region of South Africa. The ensemble classifier random forests (RF) was used to identify the environmental factors conducive to E. nitens floral bud production, and, based on these, build a predictive model deployable to the plantation forestry landscape for identifying suitable areas for E. nitens seed orchards. The RF model predicted site suitability likelihood for floral bud production with a high level of accuracy (area under the receiver operating characteristic curve = 0.83). Within the climatically optimal range for growing E. nitens, flower bud production was more abundant and consistent on cold slopes, i.e. sites experiencing lower minimum air temperatures during spring and autumn. The model was applied to the commercial plantation forestry landscape for the purpose of indicating sites climatically optimal for floral bud production in E. nitens and the establishment of breeding and seed production orchards of the same species.     

Evaluation of pedotransfer functions for estimating soil hydraulic properties of prevalent soils in a catchment of the Bavarian Alps

In this study, two types of pedotransfer functions (PTFs) were evaluated for their accuracy and applicability to a broad range of Alpine soils in the Halbammer area in southern Bavaria (Germany). The first model is ROSETTA, which is based on neural network analyses. It implements five hierarchical PTFs using limited to more extend input data. The second model is SOILPROP that is based on physical methods and predicts the soil hydraulic properties from particle size distribution and bulk density. The PTF were evaluated by comparing predicted with measured water retention values. The accuracy was quantified by direct statistical evaluation with the correlation coefficient (R), the mean error (ME) and the root mean square difference (RMSD). Additionally, a process based functional validation was performed by simulating the water flow using the measured and predicted soil hydraulic data. The RMSD values from ROSETTA models ranged from 0.068 to 0.202 cm³/cm³ for the water retention and from 0.450 to 0.579 log K_s (cm/day) concerning the hydraulic conductivity (K _s). The ME indicated underestimated water contents at high suctions and for soils with high organic content. The functional evaluation was the better as the more input data were used in the hierarchical PTFs. The RMSD of SOILPROP was 0.073 cm³/cm³ for water contents and 0.718 log K_s (cm/day) for the hydraulic conductivity. The water contents in the middle suction range were underestimated in sandy soils and overestimated in soils with low bulk density. The functional evaluation showed improved model accuracy when the predicted saturated conductivity was adjusted to more realistic values from literature showing its sensitiveness towards water flow modelling.     

基于高分二号的旺业甸林场蓄积量估测模型研究

【目的】为了探究国产高分二号(GF-2)影像在林分蓄积量估测中的潜力,并找到最佳的蓄积量估测模型。【方法】本次实验以内蒙古旺业甸林场为研究区,以高分二号卫星影像为信息源,结合2017年10月份调查的75块样地以及同时期的GF-2影像数据,提取波段特征、植被指数和纹理特征等43个遥感因子作为候选变量,利用Pearson相关系数选择出与蓄积量显著相关的6个变量,采用多元线性回归模型(MLR)、BP-神经网络模型(BP-ANN)、随机森林模型(RF)、支持向量机模型(SVM)和K邻近模型(KNN)进行蓄积量的估测。以决定系数(R^2)、均方根误差(RMSE)、相对均方根误差(RRMSE%)作为5种模型的评价指标,选择出旺业甸林场的最佳蓄积量估测模型,并绘制了研究区的森林蓄积量分布图。【结果】4种机器学习模型的结果明显优于传统的线性模型,其中随机森林(RF)模型和K邻近模型(KNN)均得到了较高的精度,其中RF模型的R^2为0.66,均方根误差为55.2 m^3/hm^2,相对均方根误差为28.1%,KNN模型的R^2为0.64,均方根误差为57.6 m^3/hm^2相对均方根误差为29.3%。【结论】在利用高分二号数据进行旺业甸林场蓄积量估测时,RF和KNN模型在估测针叶林蓄积量时相比于其他模型可以取得更好的结果。     

Distribution modelling of Japanese honeysuckle (Lonicera japonica) invasion in the Cumberland Plateau and Mountain Region, USA

Predicting the potential distribution of invasive plants within a specific region is pivotal to planning effective management but is challenged by attempting to model expanding populations that are rarely at equilibrium with their environment. We adopt an ensemble modelling approach to assess the potential distribution of Japanese honeysuckle (Lonicera japonica), a vine invasive to forests of the Cumberland Plateau and Mountain Region in the southeast of USA. The influence of disturbance, spatial and temporal heterogeneity and other landscape characteristics were assessed by creating regional level models based on occurrence records from the United States Department of Agriculture, Forest Service, Forest Inventory and Analysis (FIA) database. Logistic regression and maximum entropy (MaxEnt) models were assessed independently and evaluated as predictive tools to test the value of presence/absence and presence only data in predicting species distributions. Ensemble models were also developed that combined the predictions of the two modelling approaches to obtain a more robust prediction. While logistic and MaxEnt models were similar in their predictive ability and dominant input variables, the ensemble approach derived the best fitting model overall. The regional distribution of Japanese honeysuckle was influenced greatly by environmental conditions such as elevation, slope, and temperature with anthropogenic activity having significant, though lesser, influence. The ensemble models predict that Japanese honeysuckle has nearly reached its potential distribution. However, given the critical role of minimum temperature on Japanese honeysuckle distribution, future occupancy at higher elevations is likely to increase since January temperatures for this region are predicted to rise by 1-4 °C over the next 100 years. The models also give some indication of the likely effect of land cover change on its distribution. Japanese honeysuckle tended to be associated with a high component of farming or low component of forest within the local neighbourhood. This suggests disturbed forest and/or high fragmentation has a higher invasion potential and given past trends and expected continued population growth this disturbance and fragmentation will only increase. The models can be integrated into forest management decision support systems and assist in the development of long term management plans, integrating the impact of potential climate and land cover change scenarios.     

A trial to detect optimal pin-pricking timing in evaluating the ability to form traumatic resin canals of <Emphasis Type="Italic">Cryptomeria japonica</Emphasis> for selecting resistant trees to <Emphasis Type="Italic">Semanotus japonicus</Emphasis> (Coleoptera: Cerambycidae)

A trial to detect optimal pin-pricking timing in evaluating the ability to form traumatic resin canals (TRCs) of Cryptomeria japonica was examined to select resistant trees to Semanotus japonicus using 14 clones in 2001 and 2002. Resistance to S. japonicus and the ability to form TRCs in the phloem was evaluated by inoculating newly hatched larvae in the bark and by a pin prick, which was conducted every 10 days (four times) on the trunk around the larval phloem-feeding period, respectively. The larval survivorship varied greatly among clones for both years. The mean appearance of newly formed TRCs was generally higher in late treatment for both years, and the tangential width of them was also higher in 2001, whereas those of pre-formed TRCs were not higher for either year. The larval survivorship did not show significant correlations with the appearance and the width of pre-formed TRCs on all treatments. However, it showed significant negative correlations with the appearance of newly formed TRCs on the second and third treatments on the 2-year-old layer, although this was not significant with the width of them. This suggests that resistant clones might have the mechanism of rapidly forming TRCs when just at the stage of newly hatched larval entering the phloem. Thus, although the relationship between the appearance and the width of newly formed TRCs is not clear, pin-pricking treatment when the newly hatched larvae just enter the bark may be one of optimal times for the evaluation of the resistance of C. japonica to S. japonicus. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Spatio-temporal analysis of forest fire events in the Margalla Hills,Islamabad,Pakistan using socio-economic and environmental variable data with machine learning methods

Most forest fires in the Margalla Hills are related to human activities and socioeconomic factors are essential to assess their likelihood of occurrence.This study consid-ers both environmental (altitude,precipitation,forest type,terrain and humidity index) and socioeconomic (popula-tion density,distance from roads and urban areas) factors to analyze how human behavior affects the risk of forest fires.Maximum entropy (Maxent) modelling and random forest (RF) machine learning methods were used to predict the probability and spatial diffusion patterns of forest fires in the Margalla Hills.The receiver operating characteristic(ROC) curve and the area under the ROC curve (AUC) were used to compare the models.We studied the fire history from 1990 to 2019 to establish the relationship between the prob-ability of forest fire and environmental and socioeconomic changes.Using Maxent,the AUC fire probability values for the 1999s,2009s,and 2019s were 0.532,0.569,and 0.518,respectively;using RF,they were 0.782,0.825,and 0.789,respectively.Fires were mainly distributed in urban areas and their probability of occurrence was related to acces-sibility and human behaviour/activity.AUC principles for validation were greater in the random forest models than in the Maxent models.Our results can be used to establish preventive measures to reduce risks of forest fires by consid-ering socio-economic and environmental conditions.     

Modeling mid-rotation fertilizer responses using the age-shift approach

Growth and yield modelers have incorporated mid-rotation fertilizer responses by: modifying site index; developing new models to include fertilizer responses directly; using multipliers or additional terms to scale existing models. We investigated the use of age-shifts to model mid-rotation fertilizer responses. Age-shift prediction models were constructed from 43 installations of a nitrogen (0, 112, 224 and 336 kg ha⁻¹ elemental) by phosphorus (0, 28 and 56 kg ha⁻¹ elemental) factorial experiment established in mid-rotation loblolly (Pinus taeda L.) pine stands in the southeastern US. Age-shifts for dominant height and basal area increased with time after fertilization, to a maximum and then either remained fairly constant, or declined. The initial rate of increase, maximum age-shift and decline were functions of the rate and combinations of fertilizers applied, as well as stand density and age at fertilization. Volume age-shifts increased linearly throughout the 10-year measurement period for most treatments with the rate of increase being a function of the elements applied, stocking, site index and age at fertilization. A mid-rotation fertilizer application of 224 and 28 kg ha⁻¹ elemental N and P, respectively, resulted in age-shifts of 1.1, 1.9 and 2.4 years for dominant height, basal area and volume, respectively, 10 years after fertilization. The age-shifts were incorporated into growth and yield models.     

Modelling the influence of site and weed competition on juvenile modulus of elasticity in Pinus radiata across broad environmental gradients

Data from a nationwide set of Pinus radiata D. Don plots established at a range of conventional stand densities were analysed at age 6 to (i) determine how environment and competition from weeds influence dynamic modulus of elasticity (E) of the lower stem base, (ii) develop a predictive multiple regression model of E for basal stemwood and (iii) identify significant direct and indirect environmental influences (through stem slenderness) on E using path analysis.Site had a highly significant (P < 0.001) influence on E, which exhibited a 3-fold range from 1.6 to 5.3 GPa, across 30 sites. When compared to the weed-free controls, weed competition had a significant (P < 0.0001) and substantial effect on E, increasing values by on average 16% (2.76 GPa vs. 2.38 GPa).The positive linear relationship between stem slenderness (determined as tree height/ground-line tree diameter) and E was by far the strongest relationship (R² = 0.71; P < 0.001) among the 20 variables that were significantly related to E. A multiple regression model that included stem slenderness, mean minimum air temperature in mid-autumn, T_min, as positive linear relationships and net nitrogen (N) mineralisation in a negative linear form accounted for 86% of the variance in E. A cross-validation indicated that this model was stable and unbiased, with the validation accounting for 82% of the variance in E. The final path analysis model included T_min, net N mineralisation, below canopy solar radiation and stem slenderness as significant (P < 0.05) direct influences on E. Below canopy radiation, maximum air temperature during mid-summer, soil total phosphorus and carbon:nitrogen ratio were indirectly associated with E through their significant (P < 0.05) direct relationship with stem slenderness.These results provide considerable insight into how environment regulates E of juvenile P. radiata. Low fertility sites that have warm air temperatures and either a high canopy leaf area index, or high levels of woody weed competition, are most likely to produce trees with high stem slenderness and high E. Conversely, sites that are cool over summer and autumn and high in fertility, with low levels of intra- or inter-specific competition for light are likely to produce trees with low stem slenderness and low E.     

用直接估计法预测落叶松枯枝含水率的稳定性和外推误差分析

采用Catchpole等提出的直接估计法分析实验室中3组不同直径(0.5,1.0,1.5cm)的兴安落叶松枯枝的失水过程,通过交叉验证研究所得模型的稳定性,并对模型的外推误差进行分析。结果表明:枯枝的时滞和平衡含水率参数变化很小,结果具有良好的稳定性。同一直径的不同枝条在时滞、平衡含水率参数等方面存在差异。模型外推没有改变残差的正态分布,但外推后的残差增大。非外推残差集中出现在数值较小的区间上,而外推残差在这些区间上出现的概率下降,在非外推残差没有出现的数值较大的区间上出现的概率增加。研究给出枯枝在不同含水率预测值时不同残差出现的条件概率和均值。随含水率预测值的增加,残差有增加的趋势,不仅是平均值,大的残差出现的概率也在增加。据此可对外推模型所预测的含水率进行评判,确定其误差均值和不同残差出现的可能性,以减少据此进行的火险预报和火行为预报中的不确定性。用至少4个枯枝的混合数据在一定程度上可降低外推误差,减少这种不确定性。由于所有材料和实验条件最大程度上减少了材料和环境的差异,上述结果也可被看作用模型外推预测野外可燃物含水率时可能出现的误差的下限。     

气候敏感的马尾松生物量相容性方程系统研建

【目的】构建气候敏感的马尾松生物量相容性方程系统,分析气候因子对马尾松各分项生物量的影响,为森林碳汇监测和森林可持续经营提供技术支撑。【方法】基于150株马尾松单木生物量数据,采用非线性联立方程组法构建气候敏感的马尾松生物量相容性方程系统,各分项生物量(干材、干皮、树枝、树叶和地上总生物量)选用以直径和树高为自变量的二元生物量方程作为基础模型,利用一阶交叉验证法对所构建的生物量相容性方程系统进行评价。【结果】与传统未考虑气候因子的各分项生物量模型相比,气候敏感的马尾松生物量相容性方程系统预测精度明显提高,且该生物量相容性方程系统可定量描述不同气候带亚区生物量的差异程度,保证干材、干皮、树枝和树叶与地上总生物量相容。【结论】气候敏感的马尾松生物量相容性方程系统能有效分析气候因子对各分项生物量的影响,可应用于其他树种的生物量预估。     

Landowner interest in multifunctional agroforestry Riparian buffers

Adoption of temperate agroforestry practices generally remains limited despite considerable advances in basic science. This study builds on temperate agroforestry adoption research by empirically testing a statistical model of interest in native fruit and nut tree riparian buffers using technology and agroforestry adoption theory. Data were collected in three watersheds in Virginia’s ridge and valley region and used to test hypothesized predictors of interest in planting these buffers. Confirmatory factor analysis was used to verify independence of underlying latent measures. Multiple linear regression was used to model interest using the Universal Theory of Acceptance and Use of Technology (UTAUT). A second model that added agroforestry-specific predictors from Pattanayak et al. (Agrofor Syst 57:173–186, 2003) to UTAUT was tested and compared with the first. The first model was robust (Adj R ² = 0.49) but was improved by adding agroforestry specific predictors (Adj R ² = 0.57). Model generalizability was confirmed using double cross validation and normality indices. Social influence, risk expectancy, planting experience, performance expectancy, parcel size, and the interaction of gender and risk were significant in the final model. In addition, socioeconomic variables were used to characterize landowners according to their level of interest. Respondents with greater interest were newer owners that have higher incomes and are less active in farming. The result implies that future agroforesters may in large part consist of owners that have recently acquired land and manage their property more extensively with higher discretionary income and multiple objectives in mind.     

A comparison of seedling emergence and survival between Quercus glauca and Symplocos prunifolia

Seedling emergence and initial survival were compared for two evergreen broad-leaved species, Quercus glauca and Symplocos prunifolia. The relationships between the two seedling dynamics variables and environmental factors for the two species were also investigated. The number of seedlings that emerged in the study period was larger for S. prunifolia than for Q. glauca, while the survival rate was lower for S. prunifolia, presumably due to the closed canopy of this site. Models were selected for each species to determine the combination of variables explaining the most variation in emergence and survival of seedlings. The model selected for seedling emergence of Q. glauca showed that more seedlings emerged in lower hillslope positions and where the canopy in winter was more open. The model selected for S. prunifolia showed that fewer seedlings emerged on steeper slopes and that more emerged under a more open canopy in winter. With respect to seedling survival, models with only the proportion of open canopy in summer were selected for both species. These models showed that the survival rate was higher where the canopy in summer was more open. This is to be expected as both species are regarded as pioneer or mid-successional species. Q. glauca seemed to have the ability to persist as a dominant in the secondary forest at this site for longer than S. prunifolia due to the former’s better seedling survival rates and the rarity of climax species such as Castanopsis cuspidata.     

k-Nearest Neighbor Regression in the Estimation of Tectona Grandis Trunk Volume in the State of Pará, Brazil

Models constructed from machine learning are a potential non-parametric alternative for the prediction of biometric variables in opposition to traditional regression modeling. The hypothesis of this study was that the non-parametric approach with the k-Nearest-Neighbor algorithm (k-NN) has the possibility of presenting better accuracy with lower demand for predictor variables in the estimates of T. grandis trunk volume than the traditional volumetric models applied in forest sciences. Ten volumetric models were adjusted. In the regression k-NN was defined as maximum of 25 (25-NN), but with learning only for the odd neighbors starting at 3-NN. The optimal k nearest neighbor for two variations of predictors was obtained through repeated cross-validations. Spurr (ln) and Schumacher-Hall were the most accurate linear models that meet the linear regression assumptions. The optimal k nearest neighbor for the algorithm was k = 5 for the two variations of predictors. The use of the k-NN estimator may be a more general approach to linear regression especially when the assumptions made about the errors are not satisfied. However, its use should be considered only when traditional linear regression models or other simpler methods do not show good results.     

Performance and cost of a new mini-forwarder for use in thinning operations

Two tests were conducted with a new model of mini-forwarder, specifically designed for thinning operations. The tested machine resembles a conventional industrial forwarder, with tandem bogies and central articulation, but is much smaller and lighter. The machine was tested on forest plantations established on ex-farm land: such plantations offer favorable and homogeneous work conditions, which allowed reasonably accurate productivity figures to be obtained with a relatively small number of observation hours (about 10.5 h). Despite the relative inexperience of the driver, the tests indicated a productivity of between 3.1 and 3.8 m³ per scheduled machine hour (SMH) over an extraction distance of about 400 m. Extraction costs ranged from 12.4 to 15.1 € m⁻³ at the calculated machine rate of 47.6 € h⁻¹. Compared to older models derived from recreation vehicles or tracked wheelbarrows, the machine tested in this study offers a better performance and a much more comfortable workplace, with the operator sitting inside an enclosed and insulated cab. Fitted with four bogies and provided with a much longer wheelbase, the new forwarder is likely to be safer than tracked machines when surmounting obstacles, and it certainly offers a much smoother ride to the operator. Nevertheless, the tested machine is still much narrower than industrial forwarders and does not enjoy the same lateral stability. Hence, the machine is ideal for sneaking between trees and climbing over obstacles, but once on a slope it must be driven straight along the grade and never across it, unless with much caution. Like all hydrostatically driven vehicles, the tested mini-forwarder is not suited to long-distance extraction (>1 km): if run at high speed for too long, its hydrostatic transmission tends to overheat, forcing the operator to make frequent stops.