Boosted Regression Tree Models Research Articles

Quantifying the effects of 2D/3D urban landscape patterns on land surface temperature: A perspective from cities of different sizes

With the continuous development of urbanization, two-dimensional (2D) landscape pattern is insufficient to explain the complex urban thermal phenomenon, while three-dimensional (3D) urban composition directly affects the process of surface energy exchange and becomes an important influencing factor in regard to land surface temperature (LST). There have been some studies on the relationship between 2D/3D urban patterns and LST, however, there are relatively few studies on the effects of 2D/3D urban landscape on LST in cities of different sizes. In this research, the relationship of 2D/3D urban patterns and LST in three cities (Beijing, Shijiazhuang and Cangzhou) of different sizes were analyzed by using the boosted regression tree model. The results were as follows: 1) With the increase in city size, LST gradually increased in summer and decreased in winter. 2) The larger the size of the city was, the greater the number of factors that significantly affected LST (p < 0.05). 3) In spring, summer and autumn, volume of tree (TV), mean tree height (MTH) and the largest patch index of impervious surface (LPI_IS) had a high relative influence, and the average total relative influences of these three metrics were 80.08%, 78.99% and 45.14%, respectively. 4)There was a significant negative correlation between TV and LST, a positive correlation between LPI_IS and LST and a combination of positive and negative correlations between MTH and LST. The study of the effects of 2D and 3D urban morphology on LST could help urban planners and managers create more scientific urban planning and development management measures to mitigate urban thermal environment problems.

Species distribution models describe spatial variability in mesopelagic fish abundance in the Southern Ocean

IntroductionMesopelagic fishes play a central role in the transfer of energy through open-ocean food webs, particularly in the Southern Ocean where they are both important predators of zooplankton and a key prey group for many higher predators. However, they are notoriously difficult to sample, which has limited our understanding of the bio-physical predictors of their abundance and spatiotemporal variability. Species distribution models can be used to help understand species’ ecological requirements by relating records of their presence or abundance to environmental data.MethodsHere, we used data from Myctobase – a new circumpolar database of mesopelagic fishes – to model patterns in abundance of eight key myctophid species (family Myctophidae) and the genus Bathylagus in the Southern Ocean south of 45°S. We developed species-specific boosted regression tree models to obtain circumpolar predictions of abundance. Average daytime and night-time summer predictions for the period 1997 to 2011 at 0 to 200m depths were generated for each species.ResultsDepth and solar position were important predictors and species were stratified in their depth distribution. For all species, except for G. nicholsi, there was an interaction between depth of capture and solar position, reflecting diel vertical migration. Other important variables included sea surface temperature, dissolved oxygen at 200 m, chlorophyll a, and sea surface height, indicating an association with water mass properties. Circumpolar patterns of abundance varied between species with some displaying affinities for oceanic regions at Antarctic latitudes (e.g., E. antarctica and Bathylagus spp.) or sub-Antarctic latitudes (e.g., K. anderssoni and P. tenisoni); and affinities for shelf regions (e.g., P. boliniand G. nicholsi).DiscussionOur findings suggest that the abundance of mesopelagic fish is influenced by diel vertical migration and meso- and sub-mesoscale oceanographic features, with the Polar Front being a major delimiting feature. Our study showed contrasting patterns in community composition with higher species diversity north of the Polar Front that might be indicative of latitudinal variability in food web structure. Our spatial analysis is an important step toward resolving what determines important habitat for mesopelagic fishes, providing foundational information for understanding shifting food web dynamics into the future.

Spatiotemporal distribution of Lophius litulon in the southern Yellow Sea and East China Sea

In order to investigate the temporal-spatial distribution of yellow goosefish in the open waters of the southern Yellow Sea (SYS) and East China Sea (ECS), according to the fishery data by bottom-trawl surveys and environmental data including sea bottom temperature (SBT), sea bottom salinity (SBS), bottom concentration of dissolved oxygen (BDO) and depth during 2018-2019, we established habitat suitability index (HSI) models by arithmetic mean (AMM) and geometric mean (GMM) methods, and compared the model outputs by cross validations. Especially, the weight of each environmental factor was evaluated by boosted regression tree (BRT). Results showed that the area with the highest habitat quality varied among seasons. The yellow goosefish mainly inhabited in the adjacent area of the Yangtze River Estuary and coastal waters of Jiangsu Province with depth ranged from 22 to 49 m in spring. The optimal inhabitation was located in the SYS, with the bottom temperature ranged from 8.9-10.9 ℃ in summer and autumn. Particularly, the optimal inhabitation extended from the SYS to the ECS with bottom temperature ranged from 9.2-12.7 ℃ in winter. Results of BRT models showed that depth was the most important environmental factor in spring and bottom temperature was the crucial one in the other three seasons. Results of cross validation showed that the weighted AMM-based HSI model performed better for yellow goosefish in spring, autumn and winter. The distribution of yellow goosefish was closely related to its biological traits and environmental factors in the SYS and ECS, China.

Building a high-resolution site index map using boosted regression trees: the Norwegian case

Accurate estimation of site productivity is essential for forest projections and scenario modelling. We present and evaluate models to predict site index (SI) and whether a site is productive (potential total stem volume production ≥ 1 m3·ha−1·year−1) in a wall-to-wall high-resolution (16 m × 16 m) SI map for Norway. We investigate whether remotely sensed data improve predictions. We also study the advantages and disadvantages of using boosted regression trees (BRT), a machine-learning algorithm, to create high-accuracy SI maps. We use climatic and topographical data, soil parent material, a land resource map, and depth to water, together with Sentinel-2 satellite images and airborne laser scanning metrics, as predictor variables. We use the SI observed at more than 10 000 National Forest Inventory (NFI) sample plots throughout Norway to fit BRT models and validate the models using 5822 independent temporary plots from the NFI. We benchmark our results against SI estimates from forest monitoring inventories. We find that the SI from BRT has root mean squared error (RMSE) ranging from 2.3 m (hardwoods) to 3.6 m (spruce) when tested against independent validation data from the NFI temporary plots. These RMSEs are similar or marginally better than an evaluation of SI estimates from operational forest management plans where SI normally stems from manual photo interpretation.

A Meta-Analysis Study on the Use of Biochar to Simultaneously Mitigate Emissions of Reactive Nitrogen Gases (N2O and NO) from Soils

Nitrous oxide (N2O) and nitric oxide (NO) are detrimental reactive gaseous oxides of nitrogen. Excessive application of nitrogen fertilizers in cropping systems has significantly increased the emissions of these gases, causing adverse environmental consequences. Previous studies have demonstrated that biochar amendment can regulate soil-N dynamics and mitigate N losses, but they lacked simultaneous assessments of soil N2O and NO emissions. Thus, the factors influencing the emissions of nitrogen oxides are still unclear. Therefore, this study examined the impact of biochar application on simultaneous N2O and NO emissions based on 18 peer-reviewed papers (119 paired observations). A machine learning model (boosted regression tree model) was adopted to assess the potential influencing factors, such as soil properties, biochar characteristics, and field management conditions. The addition of biochar reduced N2O and NO emissions by 16.2% and 14.7%, respectively. Biochar with a high total carbon content and pH, from woody or herbaceous feedstock, pyrolyzed at a high temperature, applied at a moderate rate and to soil with a high-silt content, a moderate pH, and coarse texture, could simultaneously reduce soil N2O and NO emissions. Biochar amendment, thus, has the potential to lower the environmental impact of crop production. Furthermore, the influence of soil properties, biochar characteristics, and field management should be considered in the future to enhance the efficacy of biochar.

Climate change enhanced the positive contribution of human activities to net ecosystem productivity from 1983 to 2018

IntroductionAccurate assessment of the net ecosystem productivity (NEP) is very important for understanding the global carbon balance. However, it remains unknown whether climate change (CC) promoted or weakened the impact of human activities (HA) on the NEP from 1983 to 2018.MethodsHere, we quantified the contribution of CC and HA to the global NEP under six different scenarios based on a boosted regression tree model and sensitivity analysis over the last 40 years.Results and discussionThe results show that (1) a total of 69% of the areas showed an upward trend in the NEP, with HA and CC controlled 36.33 and 32.79% of the NEP growth, respectively. The contribution of HA (HA_con) far exceeded that of CC by 6.4 times. (2) The CO2 concentration had the largest positive contribution (37%) to NEP and the largest influence area (32.5%). It made the most significant contribution to the NEP trend in the range of 435–440 ppm. In more than 50% of the areas, the main loss factor was solar radiation (SR) in any control area of the climate factors. (3) Interestingly, CC enhanced the positive HA_con to the NEP in 44% of the world, and in 25% of the area, the effect was greater than 50%. Our results shed light on the optimal range of each climatic factor for enhancing the NEP and emphasize the important role of CC in enhancing the positive HA_con to the NEP found in previous studies.

Mapping soil pH levels across Europe: An analysis of LUCAS topsoil data using random forest kriging (RFK)

AbstractSoil pH affects food production, pollution control and ecosystem services. Mapping soil pH levels, therefore, provides policymakers with crucial information for developing sustainable soil use and management policies. In this study, we used the LUCAS 2015 TOPSOIL data to map soil pH at a European level. We used random forest kriging (RFK) to build a predictive model of spatial variability of soil pH, as well as random forest (RF) without co‐kriging and boosted regression trees (BRT) modelling techniques. Model accuracy was evaluated using a ten‐fold cross‐validation procedure. While we found that all methods accurately predicted soil pH, the accuracy of the RFK method was best with regression performance metrics of: R2 = 0.81 for pH (H2O) and pH (CaCl2); RMSE = 0.59 for pH (H2O) and RMSE = 0.61 in pH (CaCl2); MAE = 0.41 for pH (H2O) and MAE = 0.43 in pH (CaCl2). Dominant explanatory variables in the RF and BRT modelling were topography and remote sensing variables, respectively. The generated maps broadly depicted similar spatial patterns of soil pH, with an increasing gradient of soil pH from north to south Europe, with the highest values mainly concentrated along the Mediterranean coast. The mapping could provide spatial reference for soil pH assessment and dynamic monitoring.

Global Distribution of Aedes aegypti and Aedes albopictus in a Climate Change Scenario of Regional Rivalry.

Arboviral mosquito vectors are key targets for the surveillance and control of vector-borne diseases worldwide. In recent years, changes to the global distributions of these species have been a major research focus, aimed at predicting outbreaks of arboviral diseases. In this study, we analyzed a global scenario of climate change under regional rivalry to predict changes to these species' distributions over the next century. Using occurrence data from VectorMap and environmental variables (temperature and precipitation) from WorldClim v. 2.1, we first built fundamental niche models for both species with the boosted regression tree modelling approach. A scenario of climate change on their fundamental niche was then analyzed. The shared socioeconomic pathway scenario 3 (regional rivalry) and the global climate model Geophysical Fluid Dynamics Laboratory Earth System Model v. 4.1 (GFDL-ESM4.1; gfdl.noaa.gov) were utilized for all analyses, in the following time periods: 2021-2040, 2041-2060, 2061-2080, and 2081-2100. Outcomes from these analyses showed that future climate change will affect Ae. aegypti and Ae. albopictus distributions in different ways across the globe. The Northern Hemisphere will have extended Ae. aegypti and Ae. albopictus distributions in future climate change scenarios, whereas the Southern Hemisphere will have the opposite outcomes. Europe will become more suitable for both species and their related vector-borne diseases. Loss of suitability in the Brazilian Amazon region further indicated that this tropical rainforest biome will have lower levels of precipitation to support these species in the future. Our models provide possible future scenarios to help identify locations for resource allocation and surveillance efforts before a significant threat to human health emerges.

Spatial-temporal variations and driving factors of soil organic carbon in forest ecosystems of Northeast China

Forest soil carbon is a major carbon pool of terrestrial ecosystems, and accurate estimation of soil organic carbon (SOC) stocks in forest ecosystems is rather challenging. This study compared the prediction performance of three empirical model approaches namely, regression kriging (RK), multiple stepwise regression (MSR), random forest (RF), and boosted regression trees (BRT) to predict SOC stocks in Northeast China for 1990 and 2015. Furthermore, the spatial variation of SOC stocks and the main controlling environmental factors during the past 25 years were identified. A total of 82 (in 1990) and 157 (in 2015) topsoil (0–20 ​cm) samples with 12 environmental factors (soil property, climate, topography and biology) were selected for model construction. Randomly selected 80% of the soil sample data were used to train the models and the other 20% data for model verification using mean absolute error, root mean square error, coefficient of determination and Lin's consistency correlation coefficient indices. We found BRT model as the best prediction model and it could explain 67% and 60% spatial variation of SOC stocks, in 1990, and 2015, respectively. Predicted maps of all models in both periods showed similar spatial distribution characteristics, with the lower SOC in northeast and higher SOC in southwest. Mean annual temperature and elevation were the key environmental factors influencing the spatial variation of SOC stock in both periods. SOC stocks were mainly stored under Cambosols, Gleyosols and Isohumosols, accounting for 95.6% (1990) and 95.9% (2015). Overall, SOC stocks increased by 471 ​Tg ​C during the past 25 years. Our study found that the BRT model employing common environmental factors was the most robust method for forest topsoil SOC stocks inventories. The spatial resolution of BRT model enabled us to pinpoint in which areas of Northeast China that new forest tree planting would be most effective for enhancing forest C stocks. Overall, our approach is likely to be useful in forestry management and ecological restoration at and beyond the regional scale.

Multi-Year Environmental Trends of Shrimp Black Gill (Hyalophysa lynni) Prevalence in Texas Gulf Coast Shrimp Populations

Shrimp Black Gill, caused by the apostome ciliate Hyalophysa lynni, is an emerging disease impacting penaeid shrimp populations along the southeast Atlantic Coast and the Gulf of Mexico (GOM). Changing annual environmental conditions may drive infection levels of this parasitic ciliate in these populations, which comprise one of the largest fisheries in the United States. Hyalophysa lynni is established on the Texas Gulf Coast, and prevalence of this parasite has a strong seasonal and spatial trend, likely linked with high temperature and a wide range of estuarine salinities. Texas Parks and Wildlife Department monitored shrimp black gill in 2 penaeid shrimp species, Litopenaeus setiferus and Farfantepenaeus aztecus in 2019, with the aim of observing trends in prevalence along spatial and temporal scales. Hyalophysa lynni was found in all 7 bay systems throughout the study period, and this study is a continuation of that monitoring effort, adding 2 years of data collection (2020 and 2021) to prior research. Throughout the entire sampling period (2019–2021) and coastwide, H. lynni was found in 66% of all shrimp samples collected, although prevalence varied annually. Boosted regression tree modeling indicated that low salinity, high temperature, and time of year (late spring through fall) had a significant relationship with H. lynni prevalence in shrimp along the Texas Gulf Coast. Shrimp Black Gill is likely to continue to be present in GOM shrimp populations, and annual precipitation events and increased water temperatures may amplify the population morbidity within any given year.

Abundance of the bearded fireworm Hermodice carunculata (Polychaeta: Amphinomidae) increases across a euphotic-mesophotic depth gradient in the remote St. Peter and St. Paul’s Archipelago

The bearded fireworm Hermodice carunculata is widely distributed across tropical and subtropical Atlantic and Mediterranean Oceans and was previously considered to be mostly associated with shallow reefs. We provide here data on the distribution, abundance and habitat use of H. carunculata across a euphotic-mesophotic gradient (0-90 m) in the Saint Peter and Saint Paul’s Archipelago (SPSPA, Mid Atlantic Ridge, Brazil). Samples were obtained using SCUBA and a Remote Operated Vehicle (ROV). A total of 189 individuals were observed and a sharp increase in abundance with depth was recorded, particularly from 50 m depth onwards. In the mesophotic zone (50-90 m) individuals were closely associated with branching black-corals (Tanacetipathes spp.) and predation over black-corals and the scleractinian Madracis decactis was commonly sighted. A Boosted Regression Tree model indicated black-coral abundance as the main driver of H. carunculata abundance, suggesting that preference for optimal habitats to hide/forage is more important than depth per se on the bathymetric distribution of the fireworm. The high abundances of H. carunculata in the mesophotic zone, and its predation on keystone benthic cnidarians, suggests that this species play important roles in the dynamics of deep reefs.

Boosted Regression Tree (BRT) model for PM10 concentrations prediction in Malaysia

Air pollution in urban areas is a highly complex problem, displaying strong seasonality and dependence on meteorological factors. Urban particulate matter with an aerodynamic diameter less than 10 μm (PM10) has been identified as one of the most hazardous air pollutants to human health due to the fact that its size range overlaps with that of respirable particles. Models of air quality forecasting are used to provide forecasters with numerical guidance for issuing particulate matter concentration forecasts to human health exposure in a timely manner. The aim of the study was to propose a Boosted Regression Tree (BRT) model for predicting PM10 concentrations in the short term. Multiple Linear Regression (MLR) and Boosted Regression Tree (BRT) models for short-term PM10 predictions are provided, and performance indicators (IA, R2, RMSE, MAE, and MAPE) are used to find the appropriate model. The Department of Environment Malaysia (DOE) provided seventeen years of daily average air quality monitoring data, including eight parameters (PM10, wind speed, temperature, relative humidity, NO2, SO2, CO, and O3) and five monitoring stations (Perai, Shah Alam, Nilai, Larkin, and Pasir Gudang). The BRT model gave good results for predicting the PM10 concentrations for each station. The results indicated that for the Perai monitoring station (R2 = 0.774), Shah Alam monitoring station (R2 = 0.813), Nilai monitoring station (R2 = 0.792), Larkin monitoring station (R2 = 0.817) and the Pasir Gudang monitoring station (R2 = 0.810). According to the findings, the BRT model should be employed in air pollution research, where it is projected to outperform other methodologies in terms of predictive performance. The findings would enhance the existing air pollution research and forecasting approaches.

Landscape factors influencing roost site selection by monarch butterflies Danaus plexippus during fall migration in Ontario, Canada

Worldwide, insect populations are declining, and the eastern migratory group of the monarch butterfly Danaus plexippus in North America has not escaped this fate. The conservation of this iconic species is an international priority but requires knowledge of how monarchs interact with the landscape during different stages of the annual cycle. To better understand habitat needs of monarchs departing their core breeding range in southern Ontario, Canada, we examined how various landscape features influenced roost site selection during fall migration—an instrumental resource link between the breeding and wintering grounds. Using dedicated fall migration surveys along the Great Lakes coastlines and a citizen science dataset collected across all of Ontario, we evaluated the relationship between roost site occupancy and 18 landscape variables using a boosted regression tree (BRT) modelling approach. Results suggest that a closer distance to the Great Lakes, increased goldenrod Solidago spp. cover, moderate forest cover, rural road cover, and urban land cover are all important to roosting site selection. Our research provides important insights into the habitat characteristics of stopover sites, which will help guide future investigations and conservation actions to preserve monarch butterflies and their migratory phenomenon.

Comparative Analysis of Battery Cycle Life Early Prediction Using Machine Learning Pipeline

Lithium-Ion battery system is one of the most critical but expensive components for both electric vehicles and stationary energy storage applications. In this regard, accurate and reliable early prediction of battery lifetime is important for optimizing life cycle management of batteries from cradle to grave. In particular, accurate aging diagnostics and prognostics is crucial for ensuring longevity, performance, safety, uptime, productivity, and profitability over a battery's lifetime. However, current state-of-art methods do not provide satisfactory prediction performance (lack of uncertainty quantification) using early degradation data. In the present work, to produce the best model for both battery cycle life point prediction and range prediction (i.e., confidence intervals or prediction intervals), a pipeline-based approach is proposed, in which a full 33-feature set is generated manually based on battery degradation knowledge, and then used to learn the best model among five machine learning (ML) models that have been reported in the battery lifetime prediction literature, and two quantile regression models for battery cycle life prediction. The calibration and sharpness property of battery cycle life range prediction is properly evaluated by their coverage probability and width respectively. The experimental results show that the gradient boosting regression tree model provides the best point prediction performance, while the quantile regression forest model provides the best range prediction performance with both full 33-feature set and the MIT 6-feature set.

Inferring Students’ Self-Assessed Concentration Levels in Daily Life Using Biosignal Data From Wearables

The ability to concentrate well is an important determinant of students’ learning outcomes but remains poorly understood. In this work we investigated whether there exists a mapping between students’ biosignals and perceived concentration levels. If we succeed in this mapping, a wearable can function as a Concentration Tracker, a novel feature that is missing from current wearables. For this, a wearable wristband was used to record students’ heart rate, heart rate variability, skin temperature, skin conductivity and acceleration from body changes. Additionally, students self-assessed their concentration levels using a smartphone application. We improved the accuracy by utilizing a big amount of unlabelled biodata from outside the study sessions. Our best boosted regression tree model predicted students’ concentration level with only 1.7% NMAE error. The predictions for a user not in the training set were much weaker; the best model, a convolutional neural network, achieved a prediction NMAE error of 30.7%. This implies that the users generated biosignals highly individually. Thus, models are not well transferable from one user to another without rooting them in user-specific data. Contrary to stress research, our results showed that skin conductivity had mostly a negative correlation with students’ concentration levels. Also diverging from stress reactions, skin temperature had mainly a positive correlation. Conductivity and temperature were the two dominant predictors. Further, the results suggest that an element of deep, effortless concentration was present in the learning experience of the subjects. Altogether, our work demonstrates that a concentration tracking wearable for improving learning is technically achievable.

Regional differences in the response of California’s rangeland production to climate and future projection

Rangelands support many important ecosystem services and are highly sensitive to climate change. Understanding temporal dynamics in rangeland gross primary production (GPP) and how it may change under projected future climate, including more frequent and severe droughts, is critical for ranching communities to cope with future changes. Herein, we examined how climate regulates the interannual variability of GPP in California’s diverse annual rangeland, based on the contemporary records of satellite derived GPP at 500 m resolution since 2001. We built Gradient Boosted Regression Tree models for 23 ecoregion subsections, relating annual GPP with 30 climatic variables, to disentangle the partial dependence of GPP on each climate variable. The machine learning results showed that GPP was most sensitive to growing season (GS) precipitation, with a reduction in GPP up to 200 g cm−2 yr−1 when GS precipitation decreased from 400 to 100 mm yr−1 in one of the driest subsections. We also found that years with more evenly distributed GS precipitation had higher GPP. Warmer winter minimum air temperature enhanced GPP in approximately two-thirds of the subsections. In contrast, average GS air temperatures showed a negative relationship with annual GPP. When the pre-trained models were forced by downscaled future climate projections, changes in the predicted rangeland productivity by mid- and end of century were more remarkable at the ecoregion subsection scale than at the state level. Our machine learning-based analysis highlights key regional differences in GPP vulnerability to climate and provides insights on the intertwining and potentially counteracting effects of seasonal temperature and precipitation regimes. This work demonstrates the potential of using remote sensing to enhance field-based rangeland monitoring and, combined with machine learning, to inform adaptive management and conservation within the context of weather extremes and climate change.

Ecological engineering induced carbon sinks shifting from decreasing to increasing during 1981–2019 in China

Substantial evidence shows that most of China's terrestrial ecosystems are important carbon sinks. However, the nonlinear trend of the carbon sinks and their nonlinear response to driving factors are unclear. Taking the net ecosystem productivity (NEP) as a proxy for the ecosystem carbon sink, the nonlinear relationships between the monotonically increasing trends and decreasing to increasing shifts in the carbon sink to climate change and ecological engineering were investigated based on ensemble empirical mode decomposition (EEMD) and machine learning algorithm (boosted regression tree model, BRT). The results suggest that 16.75 % of the carbon sinks in China experienced a monotonic increase. Additionally, 20.55 % of the carbon sinks shifted from decreasing to increasing trends, primarily after 1995, and these carbon sinks were located in the key ecological engineering areas, such as the middle reaches of the Yellow River shelterbelt program area, the Liaohe shelterbelt program area, the Grain to Green program area, and the Three-North Forest shelterbelt program area. Moreover, carbon sinks exhibited strong spatial autocorrelation with low-low clustering in the north and high-high clustering in the south. The increase in CO2 (slope of CO2 < 1.8 g/m2/s/y) and solar radiation (slope of radiation >1 w/m2/y) promoted the monotonic increase in the carbon sinks in the center of China. The increase in the areas of forest and grassland shifted the carbon sink trend from decreasing to increasing in the key ecological engineering program areas, and economic development reversed the carbon sink reduction in the Pearl River shelterbelt program area. These findings highlight the positive effect of ecological engineering on carbon sinks and provide adaptation strategies and guidance for China to achieve the “carbon neutrality” target.

Machine Learning Method for Fatigue Strength Prediction of Nickel-Based Superalloy with Various Influencing Factors.

The accurate prediction of fatigue performance is of great engineering significance for the safe and reliable service of components. However, due to the complexity of influencing factors on fatigue behavior and the incomplete understanding of the fatigue failure mechanism, it is difficult to correlate well the influence of various factors on fatigue performance. Machine learning could be used to deal with the association or influence of complex factors due to its good nonlinear approximation and multi-variable learning ability. In this paper, the gradient boosting regression tree model, the long short-term memory model and the polynomial regression model with ridge regularization in machine learning are used to predict the fatigue strength of a nickel-based superalloy GH4169 under different temperatures, stress ratios and fatigue life in the literature. By dividing different training and testing sets, the influence of the composition of data in the training set on the predictive ability of the machine learning method is investigated. The results indicate that the machine learning method shows great potential in the fatigue strength prediction through learning and training limited data, which could provide a new means for the prediction of fatigue performance incorporating complex influencing factors. However, the predicted results are closely related to the data in the training set. More abundant data in the training set is necessary to achieve a better predictive capability of the machine learning model. For example, it is hard to give good predictions for the anomalous data if the anomalous data are absent in the training set.

Multiple effects of climate changes and human activities on NPP increase in the Three-north Shelter Forest Program area

In the framework of ecological engineering, vegetation net primary productivity (NPP) has grown dramatically in the Three-north Shelter Forest Program area (TSFPA), which is influenced by both human activities and climate change, as well as their interaction. However, it remains unclear how the individual (pure effects) and interactive (shared and coupling effects) impacts affect and maintain the growing vegetation NPP trend. In this study, the monotonically increasing NPP trend from nonlinear trends was detected using Ensemble empirical mode decomposition (EEMD), and its persistence was investigated using boosted regression tree (BRT) models. We investigated the relative contributions of environmental factors and variable interactions, as well as spatially identified the multiple effects of climate change and human activities on the increasing trend and its persistence over TSFPA. The results indicated that: (1) The vegetation NPP was dominated by the monotonically increasing trend (41.08 %), and 30.04 % of all pixels can still maintain NPP increment. (2) Although soil moisture exhibited the highest interaction with the precipitation seasonality (32.21 %) on the monotonically increasing NPP trend, land-use change substantially interacted with soil moisture (26.97 %), snowmelt (23.34 %), and net solar radiation (13.43 %). (3) The monotonically increasing vegetation NPP was caused not only by the pure effects of climate change (23.57 %) and human activities (11.69 %) directly but also by their coupling effects (26.64 %) or shared effects (38.10 %) interactively. (4) The shared effects (50.94 %) dominated the persistently increasing NPP trend, whereas the coupling effects (78.78 %) dominated the non-persistently increasing NPP trend. The findings highlight the importance of the pure, shared, and coupling effects of human activities and climate change interactively for elucidating the driving of the vegetation NPP increment and its persistence over ecological engineering areas.

Woody plant cover trends and potential drivers in the Horqin temperate steppe, northeast China: Remote sensing-based computation and modeling

The Horqin temperate steppe is the largest grassland in northern China. With the combined disturbance of multiple potential drivers, the native herbaceous vegetation has degraded with different intensity, and woody plants have transformed into the dominant vegetation in this area. On the one hand, woody plants play an important role in sand control, food/timber product provision, etc. On the other hand, this transformation process significantly affects the steppe ecosystem structure and function. It is necessary to clarify the patterns and determinants of woody plant cover changes. In view of the lack of high spatio-temporal resolution remote sensing data for woody cover. Here, based on high-resolution satellite images and the random forest model, the study reports the annual mean woody plant cover of the Horqin temperate steppe is 17.55 % over the past 35 years, and the significant increase area is approximately 43.7 % of the total area. Analysis of the boosted regression tree model showed potential drivers were mainly attributed to climate change (i.e., mean annual temperature, mean annual wind speed, and annual precipitation) and grazing intensity, with the cumulative contribution approaching 70 %. Furthermore, large-scale and high-density afforestation could also be responsible for the increase in woody plant cover. Against the background of frequent climate extremes, maintaining and improving steppe productivity, avoiding regional climate deterioration caused by blind afforestation, and developing a reasonable grazing management policy is the urgently needed to ensure the relative ecological stability of semi-arid temperate steppes and to achieve relatively sustainable development of the regional agro-pastoral economy.