True Skill Statistic Research Articles

Continental‐scale 1 km hummingbird diversity derived from fusing point records with lateral and elevational expert information

Anthropogenic change is affecting mountain regions worldwide. Managing this change and advancing biodiversity information for research requires spatially detailed information on species distributions which often is incomplete. Here, we provide a model‐based approach for the integration of expert‐based elevational range information with expert range maps and point occurrences to address this need. These integrated models use expert knowledge on elevational and distributional ranges as offset in a Poisson point process species distribution model (SDM). We use this approach to model the distribution of 276 hummingbird Trochilidae species at 1 km resolution and validate model performance with extensive survey data (presence–absence inventories). Models including expert elevation information consistently outperformed those lacking this information. Improvements were greatest when the number of available occurrences was small, highlighting the added value from expert elevation information, especially for data‐poor species. Separate validation data indicated significant increases in true skill statistics, based on higher specificity and slightly improved sensitivity. SDMs that included expert range information out‐performed presence‐only models based only on occurrence data in 92.5% of cases and had higher sensitivity, lower false positive rates and smaller predicted range sizes. Generally, the integrated models removed unsuitable areas from range estimates and decreased overestimates in geographic range size (false presences) inherent in expert maps and in models lacking elevation information. By stacking SDM output we provide a first, hemispheric map of predicted hummingbird species richness modelled at 1 km resolution and identify southwest Colombia as a richness hotspot. Our study highlights the value gained from integrating multiple data types in a single framework. The presented approach improved high‐resolution range predictions for single species (reducing false presences) and aggregate biodiversity patterns (e.g. reducing species richness overestimates). The method is now being implemented and expanded in Map of Life.

GIS-based multi-criteria decision making and entropy approaches for groundwater potential zones delineation

In arid and semi-arid, due to the lack of surface water, groundwater plays key role in people survival. Therefore, assessing groundwater potential zones is vital to conserve and manage this resource properly. The main object of this study was to compare two methods of AHP and entropy for delineating groundwater potential zone in arid region of Iran. For this regard, withdraw in groundwater level in 16 piezometer wells along with eight GWP conditioning factors of slope degree, stream power index (SPI), landuse, geology; rainfall volume, elevation, soil and lineament density have been considered in the training process while the discharge in the other 16 wells was involved in verifying process of the two models. GWP maps were provided by overlaying all the thematic maps in terms of the obtained weights from AHP and entropy methods. Finally, the models performance was compared through statistical measures of Sensivity, Specificity and True Skill Statistics (TSS). The results show that IOE model (TSS = 0.16) have higher prediction performance than AHP approach (TSS = 0). Therefore, IOE was more promising tools for GWP modeling.

A wrapper feature selection approach for efficient modelling of gully erosion susceptibility mapping

Identifying the vulnerability level of an area to soil erosion, particularly gully erosion, is key to the development of an efficient management strategy for policymakers. While efforts into susceptibility mapping of natural disasters have grown in recent years, understanding the most relevant predictive causal factors is still a challenge. As the selection of these factors, among many potentially relevant factors, is an important part of the model selection process, we propose a hybrid intelligent approach for the optimal selection of a set of relevant factors based on logistic regression (LR) and genetic algorithms. In order to verify the effectiveness of the proposed approach, this study also identified areas that were highly susceptible to gully erosion using three different classifiers – namely, the LR, support vector machine (SVM) and k-nearest neighbours (k-NN) techniques. We tested the approach in the Yeli Bedrag watershed in north-eastern Golestan province, Iran. The results showed that the elevation, distance to fault, slope and the index of connectivity were the most important causal factors affecting the successful prediction of gully occurrence. Comparison of maximum True Skill Statistic values showed that increased model sophistication did not necessarily result in a higher level of model performance. In terms of performance, k-NN was superior to the SVM and LR methods. This method can be effectively used for gully erosion susceptibility (GES) zonation in the study area, which is very important to support spatial planning to initiate designing mitigation strategies and conservation needs over a large area, or to plan additional conservation efforts and relocate soil conservation plans. In conclusion, our findings indicate that by incorporating the proposed hybrid intelligent approach, the number of relevant factors for GES mapping was reduced, while classification accuracy was increased.

Global ecological niche modelling of current and future distribution of peste des petits ruminants virus (PPRv) with an ensemble modelling algorithm.

Peste des petits ruminants (PPR) is a highly contagious transboundary viral disease of sheep and goats that negatively impacted the farmers and pastoralists' livelihood in Africa and Asia. To overcome the disease's consequences, the OIE and FAO are collaborating efforts to eradicate the disease once and for all. We developed a predictive model that delineates suitable territories for the virus globally in support of this eradication programme. To achieve this, we used an ecological niche modelling with an ensemble algorithm. AUC-ROC curve, true skill statistics (TSS) and Kappa values were used to evaluate the model's performance. A TSS value greater than 0.7 was used to pool outputs of the nine model. The ensemble model has better performance than individual models by every evaluation metrics (Kappa=0.82, TSS=0.88 and ROC=0.99). Annual minimum temperature (24.92%), annual maximum temperature (21.37%), goat density (18.03%) and solar radiation (14.04%) have the highest overall contribution in the ensemble model. The model indicates that India, Mongolia, Iraq, Iran, Afghanistan, Nepal, Tanzania, Uganda, Kenya, Sudan, Angola, Nigeria, DRC, Ghana, Sierra Leon, Southern Spain, France, Albania, Montenegro, Macedonia, Italy, Armenia and Azerbaijan are highly suitable for PPRv. In 2040, suitable territories for PPRv will diminish, indicating the odds are with us in eradicating disease by 2030. We believe that this model can be used as an epidemiological tool to facilitate the global eradication programme of the disease set by the OIE and FAO.

Predicting invasion potential of Senna didymobotrya (Fresen.) Irwin & Barneby under the changing climate in Africa

BackgroundSenna didymobotrya is a native African flowering shrub. It is suspected that climate change encourages the introduction and spread of invasive alien species. The possible invasion of S. didymobotrya across the continent is expected to increase in the future due to ongoing climate change. Nonetheless, there is still paucity of empirical evidence on the extent to which the changing climate contributes to the surge of the flowering shrub. This study, therefore, investigated the present and potential invasion of S. didymobotrya using the species distribution model under changing climate conditions. The two representative concentration pathways (RCP4.5 and RCP8.5) and eight bioclimatic variables and one topographic variable were used to simulate the current and future (2050s and 2070s) invasion of S. didymobotrya in Africa. The model performance was assessed using the area under the receiver operating characteristic curve (AUC) and true skill statistics (TSS).ResultsThe results of the study showed that under the current climatic conditions, 18% of Africa is suitable for the establishment and invasion of S. didymobotrya. The most suitable hotspot for S. didymobotrya invasion is eastern Africa, followed by southern Africa. The predicted model showed that by 2050, 3.3% and 3.12% of the continent would be highly suitable areas for the invasion of the species under RCP4.5 and RCP8.5, respectively. In the 2070s, under RCP4.5 and RCP8.5, the highly suitable area would be 3.13% and 2.7%, respectively. In relation to the current suitability, the cumulative projected areas of the low and moderate suitability class under RCP4.5 and RCP8.5 will rise by the years 2050 and 2070. However, under both RCPs, the non-suitable area for S. didymobotrya invasion would gradually decrease.ConclusionsFrom the findings, it can be concluded that the ecosystem’s vulnerability to S. didymobotrya invasion under future climatic conditions will proliferate significantly. Hence, to prevent the projected harm to biodiversity and ecosystem services, governments need to focus their future biodiversity management and policy directions on the means and strategies of minimizing the invasion and the distribution rate of S. didymobotrya across habitat types.

基于MaxEnt模型的甘肃安西极旱荒漠国家级自然保护区北山羊生境评估

明晰甘肃安西极旱荒漠国家级自然保护区珍稀濒危物种北山羊的分布格局，并阐释气候变化和人类活动对北山羊的影响，对今后北山羊生境管理和物种保护具有重要意义。基于北山羊实测分布点记录和环境变量数据，结合MaxEnt模型和ArcGIS空间分析功能，利用CMIP6的8个气候模式均值预测中度发展路径（SSP2-4.5）下，基准期（1970-2000年）和未来气候（2041-2060年、2081-2100年）变化情景下，甘肃安西极旱荒漠国家级自然保护区北山羊的潜在适生区分布范围及变化，并综合贡献率和置换重要性值对北山羊生境选择关键环境因子进行了分析。研究结果表明：（1） MaxEnt模型的预测精度较高，三种气候条件下ROC曲线下面积（AUC，Area Under Curve）>0.97，且真实技巧统计（TSS，True Skill Statistic）>0.90，模拟结果可靠。（2）影响北山羊生境选择的主要环境因子为气候条件（降水量季节性变异系数和等温性）、海拔和人为干扰（距泉和居名点距离）。水是保护区北山羊生存的最基本要素，气候条件共同控制北山羊生境条件。此外，北山羊习性决定其生境宜选择高海拔和远离人类活动影响地区。（3）基准期保护区北山羊主要分布在北片和南片高海拔山区，面积365.77 km²（占整个保护区的4.31%），北山羊适生区面积北片>南片、中高等适生区主要位于保护区北片。（4） CMIP6未来气候变化情景下，随着保护区生态环境改善和濒危物种保护措施的实施，北山羊潜在适生区面积呈增加趋势，但是受北山羊近亲繁殖的影响，整体上北山羊数量和适生区面积增加并不显著且有向南部及高海拔地区转移趋势。;To clarify the distribution pattern for the rare and endangered wild animals of Asiatic ibex (Capra sibirica) and determin the influence of climate change and human activities to this specises in the Anxi Extreme-Arid Desert National Nature Reserve (AEDNNR) of Gansu Province, have great importance for the future habitat management and species protection. In this paper, based on the field survey points and enviromental variables, the maximum entropy (MaxEnt) model and ArcGIS spatial analysis tools were used to predict the potential distribution and dynamic of Asiatic ibex in AEDNNR. For the nineteen bioclimatic variables, we selected an average of eight global climate models under the intermediate greenhouse gas emissions (SSP2-4.5) for the base period (the years of 1970-2000) and future climate periods (the years of 2041-2060 and 2081-2100). Additionally, the key enviromental factors for the habitat selection of Asiatic ibex were examined by combing the percentage contribution and permutation importance. The results showed that:(1) The prediction accuracy of MaxEnt model was high. The value of AUC (Area Under Curve) > 0.97, and TSS (True Skill Statistic) >0.90, which indicated that the modeling results was reliable. (2) Niche suitability of the Asiatic ibex was dominated by climate conditions (including precipitation seasonalit and Isothermaality), altitude and hunman activities (distance to springs and residential areas). As the basic element for the survival of Asiatic ibex, it control the habitat ecology and enviroment together with climate conditions. Moreover, influenced by the behaviors of Asiatic ibex, it always distributed in the regions with relatively higher altitude and aways from human activities. (3) In the baseline period, Asiatic ibex mainly distributed in the northern and southern regions with higher elevations, which occupied apprimixily 365.77 pa (accounting for 4.31% of the total area of AEDNNR). Saptially, the suitable area for Asiatic ibex in the north is larger than that in the south. The high and middle suitable habitats mian located in the northern AEDNNR. (4) In the future climate scenarios, the suitable areas appeared increasing trends by the eco-enviromental improvement and the implementation of endangered animanls protection measures. However, influenced by inbreeding of Asiatic ibex, the long-term growth of the numbers and suitable areas were not significant. The occupied area increased towards the relatively higher elevation southern regions.

Spatial prediction of shallow landslide: application of novel rotational forest-based reduced error pruning tree

Landslides are a form of soil erosion threatening the sustainability of some areas of the world. There is, therefore, a need to investigate landslide rates and behaviour. In this research, we introduced a novel hybrid artificial intelligence approach of rotation forest (RF) as a meta classifier based on reduced error pruning tree (REPTree) as a base classifier called RF-REPTree, for landslide susceptibility mapping (LSM) in the Kalaleh watershed, Golestan Province, Iran. Some benchmark models, including the open-source Java decision tree (J48), naive Bayes tree (NBTree), and REPTree were used to compare the designed model. A total of 249 landslide locations were identified and mapped. The group was split into training (70%) and testing (30%) data for modelling and reliability analysis. Based on a literature review and multi-collinearity tests, 16 landslide conditioning factors (LCFs) were selected. Of the LCFs, the topographical position index (TPI) had the highest correlation with landslide occurrence. The LSM produced by RF-REPTree revealed that nearly 29% of the study areas have high to very high landslide susceptibility (LS). Statistical analysis of the model results included the receiver operating characteristic curve (ROC), the efficiency test, the true skill statistic (TSS), and the kappa index. ROC demonstrated that the AUC values of RF-REPTree, REPTree, J48, and NBTree models were 0.832, 0.700, 0.695, and 0.759 for succession rate curves and 0.794, 0.740, 0.788, and 0.728 for prediction rate curves, respectively. Therefore, all models were judged to be acceptably accurate for LSM. Among the LS models, the RF-REPTree model achieved the highest accuracy, followed by REPTree, J48, and NBTree. The results of LSM can be used to target the mitigation of landslide hazards and provide a foundation for sustainable environmental planning.

Evaluation of re-sampling methods on performance of machine learning models to predict landslide susceptibility

This study tests the applicability of three resampling methods (i.e. bootstrapping, random-subsampling and cross-validation) for enhancing the performance of eight machine-learning models: boosted regression trees, flexible discriminant analysis, random forests, mixture discriminate analysis, multivariate adaptive regression splines, classification and regression trees, support vector machines and generalized linear models, compared to the use of the original data. The results of models were evaluated using correlation (COR), area under curve (AUC), true skill statistic (TSS), receiver-operating characteristic and the probability of detection (POD). The evaluation showed that the bootstrapping technique improved the performance of all models. The Bootstrapping-random forest (with COR = 0.75, AUC = 0.92, TSS = 0.80 and POD = 0.98) proved to be the best model for landslide prediction. Among the 18 contributing factors, distance from fault, curvature and precipitation were the most influential in all 32 models . Highlights Hazard prediction of landslide by the 8 machine-learning (ML) models. Multiple morphometric, climatic, geologic, vegetation and human factors were used. Tests the applicability of three resampling methods. The performance of the ML models and coupling models were assessed.

Using Support Vector Machine (SVM) and Ionospheric Total Electron Content (TEC) Data for Solar Flare Predictions

Predicting where and when space weather events such as solar flares and X-rays bursts are likely to occur in a specific area of interest constitutes a significant challenge in space weather research. Space weather scientists are, therefore, gradually exploring multivariate data analysis techniques from the fields of data mining or machine learning in order to approximate future occurrences of space weather events from past distribution patterns. As solar flares emit extreme ultraviolet and X-ray radiation, which leads to ionization effect in different layers of the ionosphere, most recent works related to solar flare predictions using machine learning (ML) techniques, focused on X-ray time series predictions. Here, we suggest using support vector machine for classifying subdaily and diurnal total electron content (TEC) spatial changes prior to solar flare events, in order to assess the possibility of predicting B, C, M, and X-class solar flare events. This is done as opposed to predicting TEC time series using ML techniques. The predictions are estimated up to three days before each tested class events, along with different skill scores such as precision, recall, Heidke skill score (HSS), accuracy, and true skill statistics. The results indicate that the suggested approach has the ability to predict solar flare events of X and M-class 24 h prior to their occurrence with 91% and 76% HSS skill scores, respectively, which improves over most recent related works. However, for the small-size C and B-class flares, the suggested approach does not succeed in producing similar promising results.

Combining Satellite Remote Sensing and Climate Data in Species Distribution Models to Improve the Conservation of Iberian White Oaks (Quercus L.)

The Iberian Peninsula hosts a high diversity of oak species, being a hot-spot for the conservation of European White Oaks (Quercus) due to their environmental heterogeneity and its critical role as a phylogeographic refugium. Identifying and ranking the drivers that shape the distribution of White Oaks in Iberia requires that environmental variables operating at distinct scales are considered. These include climate, but also ecosystem functioning attributes (EFAs) related to energy–matter exchanges that characterize land cover types under various environmental settings, at finer scales. Here, we used satellite-based EFAs and climate variables in species distribution models (SDMs) to assess how variables related to ecosystem functioning improve our understanding of current distributions and the identification of suitable areas for White Oak species in Iberia. We developed consensus ensemble SDMs targeting a set of thirteen oaks, including both narrow endemic and widespread taxa. Models combining EFAs and climate variables obtained a higher performance and predictive ability (true-skill statistic (TSS): 0.88, sensitivity: 99.6, specificity: 96.3), in comparison to the climate-only models (TSS: 0.86, sens.: 96.1, spec.: 90.3) and EFA-only models (TSS: 0.73, sens.: 91.2, spec.: 82.1). Overall, narrow endemic species obtained higher predictive performance using combined models (TSS: 0.96, sens.: 99.6, spec.: 96.3) in comparison to widespread oaks (TSS: 0.80, sens.: 92.6, spec.: 87.7). The Iberian White Oaks show a high dependence on precipitation and the inter-quartile range of Normalized Difference Water Index (NDWI) (i.e., seasonal water availability) which appears to be the most important EFA variable. Spatial projections of climate–EFA combined models contribute to identify the major diversity hotspots for White Oaks in Iberia, holding higher values of cumulative habitat suitability and species richness. We discuss the implications of these findings for guiding the long-term conservation of Iberian White Oaks and provide spatially explicit geospatial information about each oak species (or set of species) relevant for developing biogeographic conservation frameworks.

Mapping an Invasive Plant Spartina alterniflora by Combining an Ensemble One-Class Classification Algorithm with a Phenological NDVI Time-Series Analysis Approach in Middle Coast of Jiangsu, China

Spartina alterniflora (S. alterniflora) is one of the worst plant invaders in the coastal wetlands of China. Accurate and repeatable mapping of S. alterniflora invasion is essential to develop cost-effective management strategies for conserving native biodiversity. Traditional remote-sensing-based mapping methods require a lot of fieldwork for sample collection. Moreover, our ability to detect this invasive species is still limited because of poor spectral separability between S. alterniflora and its co-dominant native plants. Therefore, we proposed a novel scheme that uses an ensemble one-class classifier (EOCC) in combination with phenological Normalized Difference Vegetation Index (NDVI) time-series analysis (TSA) to detect S. alterniflora. We evaluated the performance of the EOCC algorithm in two scenarios, i.e., single-scene analysis (SSA) and NDVI-TSA in the core zones of Yancheng National Natural Reserve (YNNR). Meanwhile, a fully supervised classifier support vector machine (SVM) was tested in the two scenarios for comparison. With these scenarios, the crucial phenological stages and the advantage of phenological NDVI-TSA in S. alterniflora recognition were also investigated. Results indicated the EOCC using only positive training data performed similarly well with the SVM trained on complete training data in the YNNR. Moreover, the EOCC algorithm presented a more robust transferability with notably higher classification accuracy than the SVM when being transferred to a second site, without a second training. Furthermore, when combined with the phenological NDVI-TSA, the EOCC algorithm presented more balanced sensitivity–specificity result, showing slightly better transferability than it performed in the best phenological stage (i.e., senescence stage of November). The achieved results (overall accuracy (OA), Kappa, and true skill statistic (TSS) were 92.92%, 0.843, and 0.834 for the YNNR, and OA, Kappa, and TSS were 90.94%, 0.815, and 0.825 for transferability to the non-training site) suggest that our detection scheme has a high potential for the mapping of S. alterniflora across different areas, and the EOCC algorithm can be a viable alternative to traditional supervised classification method for invasive plant detection.

Mapping recreation and tourism use across grizzly bear recovery areas using social network data and maximum entropy modelling

Understanding biodiversity pressures associated with recreation and tourism is a major challenge for conservation planning and landscape management. While estimates of landscape use are often collected using mechanisms such as park entry fees and traffic density estimates, these data do not provide substantial detail about the spatial location or intensity of recreation and tourism across biodiversity management areas. To better predict patterns of recreation and tourism likelihood to support conservation planning, we used social network data from Facebook(™), Flickr(™), Google(™), Strava(™), and Wikilocs(™) along with a suite of remote-sensing-derived environmental covariates in a maximum entropy (MaxEnt) presence-only modelling framework. Social network samples were compiled and processed to reduce sampling bias and spatial autocorrelation. Road access, climate data, and remote sensing covariates describing vegetation greenness, disturbance, topography, and moisture were used as predictor variables in the MaxEnt modelling framework. Our focus site was a grizzly bear (Ursus arctos) management area in west-central Alberta, Canada. Individual models were developed for each social network dataset, as well as a combined model including all the samples . Mean cross-validated AUC, partial ROC, and true skill statistics (TSS) were used to evaluate model accuracy. Results indicated that the covariates proposed were able to best model Strava and Wikilocs activity (TSS = 0.69 and 0.50, respectively), while samples from Flickr or the combination of all social networks were least accurate (TSS = 0.32). The “access” covariate was most important for MaxEnt training gain across a number of social network models, highlighting the importance of access for recreation and tourism likelihood. The summer heat moisture index and normalized burn ratio were also useful spatial covariates in many predictions. Recreation and tourism likelihood maps were combined with grizzly bear telemetry data to examine how recreation and tourism may affect grizzly bear behaviour. All social network models found a similar influence on grizzly bear behaviour, with increasing recreation and tourism use resulting in decreased foraging behaviour and increased rapid movement, suggesting that the models developed here are useful tools for predicting grizzly bear behaviour and planning conservation strategies for the species.

Spatial thinning and class balancing: Key choices lead to variation in the performance of species distribution models with citizen science data

Abstract Spatial biases are a common feature of presence–absence data from citizen scientists. Spatial thinning can mitigate errors in species distribution models (SDMs) that use these data. When detections or non‐detections are rare, however, SDMs may suffer from class imbalance or low sample size of the minority (i.e. rarer) class. Poor predictions can result, the severity of which may vary by modelling technique. To explore the consequences of spatial bias and class imbalance in presence–absence data, we used eBird citizen science data for 102 bird species from the northeastern USA to compare spatial thinning, class balancing and majority‐only thinning (i.e. retaining all samples of the minority class). We created SDMs using two parametric or semi‐parametric techniques (generalized linear models and generalized additive models) and two machine learning techniques (random forest and boosted regression trees). We tested the predictive abilities of these SDMs using an independent and systematically collected reference dataset with a combination of discrimination (area under the receiver operator characteristic curve; true skill statistic; area under the precision‐recall curve) and calibration (Brier score; Cohen's kappa) metrics. We found large variation in SDM performance depending on thinning and balancing decisions. Across all species, there was no single best approach, with the optimal choice of thinning and/or balancing depending on modelling technique, performance metric and the baseline sample prevalence of species in the data. Spatially thinning all the data was often a poor approach, especially for species with baseline sample prevalence <0.1. For most of these rare species, balancing classes improved model discrimination between presence and absence classes using machine learning techniques, but typically hindered model calibration. Baseline sample prevalence, sample size, modelling approach and the intended application of SDM output—whether discrimination or calibration—should guide decisions about how to thin or balance data, given the considerable influence of these methodological choices on SDM performance. For prognostic applications requiring good model calibration (vis‐à‐vis discrimination), the match between sample prevalence and true species prevalence may be the overriding feature and warrants further investigation.

Accessing habitat suitability and connectivity for the westernmost population of Asian black bear (Ursus thibetanus gedrosianus, Blanford, 1877) based on climate changes scenarios in Iran.

Climate change, as an emerging phenomenon, has led to changes in the distribution, movement, and even risk of extinction of various wildlife species and this has raised concerns among conservation biologists. Different species have two options in the face of climate change, either to adopt or follow their climatic niche to new places through the connectivity of habitats. The modeling of interpatch landscape communications can serve as an effective decision support tool for wildlife managers. This study was conducted to assess the effects of climate change on the distribution and habitat connectivity of the endangered subspecies of Asian black bear (Ursus thibetanus gedrosianus) in the southern and southeastern Iran. The presence points of the species were collected in Provinces of Kerman, Hormozgan, and Sistan-Baluchestan. Habitat modeling was done by the Generalized Linear Model, and 3 machine learning models including Maximum Entropy, Back Propagation based artificial Neural Network, and Support Vector Machine. In order to achieve the ensemble model, the results of the mentioned models were merged based on the method of "accuracy rate as weight" derived from their validation. To construct pseudo-absence points for the use in the mentioned models, the Ensemble model of presence-only models was used. The modeling was performed using 15 habitat variables related to climatic, vegetation, topographic, and anthropogenic parameters. The three general circulation models of BCC-CSM1, CCSM4, and MRI-CGCM3 were selected under the two scenarios of RCP2.6 and RCP8.5 by 2070. To investigate the effect of climate change on the habitat connections, the protected areas of 3 provinces were considered as focal nodes and the connections between them were established based on electrical circuit theory and Pairwise method. The true skill statistic was employed to convert the continuous suitability layers to binary suitable/unsuitable range maps to assess the effectiveness of the protected areas in the coverage of suitable habitats for the species. Due to the high power of the stochastic forest model in determining the importance of variables, this method was used. The results showed that presence/absence models were successful in the implementation and well distinguished the points of presence and pseudo-absence from each other. Based on the random forests model, the variables of Precipitation of Driest Quarter, Precipitation of Coldest Quarter, and Temperature Annual Range have the greatest impact on the habitat suitability. Comparing the modeling findings to the realities of the species distribution range indicated that the suitable habitats are located in areas with high humidity and rainfall, which are mostly in the northern areas of Bandar Abbas, south of Kerman, and west and south of Sistan-Baluchestan. The area of suitable habitats, in the MRI-CGCM3 (189731 Km2) and CCSM4 (179007 Km2) models under the RCP2.6 scenario, is larger than the current distribution (174001 Km2). However, in terms of the performance of protected areas, the optimal coverage of the species by the boundary of the protected areas, under each of the RCP2.6 and RCP8.5 scenarios, is less than the present time. According to the electric circuit theory, connecting the populations in the protected areas of Sistan-Baluchestan province to those in the northern Hormozgan and the southern Kerman would be based on the crossing through the heights of Sistan-Baluchestan and Hormozgan provinces and the plains between these heights would be the movement pinch points under the current and future scenarios. Populations in the protected areas of Kerman have higher quality patch connections than that of the other two provinces. The areas such as Sang-e_Mes, Kouh_Shir, Zaryab, and Bahr_Aseman in Kerman Province and Kouhbaz and Geno in Hormozgan Province can provide suitable habitats for the species in the distribution models. The findings revealed that the conservation of the heights along with the caves inside them could be a protective priority to counteract the effects of climate change on the species.

Implementation of Artificial Intelligence Based Ensemble Models for Gully Erosion Susceptibility Assessment

The Rarh Bengal region in West Bengal, particularly the eastern fringe area of the Chotanagpur plateau, is highly prone to water-induced gully erosion. In this study, we analyzed the spatial patterns of a potential gully erosion in the Gandheswari watershed. This area is highly affected by monsoon rainfall and ongoing land-use changes. This combination causes intensive gully erosion and land degradation. Therefore, we developed gully erosion susceptibility maps (GESMs) using the machine learning (ML) algorithms boosted regression tree (BRT), Bayesian additive regression tree (BART), support vector regression (SVR), and the ensemble of the SVR-Bee algorithm. The gully erosion inventory maps are based on a total of 178 gully head-cutting points, taken as the dependent factor, and gully erosion conditioning factors, which serve as the independent factors. We validated the ML model results using the area under the curve (AUC), accuracy (ACC), true skill statistic (TSS), and Kappa coefficient index. The AUC result of the BRT, BART, SVR, and SVR-Bee models are 0.895, 0.902, 0.927, and 0.960, respectively, which show very good GESM accuracies. The ensemble model provides more accurate prediction results than any single ML model used in this study.

A Gray‐Box Model for a Probabilistic Estimate of Regional Ground Magnetic Perturbations: Enhancing the NOAA Operational Geospace Model With Machine Learning

AbstractWe present a novel algorithm that predicts the probability that the time derivative of the horizontal component of the ground magnetic field dB/dt exceeds a specified threshold at a given location. This quantity provides important information that is physically relevant to geomagnetically induced currents (GICs), which are electric currents associated with sudden changes in the Earth's magnetic field due to space weather events. The model follows a “gray‐box” approach by combining the output of a physics‐based model with machine learning. Specifically, we combine the University of Michigan's Geospace model that is operational at the National Oceanic and Atmospheric Administration (NOAA) Space Weather Prediction Center, with a boosted ensemble of classification trees. We discuss the problem of recalibrating the output of the decision tree to obtain reliable probabilities. The performance of the model is assessed by typical metrics for probabilistic forecasts: Probability of Detection and False Detection, True Skill Statistic, Heidke Skill Score, and Receiver Operating Characteristic curve. We show that the ML‐enhanced algorithm consistently improves all the metrics considered.

Application and Interpretation of Deep Learning for Identifying Pre-emergence Magnetic Field Patterns

Abstract Magnetic flux generated within the solar interior emerges to the surface, forming active regions (ARs) and sunspots. Flux emergence may trigger explosive events—such as flares and coronal mass ejections, and therefore understanding emergence is useful for space-weather forecasting. Evidence of any pre-emergence signatures will also shed light on subsurface processes responsible for emergence. In this paper, we present a first analysis of EARs from the Solar Dynamics Observatory/Helioseismic Emerging Active Regions dataset using deep convolutional neural networks (CNN) to characterize pre-emergence surface magnetic field properties. The trained CNN classifies between pre-emergence line-of-sight magnetograms and a control set of nonemergence magnetograms with a true skill statistic (TSS) score of approximately 85% about 3 hr prior to emergence and approximately 40% about 24 hr prior to emergence. Our results are better than a baseline classification TSS obtained using discriminant analysis (DA) of only the unsigned magnetic flux, although a multivariable DA produces TSS values consistent with the CNN. We develop a network-pruning algorithm to interpret the trained CNN and show that the CNN incorporates filters that respond positively as well as negatively to the unsigned magnetic flux of the magnetograms. Using synthetic magnetograms, we demonstrate that the CNN output is sensitive to the length scale of the magnetic regions, with small-scale and intense fields producing maximum CNN output and possibly a characteristic pre-emergence pattern. Given increasing popularity of deep learning, the techniques developed here to interpret the trained CNN—using network pruning and synthetic data—are relevant for future applications in solar and astrophysical data analysis.

Modeling Spatial Flood using Novel Ensemble Artificial Intelligence Approaches in Northern Iran

The uncertainty of flash flood makes them highly difficult to predict through conventional models. The physical hydrologic models of flash flood prediction of any large area is very difficult to compute as it requires lot of data and time. Therefore remote sensing data based models (from statistical to machine learning) have become highly popular due to open data access and lesser prediction times. There is a continuous effort to improve the prediction accuracy of these models through introducing new methods. This study is focused on flash flood modeling through novel hybrid machine learning models, which can improve the prediction accuracy. The hybrid machine learning ensemble approaches that combine the three meta-classifiers (Real AdaBoost, Random Subspace, and MultiBoosting) with J48 (a tree-based algorithm that can be used to evaluate the behavior of the attribute vector for any defined number of instances) were used in the Gorganroud River Basin of Iran to assess flood susceptibility (FS). A total of 426 flood positions as dependent variables and a total of 14 flood conditioning factors (FCFs) as independent variables were used to model the FS. Several threshold-dependent and independent statistical tests were applied to verify the performance and predictive capability of these machine learning models, such as the receiver operating characteristic (ROC) curve of the success rate curve (SRC) and prediction rate curve (PRC), efficiency (E), root-mean square-error (RMSE), and true skill statistics (TSS). The valuation of the FCFs was done using AdaBoost, frequency ratio (FR), and Boosted Regression Tree (BRT) models. In the flooding of the study area, altitude, land use/land cover (LU/LC), distance to stream, normalized differential vegetation index (NDVI), and rainfall played important roles. The Random Subspace J48 (RSJ48) ensemble method with an area under the curve (AUC) of 0.931 (SRC), 0.951 (PRC), E of 0.89, sensitivity of 0.87, and TSS of 0.78, has become the most effective ensemble in predicting the FS. The FR technique also showed good performance and reliability for all models. Map removal sensitivity analysis (MRSA) revealed that the FS maps have the highest sensitivity to elevation. Based on the findings of the validation methods, the FS maps prepared using the machine learning ensemble techniques have high robustness and can be used to advise flood management initiatives in flood-prone areas.

On the importance of predictor choice, modelling technique, and number of pseudo-absences for bioclimatic envelope model performance.

Bioclimatic envelope models are commonly used to assess the influence of climate change on species' distributions and biodiversity patterns. Understanding how methodological choices influence these models is critical for a comprehensive evaluation of the estimated impacts. Here we systematically assess the performance of bioclimatic envelope models in relation to the selection of predictors, modeling technique, and pseudo‐absences. We considered (a) five different predictor sets, (b) seven commonly used modeling techniques and an ensemble model, and (c) three sets of pseudo‐absences (1,000 pseudo‐absences, 10,000 pseudo‐absences, and the same as the number of presences). For each combination of predictor set, modeling technique, and pseudo‐absence set, we fitted bioclimatic envelope models for 300 species of mammals, amphibians, and freshwater fish, and evaluated the predictive performance of the models using the true skill statistic (TSS), based on a spatially independent test set as well as cross‐validation. On average across the species, model performance was mostly influenced by the choice of predictor set, followed by the choice of modeling technique. The number of the pseudo‐absences did not have a strong effect on the model performance. Based on spatially independent testing, ensemble models based on species‐specific nonredundant predictor sets revealed the highest predictive performance. In contrast, the Random Forest technique yielded the highest model performance in cross‐validation but had the largest decrease in model performance when transferred to a different spatial context, thus highlighting the need for spatially independent model evaluation. We recommend building bioclimatic envelope models according to an ensemble modeling approach based on a nonredundant set of bioclimatic predictors, preferably selected for each modeled species.

Climate Change May Imperil Tea Production in the Four Major Tea Producers According to Climate Prediction Models

The threat of accelerating climate change on species distribution now and in the future is a topic of increasing research interest. However, little work has been undertaken to assess how shifting climates will affect the suitability of tea cultivation. Therefore, we used MaxEnt modelling to project the impact of current and future climatic scenarios on the potential distribution of tea across the four tea-producing countries of China, India, Kenya and Sri Lanka. Projections were made for the years 2050 and 2070 with three Representative Concentration Pathways (RCPs) using seven bioclimatic predictors under three global circulation models (GCMs). The current and future habitat suitability for tea predicted by the models produced a high accuracy rate, with high areas under the receiver operating characteristic curve (AUCs) for all tested RCPs under the three GCMs for the four countries. The mean true skill statistic (TSS) values for tea in Sri Lanka, Kenya, India and China were 0.80, 0.91, 0.91, and 0.74, respectively. The kappa values (k) of the current and future models for all four countries ranged from 0.40 to 0.75, which indicates that the overall performance of the model was good. The precipitation seasonality and annual precipitation were found to be the most influential variables in Sri Lanka and India, respectively, while annual mean temperature was the most effective contributor for determining the suitability of habitat for tea in Kenya and China. An important proviso is that some existing tea-growing areas will face reduced suitability for future tea cultivation suggesting that by 2050 there will be a drastic reduction in the optimal suitability by averages of 26.2%, 14%, and 4.7% in Kenya, Sri Lanka and China, respectively. The optimal suitability will be reduced by 15.1%, 28.6% and 2.6% in Kenya, Sri Lanka and China, respectively, by 2070. India displays an advantage in projected future climates as it gains optimal suitability areas of 15% by 2050 and 25% by 2070.