The effects of model and data complexity on predictions from species distributions models

Modeling habitat suitability for the lesser-known populations of endangered mountain nyala (Tragelaphus buxtoni) in the Arsi and Ahmar Mountains, Ethiopia

Habitat suitability models have become a valuable tool for wildlife conservation and management, and are frequently used to better understand the range and habitat requirements of rare and endangered species. In this study, we employed two habitat suitability modeling techniques, namely Boosted Regression Tree (BRT) and Maximum Entropy (Maxent) models, to identify potential suitable habitats for the endangered mountain nyala (Tragelaphus buxtoni) and environmental factors affecting its distribution in the Arsi and Ahmar Mountains of Ethiopia. Presence points, used to develop our habitat suitability models, were recorded from fecal pellet counts (n = 130) encountered along 196 randomly established transects in 2015 and 2016. Predictor variables used in our models included major landcover types, Normalized Difference Vegetation Index (NDVI), greenness and wetness tasseled cap vegetation indices, elevation, and slope. Area Under the Curve model evaluations for BRT and Maxent were 0.96 and 0.95, respectively, demonstrating high performance. Both models were then ensembled into a single binary output highlighting an area of agreement. Our results suggest that 1864 km2 (9.1%) of the 20,567 km2 study area is suitable habitat for the mountain nyala with land cover types, elevation, NDVI, and slope of the terrain being the most important variables for both models. Our results highlight the extent to which habitat loss and fragmentation have disconnected mountain nyala subpopulations. Our models demonstrate the importance of further protecting suitable habitats for mountain nyala to ensure the species' conservation.

Data error propagation in stacked bioclimatic envelope models

Stacking is the process of overlaying inferred species potential distributions for multiple species based on outputs of bioclimatic envelope models (BEMs). The approach can be used to investigate patterns and processes of species richness. If data limitations on individual species distributions are inevitable, but how do they affect inferences of patterns and processes of species richness? We investigate the influence of different data sources on estimated species richness gradients in China. We fitted BEMs using species distributions data for 334 bird species obtained from (1) global range maps, (2) regional checklists, (3) museum records and surveys, and (4) citizen science data using presence-only (Mahalanobis distance), presence-background (MAXENT), and presence-absence (GAM and BRT) BEMs. Individual species predictions were stacked to generate species richness gradients. Here, we show that different data sources and BEMs can generate spatially varying gradients of species richness. The environmental predictors that best explained species distributions also differed between data sources. Models using citizen-based data had the highest accuracy, whereas those using range data had the lowest accuracy. Potential richness patterns estimated by GAM and BRT models were robust to data uncertainty. When multiple data sets exist for the same region and taxa, we advise that explicit treatments of uncertainty, such as sensitivity analyses of the input data, should be conducted during the process of modeling.

Filling the data gaps: Transferring models from data-rich to data-poor deep-sea areas to support spatial management

Spatial management of the deep sea is challenging due to limited available data on the distribution of species and habitats to support decision making. In the well-studied North Atlantic, predictive models of species distribution and habitat suitability have been used to fill data gaps and support sustainable management. In the South Atlantic and other poorly studied regions, this is not possible due to a massive lack of data. In this study, we investigated whether models constructed in data-rich areas can be used to inform data-poor regions (with otherwise similar environmental conditions). We used a novel model transfer approach to identify to what extent a habitat suitability model for Desmophyllum pertusum reef, built in a data-rich basin (North Atlantic), could be transferred usefully to a data-poor basin (South Atlantic). The transferred model was built using the Maximum Entropy algorithm and constructed with 227 presence and 3064 pseudo-absence points, and 200 m resolution environmental grids. Performance in the transferred region was validated using an independent dataset of D. pertusum presences and absences, with assessments made using both threshold-dependent and -independent metrics. We found that a model for D. pertusum reef fitted to North Atlantic data transferred reasonably well to the South Atlantic basin, with an area under the curve of 0.70. Suitable habitat for D. pertusum reef was predicted on 20 of the assessed 27 features including seamounts. Nationally managed Marine Protected Areas provide significant protection for D. pertusum reef habitat in the region, affording full protection from bottom trawling to 14 of the 20 suitable features. In areas beyond national jurisdiction (ABNJ), we found four seamounts that provided suitable habitat for D. pertusum reef to be at least partially protected from bottom trawling, whilst two did not fall within fisheries closures. There are factors to consider when developing models for transfer including data resolution and predictor type. Nevertheless, the promising results of this application demonstrate that model transfer approaches stand to provide significant contributions to spatial planning processes through provision of new, best available data. This is particularly true for ABNJ and areas that have previously undergone little scientific exploration such as the global south.

Species distribution models for predicting the habitat suitability of Chinese fire-bellied newt Cynops orientalis under climate change

Climate change influences species geographical distribution and diversity pattern. The Chinese fire-bellied newt (Cynops orientalis) is an endemic species distributed in East-central China, which has been classified as near-threatened species recently due to habitat destruction and degradation and illegal trade in the domestic and international pet markets. So far, little is known about the spatial distribution of the species. Based on bioclimatic data of the current and future climate projections, we modeled the change in suitable habitat for C. orientalis by ten algorithms, evaluated the importance of environmental factors in shaping their distribution, and identified distribution shifts under climate change scenarios. In this study, 46 records of C. orientalis from East China and 8 bioclimatic variables were used. Among the ten modeling algorithms, four (GAM, GBM, Maxent, and RF) were selected according to their predictive abilities. The current habitat suitability showed that C. orientalis had a relatively wide but fragmented distribution, and it encompassed 41,862 km2. The models suggested that precipitation of warmest quarter (bio18) and mean temperature of wettest quarter (bio6) had the highest contribution to the model. This study revealed that C. orientalis is sensitive to climate change, which will lead to a large range shift. The projected spatial and temporal pattern of range shifts for C. orientalis should provide a useful reference for implementing long-term conservation and management strategies for amphibians in East China.

Supervised Machine Learning for Estimation of Total Suspended Solids in Urban Watersheds

Machine Learning (ML) algorithms provide an alternative for the prediction of pollutant concentration. We compared eight ML algorithms (Linear Regression (LR), uniform weighting k-Nearest Neighbor (UW-kNN), variable weighting k-Nearest Neighbor (VW-kNN), Support Vector Regression (SVR), Artificial Neural Network (ANN), Regression Tree (RT), Random Forest (RF), and Adaptive Boosting (AdB)) to evaluate the feasibility of ML approaches for estimation of Total Suspended Solids (TSS) using the national stormwater quality database. Six factors were used as features to train the algorithms with TSS concentration as the target parameter: Drainage area, land use, percent of imperviousness, rainfall depth, runoff volume, and antecedent dry days. Comparisons among the ML methods demonstrated a higher degree of variability in model performance, with the coefficient of determination (R2) and Nash–Sutcliffe (NSE) values ranging from 0.15 to 0.77. The Root Mean Square (RMSE) values ranged from 110 mg/L to 220 mg/L. The best fit was obtained using the AdB and RF models, with R2 values of 0.77 and 0.74 in the training step and 0.67 and 0.64 in the prediction step. The NSE values were 0.76 and 0.72 in the training step and 0.67 and 0.62 in the prediction step. The predictions from AdB were sensitive to all six factors. However, the sensitivity level was variable.

Comparison of two commonly used methods for identifying water quality thresholds in freshwater ecosystems using field and synthetic data

Defining ecological thresholds has become increasingly relevant for water resource management. Despite the fact that there has been a rapid expansion in methods to evaluate ecological threshold responses to environmental stressors, evaluation of the relative benefits of various methods has received less attention. This study compares the performance of Gradient Forest (GF) and Threshold Indicator Taxa Analysis (TITAN) for identifying water quality thresholds in both field and synthetic data. Analysis of 14 years of macroinvertebrates data from the Mediterranean catchments of the Torrens and Onkaparinga Rivers, South-Australia, identified electrical conductivity (EC) and total phosphorus (TP) as the most important water quality variables affecting macroinvertebrates. Water quality thresholds for macroinvertebrates identified by both methods largely corresponded at low EC (GF: 400-900 μS cm^-1 vs. TITAN: 407-951 μScm^-1), total phosphorus (TP) (GF: 0.02-0.18 mg L^-1 vs. TITAN: 0.02-0.04 mg L^-1) and total nitrogen (TN) (GF: 0.2 mg L^-1 vs. TITAN: 0.28-0.67 mg L^-1) concentrations. However, multiple GF-derived thresholds, particularly at high stressor concentrations, were representative of low data distribution, and thus need to be considered with caution. In another case study of South Australian diatom data, there were marked differences in GF and TITAN identified thresholds for EC (GF: 5000 μScm^-1 vs. TITAN 1004-2440 μS cm^-1) and TP (GF: 250-500 μg L^-1 vs. TITAN: 11-329 μg L^-1). These differences were due to the fact that while TITAN parsed species responses into negative and positive taxa, GF overestimated thresholds by aggregating the response of taxa that increase and decrease along environmental gradients. Given these findings, we also evaluated the methods' performance using different distributions of synthetic data i.e. with both skewed and uniform distribution of samples and species responses. Both methods identified similar change-points in the case of a uniform environmental gradient, except when species optima were simulated at centre of the gradient. Here GF detected the change-points but TITAN failed to do so. GF also outperformed TITAN when four simulated species change-points were present. Thus, the distribution of species responses and optima and the evenness of the environment gradient can affect the models' performance. This study has shown that both methods are robust in identifying change in species response but threshold identification differs depending both on the analysis used and the nature of ecological data. We recommend the careful application of GF and TITAN, noting these differences in performance, will improve their application for water resource management.

Real-Time Identification of Irrigation Water Pollution Sources and Pathways with a Wireless Sensor Network and Blockchain Framework

Real-time identification of irrigation water pollution sources and pathways (PSP) is crucial to ensure both environmental and food safety. This study uses an integrated framework based on the Internet of Things (IoT) and the blockchain technology that incorporates a directed acyclic graph (DAG)-configured wireless sensor network (WSN), and GIS tools for real-time water pollution source tracing. Water quality sensors were installed at monitoring stations in irrigation channel systems within the study area. Irrigation water quality data were delivered to databases via the WSN and IoT technologies. Blockchain and GIS tools were used to trace pollution at mapped irrigation units and to spatially identify upstream polluted units at irrigation intakes. A Water Quality Analysis Simulation Program (WASP) model was then used to simulate water quality by using backward propagation and identify potential pollution sources. We applied a “backward pollution source tracing” (BPST) process to successfully and rapidly identify electrical conductivity (EC) and copper (Cu²⁺) polluted sources and pathways in upstream irrigation water. With the BPST process, the WASP model effectively simulated EC and Cu²⁺ concentration data to identify likely EC and Cu²⁺ pollution sources. The study framework is the first application of blockchain technology for effective real-time water quality monitoring and rapid multiple PSPs identification. The pollution event data associated with the PSP are immutable.

Mapping suitability for rice production in inland valley landscapes in Benin and Togo using environmental niche modeling

Inland valleys (IVs) in Africa are important landscapes for rice cultivation and are targeted by national governments to attain self-sufficiency. Yet, there is limited information on the spatial distribution of IVs suitability at the national scale. In the present study, we developed an ensemble model approach to characterize the IVs suitability for rainfed lowland rice using 4 machine learning algorithms based on environmental niche modeling (ENM) with presence-only data and background sample, namely Boosted Regression Tree (BRT), Generalized Linear Model (GLM), Maximum Entropy (MAXNT) and Random Forest (RF). We used a set of predictors that were grouped under climatic variables, agricultural water productivity and soil water content, soil chemical properties, soil physical properties, vegetation cover, and socio-economic variables. The Area Under the Curves (AUC) evaluation metrics for both training and testing were respectively 0.999 and 0.873 for BRT, 0.866 and 0.816 for GLM, 0.948 and 0.861 for MAXENT and 0.911 and 0.878 for RF. Results showed that proximity of inland valleys to roads and urban centers, elevation, soil water holding capacity, bulk density, vegetation index, gross biomass water productivity, precipitation of the wettest quarter, isothermality, annual precipitation, and total phosphorus among others were major predictors of IVs suitability for rainfed lowland rice. Suitable IVs areas were estimated at 155,000-225,000 Ha in Togo and 351,000-406,000 Ha in Benin. We estimated that 53.8% of the suitable IVs area is needed in Togo to attain self-sufficiency in rice while 60.1% of the suitable IVs area is needed in Benin to attain self-sufficiency in rice. These results demonstrated the effectiveness of an ensemble environmental niche modeling approach that combines the strengths of several models.

Concealed truth: Modeling reveals unique Quaternary distribution dynamics and refugia of four related endemic keystone Abies taxa on the Tibetan Plateau

Understanding the factors driving the Quaternary distribution of Abies in the Tibetan Plateau (TP) is crucial for biodiversity conservation and for predicting future anthropogenic impacts on ecosystems. Here, we collected Quaternary paleo-, palynological, and phylogeographical records from across the TP and applied ecological niche models (ENMs) to obtain a profound understanding of the different adaptation strategies and distributional changes in Abies trees in this unique area. We identified environmental variables affecting the different historical biogeographies of four related endemic Abies taxa and rebuilt their distribution patterns over different time periods, starting from the late Pleistocene. In addition, modeling and phylogeographic results were used to predict suitable refugia for Abies forrestii, A. forrestii var. georgei, A. fargesii var. faxoniana, and A. recurvata. We supplemented the ENMs by investigating pollen records and diversity patterns of cpDNA for them. The overall reconstructed distributions of these Abies taxa were dramatically different when the late Pleistocene was compared with the present. All Abies taxa gradually receded from the south toward the north in the last glacial maximum (LGM). The outcomes showed two well-differentiated distributions: A. fargesii var. faxoniana and A. recurvata occurred throughout the Longmen refuge, a temporary refuge for the LGM, while the other two Abies taxa were distributed throughout the Heqing refuge. Both the seasonality of precipitation and the mean temperature of the driest quarter played decisive roles in driving the distribution of A. fargesii var. faxoniana and A. recurvata, respectively; the annual temperature range was also a key variable that explained the distribution patterns of the other two Abies taxa. Different adaptation strategies of trees may thus explain the differing patterns of distribution over time at the TP revealed here for endemic Abies taxa.

Standards for distribution models in biodiversity assessments

Demand for models in biodiversity assessments is rising, but which models are adequate for the task? We propose a set of best-practice standards and detailed guidelines enabling scoring of studies based on species distribution models for use in biodiversity assessments. We reviewed and scored 400 modeling studies over the past 20 years using the proposed standards and guidelines. We detected low model adequacy overall, but with a marked tendency of improvement over time in model building and, to a lesser degree, in biological data and model evaluation. We argue that implementation of agreed-upon standards for models in biodiversity assessments would promote transparency and repeatability, eventually leading to higher quality of the models and the inferences used in assessments. We encourage broad community participation toward the expansion and ongoing development of the proposed standards and guidelines.

Predicting the potential distribution of the endangered red panda across its entire range using MaxEnt modeling

An upsurge in anthropogenic impacts has hastened the decline of the red panda (Ailurus fulgens). The red panda is a global conservation icon, but holistic conservation management has been hampered by research being restricted to certain locations and population clusters. Building a comprehensive potential habitat map for the red panda is imperative to advance the conservation effort and ensure coordinated management across international boundaries. Here, we use occurrence records of both subspecies of red pandas from across their entire range to build a habitat model using the maximum entropy algorithm (MaxEnt 3.3.3k) and the least correlated bioclimatic variables. We found that the subspecies have separate climatic spaces dominated by temperature-associated variables in the eastern geographic distribution limit and precipitation-associated variables in the western distribution limit. Annual precipitation (BIO12) and maximum temperature in the warmest months (BIO5) were major predictors of habitat suitability for A. f. fulgens and A. f. styani, respectively. Our model predicted 134,975 km2 of red panda habitat based on 10 percentile thresholds in China (62% of total predicted habitat), Nepal (15%), Myanmar (9%), Bhutan (9%), and India (5%). Existing protected areas (PAs) encompass 28% of red panda habitat, meaning the PA network is currently insufficient and alternative conservation mechanisms are needed to protect the habitat. Bhutan's PAs provide good coverage for the red panda habitat. Furthermore, large areas of habitat were predicted in cross-broader areas, and transboundary conservation will be necessary.

How complex should models be? Comparing correlative and mechanistic range dynamics models

Criticism has been levelled at climate-change-induced forecasts of species range shifts that do not account explicitly for complex population dynamics. The relative importance of such dynamics under climate change is, however, undetermined because direct tests comparing the performance of demographic models vs. simpler ecological niche models are still lacking owing to difficulties in evaluating forecasts using real-world data. We provide the first comparison of the skill of coupled ecological-niche-population models and ecological niche models in predicting documented shifts in the ranges of 20 British breeding bird species across a 40-year period. Forecasts from models calibrated with data centred on 1970 were evaluated using data centred on 2010. We found that more complex coupled ecological-niche-population models (that account for dispersal and metapopulation dynamics) tend to have higher predictive accuracy in forecasting species range shifts than structurally simpler models that only account for variation in climate. However, these better forecasts are achieved only if ecological responses to climate change are simulated without static snapshots of historic land use, taken at a single point in time. In contrast, including both static land use and dynamic climate variables in simpler ecological niche models improve forecasts of observed range shifts. Despite being less skilful at predicting range changes at the grid-cell level, ecological niche models do as well, or better, than more complex models at predicting the magnitude of relative change in range size. Therefore, ecological niche models can provide a reasonable first approximation of the magnitude of species' potential range shifts, especially when more detailed data are lacking on dispersal dynamics, demographic processes underpinning population performance, and change in land cover.

Uncertainty of future projections of species distributions in mountainous regions

Multiple factors introduce uncertainty into projections of species distributions under climate change. The uncertainty introduced by the choice of baseline climate information used to calibrate a species distribution model and to downscale global climate model (GCM) simulations to a finer spatial resolution is a particular concern for mountainous regions, as the spatial resolution of climate observing networks is often insufficient to detect the steep climatic gradients in these areas. Using the maximum entropy (MaxEnt) modeling framework together with occurrence data on 21 understory bamboo species distributed across the mountainous geographic range of the Giant Panda, we examined the differences in projected species distributions obtained from two contrasting sources of baseline climate information, one derived from spatial interpolation of coarse-scale station observations and the other derived from fine-spatial resolution satellite measurements. For each bamboo species, the MaxEnt model was calibrated separately for the two datasets and applied to 17 GCM simulations downscaled using the delta method. Greater differences in the projected spatial distributions of the bamboo species were observed for the models calibrated using the different baseline datasets than between the different downscaled GCM simulations for the same calibration. In terms of the projected future climatically-suitable area by species, quantification using a multi-factor analysis of variance suggested that the sum of the variance explained by the baseline climate dataset used for model calibration and the interaction between the baseline climate data and the GCM simulation via downscaling accounted for, on average, 40% of the total variation among the future projections. Our analyses illustrate that the combined use of gridded datasets developed from station observations and satellite measurements can help estimate the uncertainty introduced by the choice of baseline climate information to the projected changes in species distribution.

Predicting the distributions of Egypt's medicinal plants and their potential shifts under future climate change

Climate change is one of the most difficult of challenges to conserving biodiversity, especially for countries with few data on the distributions of their taxa. Species distribution modelling is a modern approach to the assessment of the potential effects of climate change on biodiversity, with the great advantage of being robust to small amounts of data. Taking advantage of a recently validated dataset, we use the medicinal plants of Egypt to identify hotspots of diversity now and in the future by predicting the effect of climate change on the pattern of species richness using species distribution modelling. Then we assess how Egypt's current Protected Area network is likely to perform in protecting plants under climate change. The patterns of species richness show that in most cases the A2a 'business as usual' scenario was more harmful than the B2a 'moderate mitigation' scenario. Predicted species richness inside Protected Areas was higher than outside under all scenarios, indicating that Egypt's PAs are well placed to help conserve medicinal plants.