Stephen B. Stewart

Parthenium weed (Parthenium hysterophorus L.) is one of the most noxious and fast-spreading invasive alien species, posing a major threat to ecosystems, agriculture, and public health worldwide. Mechanistic and correlative species distribution models are commonly employed to determine the potential habitat suitability of parthenium weed. However, a comparative analysis of these two approaches for parthenium weed is lacking, leaving a gap in understanding their relative effectiveness and ability to describe habitat suitability of parthenium weed. This study compared the mechanistic model CLIMEX with random forest (RF), the best-performing of a suite of correlative models. When compared against occurrence records and pseudo-absences, measured by area under the receiver operating characteristic curve, true skill statistic, sensitivity, and specificity, the results revealed higher performance of RF compared to CLIMEX. Globally, RF predicted 7 million km² (2% of the total land mass) as suitable for parthenium weed, while CLIMEX predicted 20 million km² (13%). Based on binary maps, RF and CLIMEX identified 67 and 20 countries as suitable, respectively. For Bhutan, globally trained RF predicted 8919 km² (23% of the country’s total 38,394 km²) as currently suitable, with high suitability in the southern, west–central, central, and eastern districts, particularly along major highways. For the future, the 10 general circulation models downscaled to Bhutan showed a decrease in suitability across four scenarios (SSP126, SSP245, SSP370, SSP585) and three periods (2021–2050, 2051–2080, 2071–2100), with a northward shift in suitable habitats ranging from 2 to 76 km. Additionally, 2049 (23%) km² of agricultural land is currently at risk of being invaded by parthenium weed. Correlative and mechanistic models are based on different niche concepts (i.e., realized and fundamental, respectively), and therefore combining them can provide a better understanding of actual and potential species distributions. Given the high suitability of parthenium weed under the current climate and its potential negative impacts in Bhutan, early action such as early detection and control of infested areas, regular survey and monitoring, and creating public awareness are proposed as risk mitigation strategies.

Near-surface air temperature is an essential climate variable for the study of many biophysical phenomena, yet is often only available as a daily mean or extrema (minimum, maximum). While many applications require sub-diurnal dynamics, temporal interpolation methods have substantial limitations and atmospheric reanalyses are complex models that typically have coarse spatial resolution and may only be periodically updated. To overcome these issues, we developed an hourly air temperature product for Australia with spatial interpolation of hourly observations from 621 stations between 1990 and 2019. The model was validated with hourly observations from 28 independent stations, compared against empirical temporal interpolation methods, and both regional (BARRA-R) and global (ERA5-Land) reanalysis outputs. We developed a time-varying (i.e., time-of-day and day-of-year) coastal distance index that corresponds to the known dynamics of sea breeze systems, improving interpolation performance by up to 22.4% during spring and summer in the afternoon and evening hours. Cross-validation and independent validation (n = 24/4 OzFlux/CosmOz field stations) statistics of our hourly output showed performance that was comparable with contemporary Australian interpolations of daily air temperature extrema (climatology/hourly/validation: R² = 0.99/0.96/0.92, RMSE = 0.75/1.56/1.78 °C, Bias = -0.00/0.00/-0.03 °C). Our analyses demonstrate the limitations of temporal interpolation of daily air temperature extrema, which can be biased due to the inability to represent frontal systems and assumptions regarding rates of temperature change and the timing of minimum and maximum air temperature. Spatially interpolated hourly air temperature compared well against both BARRA-R and ERA5-Land, and performed better than both reanalyses when evaluated against the 28 independent validation stations. Our research demonstrates that spatial interpolation of sub-diurnal meteorological fields, such as air temperature, can mitigate the limitations of alternative data sources for studies of near-surface phenomena and plays an important ongoing role in supporting numerous scientific applications.

Deer populations and their impacts on forest ecosystems are increasing globally. Given the imperative and expense to mitigate impacts of invasive deer, we aimed to elucidate critical drivers of (i) deer density, (ii) deer impacts, and (iii) the relationship between them, to facilitate targeted management. We used quantile regression forests to model deer density (faecal pellet counts at 1948 locations) and impacts (browsing and other impacts on > 23,000 woody plants at 343 locations) across a mosaic of agricultural and forested ecosystems in Victoria, Australia (12,775 km²). Climate, topography, vegetation cover, and distance to water features were included as model covariates. Modelled deer density (r² = 0.71, MAE = 0.56 pellets/m²) was most influenced by distance to waterbodies (> 10 ha, 31.2%), elevation (14.3%) and woody vegetation cover (12.9%). Modelled deer impact (r² = 0.32, MAE = 6.9%) was most influenced by deer density (21.0%), mean annual precipitation (12.8%) and elevation (12.2%). Deer density was typically highest near large waterbodies, at low elevation, and with intermediate tree cover (40–70%). Impacts increased steadily with deer density up to ~ 2 pellets/m². Our study demonstrates the importance of forest water and forest agricultural interfaces for both deer density and impacts. Deer are likely to be most abundant near waterbodies due to the availability of high-quality forage and water, and prefer lowland locations that have access to both open and forested habitats. Spatial models can be used to predict deer density and associated impacts to facilitate targeted invasive deer management.

Agroforestry is one nature-based solution that holds significant potential for improving the sustainability and resilience of agricultural systems. Quantifying these benefits is challenging in agroforestry systems, largely due to landscape complexity and the diversity of management approaches. Digital tools designed for agroforestry typically focus on timber and crop production, and not the broader range of benefits usually considered in assessments of ecosystem services and natural capital. The objectives of this review were to identify and evaluate digital tools that quantify natural capital benefits across eight themes applicable to agroforestry systems: timber production and carbon sequestration, agricultural production, microclimate, air quality, water management, biodiversity, pollination, and amenity. We identified and evaluated 63 tools, 9 of which were assessed in further detail using Australia as a case study. No single tool was best suited to quantify benefits across each theme, suggesting that multiple tools or models could be combined to address capability gaps. We find that model complexity, incorporation of spatial processes, accessibility, regional applicability, development speed and interoperability present significant challenges for the tools that were evaluated. We recommend that these challenges be considered as opportunities to develop new, and build upon existing, tools to enhance decision support in agroforestry systems.

Aim: The increasing availability of regional and global climate data presents an opportunity to build better ecological models; however, it is not always clear which climate dataset is most appropriate. The aim of this study was to better understand the impacts that alternative climate datasets have on the modelled distribution of plant species, and to develop systematic approaches to enhancing their use in species distribution models (SDMs).Location: Victoria, southeast Australia and the Himalayan Kingdom of Bhutan.Methods: We compared the statistical performance of SDMs for 38 plant species in Victoria and 12 plant species in Bhutan with multiple algorithms using globally and regionally calibrated climate datasets. Individual models were compared against one another and as SDM ensembles to explore the potential for alternative predictions to improve statistical performance. We develop two new spatially continuous metrics that support the interpretation of ensemble predictions by characterizing the per‐pixel congruence and variability of contributing models.Results: There was no clear consensus on which climate dataset performed best across all species in either study region. On average, multi‐model ensembles (across the same species with different climate data) increased AUC/TSS/Kappa/OA by up to 0.02/0.03/0.03/0.02 in Victoria and 0.06/0.11/0.11/0.05 in Bhutan. Ensembles performed better than most single models in both Victoria (AUC = 85%; TSS = 68%) and Bhutan (AUC = 86%; TSS = 69%). SDM ensembles using models fitted with alternative algorithms and/or climate datasets each provided a significant improvement over single model runs.Main conclusions: Our results demonstrate that SDM ensembles, built using alternative models of the same climate variables, can quantify model congruence and identify regions of the highest uncertainty while mitigating the risk of erroneous predictions. Algorithm selection is known to be a large source of error for SDMs, and our results demonstrate that climate dataset selection can be a comparably significant source of uncertainty.