Interpreting random forest analysis of ecological models to move from prediction to explanation

Rise in water vapour driven by moisture transport facilitates water availability for the greening of global deserts

There are substantial changes in the global drylands owing to climate change and anthropogenic activities. However, this aspect is not adequately explored in the context of recent climate change and global warming. Therefore, we analyse the role of water vapour in driving precipitation and corresponding surface greenness in the global deserts using satellite measurements. Statistical techniques such as partial correlation and Randon Forest (RF) are employed to understand the relationship among the physical processes that drive water availability for desert greening. Our analysis shows that water vapour is relatively lower (<25 kg/m2) in the deserts than rest of the globe, but comparable to the polar and high-altitude regions. Among the deserts, Thar (25 kg/m2) and Sahara (15-20 kg/m2) show higher water vapour, in contrast to the American and Gobi deserts (<10 kg/m2). Trajectory analysis reveals that water vapour transport from the south Atlantic Ocean is very high (90 kg/m/s) to the Sahel region of Sahara. Similarly, water vapour comes from Arabian Sea and Indian Ocean to Thar, mainly during Indian Summer Monsoon (ISM). There is an increase in water vapour driven by a rise in moisture transport to the American, Arabian, Thar and Sub-Sahara deserts during the period 2003-2020. The rise in moisture transport and associated water vapour in the deserts enhance water availability through precipitation and soil moisture, influencing surface greenness, as illustrated by the partial correlation and RF analyses. Enhanced water vapour and water availability, together with anthropogenic activities such as agriculture and afforestation in the deserts drive greening, which is more pronounced in Thar and Sub-Sahara. This study, thus, reveals the role of atmospheric moisture in regulating the terrestrial water availability and surface greenness in the extreme arid regions on the earth.

Gaining Biological Insights through Supervised Data Visualization

Dimensionality reduction-based data visualization is pivotal in comprehending complex biological data. The most common methods, such as PHATE, t-SNE, and UMAP, are unsupervised and therefore reflect the dominant structure in the data, which may be independent of expert-provided labels. Here we introduce a supervised data visualization method called RF-PHATE, which integrates expert knowledge for further exploration of the data. RF-PHATE leverages random forests to capture intricate featurelabel relationships. Extracting information from the forest, RF-PHATE generates low-dimensional visualizations that highlight relevant data relationships while disregarding extraneous features. This approach scales to large datasets and applies to classification and regression. We illustrate RF-PHATE's prowess through three case studies. In a multiple sclerosis study using longitudinal clinical and imaging data, RF-PHATE unveils a sub-group of patients with non-benign relapsingremitting Multiple Sclerosis, demonstrating its aptitude for time-series data. In the context of Raman spectral data, RF-PHATE effectively showcases the impact of antioxidants on diesel exhaust-exposed lung cells, highlighting its proficiency in noisy environments. Furthermore, RF-PHATE aligns established geometric structures with COVID-19 patient outcomes, enriching interpretability in a hierarchical manner. RF-PHATE bridges expert insights and visualizations, promising knowledge generation. Its adaptability, scalability, and noise tolerance underscore its potential for widespread adoption.

Leaf traits linked to structure and palatability drive plant-insect interactions within three forested ecosystems

PREMISE

Plant traits and insect herbivory have been highly studied within the modern record but only to a limited extent within the paleontological. Preservation influences what can be measured within the fossil record, but modern methods are also not compatible with paleobotanical methods. To remedy this knowledge gap, a comparable framework was created here using modern and paleobotanical methods, allowing for future comparisons within the fossil record.

METHODS

Insect feeding damage on selected tree species at Harvard Forest, the Smithsonian Environmental Research Center, and La Selva were characterized using the damage type system prevalent within paleobotanical studies and compared with leaf traits. Linear models and random forest analyses tested the influence of leaf traits on total, specialized, gall, and mine frequency and diversity.

RESULTS

Structural traits like leaf dry mass per area and palatability traits, including lignin and phosphorus concentrations, are important variables affecting gall and mine damage. The significance and strength of trait-herbivory relationships varied across forest types, which is likely driven by differences in local insect populations.

CONCLUSIONS

This work addresses the persistent gap between modern and paleoecological studies focusing on the influence of leaf traits on insect herbivory. This is important as modern climate change alters our understanding of plant-insect interactions, providing a need for contextualizing these relationships within evolutionary time. The fossil record provides information on terrestrial response to past climatic events and, thus, should be implemented when considering how to preserve biodiversity under current and future global change.

Land Use and Land Cover Change Detection Using the Random Forest Approach: The Case of The Upper Blue Nile River Basin, Ethiopia

Monitoring land use change dynamics is critical for tackling food security, climate change, and biodiversity loss on a global scale. This study is designed to classify land use and land cover in the upper Blue Nile River Basin (BNRB) using a random forest (RF) algorithm. The Landsat images for Landsat 45, Landsat 7, and Landsat 8 are used for classification purposes. The study area is classified into seven land use/land cover classes: cultivated lands, bare lands, built-ups, forests, grazing lands, shrublands, and waterbodies. The accuracy of classified images is 83%, 85%, and 91% using the Kappa index of agreements. From 1983 to 2022 periods, cultivated lands and built-up areas increased by 47541 and 1777 km2, respectively, at the expense of grazing lands, shrublands, and forests. Furthermore, the area of water bodies has increased by 662 km2 due to the construction of small and large-scale irrigation and hydroelectric power generation dams. The main factors that determine agricultural land expansion are related to population growth. Therefore, land use and land cover change detection using a random forest is an important technique for multispectral satellite data classification to understand the optimal use of natural resources, conservation practices, and decision-making for sustainable development.