Daniel Lubin | Search USGS

USGS2023

Stream temperature prediction in a shifting environment: The influence of deep learning architecture

Stream temperature is a fundamental control on ecosystem health. Recent efforts incorporating process guidance into deep learning models for predicting stream temperature have been shown to outperform existing statistical and physical models. This performance is in part because deep learning architectures can actively learn spatiotemporal relationships that govern how water and energy propagate through a river network. However, exploration of how spatiotemporal awareness and process guidance influence a model's generalizability under shifting environmental conditions such as climate change is limited. Here, we use Explainable Artificial Intelligence (XAI) to interrogate how differing deep learning architectures affect a model's learned spatial and temporal dependencies, and how those learned dependencies affect a model's ability to maintain high accuracy when applied to unseen environmental conditions. Using the Delaware River Basin in the northeastern United States as a test case, we compare two spatiotemporally aware process-guided deep learning models for predicting stream temperature (a recurrent graph convolution network—RGCN, and a temporal convolution graph model—Graph WaveNet). Both models achieve equally high predictive performance when testing data are well represented in the training data (test root mean squared errors of 1.64°C and 1.65°C); however, Graph WaveNet significantly outperforms RGCN in 4 out of 5 experiments where test partitions represent different types of unseen environmental conditions. XAI results show that the architecture of Graph WaveNet leads to learned spatial relationships with greater fidelity to physical processes, and that this fidelity improves the generalizability of the model when applied to shifting and/or unseen environmental conditions.

Delaware River Basin↗

USGS2003

Fire and grazing impacts on plant diversity and alien plant invasions in the southern Sierra Nevada

Patterns of native and alien plant diversity in response to disturbance were examined along an elevational gradient in blue oak savanna, chaparral, and coniferous forests. Total species richness, alien species richness, and alien cover declined with elevation, at scales from 1 to 1000 m 2 . We found no support for the hypothesis that community diversity inhibits alien invasion. At the 1-m 2 point scale, where we would expect competitive interactions between the largely herbaceous flora to be most intense, alien species richness as well as alien cover increased with increasing native species richness in all communities. This suggests that aliens are limited not by the number of native competitors, but by resources that affect establishment of both natives and aliens. Blue oak savannas were heavily dominated by alien species and consistently had more alien than native species at the 1-m 2 scale. All of these aliens are annuals, and it is widely thought that they have displaced native bunchgrasses. If true, this means that aliens have greatly increased species richness. Alternatively, there is a rich regional flora of native annual forbs that could have dominated these grasslands prior to displacement by alien grasses. On our sites, livestock grazing increased the number of alien species and alien cover only slightly over that of sites free of livestock grazing for more than a century, indicating some level of permanency to this invasion. In chaparral, both diversity and aliens increased markedly several years after fire. Invasive species are rare in undisturbed shrublands, and alien propagules fail to survive the natural crown fires in these ecosystems. Thus, aliens necessarily must colonize after fire and, as a consequence, time since fire is an important determinant of invasive presence. Blue oak savannas are an important propagule source for alien species because they maintain permanent populations of all alien species encountered in postfire chaparral, and because the vegetation mosaic in this region places them in proximity to chaparral. The speed at which alien propagules reach a burned site and the speed at which the shrublands return to their former closed-canopy condition determine alien invasion. Frequent burning of this vegetation alters the balance in favor of alien invasion. In the higher-elevation coniferous forests, species diversity was a function of fire severity and time since fire. High-intensity fires create gaps that decrease canopy coverage and increase light levels and nutrients for an ephemeral successional flora. Few species have persistent seed banks, so the time since fire is an important determinant of colonization success. There was a highly significant interaction between fire severity and time since fire for understory cover, species richness, and alien richness and cover. Understory was sparse in the first year after fire, particularly in low-severity burns, and increased substantially several years after fire, particularly on high-severity burns. Both fire severity and time since fire affected alien species richness and dominance. Coniferous forests had about one-third as many alien species as the foothill oak savannas, and fewer than half of the species were shared between these communities. Unburned coniferous forests were largely free of alien species, whereas some burned sites had a significant alien presence, which presents a challenge for fire restoration of these forests.

Ecological Applications↗