Improved Yield Prediction of Winter Wheat Using a Novel Two-Dimensional Deep Regression Neural Network Trained via Remote Sensing.

In recent years, the use of remotely sensed and on-ground observations of crop fields, in conjunction with machine learning techniques, has led to highly accurate crop yield estimations. In this work, we propose to further improve the yield prediction task by using Convolutional Neural Networks (CNNs) given their unique ability to exploit the spatial information of small regions of the field. We present a novel CNN architecture called Hyper3DNetReg that takes in a multi-channel input raster and, unlike previous approaches, outputs a two-dimensional raster, where each output pixel represents the predicted yield value of the corresponding input pixel. Our proposed method then generates a yield prediction map by aggregating the overlapping yield prediction patches obtained throughout the field. Our data consist of a set of eight rasterized remotely-sensed features: nitrogen rate applied, precipitation, slope, elevation, topographic position index (TPI), aspect, and two radar backscatter coefficients acquired from the Sentinel-1 satellites. We use data collected during the early stage of the winter wheat growing season (March) to predict yield values during the harvest season (August). We present leave-one-out cross-validation experiments for rain-fed winter wheat over four fields and show that our proposed methodology produces better predictions than five compared methods, including Bayesian multiple linear regression, standard multiple linear regression, random forest, an ensemble of feedforward networks using AdaBoost, a stacked autoencoder, and two other CNN architectures.

Pinus contorta invasions increase wildfire fuel loads and may create a positive feedback with fire.

Invasive plant species that have the potential to alter fire regimes have significant impacts on native ecosystems. Concern that pine invasions in the Southern Hemisphere will increase fire activity and severity and subsequently promote further pine invasion prompted us to examine the potential for feedbacks between Pinus contorta invasions and fire in Patagonia and New Zealand. We determined how fuel loads and fire effects were altered by P. contorta invasion. We also examined post-fire plant communities across invasion gradients at a subset of sites to assess how invasion alters the post-fire vegetation trajectory. We found that fuel loads and soil heating during simulated fire increase with increasing P. contorta invasion age or density at all sites. However, P. contorta density did not always increase post-fire. In the largest fire, P. contorta density only increased significantly post-fire where the pre-fire P. contorta density was above an invasion threshold. Below this threshold, P. contorta did not dominate after fire and plant communities responded to fire in a similar manner as uninvaded communities. The positive feedback observed at high densities is caused by the accumulation of fuel that in turn results in greater soil heating during fires and high P. contorta density post-fire. Therefore, a positive feedback may form between P. contorta invasions and fire, but only above an invasion density threshold. These results suggest that management of pine invasions before they reach the invasion density threshold is important for reducing fire risk and preventing a transition to an alternate ecosystem state dominated by pines and novel understory plant communities.

Global Invader Impact Network (GIIN): toward standardized evaluation of the ecological impacts of invasive plants.

Terrestrial invasive plants are a global problem and are becoming ubiquitous components of most ecosystems. They are implicated in altering disturbance regimes, reducing biodiversity, and changing ecosystem function, sometimes in profound and irreversible ways. However, the ecological impacts of most invasive plants have not been studied experimentally, and most research to date focuses on few types of impacts, which can vary greatly among studies. Thus, our knowledge of existing ecological impacts ascribed to invasive plants is surprisingly limited in both breadth and depth. Our aim was to propose a standard methodology for quantifying baseline ecological impact that, in theory, is scalable to any terrestrial plant invader (e.g., annual grasses to trees) and any invaded system (e.g., grassland to forest). The Global Invader Impact Network (GIIN) is a coordinated distributed experiment composed of an observational and manipulative methodology. The protocol consists of a series of plots located in (1) an invaded area; (2) an adjacent removal treatment within the invaded area; and (3) a spatially separate uninvaded area thought to be similar to pre-invasion conditions of the invaded area. A standardized and inexpensive suite of community, soil, and ecosystem metrics are collected allowing broad comparisons among measurements, populations, and species. The method allows for one-time comparisons and for long-term monitoring enabling one to derive information about change due to invasion over time. Invader removal plots will also allow for quantification of legacy effects and their return rates, which will be monitored for several years. GIIN uses a nested hierarchical scale approach encompassing multiple sites, regions, and continents. Currently, GIIN has network members in six countries, with new members encouraged. To date, study species include representatives of annual and perennial grasses; annual and perennial forbs; shrubs; and trees. The goal of the GIIN framework is to create a standard yet flexible platform for understanding the ecological impacts of invasive plants, allowing both individual and synthetic analyses across a range of taxa and ecosystems. If broadly adopted, this standard approach will offer unique insight into the ecological impacts of invasive plants at local, regional, and global scales.

Physical disturbance shapes vascular plant diversity more profoundly than fire in the sagebrush steppe of southeastern Idaho, U.S.A.

Fire is thought to profoundly change the ecology of the sagebrush steppe. The Idaho National Laboratory provides an ideal setting to compare the effects of fire and physical disturbance on plant diversity in high-native-cover sagebrush steppe. Seventy-eight 1-hectare transects were established along paved, green-striped, gravel, and two-track roads, in overgrazed rangeland, and within sagebrush steppe involving different fire histories. Transects were sampled for the diversity and abundance of all vascular plants. Alpha, beta, and phylogenetic beta diversity were analyzed as a response to fire and physical disturbance. Postfire vegetation readily rebounds to prefire levels of alpha plant diversity. Physical disturbance, in contrast, strongly shapes patterns of alpha, beta, and especially phylogenetic beta diversity much more profoundly than fire disturbance. If fire is a concern in the sagebrush steppe then the degree of physical-disturbance should be more so. This finding is probably not specific to the study area but applicable to the northern and eastern portions of the sagebrush biome, which is characterized by a pulse of spring moisture and cold mean minimum winter temperatures. The distinction of sagebrush steppe from Great Basin sagebrush should be revised especially with regard to reseeding efforts and the control of annual grasses.

Control effort exacerbates invasive-species problem.

Ecosystem managers face a difficult decision when managing invasive species. If they use aggressive practices to reduce invader abundances, they will likely reduce invaders' competitive impacts on natives. But it is often difficult or impossible to reduce invaders without damaging natives. So a critical question becomes: Which is worse for native biota, invaders or things done to control invaders? We attempted to answer this question for a common scenario. We studied several grassland natives exhibiting long-term coexistence with an invader and asked how aggressive management (herbicide use) affected the natives. Whether or not grazing was excluded, one-time herbicide use made two native forbs exceedingly rare for our entire 16-year study period. Herbicide also made several other native forbs rare, but only when grazing was excluded, and there is evidence that the dominant invader became more abundant in response to the decreases in native-forb abundances. Throughout the world, terrestrial and aquatic ecosystems are receiving herbicide applications for exotic-species control. Some of the applications are doubtless warranted because they target small invader patches or larger areas with virtually no remaining natives. However, other herbicide applications occur where large native populations occur, and our data suggest that these applications can be ill advised. Our cautionary tale is told using an herbicide-treated grassland, but our results should be considered wherever invasive-species management damages native species.

Vulnerability of rehabilitated agricultural production systems to invasion by nontarget plant species.

Vast areas of arable land have been retired from crop production and "rehabilitated" to improved system states through landowner incentive programs in the United States (e.g., Conservation and Wetland Reserve Programs), as well as Europe (i.e., Agri-Environment Schemes). Our review of studies conducted on invasion of rehabilitated agricultural production systems by nontarget species elucidates several factors that may increase the vulnerability of these systems to invasion. These systems often exist in highly fragmented and agriculturally dominated landscapes, where propagule sources of target species for colonization may be limited, and are established under conditions where legacies of past disturbance persist and prevent target species from persisting. Furthermore, rehabilitation approaches often do not include or successfully attain all target species or historical ecological processes (e.g., hydrology, grazing, and/or fire cycles) key to resisting invasion. Uncertainty surrounds ways in which nontarget species may compromise long term goals of improving biodiversity and ecosystem services through rehabilitation efforts on former agricultural production lands. This review demonstrates that more studies are needed on the extent and ecological impacts of nontarget species as related to the goals of rehabilitation efforts to secure current and future environmental benefits arising from this widespread conservation practice.