Abstract. Climate change threatens our ability to grow food for an ever-increasing population. There is aneed for high-quality soil moisture predictions in under-monitored regionslike Africa. However, it is unclear if soil moisture processes are globallysimilar enough to allow our models trained on available in situ data tomaintain accuracy in unmonitored regions. We present a multitask longshort-term memory (LSTM) model that learns simultaneously from globalsatellite-based data and in situ soil moisture data. This model is evaluated inboth random spatial holdout mode and continental holdout mode (trained onsome continents, tested on a different one). The model compared favorably tocurrent land surface models, satellite products, and a candidate machinelearning model, reaching a global median correlation of 0.792 for the randomspatial holdout test. It behaved surprisingly well in Africa and Australia,showing high correlation even when we excluded their sites from the trainingset, but it performed relatively poorly in Alaska where rapid changes areoccurring. In all but one continent (Asia), the multitask model in theworst-case scenario test performed better than the soil moisture activepassive (SMAP) 9 km product. Factorial analysis has shown that the LSTM model'saccuracy varies with terrain aspect, resulting in lower performance for dryand south-facing slopes or wet and north-facing slopes. This knowledgehelps us apply the model while understanding its limitations. This model isbeing integrated into an operational agricultural assistance applicationwhich currently provides information to 13 million African farmers.
more »
« less
What processes must we understand to forecast regional-scale population dynamics?
An urgent challenge facing biologists is predicting the regional-scale population dynamics of species facing environmental change. Biologists suggest that we must move beyond predictions based on phenomenological models and instead base predictions on underlying processes. For example, population biologists, evolutionary biologists, community ecologists and ecophysiologists all argue that the respective processes they study are essential. Must our models include processes from all of these fields? We argue that answering this critical question is ultimately an empirical exercise requiring a substantial amount of data that have not been integrated for any system to date. To motivate and facilitate the necessary data collection and integration, we first review the potential importance of each mechanism for skilful prediction. We then develop a conceptual framework based on reaction norms, and propose a hierarchical Bayesian statistical framework to integrate processes affecting reaction norms at different scales. The ambitious research programme we advocate is rapidly becoming feasible due to novel collaborations, datasets and analytical tools.
more »
« less
- NSF-PAR ID:
- 10204998
- Date Published:
- Journal Name:
- Proceedings of the Royal Society B: Biological Sciences
- Volume:
- 287
- Issue:
- 1940
- ISSN:
- 0962-8452
- Page Range / eLocation ID:
- 20202219
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
more » « less
Abstract Predicting microbial metabolic rates and emergent biogeochemical fluxes remains challenging due to the many unknown population dynamical, physiological and reaction‐kinetic parameters and uncertainties in species composition. Here, we show that the need for these parameters can be eliminated when population dynamics and reaction kinetics operate at much shorter time scales than physical mixing processes. Such scenarios are widespread in poorly mixed water columns and sediments. In this ‘fast‐reaction‐transport’ (FRT) limit, all that is required for predictions are chemical boundary conditions, the physical mixing processes and reaction stoichiometries, while no knowledge of species composition, physiology or population/reaction kinetic parameters is needed. Using time‐series data spanning years 2001–2014 and depths 180–900 m across the permanently anoxic Cariaco Basin, we demonstrate that the FRT approach can accurately predict the dynamics of major electron donors and acceptors (Pearson
r ≥ 0.9 in all cases). Hence, many microbial processes in this system are largely transport limited and thus predictable regardless of species composition, population dynamics and kinetics. Our approach enables predictions for many systems in which microbial community dynamics and kinetics are unknown. Our findings also reveal a mechanism for the frequently observed decoupling between function and taxonomy in microbial systems. -
Dr Andrea E. A. Stephens (Ed.)To forecast extinction risks of natural populations under climate change and direct human impacts, an integrative understanding of both phenotypic plasticity and adaptive evolution is essential. To date, the evidence for whether, when, and how much plasticity facilitates adaptive responses in changing environments is contradictory. We argue that explicitly considering three key environmental change components – rate of change, variance, and temporal autocorrelation – affords a unifying framework of the impact of plasticity on adaptive evolution. These environmental components each distinctively effect evolutionary and ecological processes underpinning population viability. Using this framework, we develop expectations regarding the interplay between plasticity and adaptive evolution in natural populations. This framework has the potential to improve predictions of population viability in a changing world.more » « less
-
A formidable challenge for global change biologists is to predict how natural populations will respond to the emergence of conditions not observed at present, termed novel climates. Popular approaches to predict population vulnerability are based on the expected degree of novelty relative to the amplitude of historical climate fluctuations experienced by a population. Here, we argue that predictions focused on amplitude may be inaccurate because they ignore the predictability of environmental fluctuations in driving patterns of evolution and responses to climate change. To address this disconnect, we review major findings of evolutionary theory demonstrating the conditions under which phenotypic plasticity is likely to evolve in natural populations, and how plasticity decreases population vulnerability to novel environments. We outline key criteria that experimental studies should aim for to effectively test theoretical predictions, while controlling for the degree of climate novelty. We show that such targeted tests of evolutionary theory are rare, with marine systems being overall underrepresented in this venture despite exhibiting unique opportunities to test theory. We conclude that with more robust experimental designs that manipulate both the amplitude and predictability of fluctuations, while controlling for the degree of novelty, we may better predict population vulnerability to climate change.more » « less
-
more » « less
Abstract When aiming to make predictions over targets in the pharmacological setting, a data‐focused approach aims to learn models based on a collection of labeled examples. Unfortunately, data sharing is not always possible, and this can result in many different models trained on disparate populations, leading to the natural question of how best to use and combine them when making a new prediction. Previous work has focused on global model selection or ensembling, with the result of a single final model across the feature space. Machine‐learning models perform notoriously poorly on data outside their training domain, however, due to a problem known as covariate shift, and so we argue that when ensembling models the weightings for individual instances must reflect their respective domains—in other words, models that are more likely to have seen information on that instance should have more attention paid to them. We introduce a method for such an instance‐wise ensembling of models called Synthetic Model Combination (SMC), including a novel representation learning step for handling sparse high‐dimensional domains. We demonstrate the use of SMC on an example with dosing predictions for vancomycin, although emphasize the applicability of the method to any scenario involving the use of multiple models.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1098/rspb.2020.2219
