Artificial neural networks for monitoring network optimisation—a practical example using a national insect survey

Abstract

Monitoring networks are improved by additional sensors. Optimal configurations of sensors give better representations of the process of interest, maximising its exploration while minimising the need for costly infrastructure. By modelling the monitored process, we can identify gaps in its representation, i.e. uncertain predictions, where additional sensors should be located. Here, with data collected from the Rothamsted Insect Survey network, we train an artificial neural network to predict the seasonal aphid arrival from environmental variables. We focus on estimating prediction uncertainty across the UK to guide the addition of a sensor to the network. We first illustrate how to estimate uncertainty in neural networks, hence making them relevant for model-based monitoring network optimisation. Then we highlight critical areas of agricultural importance where additional traps would improve decision support and crop protection in the UK. Possible applications include most ecological monitoring and surveillance activities, but also the weather or pollution monitoring.

Testing machine learning algorithms on a binary classification phenological model

“…Compared with the ecophysiological models, the binary classification ML model showed markedly improved prediction accuracy. Many studies have also demonstrated that the ML phenological model improves the accuracy of phenological prediction (Bourhis et al, 2021; Dai et al, 2019; Holloway et al, 2018; Oses et al, 2020). Based on the comparison of the prediction results of ML regression models for the same species (LUD prediction), the prediction RMSE of the RF regression model on the test set (5.4 days) was greater than that of the GDD model (4.3 days) (Dai et al, 2019).…”

“…Although this phenological model can facilitate the development of biologically realistic process‐based experimental designs by using phenological observations in natural conditions as the main biological information (Janosi et al, 2020), like other statistical models it still faces great challenges in providing mechanistic explanations for plant growth and development processes (Hänninen, 2016; Hänninen et al, 2019). In ML methods, it is often difficult to relate the predictions to variables directly and to quantify the relationship between labels and features (Bourhis et al, 2021). Although machine learning is statistically advanced, it cannot provide a complete solution.…”

“…In addition, when the observation data are split into training and test sets, the correlation between phenological events and climatic factors might be inconsistent in different random sampling results, especially when there are few observations for focal species. The model parameters fitted by various statistical methods might vary greatly from the sampling data, affecting the ability to understand the complex relationship between plant phenology and climate variables (Bourhis et al, 2021).…”