Search

In a recent paper in this journal, Bradshaw and colleagues analyse country statistics on flood characteristics, land cover and land cover change, and conclude that deforestation amplifies flood risk and severity in the developing world. The study addresses an important and long-standing question, but we identify important flaws. Principal among these are difficulties in interpreting country statistics and the correlation between population and floods. We review current knowledge, which suggests that the removal of trees does not affect large flood events, although associated landscape changes can under some circumstances. Reanalysis of the data analysed by Bradshaw and colleagues shows that population density alone already explains up to 83% of the variation in reported flood occurrences, considerably more than forest cover or deforestation (<10%). Feasible explanations for this statistical finding – whether spurious or causative – are not difficult to conceive. We, therefore, consider the conclusion of Bradshaw and colleagues to be unsupported. However, their study is a valuable first step to show how these or similar flood data might be used to further explore the relationship between land cover and flooding. [ABSTRACT FROM AUTHOR]

The net ecosystem exchange (NEE) of forests represents the balance of gross primary productivity (GPP) and respiration (R). Methods to estimate these two components from eddy covariance flux measurements are usually based on a functional relationship between respiration and temperature that is calibrated for night-time (respiration) fluxes and subsequently extrapolated using daytime temperature measurements. However, respiration fluxes originate from different parts of the ecosystem, each of which experiences its own course of temperature. Moreover, if the temperature-respiration function is fitted to combined data from different stages of biological development or seasons, a spurious temperature effect may be included that will lead to overestimation of the direct effect of temperature and therefore to overestimates of daytime respiration. We used the EUROFLUX eddy covariance data set for 15 European forests and pooled data per site, month and for conditions of low and sufficient soil moisture, respectively. We found that using air temperature (measured above the canopy) rather than soil temperature (measured 5 cm below the surface) yielded the most reliable and consistent exponential (Q10) temperature-respiration relationship. A fundamental difference in air temperature-based Q10 values for different sites, times of year or soil moisture conditions could not be established; all were in the range 1.6–2.5. However, base respiration (R0, i.e. respiration rate scaled to 0 °C) did vary significantly among sites and over the course of the year, with increased base respiration rates during the growing season. We used the overall mean Q10 of 2.0 to estimate annual GPP and R. Testing suggested that the uncertainty in total GPP and R associated with the method of separation was generally well within 15%. For the sites investigated, we found a positive relationship between GPP and R, indicating that there is a latitudinal trend in NEE because the absolute decrease in GPP towards the pole is greater than in R. [ABSTRACT FROM AUTHOR]

This work introduces a methodology for optimizing neural network models using a combination of continuous and categorical binary indices in the context of precipitation forecasting. Probability of detection and false alarm rate are popular metrics used in the verification of precipitation models. However, machine learning models trained using gradient descent cannot be optimized based on these metrics, as they are not differentiable. We propose an alternative formulation for these categorical indices that are differentiable and we demonstrate how they can be used to optimize the skill of precipitation neural network models defined as a multiobjective optimization problem. To our knowledge, this is the first proposal of a methodology for optimizing weather neural network models based on categorical indices. Plain Language Summary: Deep neural networks have recently demonstrated great versatility and an unprecedented capacity to model complex problems. In weather modeling, these algorithms have been applied to solve different problems. This is a promising area of research, given the availability of large volumes of weather data and increasingly powerful computers. Neural network models can learn to solve problems based on a metric, which the model tries to optimize. However, the quality of weather models is measured using a large variety of metrics, which can be a challenge when choosing which metric the model should optimize. In the case of precipitation, categorical binary metrics are a popular choice to assess the quality of a model. These metrics reduce precipitation to a "yes" or "no" event, and the results of the predicting model can be compared with the actual observations. This method is simple, yet powerful and a large number of indices and statistics have been developed to assess different aspects of the quality of precipitation models. As precipitation models are commonly assessed using these categorical binary metrics, it would be very convenient to optimize models based on them. Unfortunately, the mathematical nature of these metrics makes them unsuitable for optimizing deep learning models. In this work we present an alternative formulation for these categorical binary indices which can be used to train models. We demonstrate how a deep learning model can be trained to generate better quality precipitation data. Key Points: A methodology for optimizing neural network models based on categorical binary indices is introducedA novel multiobjective loss function combining continuous and categorical binary indices is presentedThe experimental results are tested using well‐known metrics in weather analysis, such as ROC curves [ABSTRACT FROM AUTHOR]