|
Wallach, D., Thorburn, P., Asseng, S., Challinor, A. J., Ewert, F., Jones, J. W., et al. (2016). Estimating model prediction error: Should you treat predictions as fixed or random. Env. Model. Softw., 84, 529–539.
Abstract: Crop models are important tools for impact assessment of climate change, as well as for exploring management options under current climate. It is essential to evaluate the uncertainty associated with predictions of these models. We compare two criteria of prediction error; MSEPfixed, which evaluates mean squared error of prediction for a model with fixed structure, parameters and inputs, and MSEPuncertain(X), which evaluates mean squared error averaged over the distributions of model structure, inputs and parameters. Comparison of model outputs with data can be used to estimate the former. The latter has a squared bias term, which can be estimated using hindcasts, and a model variance term, which can be estimated from a simulation experiment. The separate contributions to MSEPuncertain(X) can be estimated using a random effects ANOVA. It is argued that MSEPuncertain(X) is the more informative uncertainty criterion, because it is specific to each prediction situation.
|
|
|
Murat, M., Malinowska, I., Hoffmann, H., & Baranowski, P. (2016). Statistical modelling of agrometeorological time series by exponential smoothing. International Agrophysics, 30(1), 57–65.
Abstract: Meteorological time series are used in modelling agrophysical processes of the soil-plant-atmosphere system which determine plant growth and yield. Additionally, longterm meteorological series are used in climate change scenarios. Such studies often require forecasting or projection of meteorological variables, eg the projection of occurrence of the extreme events. The aim of the article was to determine the most suitable exponential smoothing models to generate forecast using data on air temperature, wind speed, and precipitation time series in Jokioinen (Finland), Dikopshof (Germany), Lleida (Spain), and Lublin (Poland). These series exhibit regular additive seasonality or non-seasonality without any trend, which is confirmed by their autocorrelation functions and partial autocorrelation functions. The most suitable models were indicated by the smallest mean absolute error and the smallest root mean squared error.
|
|
|
Ferrise, R., Toscano, P., Pasqui, M., Moriondo, M., Primicerio, J., Semenov, M. A., et al. (2015). Monthly-to-seasonal predictions of durum wheat yield over the Mediterranean Basin. Clim. Res., 65, 7–21.
Abstract: Uncertainty in weather conditions for the forthcoming growing season influences farmers’ decisions, based on their experience of the past climate, regarding the reduction of agricultural risk. Early within-season predictions of grain yield can represent a great opportunity for farmers to improve their management decisions and potentially increase yield and reduce potential risk. This study assessed 3 methods of within-season predictions of durum wheat yield at 10 sites across the Mediterranean Basin. To assess the value of within-season predictions, the model SiriusQuality2 was used to calculate wheat yields over a 9 yr period. Initially, the model was run with observed daily weather to obtain the reference yields. Then, yield predictions were calculated at a monthly time step, starting from 6 mo before harvest, by feeding the model with observed weather from the beginning of the growing season until a specific date and then with synthetic weather constructed using the 3 methods, historical, analogue or empirical, until the end of the growing season. The results showed that it is possible to predict durum wheat yield over the Mediterranean Basin with an accuracy of normalized root means squared error of <20%, from 5 to 6 mo earlier for the historical and empirical methods and 3 mo earlier for the analogue method. Overall, the historical method performed better than the others. Nonetheless, the analogue and empirical methods provided better estimations for low-yielding and high-yielding years, thus indicating great potential to provide more accurate predictions for years that deviate from average conditions.
|
|
|
Refsgaard, J. C., Madsen, H., Andréassian, V., Arnbjerg-Nielsen, K., Davidson, T. A., Drews, M., et al. (2014). A framework for testing the ability of models to project climate change and its impacts. Clim. Change, 122(1-2), 271–282.
Abstract: Models used for climate change impact projections are typically not tested for simulation beyond current climate conditions. Since we have no data truly reflecting future conditions, a key challenge in this respect is to rigorously test models using proxies of future conditions. This paper presents a validation framework and guiding principles applicable across earth science disciplines for testing the capability of models to project future climate change and its impacts. Model test schemes comprising split-sample tests, differential split-sample tests and proxy site tests are discussed in relation to their application for projections by use of single models, ensemble modelling and space-time-substitution and in relation to use of different data from historical time series, paleo data and controlled experiments. We recommend that differential-split sample tests should be performed with best available proxy data in order to build further confidence in model projections.
|
|
|
Mansouri, M., & Destain, M. - F. (2015). Predicting biomass and grain protein content using Bayesian methods. Stoch. Environ. Res. Risk Assess., 29(4), 1167–1177.
Abstract: This paper deals with the problem of predicting biomass and grain protein content using improved particle filtering (IPF) based on minimizing the Kullback-Leibler divergence. The performances of IPF are compared with those of the conventional particle filtering (PF) in two comparative studies. In the first one, we apply IPF and PF at a simple dynamic crop model with the aim to predict a single state variable, namely the winter wheat biomass, and to estimate several model parameters. In the second study, the proposed IPF and the PF are applied to a complex crop model (AZODYN) to predict a winter-wheat quality criterion, namely the grain protein content. The results of both comparative studies reveal that the IPF method provides a better estimation accuracy than the PF method. The benefit of the IPF method lies in its ability to provide accuracy related advantages over the PF method since, unlike the PF which depends on the choice of the sampling distribution used to estimate the posterior distribution, the IPF yields an optimum choice of this sampling distribution, which also utilizes the observed data. The performance of the proposed method is evaluated in terms of estimation accuracy, root mean square error, mean absolute error and execution times.
|
|