Refine
Has Fulltext
- no (2)
Year of publication
- 2022 (2) (remove)
Document Type
- Article (2)
Language
- English (2)
Is part of the Bibliography
- yes (2) (remove)
Keywords
- Crop modelling (1)
- Equivocality (1)
- Feature selection (1)
- Machine learning (1)
- Random forests (1)
- Support vector (1)
- machine (1)
Institute
- Institut für Biochemie und Biologie (2) (remove)
Provisioning a sufficient stable source of food requires sound knowledge about current and upcoming threats to agricultural production. To that end machine learning approaches were used to identify the prevailing climatic and soil hydrological drivers of spatial and temporal yield variability of four crops, comprising 40 years yield data each from 351 counties in Germany. Effects of progress in agricultural management and breeding were subtracted from the data prior the machine learning modelling by fitting smooth non-linear trends to the 95th percentiles of observed yield data. An extensive feature selection approach was followed then to identify the most relevant predictors out of a large set of candidate predictors, comprising various soil and meteorological data. Particular emphasis was placed on studying the uniqueness of identified key predictors. Random Forest and Support Vector Machine models yielded similar although not identical results, capturing between 50% and 70% of the spatial and temporal variance of silage maize, winter barley, winter rapeseed and winter wheat yield. Equally good performance could be achieved with different sets of predictors. Thus identification of the most reliable models could not be based on the outcome of the model study only but required expert's judgement. Relationships between drivers and response often exhibited optimum curves, especially for summer air temperature and precipitation. In contrast, soil moisture clearly proved less relevant compared to meteorological drivers. In view of the expected climate change both excess precipitation and the excess heat effect deserve more attention in breeding as well as in crop modelling.
Crop model intercomparison studies have mostly focused on the assessment of predictive capabilities for crop development using weather and basic soil data from the same location. Still challenging is the model performance when considering complex interrelations between soil and crop dynamics under a changing climate. The objective of this study was to test the agronomic crop and environmental flux-related performance of a set of crop models. The aim was to predict weighing lysimeter-based crop (i.e., agronomic) and water-related flux or state data (i.e., environmental) obtained for the same soil monoliths that were taken from their original environment and translocated to regions with different climatic conditions, after model calibration at the original site. Eleven models were deployed in the study. The lysimeter data (2014-2018) were from the Dedelow (Dd), Bad Lauchstadt (BL), and Selhausen (Se) sites of the TERENO (TERrestrial ENvironmental Observatories) SOILCan network. Soil monoliths from Dd were transferred to the drier and warmer BL site and the wetter and warmer Se site, which allowed a comparison of similar soil and crop under varying climatic conditions. The model parameters were calibrated using an identical set of crop- and soil-related data from Dd. Environmental fluxes and crop growth of Dd soil were predicted for conditions at BL and Se sites using the calibrated models. The comparison of predicted and measured data of Dd lysimeters at BL and Se revealed differences among models. At site BL, the crop models predicted agronomic and environmental components similarly well. Model performance values indicate that the environmental components at site Se were better predicted than agronomic ones. The multi-model mean was for most observations the better predictor compared with those of individual models. For Se site conditions, crop models failed to predict site-specific crop development indicating that climatic conditions (i.e., heat stress) were outside the range of variation in the data sets considered for model calibration. For improving predictive ability of crop models (i.e., productivity and fluxes), more attention should be paid to soil-related data (i.e., water fluxes and system states) when simulating soil-crop-climate interrelations in changing climatic conditions.