Durum wheat yield and protein content responses to meteorological conditions: improvement of Ceres-Wheat routine with a simplified forecasting index for early assessment

Durum wheat yield and protein content responses to meteorological conditions: improvement of Ceres-Wheat routine with a simplified forecasting index for early assessment. Orlandini S. a, c, Dalla Marta A. a, Baldi A. a, Orlando F. b, Guasconi F. a, Mancini M. a, c a Department of Agrifood Production and Environmental Sciences, University of Florence, Piazzale delle Cascine 18 – 50144 Florence, Italy simone.orlandini@unifi.it b Department of Agricultural and Environmental Sciences, Production, Landscape, Agroenergy – CASSANDRA Lab., University of Milan, Via Celoria 2 – 20133 Milan, Italy francesca.orlando@unimi.it c Climate and Sustainability Foundation Via Caproni 8 – 50144 Florence, Italy m.mancini@climaesostenibilita.it ////// Durum wheat (Triticum turgidum L. var. durum) grain protein content (GPC) is a key factor in determining the technological and rheological properties of flour for making high quality pasta. For this reason, a premium price is commonly paid to the farmer for wheat with a high grain protein content. The sensitiveness of durum wheat to weather conditions and the uncertainty of the Mediterranean environment affect both harvest quantity and quality, making it difficult to warrant the standard quality. Moreover, the inverse relation between GPC with grain yield, which are both important agronomic targets, makes the field management decision-making process even more complicated. In this context, an operational tool for durum wheat production estimation becomes necessary not only to forecast the final crop production, but also to better identify the driving climate variables in the specific productive area. Crop models are recognized to be useful tools able to capture and describe the interactions between environmental variables and crops, helpful for the interpretation and extrapolation of experimental results and the identification of the weather driving variables. Unfortunately, few studies on durum wheat modeling were carried out, especially in the Mediterranean and particularly for GPC simulation, therefore the modeling of the durum wheat responsiveness needs further investigations. In most wheat simulation models (CERES-Wheat, SWHEAT) the GPC is determined by the soil N availability and the plant N demand, the latter being positively related to the leaf area expansion, and thus to the leaf biomass able to store N. The models assume the "source-limited" nature of the grain protein deposition, and the LAI is the main descriptor of the source of N available for the translocation. However, while wheat models show high performances in the yield assessment, poor results are obtained for GPC estimation, also because the majority of models have been developed for soft wheat, where GPC has much less importance. Therefore, current algorithms must be revisited, and GPC modeling is still a challenge. Our goals were: to evaluate the suitability of a mechanistic and deterministic model (CERES-Wheat) to identify the forcing and status variables affecting the GPC of durum wheat; to improve the model performance, through the assessment of a new routine for GPC simulation; to develop and test a simplified forecasting index (SFI). The research was carried out in Val d'Orcia (Lat. 43.03 N, Long. 1.66 E), a rural area of Tuscany (Central Italy). Meteorological, productive and phenological data for cv. Claudio were used for calibration and validation of CERES-Wheat (DSSAT-CSM 4) (years 1998-2011), and for a long-term analysis (LTA) (1955-2011). The model performance was assessed by means of a correlation analysis between measured and simulated data. A new routine for GPC simulation during the LTA was developed: GPC = {(TN/NS*100)+ 0.5}*5.7. Where: 0.5 = additional factor due to the genetic difference between durum and soft; 5.7 = conversion factor for grain N to protein; TN = total N available for the translocation from aerial biomass into the grain; NS = grain nitrogen sink. In the long-term study, the fertilization was entered accordingly to the protocol most widespread in the study area: a total amount of N ranging from 95 to 200 kg/ha, split into one fertilization at sowing, and two applications during the crop cycle, at tillering and stem elongation stages, was adopted. Sowing and harvest dates were simulated automatically, within the period 10 Nov-30 Dec, when optimum soil conditions occur and at grain maturity, respectively. Monthly meteorological indices were computed for the main crop development stages: March (tillering), April (stem elongation, booting and ear emergence) and May (anthesis and grain filling). To identify the main environmental and crop variables affecting production, a linear regression analysis between the harvest components (yield and GPC), the meteorological indices, and the maximum LAI simulated by the model at the end of growth stage was performed. Furthermore, a multiple linear regression analysis was performed (with SPSS.18 software) to develop the simplified index for harvest forecast (SFI). Field measurements were carried out during two growing seasons (2010 and 2011) and the data collected (LAI, plants density, yield, GPC, phenology) were used to validate the SFI and the CERES performance in determining the variables affecting the GPC. A highly significant correlation (P≤0.001) was always found between CERES estimates and observed yield data. A significant correlation (P ≤ 0.01) was found between the observed and simulated GPC with both routines; however the new one was able to improve the simulation enhancing the values of all statistical indicators. The LTA highlighted that rainfall during the tillering stage affects final GPC and confirmed the well established inverse relationship between yield and GPC: rainfall during the tillering promotes yield, while drought and warm conditions during the grain-filling promote GPC. The results of LTA also confirmed the positive role of LAI at heading stage on harvest quantity and quality. Rainfall distribution at tillering and LAI at heading stage were then included in the SFI as a main status and forcing variables. SFI was validated in two conditions; the first validation was made over 56 years and SFI failed in the prediction of GPC variability (no correspondence between simulated and forecasted data). The second validation was made using data observed during 2009-2010 and 2010-2011 growing seasons. Therefore, fields were distinguished on the base of LAI value at the heading stage in intermediate (1≤LAI≤2) and extreme (2