Comparing Simulated Crop Growth Parameters with Observed, Generated and Combined Weather Data for a Brazilian Location.

Crop simulation models have been extensively used for many purposes: climatic risk studies; evaluation of crop management strategies; impact of climate change on yield; among others. However, for running these models accurately, reliable weather data are required. In many cases, the long-term weather data available contains problems, which include: data suspect to be erroneous; missing data; and the length of the weather series is too short. To minimize these problems, weather generators can be used to provide consistent long-term weather series.

In this study synthetic weather data series was generated by the weather generator WGEN, which is included in the Decision Support System for Agrotechnology Transfer (DSSAT). Weather data from 1978 to 2010 of the location of Piracicaba, state of São Paulo, Brazil was used to obtain the statistical parameters for the data generation. To evaluate the influence of the generated weather data on corn and soybean crop growth parameters, four treatments were created for corn: T1 - potential yield, sowing in Feb/01; T2 - potential yield, sowing in Oct/01; T3 - rainfed yield, sowing Feb/01; T4 - rainfed yield, sowing in Oct/01, and two for soybean: T1 - potential yield, sowing in Nov/15; T2 - rainfed yield, sowing in Nov/15. The model used for corn crop was the CERES-Maize and for soybean was the CROPGRO-Soybean, which had as input observed, generated (G) and a combination of observed-generated weather data, with cultivars adequately calibrated for the Brazilian environment. Based on that, the following scenarios were created: C1 - Tmax(G), Tmin, radiation and rain; C2 - Tmax, Tmin(G), radiation and rain; C3 - Tmax, Tmin, radiation(G) and rain; C4 - Tmax, Tmin, radiation and rain(G); C5 - Tmax(G), Tmin(G), radiation and rain; C6 - Tmax(G), Tmin, radiation(G) and rain; C7 - Tmax, Tmin(G), radiation(G) and rain; C8 - Tmax, Tmin, radiation(G) and rain(G) to evaluated the possibility of substituting observed by generated weather data series in agroclimatic studies.

By using mean error (%) as parameter of evaluation the results showed that, for the yield; maximum LAI; anthesys and maturity dates, the best fits were obtained with the scenarios C2, C3, C3 and C3 for T1. For T2, the best fits were reached with C3, G/C2, C2 and C3, whereas for T3 with C2, C2/C3/C5, C4 and C2/C3, and for T4 with C2, C1/C5, C4 and C2, respectively. Analyzing the same parameters for soybean, the best fits for T1 were C2, C5, C3 and C3 and for T2 were C1; G/C5; C4 and C3.

These results allowed us to conclude that combination between observed and generated weather data as input in crop simulation models is a possible way to minimize the problems with gaps and errors in the weather data series.