برآورد نوسانات عملكرد در مزارع گندم به وسيله متغير‌‌هاي مكاني: رهيافتي در كشاورزي دقيق

Estimating Within Field Variability of Wheat Yield Using Spatial Variables: An Approach to Precision Agriculture

در اين تحقيق نوسانات درون مزرع‌هاي عملكرد گندم و رابطه آن با توزيع مكاني تراكم علف هاي هرز و ميزان نيتروژن خاك با استفاده از روش‌هاي زمين آمار مورد ارزيابي قرار گرفت. نمونه گيري در منطقه اي با ابعاد 120×90 متر واقع در مزرعه اي به مساحت 8 / 3 هكتار انجام شد. ميزان نيتروژن خاك و تراكم علف‌هاي هرز در مرحله پنجه¬زني و عملكرد دانه در هنگام رسيدگي كامل از مساحت يك مترمربع واقع در مركز شبكه‌‌هاي 10×10 متري تعيين شد. نوسانات مكاني عملكرد گندم بين 1/5-4/9 با ميانگين 3/3 تن در هكتار و ضريب تغييرات (CV) 29 درصد بود در حالي‌كه تراكم علف‌‌هاي هرز (با ميانگين 2/2 بوته در مترمربع) و نيتروژن خاك (با ميانگين 05/0 درصد) تنوع بيشتري داشته و CV آن‌ها به ترتيب 55 و 41 درصد بود. نتايج رگرسيون چند متغيره نشان داد كه نيتروژن خاك و تراكم علف‌هاي هرز بدون در نظر گرفتن توزيع مكاني آن ها، در حدود 80 درصد از نوسانات عملكرد گندم را توصيف كردند. سمي واريوگرام4 مربوط به هر متغير در دو فاصله نمونه گيري 10 و 20 متري محاسبه و مدل مناسب به آن برازش داده شد. مقايسه خصوصيات آماري مدل‌‌هاي واريوگرم نشان داد كه افزايش فاصله نمونه گيري موجب كاهش دقت برآورد شد. نقشه‌‌هاي توزيع مكاني با ميان يابي (كريجينگ معمولي) بر مبناي مدل واريوگرام بر روي هر سه متغير و نيز ميان‌يابي توأم عملكرد با در نظر گرفتن تراكم علف هرز و ميزان نيتروژن خاك به عنوان متغير همراه تهيه شد و اعتبار مقادير پيش بيني شده مورد مقايسه آماري قرار گرفت. نتايج نشان داد كه دقت پيش بيني متغير‌ها در فاصله نمونه‌گيري 10 متري مطلوب بود و ميان يابي توأم عملكرد گندم با ميزان نيتروژن خاك باعث افزايش قدرت پيش‌بيني شد. از طرفي ميان يابي با اندازه گيري در فاصله 20 متري از دقت كافي برخوردار نبود ولي ميان‌يابي توأم عملكرد با نيتروژن خاك در اين فاصله نمونه‌گيري نيز باعث بهبود دقت پيش‌بيني شد.

Introduction In conventional crop management systems fields are considered as a homogenous environment however, because of high within field spatial variability such management is economically inefficient and provides drastic environmental consequences (Pierpaoli et al., 2013). Crop yield at any point of a field is a function of factors including planting density, weather conditions, management practices and biotic and abiotic stresses which results to spatial variability. Understanding the pattern and determinants of yield variability provides a basis for development of site specific management systems with lower inputs (Basso et al., 2012). In this study spatial variability of soil nitrogen, weed density and their effect on crop yield variation within a wheat field are surveyed and mapped using geostatistical methods. In addition the effects of sampling distance on the accuracy of results were evaluated. Materials and Methods Required data were collected from a 3.5 ha wheat field which was fully managed by owner based of local agronomic recommendations. Samples were taken from a 90×120 m area located in the field center and divided into 10×10 m grids. Soil nitrogen content and weed density at tillering and wheat yield at maturity were measured in 1 m2 plots located at the center of each grid. Semivariogarms were developed after fitting spherical model to the calculated semivariance for each spatial variable. Simple kiriging was used for spatial interpolation and mapping spatial variability of soil nitrogen, weed density and wheat yield and co-kriging was applied with soil nitrogen or weed density as covariates to map within field yield variation (Goovaerts, 1999; Oliver and Webster, 2014). The same analysis was repeated with 20×20 m grids to evaluate the effect of sampling distance. Predictions results were validated against measured values using standard statistical methods. GS Plus (γ- Design) ver. 9.0 was used for geostatistical analysis and mapping. Results and Discussion Grain yield was varied between 1.5-4.9 t ha-1 with coefficient of variation (CV) 0f 29%. However, weed density and soil nitrogen showed a higher spatial variation with CV of 55 and 41%, respectively. Based on the results of multiple regression, weed density and soil nitrogen accounted for 80% of the observed yield variation. Semivariance was calculated for the studied variables with 10 and 20 m lag distances and spherical model was fitted to the experimental variograms. Comparison of statistical characteristics of the variogram models indicated that precision was decreased with increasing sampling distance. Based on the modeled variograms measurements were interpolated using ordinary kriging and the resulting yield maps were reasonably mached with spatial pattern of soil nitrogen and weed density. The accuracy of interpolated yields with kriging at 10 and 20 m sampling distance was validated against the observed yields. Yield prediction accuracy was improved with cokriging particularly when soil nitrogen content was defined as covariate in both distances however, this method of interpolation was more efficient at 10 m sampling distance. Conclusion Based on the results it was concluded that spatial variability of wheat yield could be mapped with good accuracy using simple kriging when sampling distance is 10 m or lower. However, at 20 m sampling distance accurate yield maps were obtained after co-kriging with covariates such as soil nitrogen or wee density which are highly correlated with yield. Spatio-tempral yield variation could be studied by repeating such an experiment in different years.