Congress Resources: Papers, posters and presentations

< Return to abstract list

An Experience Of Modeling The Winter Wheat Yield In Ukraine Using Satellite-based Vegetation Indices

Congress: 2015
Author(s): Inna Semenova (Odessa, Ukraine)

Keyword(s): Sub-theme 17: Climate change, impacts and adaptation,
AbstractInna Semenova
Odessa State Environmental University, Department of Theoretical Meteorology and Meteorological Forecasts
Lvovskaya str., 15, Odessa, 65016, Ukraine, tel. +380482326739,

Introduction. Ukraine is one of the leading producers and exporters of cereals in Europe. Winter wheat is more than 90% of the total gross volume of this culture. Early yield crop prediction is an important aspect of economic activity and sustainable development of country, which enhances the relevance of prediction methods.
The aim of the study is to develop a regression model of the winter wheat yield and its validation for Steppe area of Ukraine.
Methods/Materials. As the predictors were used three parameters: the European continental blocking index (ECBI), which considers the features of regional atmospheric circulation in the growing season; the Vegetation Conditions Index (VCI) and proposed the new Wet Vegetation Index (WVI).
As input data were used the data of 8 and 16-day composites of vegetation index MODIS / NDVI and MODIS/NDWI, averaged for each administrative area from the database of the GLAM project (http: // The initial sets of vegetation indices and crop yield were compiled for the period 2000-2013 on the 9 administrative areas of Ukraine belonging to the Steppe zone. Data of the average winter wheat yield was obtained from the Ukrainian Statistical Administration.
The spectral index NDVI is defined as the difference of the reflected radiation intensity in the red and near infrared bands of sensing, normalized to sum of these values. Based on NDVI was created the vegetation condition index (VCI) (Kogan, 1995), which reflects the influence of meteorological conditions on vegetation. The value of VCI>80% correspond to favorable conditions of growth, while VCI <35% indicate drought of varying intensity. Therefore VCI successfully used for monitoring drought in different areas of the globe (Dąbrowska-Zielińska K. et al., 2011, Kogan F. et al , 2011, Semenova I.G., 2014).
Normalized difference water index (NDWI), proposed by B.C. Gao (1996) is used to assessment of the moisture content in vegetation. This index calculates using two narrow channels centered at wavelengths around 0.86 and 1.24 m. Both channels are sensitive to fluid content in the plant, and the index is a measure of the number of molecules of liquid water, which interact with solar radiation that falls on the vegetation.
In this study we was introduced a new index based on the NDWI - Wet Vegetation Index (WVI), which is defined as the ratio of the difference between the current value and absolute multi-year NDWI minimum for the same period and the difference between the absolute multi-year maximum and minimum NDWI at the same time. Similarly to the index VCI, we can say that high values WVI (80%) correspond to healthy and juicy vegetation and WVI value less than 50% indicate on the dry and wilted plants.
The monthly values of regional blocking index ECBI were calculated using author's methodic (Semenova I.G., 2013). The ECBI reflects the condition of zonal flow at the level 300 hPa respectively to the climatic norm. When ECBI> 0, the blocking of the zonal flow fixed, if ECBI <0, the zonal flow is stronger than normal.
Results and Discussion. Modelling the yield of winter wheat was made on the basis of constructing a linear regression equation, where the predicted values were the mean absolute crop yield (c/ha) or the mean relative crop yield (relative to trend).
Choosing periods, in which is taken value of predictors VCI, WVI and ECBI, was carried out by assessing the closeness of correlation between crop yield and the corresponding parameter in the spring and summer. We have taken as the predictors the values during those weeks when the correlation coefficient was maximized. The greatest influence on the formation of harvest the atmospheric blocking occurs in March or February, and only in the Luhansk region the maximum correlation is in May. For most areas the maximum correlation coefficient between crop yield and monthly values of ECBI ranges about -0.40...- 0.60.
For most areas the VCI has the best correlation with crop yield during period 15-23 May, and only in Kirovohrad and Lugansk regions during 9-16 June. For the WVI in different regions the best correlation starts from May, 16 till June, 8. Thus, in Odessa, Kherson, Mykolaiv regions and Crimea the yield prediction can be made immediately after May, 23. In Donetsk region the prediction is possible after May, 31, in Zaporizhia and Dnipropetrovsk regions - after June, 8, in Kirovohrad and Lugansk regions - after June, 16.
Assessment of statistical modeling for the absolute and relative yield shown that for all regions the coefficients of multiple correlation and determination are high, and constructed regressions by F-test are statistically significant (for the 5% significance level), except Lugansk region. Standard error of modeling is in averaged about 4.3 c/ha.
To evaluate the quality of the winter wheat yield prediction using the developed model, we consistently have excluded one year in the initial dataset, constructed new regression and predicted for excluded year. For eight regions alternately were excluded from regression the seven years from 2007 to 2013. This period was characterized as the dry years (2007, 2012), so favorable for the formation of high yield (2008, 2013). It has been found that the average absolute error of forecast ranges between 2.3-5.8 c/ha for the model of absolute yield and 2.3-4.7 c/ha for the model with the trend. In this case, the best indicators are fixed for Kherson region and Crimea, the worst - for Kirovohrad and Donetsk regions. The standard deviation for both models slightly exceeds 5 kg/ha. The accuracy of prediction by both models is an average of 85-86%.
Conclusion. Developed model considers the physical mechanisms of the influence of the atmospheric circulation, current weather and physiological state of plants on the yield formation. According to relative effectiveness the proposed models improves the quality of forecast yield in 2-4 times compared to the trend in almost all Steppe areas. Forecast may be issued during period May, 23 - June, 16, therefore the earliness of prediction is 1-1.5 months before harvest. 1. Gao B.C. (1996) NDWI – a normalized difference water index for remote sensing of vegetation liquid water from space. Rem. Sens. of Env. 58. 257-266.
2. Dąbrowska-Zielińska K., Ciołkosz A., Malińska A., Bartold M. (2011) Monitoring of agricultural drought in Poland using data derived from environmental satellite images. Geoinf. Iss. 3, 1 (3). 87–97.
3. Kogan F.N. (1995) Droughts of the late 1980s in the United States as derived from NOAA polar orbiting satellite data. Bull. Amer. Met. Soc. 76 (5). 655-668.
4. Kogan F., Adamenko T., Kulbida M. (2011) Satellite-Based Crop Production Monitoring in Ukraine and Regional Food Security. In: Use of Satellite and In-Situ Data to Improve Sustainability. NATO Science for Peace and Security Series C: Environmental Security. - Springer Science+Business Media B.V. 99-104.
5. Semenova I.G. (2013) Regional atmospheric blocking in the drought periods in Ukraine // J. of Eart. Sci. and Eng. 3 (5). 341-348.
6. Semenova I.G. (2014). Using of vegetation indices for drought monitoring in Ukraine // Ukr. Hydromet. J. V.14. (in Ukrainian).
7. Singh R.P., Roy S., Kogan F. (2003) Vegetation and temperature condition indices from NOAA AVHRR data for drought monitoring over India. Int. J. Rem. Sens. 24 (22). 4393–4402.
2011 IWRA - International Water Resources Association - - Admin