By Djakhangir Atakhanov (Uzbekistan)
Abstract:Crop yield forecasting is very important in every country around the world. Central Asian countries are enriched with agricultural lands. Uzbekistan’s main sector is agriculture with large variety of different crops. Nowadays, 46% of all irrigated land is utilized for cotton. Cotton is considered as most important crop in Uzbekistan. Production of cotton plays a dominant role in the economy of Uzbekistan. Remote sensing assists to create a larger scale of spatial information. Therefore it was used for prediction of cotton yield in Tashkent province, which has 15 administrative districts and 12 of them are located in the agricultural zone. The prediction was made for Tashkent province (oblast) level on administrative district level. Classification of cotton fields using MODIS (MOD13Q1) has been applied by using a decision rule and applying Normalized Difference Vegetation Index (NDVI) thresholds for day of the year 113, 161 and 225. NDVI data was acquired in order to get information about biomass condition, in a view of vegetation index. Different NDVI-based indicators were studied and analyzed in terms of choosing the best option for that region. The regression model analysis was done to investigate the relationship between official historical data on yield, weather conditions and NDVI values. A regression model between NDVI and yield was applied using an overall temporal trend component and yearly deviations from this trend. The temperature and precipitation were studied in order to find a relationship among yield, NDVI and weather conditions. The correlations between these factors were low and it is assumed that weekly observations of cotton are required for establishing a better correlation between NDVI and weather conditions. Research has been successfully done and yield for 2013 was predicted at the province level as well as at administrative districts level. Province level and administrative district level has difference in valuable indicators, relationships between weather conditions and yield or correlation between weather conditions and NDVI. As indicators maximum value of NDVI and sum of NDVI (iNDVI=NDVI integral) were used in this research. Results have shown that MODIS 250 m spatial resolution is not the most suitable satellite sensor for Tashkent region and its administrative districts. In addition, there are many varieties of fruits, vegetables and other crops that make it very difficult to classify and make an error assessment of the classification and delineation of the cotton, as the cotton growing calendar is at the same time as other crops are growing. The results obtained at the province level show a decrease of yield in 2013 to 0.2 c\ha. Model was validated by Leave-one-out cross-validation (LOOCV) procedure and the calculated RMSE for the model for Tashkent province level expresses a low value (RMSE=0.0340), which means that squared error is low and methods are accurately done. The best indicator was identified for the province level and for each administrative district level separately, subsequently the forecasting has been done for all research levels. Finally, the correlation between the ground truth data (historical data) and NDVI forecasting had been conducted and R2=0.55 which expresses, that there is good correlation. Results obtained from this research indicates, that alternative satellites and indicators have to be assessed and analyzed. The harvesting of the cotton, and time planning can be improved, as well as agricultural management in order to get better outcomes.