| Cotton is the mainstay industry in Xinjiang, forecasting of cotton production inXinjiang timely and accurately has a pivotal role in national economic development inXinjiang. Establishment of north Xinjiang cotton yield estimation model will not onlyachieve a dynamic, fast, comprehensive, accurate forecasts cotton production in northXinjiang, but also provide a theoretical basis and practical support to establishment oflarge-scale cotton yield estimation system. Our study begin in2011July13andAugust14and September15, take cotton in Wulanwusu which in north Xinjiang asthe research object, combine with satellite synchronous transit time and collect thecotton leaf area index and normalized difference vegetation Index, and collect thesample area cotton actual production data after announcing of the cotton productionstatistics in middle November. Using the same time with field measurement remotesensing data to extract NDVI and establish cotton yield estimation model. Theresearch results are as follows:(1) Comparison of cotton LAI of different growth stages show that, northernXinjiang cotton LAI first increase rapidly from bud stage to flower stage, with themaximum value of3.688678, and then reduced to2.425173in boll opening stagequickly. Combined with field observations, NDVI data were extracted from remotesensing images and corrected. Through comparative analysis, cotton NDVI reducedgradually along with cotton bud stage, flower stage and boll opening stage.Correlation analysis of cotton LAI and NDVI in all growth stages indicates that LAIand NDVI not reach a significant correlation in the bud stage, though significantcorrelation level achieved in the flower stage and boll opening stage, the correlationcoefficient is low. Heterogeneity of cotton distribution, soil disturbance factors andlarge-scale mechanical sowing harvest defoliant agent anthropogenic interference inrecent years may lead to the relatively low correlation between cotton LAI and NDVI.(2) Establish cotton yield estimation models based on cotton LAI, NDVI and yield. The results show that fitting coefficient of yield estimation model achieve asignificant level both in cotton bud stage and flower stage, with the maximum valueof0.76in latter. Through the establishment of six regression models and analysis offit accuracy, the best time for the LAI-based cotton yield estimation of northernXinjiang is flower stage, the optimum model equation is Y=42.834+153.338LAI-44.116LAI~2+4.195LAI~3. Y is cotton yield (kg), through the six regression models. Inthis study, the results obtained from the goodness of fit and significance level analysisshow that the cotton yield estimation models among bud stage, flower stage and bollopening stage are not ideal. These may caused by errors generated in the process ofsatellite image processing, non-consistency of cotton observation and image pixelregion, the heterogeneity of cotton varieties, and the widespread use of defoliants.Yield estimation based on single index need further research works.(3) Utilizing the data of cotton LAI, NDVI and production, establish amulti-compound remote sensing model in the bud stage, flower stage and boll openingstage respectively. The results show that multiple composite models have reachedsignificant levels in the cotton bud stage and flower stage, and fit goodness of flowerstage is the best one. Analysis of yield estimation model accuracy indicates thatpredicted yield in north Xinjiang cotton multi-compound remote sensing yieldestimation system in the cotton bud stage significantly higher than the actualproduction. Yield estimation model is Y=13.883LAI-393.659NDVI+461.476; In thecotton flower stage, forecast and actual yields are closer, the accuracy rate reach at95.3%, multiple composite yield estimation model is Y=17.764LAI-123.049NDVI+232.146; In the cotton boll opening, multiple composite yield estimation model is Y=-5.871LAI+89.312NDVI in+194.416.Y is cotton yield (kg), the accuracy rate ofestimated production is high as99.6%. |
PDF Full Text Request