Time-Series Analysis of Sentinel-2 Satellite Images for Sunflower Yield Forecasting
Accurate estimates and predictions of sunflower crop yields at the field level are critically important for farmers, service dealers, and policymakers. Several models based on remote sensing vegetation indices have been developed in yield assessment, but their robustness—especially in small field scale areas—needs to be examined. Here we aim to assess the damage wrought by the European common hamster to crop production and to develop a robust methodology for estimation/prediction of sunflower yield at pilot field scale using remote sensing vegetation indices generated from Sentinel-2 satellite imagery. We conducted the study in Mezőhegyes, south-eastern Hungary. In this research we derived the sunflower forecasting model by linear regression statistical analysis between the normalized difference vegetation index (NDVI) and the modified soil adjusted vegetation index 2 (MSAVI2), the fraction of absorbed photosynthetically active radiation (fAPAR), and crop yield data provided by a combine harvester equipped with a yield-monitoring system. We examined the results of the forecasting model (predicted) against the actual yield data (observed) collected by a combine harvester. We tested and validated the result of the model in 11 different parcels to evaluate the robustness of the developed method. We found a great effect of the hamster in June during the peak vegetation period, leading to crop yield reduction, when we analysed LiDAR DEM and high-resolution Google Earth images. The results demonstrated that using NDVI, MSAVI2, and fAPAR, the best time to predict sunflower yields was between 85–105 d into the growing season. MSAVI showed the strongest correlation (R=0.7), followed by fAPAR (R=0.677) and NDVI (R=0.638). Normalized root means square error (RMSE) ranged from 524 to 566.5 kg/ha for different spectral indices. To increase the model accuracy and the correlation coefficient, we developed a multiple linear regression (MLR) model using two VIs in combination with a biophysical product (NDVI+MSAVI2+fAPAR−MLR) as an independent variable against observed crop yield data. This model achieved high accuracy with value of RMSE = 516 kg/ha and R = 0.72, respectively Our results are promising because they prove the possibility of forecasting sunflower grain yield at the field level, 5–6 weeks before the harvest, which is crucial for planning food policy
Year of publication: |
[2022]
|
---|---|
Authors: | Amankulova, Khilola ; Farmonov, Nizom ; Mucsi, László |
Publisher: |
[S.l.] : SSRN |
Subject: | Prognoseverfahren | Forecasting model | Zeitreihenanalyse | Time series analysis | Theorie | Theory |
Saved in:
freely available
Saved in favorites
Similar items by subject
-
Optimal spatiotemporal prediction of karstwater levels
Berke, Olaf, (1999)
-
Rodríguez Ramos, Carlos Antonio, (2003)
-
Filtering the noise from time series and spatial data
Berke, Olaf, (1998)
- More ...