10:00 AM - 10:15 AM
[MAG33-05] Operational crops yield prediction from MODIS and VIIIRS surface reflectance in Europe and USA for summer and winter crops
Keywords:agriculture, yield estimation, remote sensing, optical sensors
Global and regional agricultural monitoring systems aim to provide up-to-date information on food production to the various actors and decision-makers in favor of global and national food security. Few systems today fulfill this role and provide production forecasts on a global scale as The Foreign Agricultural Service (FAS) of the United States Department of Agriculture (USDA) or The Global Information and Early Warning System (GIEWS) from the Food and Agriculture Organization (FAO). Even if quarterly and annual bulletins at national scale are provided, sub-national results are missing. Furthermore, there is a lack in uncertainties quantification. In this context, one of the main priorities is to provide a frequent crop yield and production estimation during all the growing season at global and regional scale with uncertainties.
In this study, we evaluated the ability of surface spectral reflectance for red and Near Infra-Red (NIR) bands (bands 1, 2) at 250 m from the Moderate Resolution Imaging Spectroradiometer (MODIS) and at 500 m from the Visible Infrared Imaging Radiometer Suite (VIIRS) to correctly estimate the production as an input of the yield forecasting model ARYA (Agriculture Remotely sensed Yield algorithm). The model is based on a Gaussian fit between the Difference Vegetation Index (DVI) and the temperature accumulation in growing degree days (GDD) estimated from ERA-5 reanalysis. Additionally, the model includes a correction for crop stress conditions captured by the difference in daily accumulated Land Surface Temperature (LST) and the air temperature from ERA-5.
First, this work presents the adaptation of the yield forecasting model to an operational model included the uncertainties. The model has been operational since 2020 for winter wheat and gave a new yield and production estimation every week from 3 to 4 months before harvest date up to the harvest date. The operational model has been calibrated and validated over the past 20 years for major winter wheat producers in Europe, Russia and the United States at regional and national level. The model succeeds in estimating the yield for wheat with an error around 5% and an RMSE around 20%.
After implementing the model for major winter wheat producers, the model was adapted for summer crops including corn, soybeans and sunflower over Ukraine and Russia. The model succeeds in estimating the yield with a coefficient of determination higher than 0.7 and an RMSE around 12%-20% (Figure 1) at national and sub-national level respectively from the planted date up to the harvested date. Due to the decommissioning of MODIS in 2023, to ensure continuity, the daily surface reflectance for the red and NIR bands of VIIRS has been implemented in the operational yield prediction model. The impact of the difference of the spatial resolution from 250 m (MODIS) to 500 m (VIIRS) and the lack of spatial and temporal coverages (1 satellite instead of 2) has been studied. The results are promising with an error from 6% with MODIS to 7% with VIIRS over Ukraine for 10 years of winter wheat estimation.
In this study, we evaluated the ability of surface spectral reflectance for red and Near Infra-Red (NIR) bands (bands 1, 2) at 250 m from the Moderate Resolution Imaging Spectroradiometer (MODIS) and at 500 m from the Visible Infrared Imaging Radiometer Suite (VIIRS) to correctly estimate the production as an input of the yield forecasting model ARYA (Agriculture Remotely sensed Yield algorithm). The model is based on a Gaussian fit between the Difference Vegetation Index (DVI) and the temperature accumulation in growing degree days (GDD) estimated from ERA-5 reanalysis. Additionally, the model includes a correction for crop stress conditions captured by the difference in daily accumulated Land Surface Temperature (LST) and the air temperature from ERA-5.
First, this work presents the adaptation of the yield forecasting model to an operational model included the uncertainties. The model has been operational since 2020 for winter wheat and gave a new yield and production estimation every week from 3 to 4 months before harvest date up to the harvest date. The operational model has been calibrated and validated over the past 20 years for major winter wheat producers in Europe, Russia and the United States at regional and national level. The model succeeds in estimating the yield for wheat with an error around 5% and an RMSE around 20%.
After implementing the model for major winter wheat producers, the model was adapted for summer crops including corn, soybeans and sunflower over Ukraine and Russia. The model succeeds in estimating the yield with a coefficient of determination higher than 0.7 and an RMSE around 12%-20% (Figure 1) at national and sub-national level respectively from the planted date up to the harvested date. Due to the decommissioning of MODIS in 2023, to ensure continuity, the daily surface reflectance for the red and NIR bands of VIIRS has been implemented in the operational yield prediction model. The impact of the difference of the spatial resolution from 250 m (MODIS) to 500 m (VIIRS) and the lack of spatial and temporal coverages (1 satellite instead of 2) has been studied. The results are promising with an error from 6% with MODIS to 7% with VIIRS over Ukraine for 10 years of winter wheat estimation.