A scalable physical model based on remote sensing in paddy yield estimation

Rice is one of the most strategic plants in Iran. On the other hand, agriculture makes a wide variety of environmental amenities and problems. Thus researches that help the production and sustainable development in this area are significant. The main purpose of this research is the design and development of a scalable remote sensing-based paddy yield model.

In this study, we used several different images available in Google Earth Engine (GEE) to estimate paddy yield at various temporal (growing seasons) and spatial scales (from 30 m resolution to regional scales). Then, a remote sensing-based light use efficiency (LUE) model integrated with inanimate environmental stressors, was implemented. This operational model was assessed against actual field-level yield data in 2016, 2017, and 2019 growing seasons across more than 691 paddy fields in Gilan province.The efficiency of the current model was evaluated through different statistical measures. The results showed a positive correlation and a signed agreement between the estimated and measured values so that in the studied growing seasons, the average correlation coefficient (R) and agreement index (d) was equal to 0.55. The average RMSE equal to 500 kg/ha, the average MAE equal to 440 kg/ha, and the average NRMSE equal to 0.12, all indicate that the accuracy of the model in estimating crop yield in these locations and years is satisfactory. Also, the submitted model showed the appropriate variability of yield values at the farm scale.

In general, this new approach has confirmed that the use of remote sensing in the GEE is appropriate for estimating crop yield at various temporal and spatial scales, as the current model can be utilized in a wide range of applications such as agricultural management and insurance.