Abstract:
In this study, we used Landsat Earth observations and gridded weather data along with global soil datasets available in Google Earth Engine (GEE) to estimate crop yield at 30 m resolution. We implemented a remote sensing and evapotranspiration-based light use efficiency model globally and integrated abiotic environmental stressors (temperature, soil moisture, and vapor deficit stressors). The operational model (Global Yield Mapper in Earth Engine (GYMEE)) was validated against actual yield data for three agricultural schemes with different climatic, soil, and management conditions located in Lebanon, Brazil, and Spain. Field-level crop yield data on wheat, potato, and corn for 2015–2020 were used for assessment. The performance of GYMEE was statistically evaluated through root-mean-square error (RMSE), mean absolute error (MAE), mean bias error (MBE), relative error (RE), and index of agreement (d). The results showed that the absolute difference between the modeled and predicted field-level yield was within ±16% for the analyzed crops in both Brazil and Lebanon study sites and within ±15% in the Spain site (except for two fields). GYMEE performed best for wheat crop in Lebanon with a low RMSE (0.6 t/ha), MAE (0.5 t/ha), MBE (−0.06 t/ha), and RE (0.83%). A very good agreement was observed for all analyzed crop yields, with an index of agreement (d) averaging at 0.8 in all studied sites. GYMEE shows potential in providing yield estimates for potato, wheat, and corn yields at a relative error of ±6%. We also quantified and spatialized the soil moisture stress constraint and its impact on reducing biomass production. A showcasing of moisture stress impact on two emphasized fields from the Lebanon site revealed that a 12% difference in soil moisture stress can decrease yield by 17%. A comparison between the 2017 and 2018 seasons for the potato culture of Lebanon showed that the 2017 season with lower abiotic stresses had higher light use efficiency, above-ground biomass, and yield by 5%, 10%, and 9%, respectively. The results show that the model is of high value for assessing global food production. © 2021 by the authors. Licensee MDPI, Basel, Switzerland.
