Abstract
Smallholder rice farmers need a multi-purpose model to forecast yield and manage limited resources such as fertiliser, irrigation water supply in-season, thus optimising inputs and increasing rice yield. Active sensing tools like Canopeo and GreenSeeker-NDVI have provided the opportunity to monitor crop health and development at different growth stages. In this study, we assessed the effectiveness of in-season estimation of rice yield in lowland fields of northwest Cambodia using weather data and vegetation cover information measured with; (1) the mobile app-Canopeo, and (2) the conventional GreenSeeker hand-held device that measures the normalised difference vegetative index (NDVI). We collected data from a series of on-farm field experiments in the rice-growing regions in 2018 and 2019. Average temperature and cumulative rainfall were calculated at panicle initiation and pre-heading stages when the crop cover index was measured. A generalised additive model (GAM) was generated using log-transformed data for grain yield, with the combined predictors of canopy cover and weather data during panicle initiation and pre-heading stages. The pre-heading stage was the best stage for grain yield prediction with the Canopeo-derived vegetation index and weather data. Overall, the Canopeo index model explained 65% of the variability in rice yield and Canopeo index, average temperature and cumulative rainfall explained 5, 65 and 56% of the yield variability in rice yield, respectively, at the pre-heading stage. The model (Canopeo index and weather data) evaluation for the training set between the observed and the predicted yield indicated an R2 value of 0.53 and root mean square error (RMSE) was 0.116 kg ha−1 at the pre-heading stage. When the model was tested on a validation set, the R2 value was 0.51 (RMSE = 925.533 kg ha−1) between the observed and the predicted yield. The NDVI-weather model explained 62% of the variability in yield, NDVI, average temperature and cumulative rainfall explained 3, 62 and 54%, respectively, of the variability in yield for the training set. The NDVI-weather model evaluation for the training set showed a slightly lower fit with R2 value of 0.51 (RMSE = 0.119 kg ha−1) between the observed and the predicted yield at pre-heading stage. The accuracy performance of the model indicated an R2 value of 0.46 (RMSE = 979.283 kg ha−1) at the same growth stage for validation set. The vegetation-derived information from Canopeo index-weather data increasingly correlated with rice yield than NDVI-weather data. Therefore, the Canopeo index-weather model is a flexible and effective tool for the prediction of rice yield in smallholder fields and can potentially be used to identify and manage fertiliser and water supply to maximise productivity in rice production systems. Data availability from more field experiments are needed to test the model’s accuracy and improve its robustness.
Highlights
Rice production is a significant contributor to the Cambodian economy for poverty reduction, income growth and national and household level food security
We developed two models from generalised additive model (GAM), which were the Canopeo-weather model and normalised difference vegetative index (NDVI)-weather model
There was a positive relationship between yield and Canopeo with simple linear regression (SLR)
Summary
Rice production is a significant contributor to the Cambodian economy for poverty reduction, income growth and national and household level food security. A past survey reported that rice yield varied from 0.6 to 5.5 t ha−1 across villages and regions in Cambodia due to variability in management practices [3]. The inability to predict yield may result in low yield given that crop yield may vary from year to year, depending on the local weather conditions, soil types and management inputs [4]. Rice farmers need scalable decision support tools to enhance resource use efficiency and intensify rice production. The discourse to increase rice productivity in smallholder fields has extended to sustainable and practices and precision agriculture [5]. Achieving an increase in rice productivity in smallholder fields will require a locally calibrated and validated model to predict yield at different rice growth stages
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have