第八届青年地学论坛

Ratoon rice (RR), a planting mode that allows paddy rice to be harvested twice in the same stubble, offers greater benefits and lower input costs compared with those of double-season rice (DR) and single-season rice. Yet, it is uncertain where RR is suitable for distribution in China. Hence, the objectives were to map the regions suitable for RR at the national level, clarify the optimal factor combinations and dominant controlling factors, and estimate potential paddy rice planting regions/areas that are more suitable for RR compared to DR. This study proposed a combination of recursive feature elimination (RFE) method and random forest (RF) model. The RFE, an effective feature selection method, was employed to filter optimal factor combinations for RR and DR, respectively. Then, with the assistance of parameter optimization and 10-fold cross-validation, the optimal RF model was utilized to determine which regions of China are suitable for RR and DR, and which regions have a suitability of RR greater than that of DR, hence are more suitable for planting RR. The results showed that the RFE-RF model was able to map accurate suitability regions with confusion matrix and AUC value of RR and DR over 0.9. Accumulated temperature greater than 0°C, mean temperature of warmest, fraction of photosynthetically active radiation and accumulated temperature greater than 10°C were the key variables affecting the suitability distribution of RR. The total area of high and very high suitable for RR and DR were 49.01 million hectares (Mha) and 37.08 Mha, respectively. Regions more suitable for RR than DR were mainly located in the Sichuan Basin, southeastern Hubei and central Anhui, with high and very high suitable area of 23.21 Mha. These results indicate that RR has a large potential planting area in China. Making full use of climate resources and expanding the planting area of RR will be beneficial to increase annual yield and reduce environmental pollution.

Suitability region,Ratoon rice,China,Random Forest,Recursive feature elimination