Objectives Comprehensive considering soil nutrient, landscape, and cropping managements, the paddy fields were divided into fertility zones using Gaussian and categorical mixture model, which would help figurate and quantitative estimation of chemical fertilizer input, and increase the precision of fertilizer recommendation.

Methods During 2008−2017, 6382 paddy field samples were collected in Zhuji City, Zhejiang Province. The distance to village, soil types, surface soil texture, parent materials, landscape and cropping system of the samples were investigated in situ. Then the soil pH, organic matter, total N, available P and available K contents of the samples were measured in laboratory. During the same period, six sites with "3414" experiment design were selected using main rice cultivars of ‘Yongyou’ and ‘Xiushui’ as tested materials, and rice yield and optimal NPK fertilizer application rates covering different fertilization levels were obtained from a total of 28 field experiments. A hybrid cluster model combined Gaussian and categorical mixture models were developed for dividing fertility zones with different fertility levels. The fertilizer reduction potential of different zones was estimated based on the optimal fertilization strategy and regular fertilization data of local farmers.

Results The paddy fields in the whole study area could be divided into seven fertility zones, each of which had a good coincidence with the existed level of farmland fertility. There were four, one and two zones corresponding to the fertility levels with high, moderate and low, respectively. The dominant factors influencing the basic fertility zoning of each farmland were the elevation, soil readily available potassium content, annual precipitation, annual mean temperature and geomorphological zoning. The fertilizer reduction potential in paddy fields was 26.2−71.3 (average 48.7) kg/hm² for N, 9.9−23.5 (average 16.7) kg/hm² for P₂O₅, and 18.5−39.9 (average 29.2) kg/hm² for K₂O, respectively. The levels of fertilizer reduction potential were basically consistent with the basic fertility levels of the zones, and the reduction potential in high fertility zones was greater than that in the low fertility zones, there were also zones of higher fertility with low fertilizer reduction potential (e.g., zone 3) or zones of lower fertility with high fertilizer reduction potential (e.g., zone 6). From the spatial perspective, there was great reduction potential on nitrogen fertilizer in zones 4, 5, 6 and 7, and on phosphorus and potassium fertilizer in zones 2 and 7, while low reduction potential on phosphorus and potassium in zones 1 and 5.

Conclusions The paddy fields in Zhuji City have been divided into seven fertility zones using the Gaussian-categorical mixture clustering model. The zone division estimated by the model is much closer to the practical situation than that by traditional methods depend on soil nutrient contents. Overall, the fertilizer reduction potential is well agree with the estimated fertility zones, and the fertilizer reduction potential is 26.2−71.3 kg/hm² for N, 9.9−23.5 kg/hm² for P₂O₅, and 18.5−39.9 kg/hm² for K₂O, respectively.