Phytolith assemblage analysis for the identification of rice paddy

The rice arable system is of importance to both society and the environment. The emergence of rice paddies was a crucial step in the transition from pre-domestic cultivation to systematic land use and management. However, many aspects of the formation of rice farming systems remain unclear. An important reason is the lack of reliable methods for identifying early rice paddies. One possible means of remedying this knowledge deficit is through analysis of phytolith assemblages, which are closely related to their parent plant communities. In this study, phytolith assemblages from 27 surface soil samples from wild rice fields, 91 surface soil samples from modern rice paddies, and 50 soil samples from non-rice fields were analysed to establish a discriminant function. This discriminant function enabled classification of 89.3% of the samples into appropriate groups. Further, the results suggested that phytolith assemblages can be used to identify rice fields and differentiate between wild rice fields and domesticated rice fields. The method was demonstrated to be an effective way of utilising the large amounts of unidentifiable phytoliths discovered at archaeological sites to provide a modern analogue that may be a valuable key to unlocking the past.

society 3,51-54 and the environment [55][56][57] . As rice arable systems are associated with specific grasses, weed assemblages [58][59][60][61] can be used to determine changes in arable systems. Recently, phytolith assemblages, based on the proportion of fixed and sensitive types have been used to distinguish between wet and dry cultivation systems [62][63][64] ; however, similar studies using phytolith assemblages to differentiate wild and domesticated rice paddies are rare.
In the present study, a detailed discriminant analysis was conducted on modern soil samples ( Fig. 1) to develop a method for identifying rice fields using phytolith assemblages as indicators. The results demonstrated that modern phytolith assemblages can be used to differentiate between wild rice fields, domesticated rice paddies, and non-rice fields.

Results
Phytolith assemblages in selected samples. A total of 170 soil samples were collected and 168 of these contained abundant phytoliths. Twenty-one phytolith morphotypes were identified (Fig. 2), including bilobate, cylindrical polylobate, cross, long saddle, short saddle, rondel, trapeziform sinuate, acicular hair cell, elongate psilate, elongate echinate, square, rectangle, bulliform, Cyperaceae-type, globular echinate and barnyard grass husk-type. Soil samples from wild rice and domesticated rice fields were characterised by the presence of bulliform phytoliths with fish-scale decorations. Characteristic phytolith types are shown in Fig. 3.
Phytolith assemblages from the 50 non-rice field samples were different from those of rice field samples in that high rates of square (22.14%), elongate psilate (13.67%), bulliform (12.78%) and bilobate (11.47%) phytoliths were present. In particular, four samples (10CL-FB1, 10CL-FB2, WN-BT5, and ZX-4) in this group contained a low rate of rice bulliform phytolith particles. Although the four soil samples did not come from wild rice or domesticated rice fields, their sampling locations were extremely close to wild or domesticated rice fields. Therefore, it is possible that these samples contained rice bulliform phytoliths because of soil tillage or other soil-disturbing activities.
Numerical analysis of modern phytolith data. We used the data on 21 phytolith types (Supplementary   Table S2) in a discriminant analysis. The discriminant scores of 168 samples were plotted along the first two discriminant functions, accounting for 64% and 36% of variance, respectively (Fig. 4). The results showed that the three group centroids were clearly separated, marginally overlapping with each other ( Fig. 4; black square). The results of the discriminant analysis showed that 89.3% (150 out of 168) of the original grouped cases were classified correctly (Table 1). Only two of the 27 samples from the wild rice field were misclassified. Of the samples from the domesticated rice paddies, 90.1% were classified correctly and nine samples were misclassified. Seven samples in the non-rice field group were classified in the rice-field group. Regarding cross-validation (each case was classified using the functions derived from all the other cases), 82.1% of the cross-validated grouped cases were classified correctly ( Table 1). The high degree of correct classification suggested that the classification of the surface soil samples into the three groups was statistically robust.

Discussion and Conclusions
Earlier studies on phytolith assemblage focused mainly on palaeoenvironment reconstruction 65,66 . Only a few specific crop-related phytoliths are highly valued for their application in archaeology 41,43,[67][68][69][70][71] . Therefore, a large proportion of unidentifiable phytoliths discovered at archaeological sites are undervalued. Excluding a few cases [62][63][64] , phytoliths could be exploited differently for archaeological applications by combining with statistical analyses, such as correspondence analysis, discriminant analysis, and principal component analysis.
This study demonstrated that rice paddies could be distinguished by their phytolith assemblages. Soil samples from both the wild rice fields and domesticated rice paddies were characterised by high proportions of bilobate, square and elongate psilate phytoliths, whereas the non-rice field samples exhibited diverse phytolith assemblages dominated by the rondel type. Furthermore, according to phytolith grouping for fixed and sensitive types 64 , a higher proportion of bilobate phytoliths (a fixed type) was observed in the soil samples from the domesticated rice paddies, whereas a higher rate of elongate psilate (a sensitive type) phytoliths was obtained from the wild rice fields. This probably indicates that wild rice habitats are swampier than those of domesticated rice. The proportion of rice bulliform phytoliths is informative, although it is relatively low. Rice bulliform phytoliths are unique to rice plants and the decorations on the edges of rice bulliform phytoliths can be used to differentiate wild rice from domesticated rice 41,42 . However, because of the limited number of samples from Southern China, the results were not compared with samples from Northern China, South Asia, and Southeast Asia, which are also major rice-producing areas.
This study analysed the phytoliths present in 168 soil samples collected from wild and domesticated rice fields and non-rice fields to establish a discriminant function, and demonstrated that these phytolith assemblages correctly classified 89.3% of the samples. The results statistically support the development of a robust method for identifying rice paddies and distinguishing between wild and domesticated rice fields. The method was demonstrated to be an effective way of utilising the large numbers of unidentifiable phytoliths from archaeological sites and this modern analogue should provide a valuable key to unlocking the past.

Methods
In this study, a total of 170 soil samples were collected for phytolith analysis (Fig. 1). These included (1) Supplementary Table S1. All necessary permits for sample collection from Jiangxi were obtained from the Jiangxi Academy of Agricultural Sciences. No permit was required to collect samples from the other locations.
Phytoliths were extracted from the soil samples according to the procedure described by Zhang 72 , with minor modifications. Initially, 5 g of soil samples was weighed. Subsequently, 30% H 2 O 2 and 15% HCl were added to the samples to remove organic matter and carbonates. The samples were then subjected to heavy liquid flotation using ZnBr 2 (density, 2.35 g/cm 3 ) to separate the phytoliths, which were subsequently mounted on a slide with Canada Balsam. The phytoliths were counted and identified under a Leica microscope at 400X magnification. More than 400 phytolith particles from grass were counted in each sample. A few woody phytoliths were identified in this study; however, because these can be highly influenced by local environment and climatic factors, they were excluded to minimize possible bias 73 . Identification of phytoliths was performed using previous studies as references 39,41,67,74 .
Discriminant analysis was performed using SPSS version 22.0. It is a method of statistical inference for deriving linear combinations of variables, called discriminant functions, which are mutually independent 75,76 . These discriminant functions ensure maximum separation among priori sample groups and can also be used to classify new samples with unknown group memberships into one of the priori groups [75][76][77] . In this study, 168 samples were initially divided into three priori groups (actual); subsequently percentage data of 21 phytolith types were used to establish the discriminant functions and the samples were classified into predicted groups.