Supporting Information for the publication: Title: Predicting plant water availability from phytolith assemblages in finger millet, pearl millet and sorghum Authors: D’Agostini Francesca, Ruiz-Pérez Javier, Madella Marco, Vadez Vincent, Lancelotti Carla This repository contains all the supplementary information cited in the manuscript. Contents: File S1: Experimental cultivation.xlsx- Excel file that includes the database of the experimental cultivations (both 2019 and 2020) with the specification of the landraces tested, experimental layouts, physiological data (counting weekly weights, transpiration rate, biomass values, water added and flowering time), soils elemental composition and metadata. File S2: Phytolith data.xlsx- Excel file that includes the database of the phytolith counting (both 2019 and 2020), inclusive of row data, morphotype concentrations, ratios and metadata. File S3: R script - R code that includes all the scripts used for statistics and data visualisation. Figure S1: weekly transpiration trend of crops grown in a) 2019 and b) 2020. Figure S2: boxplots and linear regressions (in relation to total water transpired in L) of the normalised phytolith concentration and ratio of sensitive to fixed morphotypes for sorghum, sorted per species and tissue of deposition. Figure S3: boxplots and linear regressions (in relation to total water transpired in L) of the normalised phytolith concentration and ratio of sensitive to fixed morphotypes for finger millet, sorted per species and tissue of deposition. Figure S4: boxplots and linear regressions (in relation to total water transpired in L) of the normalised phytolith concentration and ratio of sensitive to fixed morphotypes for pearl millet, sorted per species and tissue of deposition. Figure S5: logistic regression plots (predictive curves) for acute bulbosus, for the all dataset and divided per species. Figure S6: logistic regression plots (predictive curves) for bulliforms flabellate, for the all dataset and divided per species. Figure S7: logistic regression plots (predictive curves) for bulliforms (general category), for the all dataset and divided per species. Figure S8: logistic regression plots (predictive curves) for crosses, for the all dataset and divided per species. Figure S9: logistic regression plots (predictive curves) for elongates clavate, for the all dataset and divided per species. Figure S10: logistic regression plots (predictive curves) for elongates dentate, for the all dataset and divided per species. Figure S11: logistic regression plots (predictive curves) for elongates entire, for the all dataset and divided per species. Figure S12: logistic regression plots (predictive curves) for elongates sinuate, for the all dataset and divided per species. Figure S13: logistic regression plots (predictive curves) for polylobates, for the all dataset and divided per species. Figure S14: logistic regression plots (predictive curves) for rondels, for the all dataset and divided per species. Figure S15: logistic regression plots (predictive curves) for saddles, for the all dataset and divided per species. Figure S16: logistic regression plots (predictive curves) for stomata, for the all dataset and divided per species. Abstract: This work investigates the relationship between phytolith deposition and water availability in finger millet, pearl millet and sorghum, by providing new insights about the role phytoliths may play in response to water-deficit. Two phytolith proxies were tested as indicators of water availability: the ratio of sensitive/fixed morphotypes and a new predictive model built on the complete assemblage of the three species. Results show a relationship between transpiration and phytoliths deposition in the three species analysedthat can be better predicted by the application of a logistic regression model than with the use of the ratio of sensitive to fixed morphotypes. Some morphotypes are positively correlated to water availability and some are negatively correlated. Indeed, this result enhances the model’s strength. The significance of certain morphotypes in water availability response varies based on the species and the tissue of deposition. Finger millet, pearl millet and sorghum phytolith production is not commensurate, as water stress prompts each species to alter their phytolith production in a distinct manner. The model proposed has a great archaeological application. The outcomes of this investigation are of interest to archaeologists, seeking for a proxy to detect past growing conditions of C4 crops.