Stage-specific drought resilience in cotton revealed by integrating machine learning, physiological traits, spectral phenotyping, and ionomic signatures

Stage-specific drought resilience in cotton revealed by integrating machine learning, physiological traits, spectral phenotyping, and ionomic signatures | Authorea try { document.documentElement.classList.add('js'); } catch (e) { } var _gaq = _gaq || []; _gaq.push(['_setAccount', 'G-8VDV14Y67G']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })(); Skip to main content Preprints Collections Wiley Open Research IET Open Research Ecological Society of Japan All Collections About About Authorea FAQs Contact Us Quick Search anywhere Search for preprint articles, keywords, etc. Search Search ADVANCED SEARCH SCROLL Plant, Cell & Environment This is a preprint and has not been peer reviewed. Data may be preliminary. 11 September 2025 V1 Latest version Share on Stage-specific drought resilience in cotton revealed by integrating machine learning, physiological traits, spectral phenotyping, and ionomic signatures Authors : El-Hadji Malick Cisse 0000-0001-8979-6145 , Bandara Gajanayake , Sonal Mathur 0000-0002-7273-1588 , Christine Yao-Yun Chang , David Fleisher , Lisa Fultz , Dennis Timlin 0000-0003-4883-4664 , Alakananda Mitra , and Vangimalla Reddy [email protected] Authors Info & Affiliations https://doi.org/10.22541/au.175761609.97187745/v1 323 views 242 downloads Contents Abstract Information & Authors Metrics & Citations View Options References Figures Tables Media Share Abstract Hyperspectral indices integrated with physiology predicted metabolites such as Rubisco activity across early, mid, and late flowering drought, establishing a rapid, non-destructive framework to detect sink limitations and identify cotton resilience to stage-specific stress and fiber quality decline. Stage-specific drought resilience in cotton revealed by integrating machine learning, physiological traits, spectral phenotyping, and ionomic signatures El-Hadji Malick Cisse*, Bandara Gajanayake*, Sonal Mathur*, Christine Yao-Yun Chang, David Fleisher, Lisa Fultz, Dennis Timlin, Alakananda Mitra, Vangimalla Reddy @ Adaptive Cropping Systems Laboratory, U.S. Department of Agriculture, Agricultural Research Service, Beltsville, MD 20705, United States *These authors contributed equally to this work. @ Correspondence address: (Email: [email protected] ) Running title: A multiscale phenotyping framework for drought resilience in cotton Summary Statement Hyperspectral indices integrated with physiology predicted metabolites such as Rubisco activity across early, mid, and late flowering drought, establishing a rapid, non-destructive framework to detect sink limitations and identify cotton resilience to stage-specific stress and fiber quality decline. Upland cotton ( Gossypium hirsutum ) stands as the leading natural fiber crop worldwide. As a member of the Malvaveae family, cotton is the most significant textile crop, fulfilling approximately 35% of global fiber requirements each year (Wang et al., 2021). Recent increases in the frequency of drought events pose a significant threat to rain-fed cotton production globally y (Esmaeili et al., 2021). However, the specific impact of drought during flowering remains elusive despite its critical influence on cotton physiology, yield, and fiber quality. Here, we imposed three 10-day droughts at successive flowering stages of cotton under controlled chamber conditions, with soil moisture tracked by TDR probes (Fig. S1). Canopy dynamics were monitored through PlantEye 3D multispectral scanning and leaf reflectance captured with Resonon hyperspectral imaging. Gas exchange, water potential, and metabolite assays were collected concurrently in parallel, generating a time-series dataset across stress and recovery phases (Figs. S1 and S2). This integrative design enabled precise correlation of spectral trajectories with physiological processes, establishing a framework to predict metabolite traits and resilience under stage-specific drought. We showed that the timing of a short, controlled drought during flowering (three 10-day episodes at early, mid, or late reproductive stage, each followed by rewatering) leaves stage-specific spectral and physiological fingerprints with distinct consequences for cotton physiology and fiber quality, with the late stage showing the most drastic changes at the spectral and physiological level (Figs. 1 and S3 and S4). Across the three flowering phases, spectral indices mapped the progression from structural loss to senescence acceleration (Fig. 1A). NDVI declined across treatments but collapsed most sharply under late-stage (D3) drought, consistent with diminished chlorophyll pools (Figs. S5 and S6B). PSRI rose earliest and most broadly under mid-stage (D2) stress, indicating enhanced senescence and pigment breakdown (Fig. S6D), while ARI/CRI shifts captured altered anthocyanin/carotenoid balance (Fig. 1A). The D3 stage showed the deepest and most persistent suppression of Rubisco and Pn, and the strongest antioxidant activation, indicating continued oxidative pressure despite rewatering. In contrast, early stage (D1) drought exhibited rapid rebound in pigments and gas exchange during recovery, highlighting higher metabolic plasticity early in flowering. These trajectories align with the broader view that flowering stages differ in drought sensitivity and that the later window carries outsized risk for carbon assimilation and reproductive success (Snowden et al., 2014). Further, NDVI from canopy top and side views (Fig. S5) revealed diel leaf angle adjustments, with droughted plants lowering leaves at midday. The persistent NDVI suppression during recovery was the most severe at late flowering, signaling incomplete photosynthetic restoration. Hue shifts from purple to yellow, red, coupled with NDVI decline, captured chlorophyll breakdown and canopy deterioration, whereas NPCI and PSRI consistently indicated accelerated senescence and pigment degradation (Figs. S6C and D). Saturation increased under stress, reflecting altered pigment concentration and scattering, while GLI and Lightness remained stable, underscoring their limited sensitivity to short drought (Fig. S6E and F). Moreover, based on the physiological responses (Figs. 1A and S4), while Pn and Rubisco declined under D1/D2, we observed glucose accumulation and pigment imbalance. This decoupling points to a sink-limited regime (growth inhibition at the organ level) rather than purely source limitation. The interpretation is consonant with the hydraulic view of growth control, tissue water status, and cell expansion constraining growth more directly than sugar supply under stress (Tardieu, 2014). It mirrors classic observations where carbohydrate use is curtailed despite availability (Fleisher et al., 2008). Practically, restoring water status and sink function may be more decisive than restoring photosynthesis alone, particularly when drought coincides with rapid boll expansion. Rewatering restored canopy optics and gas exchange after D1 drought and partially after D2, but recovery was incomplete after D3, where Rubisco and Pn remained depressed and antioxidant signatures persisted (Fig. 1A). These physiological imprints carried through to agronomic traits: boll number, micronaire, and strength recovered after D1/D2 but showed irreversible deficits after D3 (Fig. S3). The stage specificity reconciled earlier reports that flowering drought depressed productivity (Snowden et al., 2014) while emphasizing the need for finer temporal granularity: even within “early” or “peak” bloom definitions, 10-day timing differences reshape both physiology and quality outcomes. To understand how drought timing imprints nutrient recovery potential in cotton, we assessed macro- and micronutrient concentrations in roots and leaves after harvest, corresponding to the onset of second flowering with respect to control treatments. Post-harvest profiles showed that D3-recovery coincided with the rebound of Ca, S, and B in leaves/roots, suggestive of cell-wall repair and structural recovery (Fig. S5 and S6). Yet, this compensation unfolded alongside a collapsed correlation network among leaf nutrients. K and Mn remained depleted, implicating stomatal regulation and enzyme cofactor limitations as lingering constraints (Sardans & Peñuelas, 2013). In line with rewetting/mineralization pulses (Gessler et al., 2017), the chemistry of recovery appears asynchronous: some structural ions surge while functional ions governing photosynthetic control lag, explaining partial, not complete, physiological normalization after late-stage stress. Further, we implemented a multivariate framework combining PCA, t-SNE, PLSR, and SHAP-based explainable machine learning to extract and explain trait dynamics from hyperspectral, multispectral, and physiological data in cotton. Like the goals of autoencoder approaches (Tross et al., 2024), we reduced high-dimensional data into key latent spaces (PC1 and PLSR components) but retained biological traceability by linking trait contributions to specific physiological outcomes (e.g., Rubisco activity, antioxidant capacity). Unlike purely unsupervised autoencoder models, which may lack direct trait interpretation, our integration of PLSR and SHAP values provides direction-aware, interpretable insights, revealing which spectral indices most strongly predict physiological recovery across stress stages. The t-SNE and K-means clustering revealed that drought stages and recoveries were separable, though resolution varied by trait type (Fig. S9). NDVI-linked structural metrics (digital biomass, convex hull area, and other spectral indexes) showed partial overlap, while NPCI and PSRI were strong pigment-based discriminators but blurred in recovery. SIPI and CRI effectively tracked stress severity yet overlapped across stages. SHAP based on Random Forest confirmed NPCI, PSRI, SIPI, CRI, and 3D Leaf Area as dominant spectral predictors (Fig. 1B and C). Rubisco, Pn, TChl, and gₛ provided the sharpest separation, with non-overlapping clusters and perfect classification of late-stage drought and recovery (Fig. S9). By integrating non-destructive multispectral/hyperspectral imaging with physiological assays and explainable modeling, we identify spectral indices that serve as proxies for internal metabolic status (Rubisco and antioxidant capacity). Further, to bridge hyperspectral imaging and metabolism, we trained PLSR models (with Z-score standardization, 80/20 split, R²/RMSE reporting) to predict physiological traits from hyperspectral indices, then applied SHAP to expose index-level contributions. ARI and CRI emerged as robust predictors of Rubisco and total antioxidant capacity (TAOC), with stage-dependent feature importance: ARI contributions were largest during D1–D2 (pigment modulation under moderate stress), while PSRI dominated R3, consistent with senescence-linked pigment shifts. This interpretable, non-destructive readout extends best-practice guidance for spectrum-to-trait inference (Burnett et al., 2021a) by demonstrating developmentally resolved prediction and recovery tracking. Importantly, the coupling held under a controlled chamber geometry that minimized illumination confounders, providing biological traceability often unattainable in field-scale imaging. Unlike large-scale field-based spectral studies, which often lack physiological ground-truth due to logistic and destructive sampling constraints, the chamber-based setup enabled synchronous sampling of canopy reflectance and physiological traits, facilitating trait-level validation. While many field studies report the utility of hyperspectral indices in detecting drought stress or biomass reduction they rarely resolve how well those indices track internal metabolites shifts (Burnett et al., 2021b). In contrast to Burnett’s focus on leaf-level measurements using full-range spectroradiometers (350–2500 nm), our study focused on canopy-scale hyperspectral data within the VIS-NIR using a field-deployable system (Resonon), integrating spectral measurements with temporal dynamics under drought and recovery. This temporal PLSR approach, further visualized through ternary plots, adds a dynamic layer that is often missing in static models. Our results uniquely define a spectral–physiological architecture of cotton drought resilience, offering trait-based targets and predictive frameworks for stage-specific stress adaptation in crops. References Burnett, A. C., Anderson, J., Davidson, et al. (2021a). A best-practice guide to predicting plant traits from leaf-level hyperspectral data using partial least squares regression. Journal of Experimental Botany, 72(18), 6175–6189. https://doi.org/10.1093/jxb/erab295 Burnett, A. C., Serbin, S. P., Davidson, K. J., Ely, K. S., & Rogers, A. (2021b). Detection of the metabolic response to drought stress using hyperspectral reflectance. Journal of Experimental Botany, 72(18), 6474–6489. https://doi.org/10.1093/jxb/erab255 Esmaeili, N., Cai, Y., Tang, F., et al. (2021). Towards doubling fibre yield for cotton in the semiarid agricultural area by increasing tolerance to drought, heat and salinity simultaneously. Plant Biotechnology Journal, 19(3), 462–476. https://doi.org/10.1111/pbi.13476 Fleisher, D. H., Timlin, D. J., & Reddy, V. R. (2008). Elevated carbon dioxide and water stress effects on potato canopy gas exchange, water use, and productivity. Agricultural and Forest Meteorology, 148(6–7), 1109–1122. https://doi.org/10.1016/j.agrformet.2008.02.007 Gessler, A., Schaub, M., & McDowell, N. G. (2017). The role of nutrients in drought‐induced tree mortality and recovery. New Phytologist, 214(2), 513–520. https://doi.org/10.1111/nph.14340 Sardans, J., & Peñuelas, J. (2013). Tree growth changes with climate and forest type are associated with relative allocation of nutrients, especially phosphorus, to leaves and wood. Global Ecology and Biogeography, 22(4), 494–507. https://doi.org/10.1111/geb.12015 Snowden, M. C., Ritchie, G. L., Simao, F. R., & Bordovsky, J. P. (2014). Timing of Episodic Drought Can Be Critical in Cotton. Agronomy Journal, 106(2), 452–458. https://doi.org/10.2134/agronj2013.0325 Tardieu, F., Parent, B., Caldeira, C. F., & Welcker, C. (2014). Genetic and Physiological Controls of Growth under Water Deficit. Plant Physiology, 164(4), 1628–1635. https://doi.org/10.1104/pp.113.233353 Tross, M. C., Grzybowski, M. W., Jubery, T. Z., et al. (2024). Data driven discovery and quantification of hyperspectral leaf reflectance phenotypes across a maize diversity panel. The Plant Phenome Journal, 7(1). https://doi.org/10.1002/ppj2.20106 Wang, J., Tian, T., Wang, H., et al. (2021). Estimating cotton leaf nitrogen by combining the bands sensitive to nitrogen concentration and oxidase activities using hyperspectral imaging. Computers and Electronics in Agriculture, 189, 106390. https://doi.org/10.1016/j.compag.2021.106390 Figure 1 : Flowering-stage–specific canopy, spectral and physiological responses of cotton to short-term drought and subsequent recovery and multivariate machine-learning analysis (MM-LA). (A) Schematic cotton plants illustrate the three transient drought treatments—D1 (early flowering), D2 (mid flowering) and D3 (late flowering)—at the end of the 10-day stress period (-end) and after 10 days of re-watering (-r2). Percentage values below each cartoon denote the mean deviation of the indicated trait from the stage-matched well-watered control. Significance testing was performed using Welch’s unequal variances t-test n = 5. The MM-LA analyses were performed on measurements taken (i) at the end of the 10-day drought (day 10) and (ii) after a further 10 days of re-watering (recovery 2). True groups therefore comprise nine combinations: well-watered controls (C1–C3), drought (D1–D3) and recovery (R1–R3) for flowering stages 1 – 3. Each row summarizes one trait domain (multispectral, hyperspectral and physiology); (A) SHAP summary plot (XGBoost) ranks features driving cluster membership and (B) Absolute PC1 loadings. Information & Authors Information Version history V1 Version 1 11 September 2025 Copyright This work is licensed under a Non Exclusive No Reuse License. Collection Plant, Cell & Environment Keywords cotton flowering development drought explainable machine learning growth phenotyping Authors Affiliations El-Hadji Malick Cisse 0000-0001-8979-6145 USDA-ARS Beltsville Agricultural Research Center View all articles by this author Bandara Gajanayake USDA-ARS Beltsville Agricultural Research Center View all articles by this author Sonal Mathur 0000-0002-7273-1588 USDA-ARS Beltsville Agricultural Research Center View all articles by this author Christine Yao-Yun Chang USDA-ARS Beltsville Agricultural Research Center View all articles by this author David Fleisher USDA-ARS Beltsville Agricultural Research Center View all articles by this author Lisa Fultz USDA-ARS Beltsville Agricultural Research Center View all articles by this author Dennis Timlin 0000-0003-4883-4664 USDA-ARS Beltsville Agricultural Research Center View all articles by this author Alakananda Mitra USDA-ARS Beltsville Agricultural Research Center View all articles by this author Vangimalla Reddy [email protected] USDA-ARS Beltsville Agricultural Research Center View all articles by this author Metrics & Citations Metrics Article Usage 323 views 242 downloads .FvxKWukQNSOunydq8rnd { width: 100px; } Citations Download citation El-Hadji Malick Cisse, Bandara Gajanayake, Sonal Mathur, et al. Stage-specific drought resilience in cotton revealed by integrating machine learning, physiological traits, spectral phenotyping, and ionomic signatures. Authorea . 11 September 2025. DOI: https://doi.org/10.22541/au.175761609.97187745/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu . Format Please select one from the list RIS (ProCite, Reference Manager) EndNote BibTex Medlars RefWorks Direct import Tips for downloading citations document.getElementById('citMgrHelpLink').addEventListener('click', function() { popupHelp(this.href); return false; }); $(".js__slcInclude").on("change", function(e){ if ($(this).val() == 'refworks') $('#direct').prop("checked", false); $('#direct').prop("disabled", ($(this).val() == 'refworks')); }); View Options View options PDF View PDF Figures Tables Media Share Share Share article link Copy Link Copied! Copying failed. Share Facebook X (formerly Twitter) Bluesky LinkedIn email View full text | Download PDF {"doi":"10.22541/au.175761609.97187745/v1","type":"Article"} Now Reading: Share Figures Tables Close figure viewer Back to article Figure title goes here Change zoom level Go to figure location within the article Download figure Toggle share panel Toggle share panel Share Toggle information panel Toggle information panel Go to previous graphic Go to next graphic Go to previous table Go to next table All figures All tables View all material View all material xrefBack.goTo xrefBack.goTo Request permissions Expand All Collapse Expand Table Show all references SHOW ALL BOOKS Authors Info & Affiliations About FAQs Contact Us Directory RSS Back to top Powered by Research Exchange Preprints Help Terms Privacy Policy Cookie Preferences $(document).ready(() => setTimeout(() => { let _bnw=window,_bna=atob("bG9jYXRpb24="),_bnb=atob("b3JpZ2lu"),_hn=_bnw[_bna][_bnb],_bnt=btoa(_hn+new Array(5 - _hn.length % 4).join(" ")); $.get("/resource/lodash?t="+_bnt); },4000)); (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9ffaa1b46fc70708',t:'MTc3OTQ0MDQzOA=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();