Enhancing Spaceflight Imaging Data Using Simple Online Automated Plant Phenomics (SOAPP)

Goals and summary of application architecture

The goal of developing SOAPP was to implement a user-friendly Streamlit (www.streamlit.io) web application interface for creating both single-measurement and high-throughput phenomics analysis pipelines for the aerial tissue in plant images based on PlantCV and OpenCV functionality (Fig. 2A). It is important to note that this software is designed primarily to support analysis of aerial tissues. Quantifying root system phenotypes introduces an additional set of challenges, with the roots often embedded in opaque substrates that make analyses difficult. When root systems are visible, accurately mapping their highly branched architectures and distinguishing individual features within dense, often overlapping root structures presents a complex analytical problem. The latter challenge is particularly pronounced in spaceflight-grown samples, where constrained growth volumes and the absence of gravitropic cues can cause the root systems of neighboring plants to intermingle after only a few days of growth. Fortunately, image analysis tools that effectively quantify root phenotypes such as RhizoVision Explorer (Seethepalli et al., 2021), faRIA (Narisetti et al., 2021), and RootNav 2.0 (Yasrab et al., 2019) have already been developed. SOAPP is designed to provide complementary data on shoot growth and development.

Figure 2.

Overview of Using SOAPP

To build an analysis pipeline, users upload images to SOAPP and manually define measurement parameters for the first image. Once the first image is measured, the image analysis workflow can be automatically repeated on a set of user-uploaded images.

To guide users through the phenomics configuration process, SOAPP defines five steps in the phenomics process: (i) preprocessing, (ii) masking, (iii) defining regions of interest, (iv) image analysis, and (v) data retrieval (Fig. 2B). An additional optional step, that of high-throughput, allows the user to apply the analysis pipeline established in steps 1–5 to a series of images currently uploaded to SOAPP. Images with three (red, green, and blue) 8-bit color channels (24 bit-color), formatted as a Portable Network Graphic (PNG), Tagged Image File Format (TIFF), or Joint Photographic Experts Group format (JPEG) image files can be uploaded to SOAPP as the input data, regardless of original source.

(i). Preprocessing Images

SOAPP implements a modular and extensible image preprocessing system that enables user-directed, sequential application of discrete preprocessing modules through an intuitive interface (Fig. 2C). These modules include standard image formatting tools such as cropping to remove unwanted regions and image rotation, as well as application-specific functions like keystone correction for perspective distortions, color correction and standardization methods, and both manual and automated scale determination methods. All currently available preprocessing modules are listed in Table 2. The standardized structure of these modules facilitates easy development and integration of additional tools, making the system both flexible and customizable. This comprehensive and user-friendly approach ensures images are properly prepared for downstream phenotypic analysis, enhancing accuracy and supporting SOAPP's mission as an open-source solution for plant phenomics. The modular design of the preprocessing workflow architecture enables future expansion of preprocessing tools, such as camera lens distortion correction and defect-specific noise adjustments.

(ii). Masking Plant Foliage

SOAPP guides users through creating a binary mask that separates plant material from the background by converting the preprocessed image from RGB into the HSV (Hue, Saturation, Value) and LAB (Lightness, a, b) color spaces using PlantCV's rgb2gray_hsv and rgb2gray_lab functions. For each color channel, the user can apply binary, range, or Otsu thresholding via PlantCV's respective threshold. binary, threshold.range, and threshold.otsu functions. Binary thresholding converts pixels above/below a user-defined value to white/black, whereas range defines an upper and lower limit for valid pixel values, and Otsu automatically determines an optimal threshold. The resulting binary masks from different channels are then combined using Boolean operations through PlantCV's logical_and, logical_or, and logical_xor functions to create a final mask that accurately captures the plant material. This interactive masking process enables users to visualize and iteratively refine the masks until the plant is properly segmented from the background for downstream analysis steps (Fig. 2D). While currently relying on PlantCV's thresholding methods, SOAPP's modular architecture enables future integration of additional masking algorithms beyond PlantCV's offerings, including machine learning-based approaches such as semantic segmentation (Cheng et al., 2023).

(iii). Defining Regions of Interest

After creating the binary mask, SOAPP allows users to define regions of interest (ROIs) for analysis (Fig. 2E). This is accomplished by selecting a grid of ROIs overlaid on the image using PlantCV's roi.multi function. The defined ROIs are displayed as colored circles on the original image, and users can iteratively add, remove, or resize ROIs as needed to accurately capture the desired samples. While SOAPP currently uses circular ROIs, its modular design allows for integrating custom ROI selection methods in the future, such as advanced techniques like polygon or freehand ROI drawing tools. By providing both grid-based and individual ROI options along with visualization and adjustment capabilities, SOAPP ensures precise ROI definition for diverse sample arrangements, guiding downstream quantification of phenotypic traits for individual plants or groups simultaneously.

(iv). Running Image Analysis

After defining the ROIs, SOAPP allows users to select from a range of analysis methods to quantify phenotypic traits of the plant samples (Fig. 2F). The “Analyze Object” function of PlantCV extracts shape descriptors like area, perimeter, height, and width from the plant object within each ROI, providing insights into overall plant size and morphology. The “Analyze Color” function quantifies the color composition of plant material. It generates histograms and computes values such as hue, saturation, and value from the different RGB, HSV, and LAB color spaces, exposing changes in pigmentation that can potentially reveal features such as stress responses (Gil et al., 2022). The “Watershed Segmentation” analysis segments the plant into individual leaves/parts geometrically, enabling the counting of leaves/branches and measuring traits for each segment, facilitating the study of leaf-level phenotypes. While SOAPP currently implements the core PlantCV analysis suite, its modular architecture allows integrating additional analysis methods tailored to specific applications, including advanced techniques like machine learning-based feature extraction. Users can select any combination of these analyses based on their requirements. The analysis is performed on the masked plant regions using the corresponding PlantCV functions. All analysis outputs are generated in PlantCV's standard JSON format, providing a structured representation of the extracted phenotypic data. This consistent formatting enables integration with downstream data processing pipelines or visualization tools. By offering a range of analysis options along with the ability to extend with custom methods, SOAPP provides a flexible and powerful approach to quantifying diverse aerial phenotypic traits from plant images in a user-friendly manner.

(v). Data Retrieval

SOAPP provides users with multiple options for retrieving phenotypic analysis results. In addition to exporting data in PlantCV's native structured JSON format, data can be stored as a flattened comma-separated value (CSV) format for easy import into spreadsheets or statistical software (Fig. 2G). If scale information is available from manual input or automated scale-finding tools, all size-based measurements, such as area and perimeter, are automatically scaled to real-world units in the CSV export (Fig. 2G). When using unit scaling in morphological analysis, appropriately scaled phenotypic measurements will be reported in the morphology data output CSV file, identifiable with the suffix “_scaled” in the variable name. When color analysis is performed, SOAPP generates color histogram visualizations for each analyzed plant object using PlantCV's analyze_color function output to aid quantitative interpretation of color composition data. SOAPP also calculates mean and standard deviation metrics from these histograms using either the arithmetic mean for linear values, the circular mean for circular values such as hue, and the root mean square, which accounts for non-linear color encodings within images. All values are reported using the arithmetic mean (indicated with no suffix) and root mean square (indicated with measurement suffix “_sq”); however, the appropriate values should be used given the colorspace encoding of the user images. When using a color standard preprocessing model, standardized values are reported as either linear distance between calculated means and the standard value (indicated with measurement suffix “_corr”) or with the distance between observed and expected standard values being applied to the phenotypic measurement (indicated with measurement suffix “_corr_wb”). Additionally, SOAPP calculates a suite of RGB vegetative indices previously reported, such as the Red Green Blue Vegetation Index, listed in Table 3.

Reproducing and Automating Analysis

During any point of analysis, users can download an exact configuration file of their defined workflow. This configuration file is formatted as a human-readable JSON file. At the start of a subsequent analysis, the user can upload a configuration file to SOAPP, populating all previously defined parameters throughout all stages of analysis. Additionally, users can also download intermediate images from each major analysis pipeline step, such as the original, preprocessed, masked, and analyzed images with overlaid ROIs and measurements for visual verification. This recording of each pipeline step is also critical for compliance with the FAIR standards (www.go-fair.org) that require data and analyses to be Findable, Accessible, Interoperable, and Reusable.

Configuring the initial analysis usually takes only a few minutes. Once the analysis workflow is defined, the process can be automatically applied to all images currently uploaded in the user's application instance. It is important to note that such automated analysis is dependent on a consistent image format across the uploaded user images. However, this requirement is usually the norm when replicate experiments or analyses made on datasets from the same growth and imaging hardware are being compared. After such automated processing has been performed, bulk analysis results can be downloaded as a .ZIP file containing results and intermediate data for each image analyzed.

Case Study 1: Validating SOAPP Measurements Using Analysis of Arabidopsis Grown on Agar Plates

Hughes et al. (2025) reported SOAPP could make highly reproducible measurements of hypocotyl length from images of seedlings across multiple accessions of Arabidopsis thaliana, with essentially no measurement-to-measurement variation across multiple rounds of analysis. In comparison, these researches found an average variance of 5.47% in results when equivalent measurements were performed manually using the ImageJ image analysis software (Schneider et al., 2012). To extend this validation of the accuracy of measurements made using SOAPP, we reanalyzed image data from five of the Arabidopsis ecotypes used in the Hughes et al. (2025) study: Doubravnik-7, IP-Deh-1, Ha-HBT1-2, Ler-1, and Li-7. Instead of hypocotyl length, we used SOAPP to extract the area of the foliage as a metric likely to be widely used by plant space biology researchers as a proxy for biomass accumulation. Measurements were performed using both manual analysis through ImageJ and via SOAPP to ask how closely these two measurement approaches correlated. Briefly, seeds were sown onto Petri plates containing gel growth medium consisting of 50% Murashige and Skoog nutrients (w/v), 1% sucrose (w/v), and 1.2% of agar (w/v). Plants were then grown under both 1 x g gravity (vertically oriented plates) and simulated microgravity conditions created using a rotating 2D clinostat running at 1 rpm (Kiss et al., 2019) for seven days (see Hughes et al., 2025 for detailed growth parameters). Following this growth period, plates were imaged using a high-resolution Epson scanner. A representative image of this dataset is presented in Figure 3A. The goal of including multiple ecotypes and two conditions known to alter plant growth profiles was to generate a wide range of plant sizes. Vegetative tissue area was measured both by hand using ImageJ and via SOAPP to assess consistency of vegetative tissue measurement between these two analytical approaches. As manual analyses using ImageJ require some practice and skill for accuracy, to ensure the robustness and consistency comparative data, all hand-measurements were collected by a single researcher with several years of experience in these kinds of manual plant image analysis tasks. A representative output image is presented in Figure 3B. Scale was measured manually with the ruler included in each image using ImageJ's scale function, and area measurements reported as square millimeters for both SOAPP and hand measurement data. Manual ImageJ and SOAPP measurements for 214 total plants were compared using an individual plant-wise linear regression performed with Python package SciPy (version 1.12.0; Scipy.org). Figure 3C shows results of individual plant comparisons versus an equivalent reference reflecting a theoretical complete agreement between the two approaches. The linear regression (y = 0.01839 + 1.039x; p = 3.262e-204) yielded r = 0.9938, indicating SOAPP was able to effectively match hand measured values.

Figure 3.

Case Study 2: Adding Value to Existing Datasets — Reanalysis of OSD476 - Arabidopsis Grown on Lunar Regolith

The accumulating imaging data from spaceflight-related plant biology experiments appearing in publications and being deposited in NASA's Open Science Data Archive offers a rich resource for extracting further insight into plant responses to growing in space. SOAPP provides a simple approach to rapidly mining these existing datasets. For example, daily images of Arabidopsis grown on both true lunar regolith and in the JSC1A simulant (Paul et al., 2022) were obtained from the NASA Open Science Data Repository (OSD-476). Images acquired from July 2–16, 2021 (coded as 210502-210516) were analyzed for plant shape characteristics of green tissue using SOAPP for comparison to those of the original publication (Paul et al., 2022).

SOAPP's automated phenotyping of these images (Fig. 4A–D) enabled comprehensive quantification of multiple traits, including plant area, perimeter, solidity (object area/convex hull area), convex hull area and vertices, longest path, width, and height, as well as ellipse properties like angle and eccentricity. The extracted multi-trait data facilitated exploratory data analysis and pairwise correlation studies (Fig. 4E) using R library “corrplot” version 0.95. These analyses offer the possibility to uncover new potential relationships between different phenotypic parameters, such as correlations between plant area, perimeter, and convex hull properties that may help provide new insights into growth patterns and responses to the regolith substrate. Leaf area for each plant was plotted over time, showing the increase in healthy tissue in JSC1A substrate-grown plants in comparison to the three Apollo mission true regolith samples (Fig. 4F). SOAPP measurements were averaged plate-wise and matched with their corresponding measurements of the widest spread of the rosette made by hand using ImageJ. The associations between these manual measurements of rosette diameter (variable name “leaf_spread”) and SOAPP measured values reported on a per-time, per plate basis are shown in Figure 4G.

Figure 4.

The original analysis of these plants growing in regolith and regolith simulant by Paul et al. (2022) used a manual approach similar to that described above to quantify rosette diameter. Their analysis showed that regolith leads to slower plant growth that plateaus after approximately 15 days post germination. These researchers further showed that plants grown in JSC1A lunar regolith simulant outperformed those in the true lunar regolith. The rapid foliage morphology quantification with SOAPP allowed us to both confirm and extend these previously reported rosette diameter-based conclusions. Thus, as noted above, SOAPP measures a range of rosette shape-related parameters in addition to rosette diameter such as “area” and “longest_path”. Figure 4G shows a strong correlation between these other parameters and rosette diameter (defined by the parameter “leaf_spread”) indicating that the original conclusions drawn by Paul et al. (2022) based on rosette size alone likely robustly reflected an effect also captured in these other size and morphology characteristics. Figure 4E shows that most of these shape-related characteristics strongly correlate with each other and so we therefore concentrated on using one representative parameter, green leaf area (the “area” variable in the figure), as an example of extracting additional insights from these datasets. Thus, by measuring green tissue area, SOAPP identified day 15 (210515 in Fig. 4F) as the timepoint where JSC1A-grown plants begin to exhibit morphological features of worsening health outcomes beyond slower growth, as inferred from a reduction in green leaf area within the rosette. This phenomenon is not observed in the rosette diameter analysis in Paul et al. (2022), where diameter measurement plateaus beginning at day 13 but does not decline. This reduction in the area parameter revealed by SOAPP reflects a loss of green tissue within the leaf structures of the rosette, likely reflecting the onset of senesce. Figure 4F further indicates that a similar trend in reduction in green tissue may be evident in the Apollo 12- and Apollo17-regolith grown plants. Unfortunately, the small size of the rosette and large variability in the plants in these samples means these differences are not significant. In this context, the “solidity” parameter provides SOAPP with a further measure of green leaf fill density within the rosette. The weak correlation between rosette diameter and the solidity parameter (Fig. 4G) reinforces the idea that measurements based on rosette diameter alone do not capture these kinds of changes in overall plant crown closure.

Case Study 3: Using SOAPP to Rapidly Screen for Stress Responses — Analysis of Flooding and Drought Responses in Single Leaves

One potentially important spaceflight-related application of image analysis is in assessment of physiological responses— for example, detecting visual clues as to stress-related patterns of development such as the reduction in green leaf area described above. These kinds of analyses can not only provide important insights into responses during post-experiment analyses of plant flight experiments but also hold great potential for the crew to perform real-time diagnosis of plant vigor from relatively simple digital photography. Indices such as Normalized Difference Vegetative Index (NDVI) and Enhanced Vegetation Index (EVI; Huete et al., 2002), or even following the accumulation of the purple stress pigment anthocyanin (Li & Ahammed, 2023), are all optical techniques widely used in making these kinds of measurements and are implemented in SOAPP. Accurately applying these approaches becomes challenging under the varying lighting regimes inherent in crew photography. However, SOAPP can quantitatively extract and correct color data to help make these measurements robustly. The following analysis demonstrates SOAPP's use of a simple internal color standard to make measurements that accurately reflect true differences in sample color data in plants exposed to water stress.

Lettuce (Lactuca sativa cv. Outredgeous) was grown at Kennedy Space Center in an Advanced Plant Habitat (APH; NASA, 2017)-analog growth system. The growth hardware consisted of a substrate-filled science carrier housed within a standard Percival environmental control chamber that was designed to mimic the APH growth system. Science carrier quadrants were packed with the APH-standard porous clay substrate (arcillite; Monje et al., 2020) to ensure methodological consistency with International Space Station plant-growth protocols. Plants were cultivated under an 18/6 h (light/dark) photoperiod with an average photosynthetic photon flux density of 300 μmol·m⁻²·s⁻¹. The lighting spectrum consisted of 16% blue, 24% green, and 60% red light, with an additional 36 μmol·m⁻²·s⁻¹ of far-red light (700–800 nm). All plants were watered daily from germination until 14 days after planting (DAP). After this initial period, four watering regimes were implemented: flood (specifically, waterlogging of the root system), control, intermittent drought, and chronic drought. The root waterlogging treatment was applied by filling the science carrier with water to the top of the arcillite layer and allowing the substrate to dry to only ~2% below this flood fill level before re-watering. The control treatment was allowed to dry to ~52% substrate moisture and maintained at that level. Plants subjected to intermittent drought followed the control regime for the first 14 DAP, after which irrigation was withheld until visible leaf wilting occurred (typically after 5–6 days). These lower water levels were maintained for 24 h, and then the science carrier was re-watered to control levels before the cycle was repeated. Chronic drought plants were gradually reduced to ~35% substrate moisture content and maintained by minimal daily watering.

At the end of the experiment, individual leaves from each plant were harvested and photographed on a neutral white background (Fig. 5A). These images were then analyzed for shape and color using SOAPP (Fig. 5B). Binary masks were constructed from the L, A, and B channels to account for inter-image variation in leaf color relative to the neutral background. Image scale values were measured from the AIRI Bio-Imaging Spectrum Calibration Scale (Astrobotany. com) included with each image and applied to analysis using SOAPP's “Set Image Scale” preprocessing module. To account for the non-linear color values of the images, SOAPP-generated root mean squared-derived RGB and Vegetative Index values were chosen for final analysis.

Figure 5.

Dimensional reduction of the high-dimensional colorimetric and morphological phenotypic data was performed using Principal Component Analysis. Principal Component (PC) 1 and PC2 were then plotted with ellipses capturing the 90% confidence interval for each watering regime (Fig. 5C). This analysis revealed that while the two drought regimes clustered together, there is little discrimination between the control and the flooding treatment. To gain further insight into the features driving this clustering, phenotype variable loadings for PC1 and PC2 were plotted alongside eigenvalues in a biplot (Fig. 5D). In this representation of the data, the eigenvalues reflect the absolute amount of variance explained by each PC and the loadings (vector lengths) show the degree of contribution of each variable to the PC. This analysis revealed some of the color features driving the separation of the drought phenotypes from the control and waterlogging treatments. These factors included: variance in the color saturation and the circular mean of the hue, the mean of the green-magenta signal, and the root mean square of the Normalized Difference Anthocyanin Index (NDAI) and of the blue reflectance (Rb) parameters, providing both clues as to mechanistic changes, such as potential accumulation of anthocyanins, and directing the next steps of analysis.

Therefore, leaf area and the LAB color space channel “A” that represents the green-to-magenta color axis were next plotted individually (Fig. 5E, F) and tested for significant effects on phenotype by water interaction using a One-Way ANOVA with Tukey Post-Hoc test using R. This analysis found significant differences in leaf area [F(3,35) = 30.23, p = 7.94e-10] and green-magenta value [F(3,35) = 3.227, p = 0.0341] likely attributable to changes in condition. Thus, both the control and waterlogged treatment groups produced comparable leaf area and green-magenta values (Fig. 5F). In comparison, the intermittent drought and drought groups displayed progressive loss of overall leaf area as water quantity declined (Fig. 5E). Additionally, the range of leaf area and green-magenta values was smaller in the reduced watering conditions compared to the control and waterlogged conditions, likely reflecting reduced growth coupled to the production of stress-related pigments such as anthocyanin (Fig. 5E, F). These observations suggest that, under the growth conditions used, drought, rather than root system waterlogging, may be more likely to result in changes in plant health and yield outcomes, as only these conditions led to large-scale, visible changes relative to the control. Regarding lettuce's potential in a cut-and-come-again cultivation context for spaceflight, the reduced sensitivity to root system waterlogging relative to drought has important connotations. Thus, root system waterlogging is thought to be a major potential stress associated with space-based plant cultivation due to the effect of microgravity on the dynamics of water movement (Poulet et al., 2016). However, our SOAPP analysis suggests underwatering may be a more significant issue to address in a spaceflight-oriented cut-and-come-again scenario. Consistent with these ideas, lettuce is known to be compatible with non-aerated, non-circulating hydroponics systems on Earth (Kratky, 2004) whereas even a brief water deficit quickly stunts leaf growth and reduces overall yield (Li et al., 2023).

Case Study 4: Applying SOAPP for Automated Keystone Correction and Image Standardization Using the AIRI Bio-Imaging Spectrum Calibration Sticker in Tomato Imaging

One other challenge inherent in crew photography is the difficulty in setting a uniform distance and angle from the subject to standardize subsequent morphological measurements. In particular, off-axis camera alignments lead to keystoning of images, distorting size measurements. Although cameras permanently mounted in growth facilities can alleviate this issue by having a known, fixed position, this solution limits the scope of image data collection. Cameras with internal position and attitude detectors provide one solution, however, SOAPP can automatically extract scale and keystone corrections from images that incorporate a simple reference sticker. The following demonstrates these in-built corrections in an analysis of tomato plant morphology. A tomato plant from line 1 of the Solanum lycopersicum (NCEBR-1) × Solanum pimpinellifolium L. (LA2093) recombinant inbred line (RIL) population was grown in a controlled environment chamber under a 12:12 h (light:dark) photoperiod at 28 °C day and 20 °C night temperatures, with 475 ppm CO₂ and 70% relative humidity. Following harvest and washing, the plant was imaged against a neutral white background under mixed warm-white LED and fluorescent lighting. The imaging background included ArUco markers (IDs 6 and 7, from OpenCV's ArUco “DICT_4X4_50” dictionary), placed in a square pattern with a 1:3 identity ratio to enable automated keystone correction (Fig. 6A).

Figure 6.

Keystone correction was performed using the SOAPP preprocessing module “Marker Frame Adjust,” with the “SQUARE” correction method and a 300-pixel inset. This centered the subject within a corrected image frame, preserving the corner markers (Fig. 6B). The correction resulted in a reduction in marker area variability with the relative range of mean marker size decreasing from 15.69% to 4.663%, a 70.29% reduction in this measure of image skew. Changes in individual marker area relative to the image-wide mean are shown in Figure 6C.

An AIRI Bio-Imaging Spectrum Calibration sticker was included in the image and automatically detected (Fig. 6D), providing standardized post-correction scale and color references. Image scale was computed using the SOAPP “Astrobotany Sticker - Scale” module with the “STICKER” correction method, yielding a post-correction resolution of 7637.356 pixels per meter. Color reference values were extracted using the “Astrobotany Sticker - Color Reference” module, producing standardized RGB values of 173.03 (R), 142.83 (G), and 114.19 (B). These standards were used automatically by SOAPP to generate scaled linear and area measurements in the morphological results file, and to apply color correction in the colorimetric results file. The final phenotypic analysis image is shown in Figure 6E.