Over the last century, the management of pastoral systems has undergone major changes to meet the livelihood needs of alpine communities. Faced with the changes induced by recent global warming, the ecological status of many pastoral systems has seriously deteriorated in the western alpine region. We assessed changes in pasture dynamics by integrating information from remote-sensing products and two process-based models, i.e. the grassland-specific, biogeochemical growth model PaSim and the generic crop-growth model DayCent. Meteorological observations and satellite-derived Normalised Difference Vegetation Index (NDVI) trajectories of three pasture macro-types (high, medium and low productivity classes) in two study areas - Parc National des Écrins (PNE) in France and Parco Nazionale Gran Paradiso (PNGP) in Italy - were used as a basis for the model calibration work. The performance of the models was satisfactory in reproducing pasture production dynamics (R2 = 0.52 to 0.83). Projected changes in alpine pastures due to climate-change impacts and adaptation strategies indicate that: i) the length of the growing season is expected to increase between 15 and 40 days, resulting in changes in the timing and amount of biomass production, ii) summer water stress could limit pasture productivity; iii) earlier onset of grazing could enhance pasture productivity; iv) higher livestock densities could increase the rate of biomass regrowth, but major uncertainties in modelling processes need to be considered; and v) the carbon sequestration potential of pastures could decrease under limited water availability and warming.

Image-based estimation of above-ground biomass accumulation is recognized as the predominant asset of breeding programs for accelerating gains in crop adaptation and productivity. High-throughput phenotyping (HTP) has the potential to greatly facilitate genetic improvements by dissecting morphological traits which can serve as accurate predictors of optically sensed plant biomass. Thus, various high-throughput data acquisition methods have been recently developed to quantify desirable phenotypes from images. Novel insights are essential to provide helpful guidelines to breeders for the optimal selection of phenotyping approaches aimed at estimating plant biomass. In this study, three representative HTP data acquisition methods based on two-dimensional (2D) image analysis, Multi View Stereo (MVS)-Structure from Motion (SfM) three-dimensional (3D)-reconstruction and Structured Light (SL) 3D-scanning were compared for estimating fresh (FAGB) and dry above-ground biomass (DAGB) weight of potted plants at early growth stages. Two crop species with contrasting canopy shapes and architectures, namely maize (Zea mais L.) and tomato (Solanum lycopersicum L.), were used as model plants. First, the performances of each sensing approach were tested in the accurate reproduction of the major phenotypic traits and, secondly, in the reliable fresh/dry AGB estimation from the relevant allometric equations calibrated according multi- (six sampling dates, once a week) and mono-temporal (one sampling date at harvest time) datasets. The overall results demonstrated the effectiveness of the tested methods in reproducing the salient features of canopies with increasing architectural complexity, including plants’ height (R2 = 0.98, rRMSE = 7.73 % and AIC = 475.07), shoot area (R2 = 0.91, rRMSE = 29.53 % and AIC = 1369.77) and convex hull volume (R2 = 0.88, rRMSE = 27.32 % and AIC = 818.19). In this context, the shoot area associated with the age of the plant was found to be the most indicative phenotypic determinant for an accurate estimation of FAGB and DAGB. Accordingly, the greater ability of the 2D image analysis in quantifying canopies of elongated plants characterized by thin organs ensured the best estimates of fresh/dry biomass accumulation in maize (0.98 ≤ R2 ≤ 0.99 % and 8.98 % ≤ rRMSE ≤ 16.03 %, considering multi- and mono-temporal calibrated models). Contrariwise, the MVS-SfM 3D-reconstruction of more complex canopies with compact habit was advantageous for the accurate prediction of above-ground DAGB and FAGB dynamics in tomato (R2 = 0.99 % and 6.70 % ≤ rRMSE ≤ 15.82 %, considering multi- and mono-temporal calibrated models). These findings provide references to carefully select the best suited HTP data acquisition approach for the accurate estimation of biomass accumulation across plants of different canopy complexity, thereby paving the way to break through current phenotyping bottlenecks in breeding applications for current and future food security.

High-throughput plant phenotyping requires integrated image-based tools for automated and simultaneous quantification of multiple morphological and physiological traits, which are valuable indicators of plant sensitivity to limiting environmental conditions. In this study, we proposed a novel segmentation algorithm for the automatic collection of plant structural parameters based on three-dimensional (3D)-modeling obtained through a phenotyping platform and a Structure from Motion (SfM) approach. The algorithm was initially tested on a 3D-reconstruction of four potted commercial tomato cultivars, namely “Saint Pierre” (S), “Costoluto Fiorentino” (C), “Reginella” (R), and “Gianna” (G) for the identification of the main phenotypic plant traits (heights, angles and areas). The results pointed out that the proposed algorithm was able to automatically detect and measure the plant height (R2 = 0.98, RMSE = 0.34 cm, MAPE = 3.12% and AIC = 6.03), petioles inclination (R2 = 0.96, RMSE = 1.35°, MAPE = 3.64% and AIC = 22.16), single-Leaf Area (R2 = 0.98, RMSE = 0.95 cm2, MAPE = 7.40% and AIC = 14.91) and single-leaf angle (R2 = 0.84, RMSE = 1.43°, MAPE = 2.17% and AIC = 15.83). As a study case, the algorithm was applied for monitoring plant’s dynamic responses to early water stress, measured according to Fraction of Transpirable Soil Water (FTSW), of the same tomato varieties, grown in pots, during 20 consecutive days under three treatments (full-irrigation, 50% deficit irrigation and no-irrigation). The results showed that for R and G cv., plant height was the phenotypic trait most sensitive to water stress (plant growth inhibition at 0.58 of FTSW value), while Total Leaf Area and transpiration rates started to be affected at lower FTSW (0.52 and 0.40, respectively). Conversely, S and C cv. did not exhibit any significant change in phenotypic traits under analysis, likely because these varieties exhibited a slow growth rate, allowing them to consume less water and therefore not reach a water stress threshold. The results indicated that plant height trait might be used in subsequent analyses to facilitate the rapid identification of tomato varieties resistant to water stress, thus enhancing crossbreeding programmes.

This study aims to test the performances of a low-cost and automatic phenotyping platform, consisting of a Red-Green-Blue (RGB) commercial camera scanning objects on rotating plates and the reconstruction of main plant phenotypic traits via the structure for motion approach (SfM). The precision of this platform was tested in relation to three-dimensional (3D) models generated from images of potted maize, tomato and olive tree, acquired at a different frequency (steps of 4°, 8° and 12°) and quality (4.88, 6.52 and 9.77 µm/pixel). Plant and organs heights, angles and areas were extracted from the 3D models generated for each combination of these factors. Coefficient of determination (R2), relative Root Mean Square Error (rRMSE) and Akaike Information Criterion (AIC) were used as goodness-of-fit indexes to compare the simulated to the observed data. The results indicated that while the best performances in reproducing plant traits were obtained using 90 images at 4.88 µm/pixel (R2 = 0.81, rRMSE = 9.49% and AIC = 35.78), this corresponded to an unviable processing time (from 2.46 h to 28.25 h for herbaceous plants and olive trees, respectively). Conversely, 30 images at 4.88 µm/pixel resulted in a good compromise between a reliable reconstruction of considered traits (R2 = 0.72, rRMSE = 11.92% and AIC = 42.59) and processing time (from 0.50 h to 2.05 h for herbaceous plants and olive trees, respectively). In any case, the results pointed out that this input combination may vary based on the trait under analysis, which can be more or less demanding in terms of input images and time according to the complexity of its shape (R2 = 0.83, rRSME = 10.15% and AIC = 38.78). These findings highlight the reliability of the developed low-cost platform for plant phenotyping, further indicating the best combination of factors to speed up the acquisition and elaboration process, at the same time minimizing the bias between observed and simulated data.