A novel NIR-image segmentation method for the precise estimation of above-ground biomass in rice crops

Traditional methods to measure spatio-temporal variations in biomass rely on a labor-intensive destructive sampling of the crop. In this paper, we present a high-throughput phenotyping approach for the estimation of Above-Ground Biomass Dynamics (AGBD) using an unmanned aerial system. Multispectral imagery was acquired and processed by using the proposed segmentation method called GFKuts, that optimally labels the plot canopy based on a Gaussian mixture model, a Montecarlo based K-means, and a guided image filtering. Accurate plot segmentation results enabled the extraction of several canopy features associated with biomass yield. Machine learning algorithms were trained to estimate the AGBD according to the growth stages of the crop and the physiological response of two rice genotypes under lowland and upland production systems. Results report AGBD estimation correlations with an average of r = 0.95 and R2 = 0.91 according to the experimental data. We compared our segmentation method against a traditional technique based on clustering. A comprehensive improvement of 13% in the biomass correlation was obtained thanks to the segmentation method proposed herein.


1) Field Experiments Description
• Method Validation Experiment: the images will be taken in 6 different phases of the crop: germination, maximum tillering, panicle start, flowering, 10 days after flowering and harvesting. Simultaneously, 1m linear samples will be taken per plot. • Different Densities Experiments: these experiments will be applied in a 700 m 2 area, evaluating 25 lines in 3 m 2 plots with 3 repetitions. These 25 lines are CIAT elite lines which are part of the rice improvement program and from farms of the Tolima's Fedearroz.

2) Experiments for different densities, and nitrogen and water doses
Experiments 4, 5, 6: Palmira -Valle (July -November 2018) The genetic material that was used consisted in 6 very different genotypes in terms of cumulative dry biomass and vegetative cycle (Table 5) (including Line 23 and IR64). The reference range for cumulative dry biomass (stems and leaves) and total dry biomass (stems, leaves and panicles) in previous experiments where Line 23 and IR64 were evaluated showed a variation of 15.56 to 3650 g m 2 of dry biomass. With this genotype biodiversity, a goal was to have a gradient in the cumulative dry mass and for the algorithm for dry biomass estimation to be able to detect the differences between genotypes and phenological stages. Experiments 4, 5 and 6 were established following a random complete block design with 3 repetitions and 8 genotypes for each repetition. Each experimental unit had an area of 5.77 m 2 (2.10 m x 2.75 m) and had 7 furrows (Figure 7). The sowing was performed using the transplanting method with a distance of 0.3m between furrows and 0.25m between plants, with 77 plants in each plot. Each sampling unit was composed by 4 plants by linear meter. The sowing and transplanting of the genotypes was interleaved (Table 6), starting with the sowing of the genotype with the highest cumulative degrees day and finalizing with the one with lowest cumulative value; all this was done looking to synchronize the reproductive and ripening stages and for the samples to be homogeneous in terms of the development of the genotype.  Experiments 4, 5 and 6 showed a variation in terms of fertilizer doses and the availability of irrigation: • Experiment 4: 200kg of nitrogen and no irrigation from vegetative to ripening stages ( Figure 8).
To evaluate the effect of irrigation limitation from vegetative to ripening stages for experiment 4, the leaf's temperature was evaluated utilizing a MultispeQ (also applicable for experiments 5 and 6).   The fertilization plan for experiments 4, 5 and 6 are presented in Table 3, Table 4 and Table 5.    The weed control consisted in the usage of the herbicides Butaclor (4 L ha -1 ) and Basagran (3 L ha -1 ) before transplanting, while for insect control Regent (0.3 L ha -1 ) was used. No diseases were detected.
During the experiments, pictures were taken at the start of the vegetative stage and 3 samples: one sample during blooming and two during maturity stages, with their corresponding pictures. The results of the samples and measurements are included in Appendix 4 (experiment 4), Appendix 5 (experiment 5) and Appendix 6 (experiment 6).

3) Validation experiments and evaluation of 330 varieties Experiment 7: Palmira -Valle (August -December 2018)
The seventh experiment was carried out in order to validate the algorithm, in terms of dry matter content, water content and nitrogen content in a population of 330 contrasting genotypes. This experiment has focused on the flowering cycle, cumulative dry biomass, leaf coloration and modifying the sowing density. This experiment was established under a partially replicated (p-rep) design. Each experimental unit had an area of 4 m 2 (4.0 m x 1.0 m) and had 5 furrows ( Figure 5). The sowing was performed using the transplanting method with a distance of 0.20 m between furrows and 0.20 m between plants, with 100 plants per plot ( Figure 12) For this experiment, images were taken with multispectral cameras 94 days after sowing, when the first genotypes began flowering. Afterwards, three samplings were performed including their respective images. 48 of the 330 genotypes were harvested. These will be used to train the algorithm developed by the Javeriana University. The remaining 282 genotypes will be used for validation. The results of the samples and measurements are included in Appendix 7 (experiment 7). The fertilization plan for the genotypes in Palmira -Valle is presented in Table 10 in days after transplantation (DAT)  The weed control consisted in the usage of the herbicides Butaclor (4 L ha -1 ) and Basagran (3 L ha -1 ) before transplanting, while for insect control Regent (0.3 L ha -1 ) was used. No diseases were detected.