Machine learning methods for efficient and automated in situ monitoring of peach flowering phenology

Highlights

• Image colors vary dramatically at different BBCH flowering phenological stages.

• A random forest model classifies BBCH stages by image color ranges accurately.

• Growing degree days and biological growth time improve the model accuracy.

• In situ flowering phenology monitoring is applicable for peach breeding management.

Abstract

Accurate knowledge of peach flowering phenology is essential for scheduling precise irrigation and managing artificial pollination for breeding. However, in situ monitoring of peach flowering phenology is laborious and challenging. This study divided peach flowering phenology into eight stages covering inflorescence emergence and flowering using BBCH scales. Web-connected recording cameras and weather stations automatically-collected flower images and meteorological data from three peach cultivars. Daily variables, including area proportion of specific color ranges in the images, temperature, illuminance and heat/chilling requirement data were applied for training and testing four machine learning models: random forest (RF), support vector machine (SVM), naïve Bayes (NB), and k-Nearest Neighbors (KNN). Grid tuning and 10-fold cross-validations were conducted on these models to determine the optimal method for identifying peach flowering phenological stages. The RF model obtained highest optimal F1 score of 98.82 % (harmonic average of model accuracy and recall) among the models on the testing set. The area proportion of specific color ranges in the real-time images are essential for model performance. Heat requirement data (biological growth time and growing degree days) can improve the overall accuracy. The classification accuracy at different BBCH phenological stages of the three cultivars is acceptable for real-time monitoring of peach flowering phenology. These are applicable for peach breeding, heat stress management, and irrigation scheduling. The RF model also shows acceptable accuracy with flower images and public meteorological data input, which reveals the extendibility for phenology monitoring of other trees or crops.

Introduction

Plant phenology is the seasonal calendar of plant biological events, including sprouting, shoot development, inflorescence emergence, flowering, fruit development, and senescence (Jochner and Menzel, 2015). Accurate phenology of plants is crucial for agricultural production and operations because it guides specific schedules for fertilizing, irrigation, pest control, harvesting, and breeding (Fu et al., 2019, Piao et al., 2019). Therefore, updated descriptions of crop growth stages provide universal, and standardized phenological scales for major crop species (Hill et al., 2015). The methods for collecting and compiling phenological scales are becoming standardized and objective (d’Andrimont et al., 2020).

The Biologische Bundesanstalt, Bundessortenamt und CHemische Industrie (BBCH) scale, developed by the Federal Biological Research Centre for Agriculture and Forestry (Meier, 2001), outperforms many plant phenological scales and is considered a standard plant phenological scale criterion in plant phenology studies (Stucky et al., 2018). The BBCH method describes plant phenological stages as a decimal system with a two-digit code, from zero to nine. The first digit defines the principal stage, such as leaf development and flowering. In contrast, the second digit denotes the secondary stage, including the degree of development or characteristic changes of a major stage. The BBCH scale has been extended for specific crops, each having specific coding systems. In fruit trees, pome fruit, stone fruit, currants (Meier et al., 1994), citrus (Agustí et al., 1995), and mango (Hernández Delgado et al., 2011), have specific BBCH scales.

Peach is botanically classified as stone fruit. Peach flowering involves two of the eight stone fruit phenological stages in the BBCH scale. The peach flowering stages are inflorescence emergence and flowering, BBCH principal stage codes 5 and 6 (Meier, 2001). Genotype and phenotype differences between early-blossom and late-blossom peaches are significant for the peach industry because blossom timing is linked to fruit maturity timing (Oussama et al., 2008). In south China, fruits of early-blossom peach cultivars tend to mature late. Many peach orchards cultivate two to five cultivars with different fruit maturity timings to elongate the sales season. Therefore, knowing precise peach flowering phenological stages facilitates accurate fruit harvest timing and scheduling harvest (Sánchez-Pérez et al., 2014).

Peach flowers are self-compatible with very short pollination receptive periods. Pollinators such as honey bees facilitate natural pollination for fruit production during the receptive period (Penso et al., 2020). However, when cross-breeding between cultivars with different flowering timings, artificial pollination is conducted during the receptive period (Larue et al., 2021). Male anthers are collected and dried to extract pollen, and female flowers are emasculated before petal opening to prevent natural pollination from other pollen. Additionally, the change of flowering phenology reflects potential heat stress the trees are suffering, which is useful information to make precise irrigation schedules (Vanalli et al., 2021). Appropriate applying of techniques of heat stress management requires comprehensive understanding of phenological changes and the corresponding heat tolerance features (Akter and Rafiqul Islam, 2017). Thus, the precise knowledge of flowering phenological stages is essential for scheduling artificial pollination and precise irrigation. Precise and efficient monitoring methods for peach flowering phenological stages must be developed rather than the current arbitrary manual observations (Prudencio et al., 2018).

Manual observation and determination of peach phenological stages is laborious and empirical; hence, artificial estimates of pollination and harvest timing are inevitable. Several studies have monitored and predicted flowering phenological stages in many plants, including apple, pear, and soybean. For strawberry, a combination of networked microsensors, weather models, and machine learning predicted flowering phenological stages and fruit yield (Lee et al., 2020). Deep learning methods detected apple flower bloom intensity and rice phenology from in-field images using convolutional neural networks to monitor flowering phenological stages (Dias et al., 2018, Han et al., 2021). A similar machine learning method detected and monitored pear flowering phenological stages from pear tree images (Sun et al., 2021). A simple linear iterative clustering (SLIC)-based machine learning method monitored and predicted the soybean flowering timing (So et al., 2017). All the above studies involved machine learning using crop flowering images. These models performed well in different crops and contexts because they were built with various statistical and mathematical principles using training data rather than being explicitly designed for specific tasks (Zhu et al., 2022, Agudelo-Rodríguez et al., 2020). Peach flowers have distinct colors; hence, machine learning based on flower colors is applicable for estimating and monitoring peach flowering phenological stages.

This study analyzed feature differences in peach flower images from several flowering phenological stages. The performances of various machine learning methods (random forest, support vector machine, naïve Bayes, and k-Nearest Neighbors) for estimating and monitoring peach flowering phenological stages were tested using peach flower images and meteorological data. Then, the most optimal and accurate method for automated monitoring of peach flowering phenology is proposed, and the effects of input variables on monitoring accuracy are discussed. Finally, the perspectives and remarks of optimal methods are presented.