Machine learning approach to predict leaf colour change in Fagus sylvatica L. (Spain)
Graphical Abstract
Introduction
The European beech (Fagus sylvatica L.) is one of the most important deciduous forest tree species in Europe, covering up to 12 million ha (Rezaie et al., 2018). International efforts such as GENTREE-h2020 warn about the threats that climate change exerts over many habitats of deciduous tree species in Southern Europe (beech included), which are likely to change by the end of 21st century (Schueler et al., 2014).
According to Lindner et al. (2010), climate regional changes are particularly visible in forests, which can alter tree growth rates and its reproduction (Jump et al., 2006). The generally long-life of trees impedes a rapid adaptation to relatively quick environmental changes (Lindner, 2010). In the Iberian Peninsula, the beech habitat has been mainly restricted to mountainous areas where climatic conditions are more favourable (Blanco et al., 1997), showing sensitivity to climate changes (Dulansurem et al., 2017). For instance, lower growth in mature beech trees was observed at lower elevations (Jump et al., 2006). This fact suggests that warming temperatures and drought periods greatly condition beech distribution. Geßler et al. (2007) proved that beech trees were sensitive to water-logging and flooding against habitat competitors such as oaks, which are more resilient to those circumstances. Rezaie et al. (2018) demonstrated an active response of beech trees to atmospheric CO2 increase, with a positive correlation between CO2 and intrinsic water use efficiency. Thus, the physiological performance, competitive and growth rates of beech trees are closely linked to the environmental conditions of their habitat and therefore, subjected to regional climate changes (Peuke et al., 2002).
Measuring the phenology of colour and fall leaf dynamics may be of great interest in studies of ecology, carbon cycle, and even tourist industry (Zhang and Goldberg, 2011). In addition, long-term records in leaf coloration and senescence can provide useful information concerning global climate changes (Myneni et al., 1997) since autumn phenology is thought to be influenced by factors such as temperature, precipitation or photoperiod (Estrella and Menzel, 2006; Archetti et al., 2013; Keenan and Richardson, 2015). However, the knowledge about the relationship between autumn phenological responses of deciduous trees species and climate change remains somewhat limited (Zhang et al., 2020). The physiological process by which leaves change colour during autumn is well documented by Harvard University (2020). It is merely related to pigment variations before leaves falling from the trees. During the senescence period, the leaf chlorophyll a/b concentration decreases while carotenoids remain relatively constant (resulting in yellowish colours) until the leaves fall (Zhang et al., 2020). In some tree species, leaves may turn red as a result of the active synthesis of anthocyanin pigments (Harvard, 2020).
Some studies aimed to monitor leaf dynamics using remote sensing methods. Richardson et al. (2009) used a sensor with red, green and blue channels to observe how the timing and rate of autumn senescence varied across the canopy of different species, with greater variability in autumn than spring. The temporally-normalized brownness index is another interesting approach developed by Zhang and Goldberg (2011), which detected fall foliage coloration with an overall absolute mean difference of less than 5 days. Other approaches used ground-based data to predict the duration and amount of autumn colours in mixed forests (Archetti et al., 2013; Xie et al., 2018).
Machine learning is a popular set of techniques that build computer programs which automatically improve through experience (Jordan & Mitchell, 2015). As demonstrated by Ameida et al. (2012), these machine learning algorithms can be used to identify or detect some phenological patterns in imagery, relate vegetation status with meteorological data (Czernecki et al., 2018) or analyse photosynthesis data with predictions purposes (Zhang et al., 2020b).
The aim of this study is to develop a model to predict leaf colour change in beech forests in Spain. Thus, we can evaluate the temporal variability of the leaf colour change over 17 years (2001-2017) and assess the impact of regional climate changes over the study area.
Section snippets
Study area
We studied the beech forests contained in the Spanish Forest Dataset -MFE50, 1:50000- (MITECO, 2006). They are mainly located in northern Spain and cover a total extension of 398738 ha (Fig. 1). Based on the elevation data provided by the Shuttle Radar Topographic Mission -SRTM, 90 m- (Jarvis et al., 2008), they are located at a mean elevation of 910 m (standard deviation -std- = 42 m). The highest forest is located in the Pyrenees at 1950 m (0°40`26``E, 42°48´25´´N), and the lowest in the
Peak leaf colour change
For the period 2001 to 2017, the earliest date in which beech forests obtained 75CF was on 281 DOY (∼ 8th October), the median on 317 DOY (∼ 13th November) and the latest on 351 DOY (∼ 17th December). The visual inspection of the time series data presented in Fig. 4 does not evidence any acceleration or deceleration in terms of 75CF over time. The median 75CF values do not follow any discernible trend over time, with scores ranging from 312 to 322 DOY.
We used the SRTM (90 m) elevation map to
Discussion
Leaf colour dynamics can provide very relevant information about beech forests because of their sensitivity to subtle changes in climate. However, studies based on autumn plant phenology are scarce due to the limited or unappropriate amount of ground data (Lang et al., 2019).
In our study, we used a time-series satellite-based method (Zhang and Goldberg, 2011) to infer leaf colour change in beech trees. We applied an EDA step to preprocess the data before fitting the models. Firstly, we explore
Conclusions
In this study, we developed a methodology to estimate leaf colour change in beech forests, 10 days in advance, using climate data.
The time series analysis of the actual 75CF did not show any acceleration or deceleration over time. Nevertheless, we noticed a decreasing negative trend between the mean elevation of the forests and their actual 75CF in terms of Pearson correlation coefficient.
The performance of nine ML algorithms were compared to evaluate their predictive capacity. We used an EDA
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We acknowledge the E-OBS dataset from the EU-FP6 project UERRA (https://www.uerra.eu) and the Copernicus Climate Change Service, and the data providers in the ECA&D project (https://eca.knmi.nl).
References (72)
- et al.
Modelling and mapping beech forest distribution and site productivity under different climate change scenarios in the Cantabrian Range (North-western Spain)
Forest Ecol. Manag.
(2019) - et al.
Modelling interannual and spatial variability of leaf senescence for three deciduous tree species in France
Agric. For. Meteorol.
(2009) Complex network-based time series remote sensing model in monitoring the fall foliage transition date for peak coloration
Remote Sens. Environ.
(2019)- et al.
Modelling carbon and water cycles in a beech forest: Part I: Model description and uncertainty analysis on modelled NEE
Ecol. Modell.
(2005) - et al.
Monitoring elevation variations in leaf phenology of deciduous broadleaf forests from SPOT/VEGETATION time-series
Remote Sens. Environ.
(2011) - et al.
A new process-based model for predicting autumn phenology: How is leaf senescence controlled by photoperiod and temperature coupling?
Agric. For. Meteorol.
(2019) - et al.
Long-term impacts of drought on growth and forest dynamics in a temperate beech-oak-birch forest
Agric. For. Meteorol.
(2018) - et al.
Drought matters–Declining precipitation influences growth of Fagus sylvatica L. and Quercus robur L. in north-eastern Germany
Forest Ecol. Manag.
(2011) - et al.
MODIS vegetation index compositing approach: A prototype with AVHRR data
Remote Sens. Environ.
(1999) - et al.
Predicting autumn phenology: how deciduous tree species respond to weather stressors
Agric. For. Meteorol.
(2018)
Monitoring fall foliage coloration dynamics using time-series satellite data
Remote Sens. Environ.
Responses of Autumn phenology to climate change and the correlations of plant Hormone Regulation
Sci. Rep.
Photoperiod controls vegetation phenology across Africa
Commun. biol.
Beware of R 2: simple, unambiguous assessment of the prediction accuracy of QSAR and QSPR models
J. Chem. Inf. Model.
Remote phenology: Applying machine learning to detect phenological patterns in a cerrado savanna
Predicting climate change impacts on the amount and duration of autumn colors in a New England forest
PLoS One
Displaying remotely sensed vegetation dynamics along natural gradients for ecological studies
Int. J. Remote Sens.
A random forest guided tour
Test
Los bosques ibéricos. Una interpretación geobotánica
Random forests
Machine learning
Variable selection and importance in presence of high collinearity: an application to the prediction of lean body mass from multi-frequency bioelectrical impedance
J. Appl. Statist.
Machine learning modeling of plant phenology based on coupling satellite and gridded meteorological dataset
Int. J. Biometeorol.
An Ensemble Version of the E-OBS Temperature and Precipitation Datasets
J. Geophys. Res. Atmos.
Autumn leaf phenology: discrepancies between in situ observations and satellite data at urban and rural sites
Int. J. Remote Sens.
IPOPP 2.3 User's Guide
European beech responds to climate change with growth decline at lower, and growth increase at higher elevations in the center of its distribution range (SW Germany)
Trees
Alteration of the phenology of leaf senescence and fall in winter deciduous species by climate change: effects on nutrient proficiency
Global change biology
Responses of leaf colouring in four deciduous tree species to climate and weather in Germany
Climate Res.
WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas
Int. J. Climatol.
The control of autumn senescence in European aspen
Plant Physiol.
Potential risks for European beech (Fagus sylvatica L.) in a changing climate
Trees
Cited by (2)
Interpretable machine learning algorithms to predict leaf senescence date of deciduous trees
2023, Agricultural and Forest MeteorologyMachine learning-based risk prediction model for cardiovascular disease using a hybrid dataset
2022, Data and Knowledge EngineeringCitation Excerpt :It is regarded as a component of artificial intelligence. It has a wide range of applications in the fields of electrical [8], health care [9], agriculture [10], meteorology [11], and so on. The HD risk prediction model was developed by Shah et al. [12] using 14 essential attributes.