Document [original]

Remote Sensing of Environment 307 (2024) 114162

Available online 17 April 2024

Spectral-temporal traits in Sentinel-1 C-band SAR and Sentinel-2

multispectral remote sensing time series for 61 tree species in

Central Europe

Christian Schulz

, Michael F¨

orster

, Stenka Valentinova Vulova

, Alby Duarte Rocha

Birgit Kleinschmit

Geoinformation in Environmental Planning Lab, Technische Universit¨

at Berlin, Straße des 17. Juni 145, 10623 Berlin, Germany

Department of Environmental Meteorology, Institute for Landscape Architecture and Landscape Planning, University of Kassel, 34127 Kassel, Germany

ARTICLE INFO

Edited by: Marie Weiss

Keywords:

Temperate forests

Principal component analysis

Spectral fingerprints

Tree species mapping

Machine learning

Germany

ABSTRACT

Tree species maps derived from satellite imagery increasingly support forest administrations and nature con-

servation authorities with large-scale and up-to-date information. However, many species are often excluded or

aggregated in classification tasks due to a limited knowledge of the most suitable predictors. Our study aims to

gain a better understanding of optical and polarimetric traits for tree species mapping by examining Sentinel-1

and Sentinel-2 time series from 61 tree species in temperate Europe. For a selection of 32 optical, polarimetric

and structural variables, the principal component analysis revealed that Sentinel-2 variables mainly explain the

variance in the data by contributing to the “seasonality” and “foliage color” components. Sentinel-1 contribute

most to the “texture” component. The Normalized Difference Vegetation Index (NDVI), Tasseled Cap Greenness

(TCG) and Radar Vegetation Index (RVI) were chosen as key variables for further analysis. Seasonality was found

to be the most dominant aspect in all vegetation indices. Furthermore, the TCG was found to be useful to

distinguish between early and late budding species. The RVI showed a large potential to discriminate conifers,

which is attributed to the crown volume effect of C-band SAR. Using exploratory data analysis, we further

examined the influence of management, biogeographical and meteorological factors on the time series from

Fagus sylvatica, Pinus sylvestris, and Picea abies. The NDVI and TCG are relatively robust to different conditions.

For the two conifer species however, we found strong spatial variations of the RVI which are presumably caused

by different crown conditions across the study area. Using Sentinel-1 data could therefore lead to uncertainties in

tree species mapping across large biogeographical gradients. This study contributes to the improvement of tree

species mapping based on optical and dual-polarimetric data and thus benefits forest authorities and other

stakeholders in their monitoring tasks and decision-making.

1. Introduction

Worldwide, forests provide numerous ecosystem services to society

(Millennium Ecosystem Assessment, 2005). However, forest ecosystems

are increasingly affected by the direct and indirect effects of climate

change leading to large-scale disturbances and shifts in species compo-

sition (MacDicken et al., 2016; Seidl et al., 2017; McDowell et al., 2020).

For large-scale mapping and monitoring of forest changes, multi-

temporal remote sensing from space is a key resource (Lausch et al.,

2016) and thus supports decision makers in the forest management and

conservation sectors (Fassnacht et al., 2016, 2023). In particular, forest

type and tree species mapping is increasingly needed for the update of

national forest inventories (Simons et al., 2021). Knowledge about shifts

in species composition is essential for biodiversity monitoring (Kollert

et al., 2021) and sustainable forest management (Fassnacht et al., 2016;

MacDicken et al., 2016).

Remote sensing covers a broad range of space-borne optical and

radar sensors capturing multiple biochemical and biophysical charac-

teristics of forests (Fassnacht et al., 2016; Lausch et al., 2016; Holzwarth

et al., 2020; Pu, 2021). Multispectral data is used on the global to local

scale (Buras et al., 2021; Hansen et al., 2013; Schuldt et al., 2020; Senf

et al., 2020). Radar data, e.g. from synthetic aperture radar (SAR)

* Corresponding author.

E-mail address: [email protected] (C. Schulz).

Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

https://doi.org/10.1016/j.rse.2024.114162

Received 13 September 2023; Received in revised form 11 March 2024; Accepted 12 April 2024

Remote Sensing of Environment 307 (2024) 114162

sensors, is primarily used on the regional to local scale, but have also

become relevant for the continental scale (Baron and Erasmi, 2017;

Rüetschi et al., 2017; Vreugdenhil et al., 2020; Dost´

alov´

a et al., 2021;

dos Santos et al., 2021; Soudani et al., 2021).

Since the launch of the Copernicus program in 2014, multispectral

and SAR remote sensing have become indispensable for European forest

research (Holzwarth et al., 2020). In particular, the freely available data

from Sentinel-2 (S2) and Sentinel-1 (S1) are increasingly supporting

forest authorities and monitoring agencies to provide up-to-date infor-

mation on tree species distribution (Holzwarth et al., 2020; Pu, 2021).

Early tree species classification studies focused on selecting optimal

input features from S2 data with up to twelve classes (e.g. Immitzer

et al., 2016, 2019). More recent studies with a similar scope use longer

time periods, larger study areas and data from both Copernicus sensors

for tree species detection (e.g. Lechner et al., 2022; Welle et al., 2022;

Preidl and Faude, 2023; Blickensd¨

orfer et al., 2024). SAR data has rarely

been assessed for their capacity to differentiate tree species (Fassnacht

et al., 2016). However, within the last years, S1 is more frequently used

for classification tasks as an additional data source (e.g., Bjerreskov

et al., 2021; Dobrini´

c et al., 2021; Meroni et al., 2021; Lechner et al.,

2022; Blickensd¨

orfer et al., 2024).

In optical remote sensing, the spectra in the visible and infrared

wavelengths are especially well-suited for quantifying changes in foliage

pigments such as chlorophyll (Lausch et al., 2016; Misra et al., 2020).

Multispectral indices, such as the Normalized Difference Vegetation

Index (NDVI) (Rouse et al., 1974) and Tasseled Cap Greenness (TCG)

(Crist and Cicone, 1984), are well-established indicators for vegetation

monitoring. Furthermore, by using time series, temporal features

describing the phenology of forests can be derived (J¨

onsson and

Eklundh, 2002; Mayr et al., 2019; Kowalski et al., 2020; Misra et al.,

2020; Kollert et al., 2021; Soudani et al., 2021). However, one of the

main limitations of optical data are cloud-induced data gaps. Across

Central Europe the average gap length in S2 data lies between 18 and 46

days within a year (Sudmanns et al., 2020).

In radar remote sensing, the volume scattering effect in SAR data is

highly relevant for biomass estimation and deforestation mapping of

tropical rainforests (Flores-Anderson et al., 2019). For temperate forests,

relevant applications of SAR data include storm damage and disease

monitoring (Baron and Erasmi, 2017; Holzwarth et al., 2020; Vreug-

denhil et al., 2020; K¨

onig et al., 2023). Furthermore, polarimetric

vegetation indices, such as the Cross Ratio (Paloscia et al., 1999) or the

Radar Vegetation Index (RVI) (Kim et al., 2012), have been investigated

for crop growth assessment (Mandal et al., 2020; Bhogapurapu et al.,

2021) and land cover characterization (Dubois et al., 2020). Similar to

the optical vegetation indices, polarimetric indices can be used to derive

phenological features (Rüetschi et al., 2017; Frison et al., 2018;

Vreugdenhil et al., 2018; Dubois et al., 2020; Bjerreskov et al., 2021;

Soudani et al., 2021). One major advantage of SAR sensors is the pro-

vision of data regardless of cloud cover (Bjerreskov et al., 2021; dos

Santos et al., 2021). However, SAR remote sensing is also subject to

limitations. For instance, backscattering is sensitive to soil moisture,

which leads to anomalies in the signal after heavy rain events (Holtgrave

et al., 2018; Vreugdenhil et al., 2020).

The latest research shows great potential of optical and radar satel-

lite imagery time series for accurately mapping tree species at different

spatial scales. However, there are still limitations to solve ‘real world’

problems (Fassnacht et al., 2023). To successfully support climate

change adaptation and biodiversity protection strategies, the selected

target classes must also include less dominant and ecologically relevant

tree species. This requirement is often hampered by missing or limited

label information about the species in the reference data, which often

leads to a reduced number of target classes to the most dominant species

or species groups with similar taxa (Lines et al., 2022). Hence, a better

understanding of the relationship between the spectral traits and the

remotely sensed signal is necessary to assess whether and how tree

species can be classified under given conditions (Fassnacht et al., 2016).

Prediction modelling often underperforms for the task of tree species

classification due to a limited knowledge about highly informative

features that can be derived from remote sensing data. Furthermore, the

influence of different forest conditions (e.g. stand age, stock density,

climate, altitude) on the species-specific traits is still largely undeter-

mined. To support the feature extraction process for tree species map-

ping, we conducted an exploratory data analysis for a large number of

tree species using time series data from optical and SAR remote sensing

data in Lower Saxony, Germany. We first examined spectral-temporal

traits of SAR S1 and optical S2 time series of 61 tree species. We then

investigated the influence of management, biogeographic and meteo-

rological conditions on the spectral-temporal characteristics of three

important tree species (Fagus sylvatica, Picea abies, Pinus sylvestris).

We explicitly answer the research questions: a) Which sensor types,

bands and indices explain the largest variance in the time series data for a

variety of tree species? b) Does the spectral-temporal information allow for

differentiation of forest classes beyond the genus-level? c) Which stand fac-

tors influence the remote sensing signals the most? Based on our analysis, we

derive recommendations for upcoming tree species classification using

optical and SAR time series data.

2. Data

2.1. Study area and time frame

Our study area covers the state-owned forests of the federal country

of Lower Saxony, Germany (Fig. 1). The North German region comprises

the Northwestern and Northeastern German Flatlands, which are charac-

terized by an Atlantic climate (i.e., wet and winter-mild) and the Central

German Uplands with a continental climate (i.e., dry and winter-cold) in

the East and South-East (Ssymank, 1994; Beck et al., 2018). Daily air

temperatures reach a mean annual value of 10 to 11.5 ◦C with extremes

of −10 ◦C in winter and 35 ◦C in summer (Deutscher Wetterdienst,

2022). The annual precipitation sum ranges between 300 and 900 mm

(Deutscher Wetterdienst, 2022).

Approximately 21.6% of Lower Saxony consists of forests of which

most are grown for commercial purposes (Nordwestdeutsche Forstliche

Versuchsanstalt and Nieders¨

achsisches Ministerium für Ern¨

ahrung,

Landwirtschaft und Verbraucherschutz, 2021). Up to 89 tree species are

documented in our study area (Nieders¨

achsische Landesforsten, 2020a).

The three species Scots pine (Pinus sylvestris), European beech (Fagus

sylvatica), and Norway spruce (Picea abies) comprise 67% percent of the

forested area (Nordwestdeutsche Forstliche Versuchsanstalt and Nie-

ders¨

achsisches Ministerium für Ern¨

ahrung, Landwirtschaft und Ver-

braucherschutz, 2021). The majority of forests are distributed at

altitudes between 0 and 400 m (Nieders¨

achsische Landesforsten,

2020b). In general, there are larger proportions of forests in the uplands

and lower proportions in the flatlands. However, the spatial distribution

of the individual species varies (Nieders¨

achsische Landesforsten, 2020a)

(Fig. A.1, Supplement).

Our study time frame covers the years 2018 to 2021 with three

consecutive dry years and a relatively mild year (Boeing et al., 2022).

We used multiple years for our study because it allows us to generalize

temporal patterns for tree species in S1 and S2 satellite imagery across

varying meteorological conditions.

2.2. Reference data

The reference data of this study originates from the state forest

management and inventory data from the Federal Forest Service of

Lower Saxony (Nieders¨

achsische Landesforsten, 2020a, 2020b)

(Fig. 1B). The rolling archive is updated on a 10-year cycle through

annually rotating field and aerial orthophoto flight campaigns

(B¨

ockmann, 2016). It includes the management parcels (2011−2020)

from 166,421 forest stands and inventory plots (2015–2019) from

18,289 forest stands. The sizes of the parcels range between 83.8 and

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

0.05 ha with a median of approximately 0.8 ha. The size of the inventory

plots is given by the radius of 13 m (B¨

ockmann, 2016). Table 1 sum-

marizes the attributes of the two datasets used in this study, which were

derived from management plans, ground truth measurements and geo-

spatial modelling (K¨

ohler et al., 2016). The most important reference

labels are given by the attributes stand type and main tree species

(Tables A.3–4, Supplement). Further information on the reference data

can be found in Ahlswede et al. (2023).

We created two subsets from the reference data (Table A.1 and

Fig. A.1, Supplement): a) The WEFL61 dataset originates from the

forestry management data and includes 10,914 forest stands from 35

broadleaf and 26 needleleaf tree species. Parcels with no trees (i.e., long-

term clearings) were excluded from our analysis. To limit processing

time, the maximum number of parcels per class was set to 500 randomly

selected stands. The tree species labels were derived from the attribute

main tree species which depicts the species with the highest proportion at

the location. Furthermore, based on the attribute stand type, we selected

pure stands where possible. “Pure” is considered to contain a proportion

of at least 90% of the main tree species. If not available, mixed stands

were chosen. b) The BI3 dataset originates from the forest inventory data

and contains 6698 forest stands from the three species European beech

(Fagus sylvatica), Norway spruce (Picea abies) and Scots pine (Pinus syl-

vestris). It covers information on nine stand factors for each forest stand.

To avoid mixed pixel signals, only pure forest stands were used for

analysis.

2.3. Satellite imagery

For the S2 time series extraction, we used Level-2 A pre-processed

products provided by the European Space Agency. The imagery con-

tains 13 spectral bands in the visible, near-infrared and shortwave-

infrared ranges with a spatial resolution of 10 to 60 m depending on

the band (Drusch et al., 2012). The products were acquired as surface

reflectance products from the cloud computing platform Google Earth

Engine (GEE) (Gorelick et al., 2017) (image collection: COPERNICUS/

S2_SR). We masked out the clouds from all available S2 scenes using the

provided quality band. For the study time frame 2018 to 2021, the

extracted S2 time series reach 43 to 257 observations with a median

value of 182 (Fig. A.3, Supplement). Cloud-induced data gaps are more

likely to occur in the winter season and in coastal areas.

Fig. 1. Study area and reference data: A) Location of Lower Saxony, the biogeographical regions in Germany (Ssymank, 1994) and location of the detail map (red

marker), B) examples of the reference data (Nieders¨

achsische Landesforsten, 2020a, 2020b) in forest stands close to the city of Hildesheim. Background layer is a

true-color Sentinel-2 composite from April 2020 from Geodatenportal Niedersachsen (GDI-NI), Landesamt für Geoinformation und Landesvermessung Niedersachsen

(https://www.geodaten.niedersachsen.de/, last access: 31 December 2020). (For interpretation of the references to color in this figure legend, the reader is referred

to the web version of this article.)

Table 1

Available information in the reference datasets forest management parcels (WEFL61 dataset, 2011–2020) and forest inventory plots (BI3 dataset, 2015–2019). Detailed

descriptions of the attributes can be found in the Supplement (Fig. A.2, Tables A.3–7).

Name Variable

type

Data type Unit Number of

categories

Value

range

Mean

value

Forest management

parcels

Forest inventory

plots

Stand type categorical nominal – 62 – – x x

Main tree species categorical nominal – 89 – – x –

Stand age class categorical ordinal – 10 – – x x

Number of stock trees numerical discrete n/ha – 1–51 15 – x

Altitude numerical continuous m 0–300 112 – x

Terrain slope numerical continuous % – 1–44 3 – x

Ground vegetation

coverage

numerical continuous % – 0–99 55 – x

Water supply level categorical ordinal – 43 – – – x

Nutrient supply level categorical ordinal – 7 – – – x

Longitude numerical continuous decimal

degrees

– – – x x

Latitude numerical continuous decimal

degrees

– – – x x

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

For the S1 time series extraction, we used Ground Range Detected

(GRD) backscatter products provided by the European Space Agency.

The two C-Band SAR sensor platforms Sentinel-1 A and 1B have a revisit

time of <6 days in Central Europe and a spatial resolution of 10 m

(Torres et al., 2012). The products were acquired from GEE (image

collection: COPERNICUS/S1_GRD). The image collection provides

scenes from two flight directions (ascending, descending). To avoid

confusion in the backscatter signal due to changing water content of the

plants at different times of the day, we only used scenes from the

descending flight mode that were recorded in the early morning in our

study area. With GEE, the data was post-processed to Analysis Ready

Data (ARD) (Mullissa et al., 2021), which included an additional border

noise correction, a speckle filter, and a radiometric terrain normaliza-

tion. The extracted S1 time series present 462 observations evenly

distributed over the course of the study time frame (Fig. A.3,

Supplement).

For our analysis, we chose a set of 32 optical and polarimetric var-

iables that have been documented in the literature to depict various

spectral, temporal and structural traits of forests including color,

texture, volume, chlorophyll content, seasonality and phenology

(Lausch et al., 2016; Fassnacht et al., 2016; Holzwarth et al., 2020). It

includes twelve optical bands, seven spectral indices, two SAR polari-

zations, five polarimetric indices, and six texture metrics (Table 2).

Fig. 2 displays three exemplary variables that can be derived from S2

and S1. The NDVI is an index of the red and near infra-red spectra which

is mainly influenced by the chlorophyll content of vegetated surfaces

(Misra et al., 2020). The TCG is a weighted index of the visible and infra-

red spectra which is driven by foliage pigment and chlorophyll content

(Healey et al., 2005). The RVI is a ratio of the polarimetric signal and is

strongly influenced by the crown volume of trees (Meyer, 2019).

Using the Earth Engine Code Editor, we retrieved time series for the

selected samples of the WEFL61 and BI3 subsets (Fig. 2). Zonal statistics

were calculated either by using the actual parcel border in the WEFL61

data or a buffer of 30 m radius around each plot center in the BI3 data. In

the pre-processed data several quality issues across the different sensors,

variables and years were observed, including unevenly distributed ob-

servations, different scales and units of the variables and outliers.

Accordingly, the time series vectors were further post-processed using

the R statistical computing language (R Core Team, 2021) to harmonize

the time series: 1) Outlier removal: Through individual threshold setting

for each variable, we removed observations with extreme values.

Despite being rare, the reasons for outliers in S2 could be undetected

clouds. Unreasonable values in S1 are often attributed to preprocessing

errors. 2) Creation of gap-free and equidistant products: For a better

comparability of temporal patterns across different years, the temporal

resolution was set to daily by linearly interpolating missing values. 3)

Normalization: For better statistical comparison, we normalized the

scale of each variable to the range of 0 to 1.

2.4. Meteorological data

Meteorological measurements were used to further interpret spatial

and temporal patterns found in the satellite imagery. The German

Weather Service provides freely available time series data from 415

stations in the study area including daily precipitation and air temper-

ature (Climate Data Center, https://opendata.dwd.de). For each forest

stand in our reference data subsets, we downloaded the data from the

three closest weather stations and calculated a mean time series vector.

The data was derived using the R package rdwd (Boessenkool, 2023).

3. Methods

Since we aim to proactively understand and interpret the nature of

SAR and optical data on tree species in temperate Europe, we deliber-

ately avoided machine learning techniques. Rather, the chosen methods

aim at unveiling unknown structures and patterns in the labeled S1 and

S2 data to assist in extracting the most relevant features for tree species

classification and to partially solve the well-known black box problem in

machine learning (Rudin, 2019). Analyzing the highly complex spatio-

temporal data with 32 variables required the use of multivariate sta-

tistics and advanced visualization techniques. For our study, we selected

quantitative methods from explorative and descriptive statistics as well

as methods that can reduce dimensionality in large datasets.

3.1. Exploratory data analysis

The exploratory data analysis (EDA) (Tukey, 1977), covering tools

from descriptive statistics and graphical representations, aims at

developing hypotheses on the usability of datasets. In our study, EDA

was used to provide a deeper understanding of the temporal and spectral

content of each data source for the task of tree species classification. We

identified and visualized generalizable spectral-temporal patterns in the

S1 and S2 time series of each species in the WEFL61 dataset to discover

potentials and limitations of the data in discriminating tree species. To

compare the spectral and polarimetric traits, we calculated the mean

annual time series vector of each variable for each species (i.e., the

temporal profile) within the WEFL61 dataset (F¨

orster et al., 2012; Verger

et al., 2016; Veloso et al., 2017). To assess uncertainty, we further

plotted the temporal variation of the time series per class. We also used

EDA to reveal the main influences for variations within the individual S1

and S2 time series of the BI3 dataset for the broadleaf species Fagus

sylvatica and the needleleaf species Pinus sylvestris and Picea abies. Those

variations might be related to diverging forest stand properties and site

conditions (Grabska and Socha, 2021), which we refer to as stand factors.

Table 2

Sentinel-2 and Sentinel-1 variables used in this study. Further information on the

wavelengths and the formulas can be found in the Supplement (Table A.2).

Abbreviation: GLCM =Gray Level Co-occurrence Matrix.

Variable Full name Source

AEROS Aerosol

Drusch et al. (2012)

BLUE Blue

GREEN Green

RED Red

RE1 Red edge 1

RE2 Red edge 2

RE3 Red edge 3

NIR Near-infrared

RE4 Red edge 4

WVAP Water vapor

SWIR1 Short wave-infrared 1

SWIR2 Short wave-infrared 2

NDVI Normalized Difference Vegetation

Index Rouse et al. (1974)

EVI Enhanced Vegetation Index Huete et al. (2002)

MSAVI Modified Soil Adjusted Vegetation

Index Qi et al. (1994)

NDWI Normalized Difference Water Index Gao (1996)

TCB Tasseled Cap Brightness

Crist and Cicone (1984) TCG Tasseled Cap Greenness

TCW Tasseled Cap Wetness

VH Horizontally polarized backscatter Torres et al., 2012

VV Vertically polarized backscatter

RVI Radar Vegetation Index for Sentinel-1 Mandal (2020)

VHVVR VH-VV Cross-Polarization Ratio Vreugdenhil et al.

(2018)

VVVHR VV-VH Cross-Polarization Ratio Soudani et al. (2021)

NDIVH Normalized Difference VH-VV Ratio Schulz et al., 2021b

NDIVV Normalized Difference VV-VH Ratio

CONTRAST_S2 GLCM Contrast from GREEN

Haralick et al. (1973)

ENTROPY_S2 GLCM Entropy from GREEN

STDEV_S2 GLCM Standard Deviation from

GREEN

CONTRAST_S1 GLCM Contrast from VH

ENTROPY_S1 GLCM Entropy from VH

STDEV_S1 GLCM Standard Deviation from VH

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

Fig. 2. Scheme for the acquisition of daily, gap-free time series from satellite and weather stations data using the example of European beech (Fagus sylvatica). The x-

axis encodes the day of year. The y-axis shows the vegetation index time series per stand. Abbreviations: S1 =Sentinel-1, S2 =Sentinel-2, NDVI =Normalized

Difference Vegetation Index, TCG =Tasseled Cap Greenness, RVI =Radar Vegetation Index, ATEMP =Mean daily air temperature, PRECIP =Daily precipita-

tion sum.

Fig. 3. Coefficient of determination (R

) between 32 optical, polarimetric and textural variables from Sentinel-2 and Sentinel-1 time series data (2018–2021) from

the WEFL61 dataset. The red squares highlight groups of variables with high redundancy which are further described in Table 3. (For interpretation of the references

to color in this figure legend, the reader is referred to the web version of this article.)

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

3.2. Correlation analysis

Correlation analysis is a tool from descriptive statistics which

quantifies linear relationships between variables and helps in the

reduction of multicollinear features in highly dimensional datasets. For

the 32 variables in the WEFL61 dataset, we calculated Pearson’s corre-

lation coefficient (r) and R-squared (R

) to determine highly correlated

groups. We also assessed the correlations between the temporal profiles

of the 61 tree species to find species with similar characteristics.

3.3. Principal component analysis

Principal component analysis (PCA), a method from descriptive

statistics, is used to explain the variance within large multidimensional

datasets and to reduce the set of variables to a few components. First,

PCA was used to reduce the multicollinearity and the number of vari-

ables in the WEFL61 dataset for further analysis. Second, PCA was used

to transform the individual variables into independent dimensions (i.e.

principal components (PCs)) which reveal highly explainable aspects in

the data for tree species classification. In order to perform a statistically

robust analysis, the input variables were centered and scaled before

applying PCA. The results were displayed through scree plots, bi plots

and further visualizations.

4. Results

4.1. Spectral dimensionality reduction

4.1.1. Correlation analysis: groups of multicollinearity and representative

variables

Figure 3 shows that highly redundant information can be found in

variables combining similar wavelengths and originating from the same

sensors. To identify the most meaningful variables in the WEFL61

dataset, we formed groups of variables with high cross-correlation ac-

cording to their coefficient of determination (R

>0.5). Within each

group, we selected a representative variable based on the highest

average value (Table 3).

4.1.2. Principal component analysis: important dimensions and variables

With 63.1%, the major proportion of information about the variance

for the WEFL61 dataset is covered by the first three PCs (Fig. 4A). The

loading plots in Fig. 4C show a high contribution of S2 variables within

PC1 and PC2. S1 variables contribute predominantly in PC3 and

partially to PC1. The differences in the variable importance may be

explained by the sensor techniques (optical vs. SAR) and their respective

wavelength ranges (443–1610 nm vs. 3.75–7.5 cm).

Most important in PC1 are the variables from the red-edge and near-

infrared spectra (RE3, RE4, NIR, RE2), tasseled cap transformations

(TCG, TCB) and SAR vegetation indices (RVI, VHVVR, VVVHR). As these

variables mainly explain seasonal changes in the chlorophyll content or

crown volume, we consider PC1 to be the “seasonality” component. In

PC2, the most important variables are close to the visible light (RED,

RE1, GREEN, BLUE, AEROS) and short-wave infrared spectrum (SWIR2,

SWIR1) or are optical vegetation indices (MSAVI, NDVI, TCB). Since

these variables mostly explain the foliage color and brightness, we

consider the second PC to be the “foliage color” component. In PC3, the

most important variables are the GLCM texture metrics (CONTRAST_S1,

ENTROPY_S1, STDEV_S1, CONTRAST_S2) and polarimetric metrics (VH,

VV, VVVHR, NDIVV, NDIVH, RVI), which all could be used to explain

the surface structure. Thus, we consider the third PC to be the “texture”

component.

Further investigations of PCAs on a monthly basis show that the

individual importances of the variables differ across the year (Fig. 4D).

For the months April to December, the S2 variables explain most of the

variance in the first and second PCs (Fig. B.1-B.12, Supplement). The

Tasseled Cap Brightness (TCB), red edge (RE1, RE2, RE3), short wave

infrared (SWIR1, SWIR2) and visible light variables (RED, GREEN) have

the highest contributions in the first PCs. In the second PCs, the vege-

tation indices (NDVI, MSAVI, TCG) are most important. S1 variables

(VVVHR, VH, RVI) become relevant from January to March where they

contribute highly to the variance in the second PCs.

4.2. Spectral-temporal profiling

4.2.1. Exploratory data analysis: inter-comparison of the temporal profiles

The three key variables NDVI, TCG and RVI were selected for further

analysis, because they have been found to be representative for three

large groups of variables in the correlation analysis and are important

variables in the first three components of the PCA. Fig. 5 shows the

typical results for all study years: i) The NDVI shows stronger seasonal

patterns for the deciduous classes than for most evergreen species.

However, some conifers (e.g. the group of Picea and Pinus nigra) also

have a pronounced seasonality. The highest variations of the NDVI

across all classes occur in December and January. ii) The TCG shows

seasonal patterns for all classes with strong summer peaks for deciduous

and medium to weak summer peaks for evergreen classes. Furthermore,

some profiles from the group of deciduous species show a shorter

summer peak (e.g. Betula pendula and Carpinus betulus) whereas others

show a longer one (e.g. Tilia cordata and Ulmus laevis). The highest

variation of the TCG across the samples across most classes occur be-

tween May and October. iii) The RVI shows seasonal patterns that are

similar but inverted to the NDVI. Large differences between the conifers

can be observed. For example, the pines (Pinus) reach relatively high RVI

values across the year compared to the spruces (Picea). In the temporal

variation profiles, low variations were observable.

The heatmaps in Fig. 6 indicate species with similar spectral and

temporal characteristics. For example, among the needleleaf species, the

group of pines (Pinus banksiana, P. contorta, P. mugo, P. nigra, P. rigida, P.

strobus, P. sylvestris) shows noticeably high RVI and low TCG values. A

similar group can be found among the broadleaf species, but comprising

species from different genera (Alnus glutinosa, Betula pendula,

B. pubescens, Populus tremula, Prunus serotina, Salix spp.). The group of

spruces (Picea abies, P. pungens, P. omorika) stands out among the

evergreen species as it exhibits a strong seasonality in NDVI. The group

of oaks (Quercus petraea, Q. robur, Q. rubra) and other late budding

broadleaf species with comparable crown habitus (e.g. Fagus sylvatica,

Fraxinus excelsior, Tilia cordata) show almost identical spectral and

temporal traits for all key variables. Another group with common

characteristics consists of conifers with darker, bluish foliage color

(Abies nordmanniana, Chamaecyparis nootkatensis, Pinus nigra, Picea

pungens, Taxus baccata), which show higher TCG values in summer

compared to the other conifer species.

4.2.2. Correlation analysis: similarity between spectral-temporal profiles

Figure 7 shows the correlation between the temporal profiles of the

species. Within the group of deciduous classes, high cross correlation

Table 3

Groups of highly redundant variables (R

>0.5) within the WEFL61 dataset.

Group Description

1 optical vegetation index (TCG*) and bands close to the near infrared and

red-edge spectrum

2 optical vegetation indices (NDVI*, MSAVI) and bands close to the visible

light spectrum

3 polarimetric vegetation indices (RVI*, VHVVR, VVVHR, NDIVH, NDIVV)

4 texture metrics from backscatter (CONTRAST_S1*, ENTROPY_S1,

STDEV_S1)

5 optical brightness index (TCB*) and bands close to the short-wave infrared

spectrum

6 optical wetness indices (TCW*, NDWI)

7 polarimetric backscatter coefficients (VH*, VV)

representative variable for the group.

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

values for all three key variables are observed, with median r values

ranging from 0.93 (RVI) to 0.98 (TCG). Within the group of evergreen

species, a higher range with median r values ranging from 0.62 (RVI) to

0.92 (TCG) was found. Across the key variables, the RVI shows the

largest range of correlation between the temporal profiles of tree species

with r reaching a minimum of 0.23 and a maximum of 0.99. The TCG has

the lowest range, with a minimum of 0.77 and a maximum of 0.99. Thus,

RVI is expected to show the largest absolute variation between the in-

dividual species.

4.2.3. Principal component analysis: spectral-temporal traits of tree species

The distribution of the tree species samples within the first two

components of the PCA (see Ch. 4.1) were further investigated (Fig. 8).

In general, the distribution of the samples within the score plots differ

between deciduous (e.g. Fagus sylvatica, Larix decidua) and evergreen (e.

g. Pinus sylvestris) species (Fig. 8A). These differences are related to the

“seasonality” component (PC1), which has a stronger influence on the

deciduous classes, while the “foliage color” component (PC2) influences

the sample distributions of all classes (Fig. 8B). The mean distributions

of the samples for each class reveal two basic clusters that depict the

plant functional type of the species (Fig. 8C). Looking at the month-wise

PCAs, these two groups are reflected for all seasons (Fig. 8D). However,

samples from the needleleaf-deciduous classes (Larix decidua,

L. kaempferi) tend to align with the “deciduous” group in winter and to

the “evergreen” group in summer (Fig. B.15, Supplement).

4.3. Stand factor exploration

None of the management stand factors (stand age class, number of

stock trees and ground vegetation coverage) had a clear influence on the

NDVI, TCG and RVI time series of the three selected species in our study

area. Also, for the group of biogeographic stand factors (altitude, terrain

slope, latitude, longitude, water supply level and nutrient supply level), no

influences were observed on the NDVI and TCG time series. However, in

the RVI heatmaps of all study years, strong variances related to the

factors latitude, longitude and altitude are observable (Fig. 9). The maps of

mean RVI show low values in the northwest and high values in the

southeast of the study area for the two needleleaf species, whereas the

results for the broadleaf class showed no spatial variance (Fig. 10B). For

the meteorological stand factors (precipitation, air temperature), no in-

fluence on the NDVI and TCG time series was found. However, for the

RVI, similar spatial patterns to those of the meteorological data were

Fig. 4. Principal component analysis of 32 optical, polarimetric and textural variables from Sentinel-2 and Sentinel-1 time series data (2018–2021) for the WEFL61

dataset: A) Cumulative sum of the variance explained by the first ten principal components, B) Loading plot of the variables within the first (PC1) and second

principal component (PC2), C) Contributions of the variables to the first three principal components. The red dashed line shows the mean value of the contribution of

all variables. D) Standardized variance of the variables within the first and second dimensions for the subsets of four individual months. All the Sentinel-2 and

Sentinel-1 variables were included into the principal component analysis as loading variables. Sentinel-2 and Sentinel-1 variables are highlighted in blue and in

brown, respectively. The month-wise principal component analyses were run on data subsets from all study years (2018–2021). Further results can be found in the

Supplement (Fig. B.1-B.12). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

observed for the two needleleaf species (Fig. 10). We also noted that for

all study years sudden declines in the RVI coincided with temperatures

below 0 ◦C and dry periods. The most noticeable drop was found during

March 2018, where the winter high “Hartmut” prevailed (Fig. 10A). For

the full results of the stand factor exploration, we refer to the Supple-

ment (Figs. D.1-D.16).

5. Discussion

A large range of spectral and temporal features for 61 tree species in

Central Europe can be derived from S1 and S2 data. We discuss key

features and challenges for tree species classification under different

background conditions such as biogeography that may affect the

transferability of supervised machine learning algorithms.

5.1. Key features for tree species classification

5.1.1. Key components and variables

According to the PCA, optical variables explain the largest propor-

tion of the variance in the “seasonality” and “foliage color” components.

SAR variables also capture seasonality, as well as important information

within the “texture” component. For vegetation classification tasks it has

been already documented that S2 data is more informative than S1 data

(Dobrini´

c et al., 2021; Meroni et al., 2021; Song et al., 2021; Lechner

et al., 2022; Ahlswede et al., 2023). This underlines the high relevance of

optical sensors for tree species classification (Dymond et al., 2002;

Fig. 5. Temporal profiles and variation across the year 2021 from three vegetation indices based on the WEFL61 dataset: A) Heatmap of the profiles from 61 tree

species, B) Heatmaps of the variation among the time series samples of each class. A high variation is represented in red color. The x-axis stands for the day of year.

The y-axis groups the species alphabetically and by their plant functional type. The number in brackets after the species name represents the number of time series (i.

e., forest stands) which were used for the calculation of the temporal profile. For better visualization, the value ranges were normalized from 0 to 1 across each

variable. Abbreviations: NDVI =Normalized Difference Vegetation Index, TCG =Tasseled Cap Greenness, RVI =Radar Vegetation Index, BD =broadleaf-deciduous,

ND =needleleaf-deciduous and NE =needleleaf-evergreen. Results from all study years can be found in the Supplements (Fig. C.3-C.5). (For interpretation of the

references to color in this figure legend, the reader is referred to the web version of this article.)

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

Fassnacht et al., 2016; Misra et al., 2020; Pu, 2021). However, among

the studied variables derived from both sensor types, we found large

potential for discriminating between needleleaf species with SAR vari-

ables. This finding aligns with Lechner et al. (2022), who demonstrated

the utility of dual-polarimetric descriptors for tree species mapping

including five conifer species.

The RVI was demonstrated to be highly descriptive for tree species

and adds complementary information to optical vegetation indices. To

Fig. 6. Circular heatmaps of the temporal profiles from 61 tree species across the year 2018 based on the WEFL61 dataset: A) Normalized Difference Vegetation

Index (NDVI), B) Tasseled Cap Greenness (TCG), C) Radar Vegetation Index (RVI). The heatmaps are read from outer to inner circle with day of year 1 as outermost

and day of year 365 as innermost. Left column: Profiles in chronological order by annual mean value. Middle column: Profiles arranged by the genus in alphabetical

order. Right column: Profiles arranged by the plant functional type and clustered by the mean value. For better visualization, the value range of each index was

normalized from 0 to 1. Results from all study years can be found in the Supplements (Fig. C.6-C.9).

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

our knowledge, only one study specifically used the dual-polarimetric

RVI from S1 for tree species classification (Blickensd¨

orfer et al., 2024).

However, other SAR variables serve as alternatives. For instance, the

VH-VV cross-ratio and the VH polarization also showed signal drops in

summer in deciduous forests (Dubois et al., 2020; Vreugdenhil et al.,

2020; Lechner et al., 2022). Despite Lechner et al. (2022) finding only

marginal improvements for tree species classification using the S1 VH-

VV cross-ratio, similar studies in agriculture have shown that SAR de-

scriptors are essential for an improved prediction quality (Mandal et al.,

2020; Bhogapurapu et al., 2021). Therefore, future research should

systematically investigate the added value of using polarimetric vege-

tation indices for forest applications.

Fig. 7. Pearson’s correlation coefficient (r) for the temporal profiles of deciduous (left column) and evergreen species (right column). Each correlation matrix shows

20 randomly selected species and the median r. Abbreviations: NDVI =Normalized Difference Vegetation Index, TCG =Tasseled Cap Greenness, RVI =Radar

Vegetation Index.

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

Fig. 8. Distribution of the individual samples in the principal component analysis for the tree species classes in the WEFL61 dataset: A) Score plots for the exemplary

species European beech (Fagus sylvatica), European larch (Larix decidua) and Scots pine (Pinus sylvestris) (ellipse level =95%), B) Score plots of all classes drawn as

ellipses of 95% confidence intervals and grouped by the plant functional type, C) Centroid of the samples for each class, D) Centroid of the samples for each class at

four individual months. All the Sentinel-2 and Sentinel-1 variables from Table 2 were included into the principal component analysis as loading variables. For better

visualization, the set of species was reduced to the 43 most dominant tree species in the data. The month-wise principal component analyses were run on data subsets

from all study years (2018–2021). Further results can be found in the Supplement (Figs. B.1-B.15).

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

5.1.2. Influence of the season on variable importance

Strong shifts in the importances of the input variables were found in

the PCAs in monthly subsets. They are caused by the different sensor

characteristics and their sensitivity to changing biophysical traits during

the seasons. These differences may affect the transferability when

applying pre-trained machine learning models on data from different

seasons. The importance of the infra-red channels and optical vegetation

indices during summer for tree species classification is undeniable (e.g.

Immitzer et al., 2016; Persson et al., 2018). Comparable research on S1

covers so far only the forest type-level (e.g., Rüetschi et al., 2017;

Vreugdenhil et al., 2020; Bjerreskov et al., 2021; Dost´

alov´

a et al., 2021;

Waser et al., 2021). Yet, our results show an unexploited potential of

SAR to differentiate tree crown volumes in temperate forests in different

seasons. S1 also appears to be influenced by the crown texture and

branch structure which may be applicable to differentiate tree species in

winter.

5.1.3. Important components for the plant functional types

Studies relying on multispectral data found that the SWIR, optical

red and the red-edge spectra are crucial for tree species classification

(Immitzer et al., 2016, 2019; Bolyn et al., 2022; Grabska et al., 2019).

This is generally consistent with the results of the PCA for the 61 tree

species, where the most important principal component (“seasonality”)

is influenced by the infra-red spectra and optical vegetation indices.

However, the importances of the spectral ranges differ for the plant

functional types. The “seasonality” component is less relevant for the

evergreen species. They are therefore more strongly influenced by the

“foliage color” component, which primarily represents the spectrum of

visible light. This result partly agrees with Kollert et al. (2021), who

found that the red band from S2 is important for distinguishing conifers.

The “texture” component, to which the S1 variables contributed the

most, adds a third layer of information. Since crown volume influences

the strength of the backscatter, we find polarimetric variables particu-

larly relevant for distinguishing conifers.

5.2. Key challenges for tree species classification

5.2.1. Spectral-temporal similarities of tree species

Similar spectral and temporal traits are the main challenge for tree

species classification. They are often based on a common phenology,

foliage color or crown volume. For instance, the group of pines consists

of species with sparse tree crowns with low volume. Similar results were

observed at a group of deciduous species from different genera. For the

group of spruces, a generally strong seasonality was unexpectedly

observed compared to other conifer species. This could be either

explained by particularly strong changes in the chlorophyll content

during the vegetation season or by large-scale disturbance by the Eu-

ropean spruce bark beetle (Ips typographus) (e.g. Buras et al., 2021; Xu

et al., 2024), which lead to gaps in the canopy and mixed pixel signals

though understory vegetation.

For Central Europe, it is common practice in tree species classifica-

tion to limit the target classes to the most dominant species and to form

groups of species based on their taxonomic relationships (e.g. genus) or

forestry designations (e.g. hardwood versus softwood) (Immitzer et al.,

2016; Wessel et al., 2018; Grabska et al., 2020; Kollert et al., 2021;

Ahlswede et al., 2023; Kluczek et al., 2023). Due to the small number of

Fig. 9. RVI time series for the year 2020 sorted by three stand factors: A) Longitude in decimal degrees, B) Latitude in decimal degrees, C) Altitude in meters. For

better visualization, the value range was normalized from 0 to 1. The results from all study years can be found in the Supplement (Figs. D.1-D.12).

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

parcels and the uncertain proportion of the main tree species, some

temporal profiles in our results may not be highly representative.

However, they show clearly that aggregating species by their taxonomic

relationships is a suboptimal practice, because the spectral-temporal

profiles within the groups can differ (e.g. Acer campestre vs.

A. platanoides). Instead, forming groups of species based on similar

spectral-temporal characteristics is more suitable if only a few training

samples are available for the individual tree species (Hemmerling et al.,

2021).

Due to the similar traits of many species, reaching the goal of clas-

sification on the actual tree species-level is challenging. However, Her-

mosilla et al. (2022) presented nation-wide maps of 37 tree species in

Canada modeled using spaceborne optical satellite imagery and

biogeographical features. Marconi et al. (2022) also extracted highly

descriptive features for 77 tree species on the US scale from airborne

hyperspectral imagery. Similar approaches could support developers to

enhance the thematic content of large-scale forest maps to better inform

forestry and biodiversity stakeholders in Europe. A larger range of input

variables from hyperspectral imagery (Guanter et al., 2015; Clasen et al.,

2015) and the use of X- and L-band SAR imagery (Krieger et al., 2007;

Motohka et al., 2019) may also improve tree species mapping and thus

needs to be further investigated.

5.2.2. Spatial and temporal domain adaptation problems

The RVI exhibited large spatial variations related to geolocation in

our relatively flat and homogeneous study area. Therefore, it should be

considered that classification models using SAR data as an input could

potentially face spatial domain adaptation problems, particularly in

study areas with larger biogeographic gradients. Some tree species

classification studies utilize a spatial cross-validation in order to choose

the best model for mapping (e.g., Hermosilla et al., 2022). Other studies

include a spatially balanced sample selection for training data genera-

tion (e.g., Welle et al., 2022). Both approaches prevent the regional

overfitting of machine learning models and thus allow for an improved

generalization on the large-scale (Meyer and Pebesma, 2021).

The extraction of phenology features from optical and SAR data is

considered beneficial for tree species classification (e.g. Pasquarella

et al., 2018; Hemmerling et al., 2021; Kollert et al., 2021; Lechner et al.,

2022). Nonetheless, most studies did not incorporate temporal metrics

as predictors. TCG and RVI were found to be particularly useful for

discriminating between early and late budding species. However, strong

shifts in seasonal patterns due to weather conditions were also observed

over the four study years which may lead to problems in the temporal

generalization of machine learning classifiers. Schulz et al. (2021a)

demonstrated that predictive models trained with data from multiple

years perform robustly and generalize better to unknown data. Fass-

nacht et al. (2023) emphasized temporal cross-validation in upcoming

research to ensure temporally continuous forest monitoring with remote

sensing.

5.3. Influence of stand factors

5.3.1. Influence of management conditions

For the three most common species in our study area, management

conditions surprisingly did not have a discernible influence on the op-

tical and polarimetric vegetation indices. Potential errors in the analysis

caused by outdated labels or the use of mixed forest stands can be largely

excluded, since the time period of the ground truth campaigns

(2015–2019) closely aligns with the study time period (2018–2021) and

only pure forest stands (i.e., monocultures) were used. The results

partially align with Grabska and Socha (2021), who found no significant

influence from management conditions on the S2 reflectance values for

Fagus sylvatica in the Polish Carpathians. For Pinus sylvestris, however,

they found that stand density influences the traits. Müller et al. (2021)

investigated the influence of understory vegetation on S1 backscatter at

Pinus sylvestris stands in Central Germany and found a significant impact

of the eagle fern (Pteridium aquilinum) on both polarizations. A direct

comparison to our results is challenging as we based our stand factor

Fig. 10. Spatial and temporal variances in the Radar Vegetation Index (RVI), air temperature in C

◦(TEMP) and precipitation in mm (PREC) for the year 2018: A) 200

randomly selected time series from Norway spruce (Picea abies) forest stands arranged by the factor latitude, B) Mean Radar Vegetation Index (MEAN_RVI), mean air

temperature (MEAN_TEMP) and sum of precipitation in mm (SUM_PREC) for the classes European beech (Fagus sylvatica), Norway spruce (Picea abies) and Scots pine

(Pinus sylvestris), C) Mean air temperature (MEAN_TEMP) and Sum of precipitation (SUM_PREC) at the forest stands from all three species. The results from all study

years can be found in the Supplement (Figs. D.13-D.16).

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

analysis on three vegetation indices instead of the original optical or

polarimetric variables. We assume that the selected vegetation indices

are less affected by management conditions than the original S1 and S2

bands which emphasizes their potential as robust input variables for tree

species mapping.

5.3.2. Influence of biogeographic conditions

For the two conifers Pinus sylvestris and Picea abies, we found a

substantial influence of the geolocation and altitude on the RVI. These

deviations may weaken the suitability of SAR data for tree species

classification on larger scales. Three possible reasons for the found

spatial variations were investigated: i) Morning dew was suspected to be

a plausible reason for the spatial variations (Dubois et al., 2020). To

evaluate the hypothesis, we analyzed S1 data from the ascending flight

mode (derived in the evening) and found the same spatial patterns to the

results in the study. Benninga et al. (2019) also found no systematic

effect of dew on S1 in forests. Hence, we excluded dew as the reason for

the spatial differences in RVI. ii) Because S1 backscatter is influenced by

soil moisture in non-forested ecosystems (Holtgrave et al., 2018, 2020;

Klinke et al., 2018), it may also be affected in forested surfaces. How-

ever, Dubois et al. (2020) state that backscatter signals cannot penetrate

tree crowns with dense canopies. As the spatial variations occur

throughout the year and only for the conifers, soil moisture is unlikely to

explain the phenomenon. iii) Dubois et al. (2020) state that volume

backscattering may be influenced by the water content in the foliage. We

expect that due to a moister and milder climate, forests in the northwest

of our study area develop larger and denser tree crowns which leads to

high volume scattering in S1. Forest stands in the drier southeast may be

affected by drought in recent years (Zink et al., 2016). The consequent

differences in foliage density and a tree crown water content might

explain the RVI variations in the two conifer classes. However, a similar

effect on Fagus sylvatica would be expected. Local measurements,

different study regions and the analysis of other species could provide a

better understanding of the found variations.

5.3.3. Influence of meteorological conditions

In the RVI time series of all species and study years, sudden drops in

the RVI co-occur with dry frost periods. According to the literature,

these irregularities could be caused by frozen topsoil or by reduced

water content in the tree crowns (Kwok et al., 1994; Fassnacht et al.,

2016; Baghdadi et al., 2018; Benninga et al., 2019). Thus, even if the

winter season is less important for tree species mapping, frost-related

anomalies in the backscatter should be considered as a limiting factor

when using S1 time series data.

6. Conclusion

In our study, we used techniques from explorative statistics to esti-

mate the usability of C-band SAR and multispectral satellite imagery for

tree species classification in Central Europe. We compared the spectral-

temporal traits of a large number of deciduous and evergreen tree spe-

cies, including several rarely studied species. We revealed potential

strengths but also pitfalls of using the S1 and S2 time series data for

supervised classification with respect to spatial and temporal extrapo-

lation. Here, we succinctly summarize our recommendations for tree

species mapping using machine learning according to this study and

previous literature:

•Carefully decide whether to classify at the genus or species-level. We

believe that individual species mapping is the next frontier due to the

high importance of rare species mapping for biodiversity and con-

servation applications.

•Genus may not be the optimal grouping of species for classification;

we recommend groups of species with similar spectral-temporal

traits which can be used instead.

•Take at least one seasonal index, one foliage color index, and one

structural (SAR) vegetation index and take into account the vari-

ability of the whole year as they can allow for a greater number of

tree species to be distinguished.

•SAR vegetation indices such as RVI could improve the classification

of tree species, especially for conifers.

•SAR indices may be more affected by biogeographic factors leading

to differences in the crown volume; optical indices may be more

suitable for large-scale mapping.

•Compared to the NDVI, the TCG and RVI are more useful for

discriminating between early and late budding deciduous species.

In conclusion, we highly encourage the use of both optical and SAR

data for tree species mapping. By investigating the full potential of the

S1 and S2 time series data, we support the feature engineering process in

machine learning and improve forest mapping on a larger scale. The

spatial distribution of endangered or invasive species is urgently needed

to identify regions of high priority for biodiversity protection, climate

change adaptation and wildfire prevention. Our results support the

mapping of a higher number of tree species, which is essential for

environmental authorities and conservation organizations.

CRediT authorship contribution statement

Christian Schulz: Conceptualization, Formal analysis, Funding

acquisition, Methodology, Project administration, Software, Visualiza-

tion, Writing – original draft, Writing – review & editing. Michael

F¨

orster: Funding acquisition, Methodology, Supervision, Writing – re-

view & editing. Stenka Valentinova Vulova: Methodology, Writing –

review & editing. Alby Duarte Rocha: Writing – review & editing,

Methodology, Software, Supervision. Birgit Kleinschmit: Conceptual-

ization, Data curation, Funding acquisition, Project administration, Su-

pervision, Writing – review & editing.

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.

Data availability

The R and GEE scripts can be made available via email request.

Acknowledgments

This article is a contribution of the project Künstliche Intelligenz mit

Erdbeobachtungs- und Multi-Source Geodaten für das Infrastruktur-,

Naturschutz- und Waldmonitoring (TreeSatAI) funded by the German

Federal Ministry for Education and Research (01IS20014A). The authors

are very grateful for the Forest Administration of Lower Saxony (For-

stplanungsamt Wolfenbüttel) for providing forest management data. We

thank the DWD for providing free meteorological data. We also thank

the ESA and Google for providing us free and pre-processed S1 and S2

data. Two anonymous reviewers kindly provided helpful feedback

during the revisions process.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.

org/10.1016/j.rse.2024.114162.

References

Ahlswede, S., Schulz, C., Gava, C., Helber, P., Bischke, B., F¨

orster, M., Arias, F., Hees, J.,

Demir, B., Kleinschmit, B., 2023. TreeSatAI benchmark archive: a multi-sensor, multi-

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

label dataset for tree species classification in remote sensing. Earth Syst. Sci. Data 15,

681–695. https://doi.org/10.5194/essd-15-681-2023.

Baghdadi, N., Bazzi, H., El Hajj, M., Zribi, M., 2018. Detection of frozen soil using

Sentinel-1 SAR data. Remote Sens. 10, 1182. https://doi.org/10.3390/rs10081182.

Baron, D., Erasmi, S., 2017. High resolution forest maps from interferometric TanDEM-X

and multitemporal Sentinel-1 SAR data. PFG - J. Photogramm. Remote Sens. Geoinf.

Sci. 85, 389–405. https://doi.org/10.1007/s41064-017-0040-1.

Beck, H.E., Zimmermann, N.E., McVicar, T.R., Vergopolan, N., Berg, A., Wood, E.F.,

2018. Present and future K¨

oppen-Geiger climate classification maps at 1-km

resolution. Sci. Data 5, 180214. https://doi.org/10.1038/sdata.2018.214.

Benninga, H.-J., Van Der Velde, R., Su, Z., 2019. Impacts of radiometric uncertainty and

weather-related surface conditions on soil moisture retrievals with Sentinel-1.

Remote Sens. 11, 2025 https://doi.org/10.3390/rs11172025.

Bhogapurapu, N., Dey, S., Bhattacharya, A., Mandal, D., Lopez-Sanchez, J.M.,

McNairn, H., L´

opez-Martínez, C., Rao, Y.S., 2021. Dual-polarimetric descriptors from

Sentinel-1 GRD SAR data for crop growth assessment. ISPRS J. Photogramm. Remote

Sens. 178, 20–35. https://doi.org/10.1016/j.isprsjprs.2021.05.013.

Bjerreskov, K.S., Nord-Larsen, T., Fensholt, R., 2021. Classification of Nemoral forests

with fusion of multi-temporal Sentinel-1 and 2 data. Remote Sens. 13, 950. https://

doi.org/10.3390/rs13050950.

Blickensd¨

orfer, L., Oehmichen, K., Pflugmacher, D., Kleinschmit, B., Hostert, P., 2024.

National tree species mapping using Sentinel-1/2 time series and German National

Forest Inventory data. Remote Sens. Environ. 304, 114069 https://doi.org/10.1016/

j.rse.2024.114069.

B¨

ockmann, T., 2016. Warum sind Betriebsinventuren für die forstliche Praxis wichtig?

Forstarchiv 87, 31–37. https://doi.org/10.4432/0300-4112-87-31.

Boeing, F., Rakovec, O., Kumar, R., Samaniego, L., Schr¨

on, M., Hildebrandt, A.,

Rebmann, C., Thober, S., Müller, S., Zacharias, S., Bogena, H., Schneider, K.,

Kiese, R., Attinger, S., Marx, A., 2022. High-resolution drought simulations and

comparison to soil moisture observations in Germany. Hydrol. Earth Syst. Sci. 26,

5137–5161. https://doi.org/10.5194/hess-26-5137-2022.

Boessenkool, B., 2023. Rdwd: Select and Download Climate Data from “DWD” (German

Weather Service). URL https://bookdown.org/brry/rdwd.

Bolyn, C., Lejeune, P., Michez, A., Latte, N., 2022. Mapping tree species proportions from

satellite imagery using spectral–spatial deep learning. Remote Sens. Environ. 280,

113205. https://doi.org/10.1016/j.rse.2022.113205.

Buras, A., Rammig, A., Zang, C.S., 2021. The European Forest condition monitor: using

remotely sensed forest greenness to identify hot spots of Forest decline. Front. Plant

Sci. 12, 689220 https://doi.org/10.3389/fpls.2021.689220.

Clasen, A., Somers, B., Pipkins, K., Tits, L., Segl, K., Brell, M., Kleinschmit, B.,

Spengler, D., Lausch, A., F¨

orster, M., 2015. Spectral Unmixing of Forest crown

components at close range, airborne and simulated Sentinel-2 and EnMAP spectral

imaging scale. Remote Sens. 7, 15361–15387. https://doi.org/10.3390/rs71115361.

Crist, E.P., Cicone, R.C., 1984. Application of the tasseled cap concept to simulated

thematic mapper data. Photogramm. Eng. Remote. Sens. 50, 343–352.

Deutscher Wetterdienst (DWD), 2022. Climate Data Center (CDC) [WWW Document].

URL. https://opendata.dwd.de/.

Dobrini´

c, D., Gaˇ

sparovi´

c, M., Medak, D., 2021. Sentinel-1 and 2 time-series for

vegetation mapping using random Forest classification: a case study of northern

Croatia. Remote Sens. 13, 2321. https://doi.org/10.3390/rs13122321.

Dost´

alov´

a, A., Lang, M., Ivanovs, J., Waser, L.T., Wagner, W., 2021. European wide

Forest classification based on Sentinel-1 data. Remote Sens. 13, 337. https://doi.org/

10.3390/rs13030337.

Drusch, M., Del Bello, U., Carlier, S., Colin, O., Fernandez, V., Gascon, F., Hoersch, B.,

Isola, C., Laberinti, P., Martimort, P., Meygret, A., Spoto, F., Sy, O., Marchese, F.,

Bargellini, P., 2012. Sentinel-2: ESA’s optical high-resolution Mission for GMES

operational services. Remote Sens. Environ. 120, 25–36. https://doi.org/10.1016/j.

rse.2011.11.026.

Dubois, C., Mueller, M.M., Pathe, C., Jagdhuber, T., Cremer, F., Thiel, C., Schmullius, C.,

2020. Characterization of land cover seasonality in SENTINEL-1 time series data.

ISPRS Ann. Photogramm. Remote Sens. Spatial. Inform. Sci. V-3–2020, 97–104.

https://doi.org/10.5194/isprs-annals-V-3-2020-97-2020.

Dymond, C.C., Mladenoff, D.J., Radeloff, V.C., 2002. Phenological differences in tasseled

cap indices improve deciduous forest classification. Remote Sens. Environ. 80,

460–472. https://doi.org/10.1016/S0034-4257(01)00324-8.

Fassnacht, F.E., Latifi, H., Stere´

nczak, K., Modzelewska, A., Lefsky, M., Waser, L.T.,

Straub, C., Ghosh, A., 2016. Review of studies on tree species classification from

remotely sensed data. Remote Sens. Environ. 186, 64–87. https://doi.org/10.1016/j.

rse.2016.08.013.

Fassnacht, F.E., White, J.C., Wulder, M.A., Næsset, E., 2023. Remote sensing in forestry:

current challenges, considerations and directions. Int. J. For. Res., cpad024 https://

doi.org/10.1093/forestry/cpad024.

Flores-Anderson, A.I., Herndon, K.E., Thapa, R.B., Cherrington, E. (Eds.), 2019. The SAR

Handbook: Comprehensive Methodologies for Forest Monitoring and Biomass

Estimation. NASA.

F¨

orster, S., Kaden, K., F¨

orster, M., Itzerott, S., 2012. Crop type mapping using

spectral–temporal profiles and phenological information. Comput. Electron. Agric.

89, 30–40. https://doi.org/10.1016/j.compag.2012.07.015.

Frison, P.-L., Fruneau, B., Kmiha, S., Soudani, K., Dufrˆ

ene, E., Toan, T.L., Koleck, T.,

Villard, L., Mougin, E., Rudant, J.-P., 2018. Potential of Sentinel-1 data for

monitoring temperate mixed Forest phenology. Remote Sens. 10, 2049 https://doi.

org/10.3390/rs10122049.

Gao, B., 1996. NDWI—A normalized difference water index for remote sensing of

vegetation liquid water from space. Remote Sens. Environ. 58, 257–266. https://doi.

org/10.1016/S0034-4257(96)00067-3.

Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017.

Google earth engine: planetary-scale geospatial analysis for everyone. Remote Sens.

Environ. 202, 18–27. https://doi.org/10.1016/j.rse.2017.06.031.

Grabska, E., Socha, J., 2021. Evaluating the effect of stand properties and site conditions

on the forest reflectance from Sentinel-2 time series. PLoS One 16, e0248459.

https://doi.org/10.1371/journal.pone.0248459.

Grabska, E., Hostert, P., Pflugmacher, D., Ostapowicz, K., 2019. Forest stand species

mapping using the Sentinel-2 time series. Remote Sens. 11, 1197 https://doi.org/

10.3390/rs11101197.

Grabska, E., Frantz, D., Ostapowicz, K., 2020. Evaluation of machine learning algorithms

for forest stand species mapping using Sentinel-2 imagery and environmental data in

the polish Carpathians. Remote Sens. Environ. 251, 112103 https://doi.org/

10.1016/j.rse.2020.112103.

Guanter, L., Kaufmann, H., Segl, K., Foerster, S., Rogass, C., Chabrillat, S., Kuester, T.,

Hollstein, A., Rossner, G., Chlebek, C., Straif, C., Fischer, S., Schrader, S., Storch, T.,

Heiden, U., Mueller, A., Bachmann, M., Mühle, H., Müller, R., Habermeyer, M.,

Ohndorf, A., Hill, J., Buddenbaum, H., Hostert, P., van der Linden, S., Leit˜

ao, P.J.,

Rabe, A., Doerffer, R., Krasemann, H., Xi, H., Mauser, W., Hank, T., Locherer, M.,

Rast, M., Staenz, K., Sang, B., 2015. The EnMAP Spaceborne imaging spectroscopy

Mission for earth observation. Remote Sens. 7, 8830–8857. https://doi.org/

10.3390/rs70708830.

Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A.,

Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A.,

Chini, L., Justice, C.O., Townshend, J.R.G., 2013. High-resolution global maps of

21st-century forest cover change. Science 342, 850–853. https://doi.org/10.1126/

science.1244693.

Haralick, R.M., Shanmugam, K., Dinstein, I., 1973. Textural features for image

classification. IEEE Trans. Syst. Man Cybern. SMC-3, 610–621. https://doi.org/

10.1109/TSMC.1973.4309314.

Healey, S., Cohen, W., Zhiqiang, Y., Krankina, O., 2005. Comparison of tasseled cap-

based Landsat data structures for use in forest disturbance detection. Remote Sens.

Environ. 97, 301–310. https://doi.org/10.1016/j.rse.2005.05.009.

Hemmerling, J., Pflugmacher, D., Hostert, P., 2021. Mapping temperate forest tree

species using dense Sentinel-2 time series. Remote Sens. Environ. 267, 112743

https://doi.org/10.1016/j.rse.2021.112743.

Hermosilla, T., Bastyr, A., Coops, N.C., White, J.C., Wulder, M.A., 2022. Mapping the

presence and distribution of tree species in Canada’s forested ecosystems. Remote

Sens. Environ. 282, 113276 https://doi.org/10.1016/j.rse.2022.113276.

Holtgrave, A.-K., F¨

orster, M., Greifeneder, F., Notarnicola, C., Kleinschmit, B., 2018.

Estimation of soil moisture in vegetation-covered floodplains with Sentinel-1 SAR

data using support vector regression. PFG - J. Photogramm. Remote Sens. Geoinf.

Sci. 86, 85–101. https://doi.org/10.1007/s41064-018-0045-4.

Holtgrave, A.-K., R¨

oder, N., Ackermann, A., Erasmi, S., Kleinschmit, B., 2020. Comparing

Sentinel-1 and -2 data and indices for agricultural land use monitoring. Remote Sens.

12, 2919. https://doi.org/10.3390/rs12182919.

Holzwarth, S., Thonfeld, F., Abdullahi, S., Asam, S., Da Ponte Canova, E., Gessner, U.,

Huth, J., Kraus, T., Leutner, B., Kuenzer, C., 2020. Earth observation based

monitoring of forests in Germany: a review. Remote Sens. 12, 3570 https://doi.org/

10.3390/rs12213570.

Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview

of the radiometric and biophysical performance of the MODIS vegetation indices.

Remote Sens. Environ. 83, 195–213. https://doi.org/10.1016/S0034-4257(02)

00096-2.

Immitzer, M., Vuolo, F., Atzberger, C., 2016. First experience with Sentinel-2 data for

crop and tree species classifications in Central Europe. Remote Sens. 8, 166. https://

doi.org/10.3390/rs8030166.

Immitzer, M., Neuwirth, M., B¨

ock, S., Brenner, H., Vuolo, F., Atzberger, C., 2019.

Optimal input features for tree species classification in Central Europe based on

multi-temporal Sentinel-2 data. Remote Sens. 11, 2599. https://doi.org/10.3390/

rs11222599.

J¨

onsson, P., Eklundh, L., 2002. Seasonality extraction by function fitting to time-series of

satellite sensor data. IEEE Trans. Geosci. Remote Sensing 40, 1824–1832. https://

doi.org/10.1109/TGRS.2002.802519.

Kim, Y., Jackson, T., Bindlish, R., Lee, H., Hong, S., 2012. Radar vegetation index for

estimating the vegetation water content of rice and soybean. IEEE Geosci. Remote

Sens. Lett. 9, 564–568. https://doi.org/10.1109/LGRS.2011.2174772.

Klinke, R., Kuechly, H., Frick, A., F¨

orster, M., Schmidt, T., Holtgrave, A.-K.,

Kleinschmit, B., Spengler, D., Neumann, C., 2018. Indicator-based soil moisture

monitoring of wetlands by utilizing sentinel and landsat remote sensing data. PFG -

J. Photogramm. Remote Sens. Geoinf. Sci. 86, 71–84. https://doi.org/10.1007/

s41064-018-0044-5.

Kluczek, M., Zagajewski, B., Zwijacz-Kozica, T., 2023. Mountain tree species mapping

using Sentinel-2, PlanetScope, and airborne HySpex hyperspectral imagery. Remote

Sens. 15, 844. https://doi.org/10.3390/rs15030844.

K¨

ohler, M., Steinicke, C., Evers, J., Meesenburg, H., Ahrends, B., 2016. Modelling water

and nutrient regime of forests in the framework of the forest site map of Lower

Saxony. For. Ecol. Landsc. Res. Nat. Conserv. 16, 83–94.

Kollert, A., Bremer, M., L¨

ow, M., Rutzinger, M., 2021. Exploring the potential of land

surface phenology and seasonal cloud free composites of one year of Sentinel-2

imagery for tree species mapping in a mountainous region. Int. J. Appl. Earth Obs.

Geoinf. 94, 102208 https://doi.org/10.1016/j.jag.2020.102208.

K¨

onig, S., Thonfeld, F., F¨

orster, M., Dubovyk, O., Heurich, M., 2023. Assessing

combinations of Landsat, Sentinel-2 and Sentinel-1 time series for detecting bark

beetle infestations. GiSci. Remote Sens. 60, 2226515 https://doi.org/10.1080/

15481603.2023.2226515.

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

Kowalski, K., Senf, C., Hostert, P., Pflugmacher, D., 2020. Characterizing spring

phenology of temperate broadleaf forests using Landsat and Sentinel-2 time series.

Int. J. Appl. Earth Obs. Geoinf. 92, 102172 https://doi.org/10.1016/j.

jag.2020.102172.

Krieger, G., Moreira, A., Fiedler, H., Hajnsek, I., Werner, M., Younis, M., Zink, M., 2007.

TanDEM-X: a satellite formation for high-resolution SAR interferometry. IEEE Trans.

Geosci. Remote Sens. 45, 3317–3341. https://doi.org/10.1109/TGRS.2007.900693.

Kwok, R., Rignot, E.J.M., Way, J., Freeman, A., Holt, J., 1994. Polarization signatures of

frozen and thawed forests of varying environmental state. IEEE Trans. Geosci.

Remote Sens. 32, 371–381. https://doi.org/10.1109/36.295051.

Lausch, A., Erasmi, S., King, D., Magdon, P., Heurich, M., 2016. Understanding Forest

health with remote sensing -part I—A review of spectral traits, processes and remote-

sensing characteristics. Remote Sens. 8, 1029. https://doi.org/10.3390/rs8121029.

Lechner, M., Dost´

alov´

a, A., Hollaus, M., Atzberger, C., Immitzer, M., 2022. Combination

of Sentinel-1 and Sentinel-2 data for tree species classification in a central European

biosphere reserve. Remote Sens. 14, 2687. https://doi.org/10.3390/rs14112687.

Lines, E.R., Allen, M., Cabo, C., Calders, K., Debus, A., Grieve, S.W.D., Miltiadou, M.,

Noach, A., Owen, H.J.F., Puliti, S., 2022. AI Applications in Forest Monitoring Need

Remote Sensing Benchmark Datasets. https://doi.org/10.48550/arXiv.2212.09937.

MacDicken, K., Jonsson, ¨

O., Pi˜

na, L., Maulo, S., Contessa, V., Adikari, Y., Garzuglia, M.,

Lindquist, E., Reams, G., D’Annunzio, R., 2016. Global Forest Resources Assessment

2015: How Are the world’s Forests Changing?, Second edition. Food and Agriculture

Organization of the United Nations, Rome.

Mandal, D., 2020. Radar Vegetation Index for Sentinel-1 SAR Data - RVI4S1 Script

[WWW Document]. Script provided for sentinel-hub.com. URL. https://github.com

/sentinel-hub/custom-scripts/tree/master/sentinel-1/radar_vegetation_index.

Mandal, D., Kumar, V., Ratha, D., Dey, S., Bhattacharya, A., Lopez-Sanchez, J.M.,

McNairn, H., Rao, Y.S., 2020. Dual polarimetric radar vegetation index for crop

growth monitoring using sentinel-1 SAR data. Remote Sens. Environ. 247, 111954

https://doi.org/10.1016/j.rse.2020.111954.

Marconi, S., Weinstein, B.G., Zou, S., Bohlman, S.A., Zare, A., Singh, A., Stewart, D.,

Harmon, I., Steinkraus, A., White, E.P., 2022. Continental-scale hyperspectral tree

species classification in the United States National Ecological Observatory Network.

Remote Sens. Environ. 282, 113264 https://doi.org/10.1016/j.rse.2022.113264.

Mayr, S., Kuenzer, C., Gessner, U., Klein, I., Rutzinger, M., 2019. Validation of earth

observation time-series: a review for large-area and temporally dense land surface

products. Remote Sens. 11, 2616 https://doi.org/10.3390/rs11222616.

McDowell, N.G., Allen, C.D., Anderson-Teixeira, K., Aukema, B.H., Bond-Lamberty, B.,

Chini, L., Clark, J.S., Dietze, M., Grossiord, C., Hanbury-Brown, A., Hurtt, G.C.,

Jackson, R.B., Johnson, D.J., Kueppers, L., Lichstein, J.W., Ogle, K., Poulter, B.,

Pugh, T.A.M., Seidl, R., Turner, M.G., Uriarte, M., Walker, A.P., Xu, C., 2020.

Pervasive shifts in forest dynamics in a changing world. Science 368, eaaz9463.

https://doi.org/10.1126/science.aaz9463.

Meroni, M., d’Andrimont, R., Vrieling, A., Fasbender, D., Lemoine, G., Rembold, F.,

Seguini, L., Verhegghen, A., 2021. Comparing land surface phenology of major

European crops as derived from SAR and multispectral data of Sentinel-1 and -2.

Remote Sens. Environ. 253, 112232 https://doi.org/10.1016/j.rse.2020.112232.

Meyer, F., 2019. Spaceborne synthetic aperture radar: Principles, data access, and basic

processing techniques. In: Flores-Anderson, A.I., Herndon, K.E., Thapa, R.B.,

Cherrington, E. (Eds.), SAR Handbook: Comprehensive Methodologies for Forest

Monitoring and Biomass Estimation. NASA.

Meyer, H., Pebesma, E., 2021. Predicting into unknown space? Estimating the area of

applicability of spatial prediction models. Methods Ecol. Evol. 12, 1620–1633.

https://doi.org/10.1111/2041-210X.13650.

Millennium Ecosystem Assessment, 2005. Ecosystems and Human Well-Being: Synthesis.

World Resources Institute, Washington, DC.

Misra, G., Cawkwell, F., Wingler, A., 2020. Status of Phenological research using

Sentinel-2 data: a review. Remote Sens. 12, 2760. https://doi.org/10.3390/

rs12172760.

Motohka, T., Kankaku, Y., Miura, S., Suzuki, S., 2019. Alos-4 L-band SAR mission and

observation. In: IEEE International Geoscience and Remote Sensing Symposium,

Yokohama, Japan, vol. 2019, pp. 5271–5273. https://doi.org/10.1109/

IGARSS.2019.8898169.

Müller, M.M., Dubois, C., Jagdhuber, T., Pathe, C., Schmullius, C., 2021. Investigation of

Sentinel-1 time series for sensitivity to fern vegetation in an European temperate

forest. Int. Arch. Photogramm. Remote. Sens. Spat. Inf. Sci. XLIII-B3-2021, 127–134.

https://doi.org/10.5194/isprs-archives-XLIII-B3-2021-127-2021.

Mullissa, A., Vollrath, A., Odongo-Braun, C., Slagter, B., Balling, J., Gou, Y., Gorelick, N.,

Reiche, J., 2021. Sentinel-1 SAR backscatter analysis ready data preparation in

google earth engine. Remote Sens. 13, 1954. https://doi.org/10.3390/rs13101954.

Nieders¨

achsische Landesforsten, 2020a. Waldeinrichtungsfl¨

achen der Nieders¨

achsischen

Landesforsten [dataset].

Nieders¨

achsische Landesforsten, 2020b. Betriebsinventurdaten der Nieders¨

achsischen

Landesforsten [dataset].

Nordwestdeutsche Forstliche Versuchsanstalt and Nieders¨

achsisches Ministerium für

Ern¨

ahrung, Landwirtschaft und Verbraucherschutz (Ed.), 2021. Waldzustandsbericht

2021 für Niedersachsen. https://doi.org/10.5281/zenodo.5615008.

Paloscia, S., Macelloni, G., Pampaloni, P., Sigismondi, S., 1999. The potential of C- and L-

band SAR in estimating vegetation biomass: the ERS-1 and JERS-1 experiments. IEEE

Trans. Geosci. Remote Sensing 37, 2107–2110. https://doi.org/10.1109/36.774723.

Pasquarella, V.J., Holden, C.E., Woodcock, C.E., 2018. Improved mapping of forest type

using spectral-temporal Landsat features. Remote Sens. Environ. 210, 193–207.

https://doi.org/10.1016/j.rse.2018.02.064.

Persson, M., Lindberg, E., Reese, H., 2018. Tree species classification with multi-

temporal Sentinel-2 data. Remote Sens. 10, 1794 https://doi.org/10.3390/

rs10111794.

Preidl, S., Faude, U., 2023. Baumartenklassifizierung und naturschutzfachliche

Bewertung von W¨

aldern mit Fernerkundungsdaten: Entwicklung eines

prototypischen Ansatzes. BfN-Schriften 668. https://doi.org/10.19217/skr668.

Pu, R., 2021. Mapping tree species using advanced remote sensing technologies: a state-

of-the-art review and perspective. J. Remote Sens., 9812624 https://doi.org/

10.34133/2021/9812624.

Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H., Sorooshian, S., 1994. A modified soil

adjusted vegetation index. Remote Sens. Environ. 48, 119–126. https://doi.org/

10.1016/0034-4257(94)90134-1.

R Core Team, 2021. R: A Language and Environment for Statistical Computing. R

Foundation for Statistical Computing, Vienna, Austria.

Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1974. Monitoring vegetation

systems in the great plains with ERTS. In: Freden, S.C., Mercanti, E.P., Becker, M.A.

(Eds.), Third Earth Resources Technology Satellite-1 Symposium: Section A-B.

Technical Presentations, NASA, Washington, D.C., pp. 309–318

Rudin, C., 2019. Stop explaining black box machine learning models for high stakes

decisions and use interpretable models instead. Nat. Mach. Intell. 1, 206–215.

https://doi.org/10.1038/s42256-019-0048-x.

Rüetschi, M., Schaepman, M., Small, D., 2017. Using multitemporal Sentinel-1 C-band

backscatter to monitor phenology and classify deciduous and coniferous forests in

northern Switzerland. Remote Sens. 10, 55. https://doi.org/10.3390/rs10010055.

dos Santos, E.P., da Silva, D.D., do Amaral, C.H., 2021. Vegetation cover monitoring in

tropical regions using SAR-C dual-polarization index: seasonal and spatial

influences. Int. J. Remote Sens. 42, 7581–7609. https://doi.org/10.1080/

01431161.2021.1959955.

Schuldt, B., Buras, A., Arend, M., Vitasse, Y., Beierkuhnlein, C., Damm, A., Gharun, M.,

Grams, T.E.E., Hauck, M., Hajek, P., Hartmann, H., Hiltbrunner, E., Hoch, G.,

Holloway-Phillips, M., K¨

orner, C., Larysch, E., Lübbe, T., Nelson, D.B., Rammig, A.,

Rigling, A., Rose, L., Ruehr, N.K., Schumann, K., Weiser, F., Werner, C.,

Wohlgemuth, T., Zang, C.S., Kahmen, A., 2020. A first assessment of the impact of

the extreme 2018 summer drought on central European forests. Basic Appl. Ecol. 45,

86–103. https://doi.org/10.1016/j.baae.2020.04.003.

Schulz, C., Holtgrave, A.-K., Kleinschmit, B., 2021a. Large-scale winter catch crop

monitoring with Sentinel-2 time series and machine learning–an alternative to on-

site controls? Comput. Electron. Agric. 186, 106173 https://doi.org/10.1016/j.

compag.2021.106173.

Schulz, D., Yin, H., Tischbein, B., Verleysdonk, S., Adamou, R., Kumar, N., 2021b. Land

use mapping using Sentinel-1 and Sentinel-2 time series in a heterogeneous

landscape in Niger, Sahel. ISPRS J. Photogramm. Remote Sens. 178, 97–111. https://

doi.org/10.1016/j.isprsjprs.2021.06.005.

Seidl, R., Thom, D., Kautz, M., Martin-Benito, D., Peltoniemi, M., Vacchiano, G., Wild, J.,

Ascoli, D., Petr, M., Honkaniemi, J., Lexer, M.J., Trotsiuk, V., Mairota, P.,

Svoboda, M., Fabrika, M., Nagel, T.A., Reyer, C.P.O., 2017. Forest disturbances

under climate change. Nat. Clim. Chang. 7, 395–402. https://doi.org/10.1038/

nclimate3303.

Senf, C., Buras, A., Zang, C.S., Rammig, A., Seidl, R., 2020. Excess forest mortality is

consistently linked to drought across Europe. Nat. Commun. 11, 6200. https://doi.

org/10.1038/s41467-020-19924-1.

Simons, N.K., Felipe-Lucia, M.R., Schall, P., Ammer, C., Bauhus, J., Blüthgen, N.,

Boch, S., Buscot, F., Fischer, M., Goldmann, K., Gossner, M.M., H¨

ansel, F., Jung, K.,

Manning, P., Nauss, T., Oelmann, Y., Pena, R., Polle, A., Renner, S.C., Schloter, M.,

Sch¨

oning, I., Schulze, E.-D., Solly, E.F., Sorkau, E., Stempfhuber, B., Wubet, T.,

Müller, J., Seibold, S., Weisser, W.W., 2021. National Forest Inventories capture the

multifunctionality of managed forests in Germany. For. Ecosyst. 8, 5. https://doi.

org/10.1186/s40663-021-00280-5.

Song, X.-P., Huang, W., Hansen, M.C., Potapov, P., 2021. An evaluation of Landsat,

Sentinel-2, Sentinel-1 and MODIS data for crop type mapping. Sci. Remote Sens. 3,

100018 https://doi.org/10.1016/j.srs.2021.100018.

Soudani, K., Delpierre, N., Berveiller, D., Hmimina, G., Vincent, G., Morfin, A.,

Dufrˆ

ene, ´

E., 2021. Potential of C-band synthetic aperture radar Sentinel-1 time-series

for the monitoring of phenological cycles in a deciduous forest. Int. J. Appl. Earth

Obs. Geoinf. 104, 102505 https://doi.org/10.1016/j.jag.2021.102505.

Ssymank, A., 1994. Neue Anforderungen im europ¨

aischen Naturschutz - Das

Schutzgebietssystem Natura 2000 und die “FFH-Richtlinie” der EU. Natur und

Landschaft 69, 395–406.

Sudmanns, M., Tiede, D., Augustin, H., Lang, S., 2020. Assessing global Sentinel-2

coverage dynamics and data availability for operational Earth observation (EO)

applications using the EO-compass. Int. J. Digital Earth 13, 768–784. https://doi.

org/10.1080/17538947.2019.1572799.

Torres, R., Snoeij, P., Geudtner, D., Bibby, D., Davidson, M., Attema, E., Potin, P.,

Rommen, B., Floury, N., Brown, M., Traver, I.N., Deghaye, P., Duesmann, B.,

Rosich, B., Miranda, N., Bruno, C., L’Abbate, M., Croci, R., Pietropaolo, A.,

Huchler, M., Rostan, F., 2012. GMES Sentinel-1 mission. Remote Sens. Environ. 120,

9–24. https://doi.org/10.1016/j.rse.2011.05.028.

Tukey, J.W., 1977. Exploratory Data Analysis, Addison-Wesley Series in Behavioral

Science. Addison-Wesley Pub. Co, Reading, Mass.

Veloso, A., Mermoz, S., Bouvet, A., Le Toan, T., Planells, M., Dejoux, J.-F., Ceschia, E.,

2017. Understanding the temporal behavior of crops using Sentinel-1 and Sentinel-2-

like data for agricultural applications. Remote Sens. Environ. 199, 415–426. https://

doi.org/10.1016/j.rse.2017.07.015.

Verger, A., Kandasamy, S., Baret, F., 2016. Temporal techniques in remote sensing of

global vegetation. In: Ban, Y. (Ed.), Multitemporal Remote Sensing. Springer

International Publishing, Cham, pp. 217–232. https://doi.org/10.1007/978-3-319-

47037-5_11.

C. Schulz et al.

Remote Sensing of Environment 307 (2024) 114162

Vreugdenhil, M., Wagner, W., Bauer-Marschallinger, B., Pfeil, I., Teubner, I., Rüdiger, C.,

Strauss, P., 2018. Sensitivity of Sentinel-1 backscatter to vegetation dynamics: an

Austrian case study. Remote Sens. 10, 1396. https://doi.org/10.3390/rs10091396.

Vreugdenhil, M., Navacchi, C., Bauer-Marschallinger, B., Hahn, S., Steele-Dunne, S.,

Pfeil, I., Dorigo, W., Wagner, W., 2020. Sentinel-1 cross ratio and vegetation optical

depth: a comparison over Europe. Remote Sens. 12, 3404 https://doi.org/10.3390/

rs12203404.

Waser, L.T., Rüetschi, M., Psomas, A., Small, D., Rehush, N., 2021. Mapping dominant

leaf type based on combined Sentinel-1/−2 data – challenges for mountainous

countries. ISPRS J. Photogramm. Remote Sens. 180, 209–226. https://doi.org/

10.1016/j.isprsjprs.2021.08.017.

Welle, T., Aschenbrenner, L., Kuonath, K., Kirmaier, S., Franke, J., 2022. Mapping

dominant tree species of German forests. Remote Sens. 14, 3330 https://doi.org/

10.3390/rs14143330.

Wessel, M., Brandmeier, M., Tiede, D., 2018. Evaluation of different machine learning

algorithms for scalable classification of tree types and tree species based on Sentinel-

2 data. Remote Sens. 10, 1419 https://doi.org/10.3390/rs10091419.

Xu, C., F¨

orster, M., Gr¨

anzig, T., May, J., Kleinschmit, B., 2024. Relating soil moisture and

Sentinel-2 vegetation index patterns to spruce bark beetle infestations prior to

outbreak. Forest.: Intern. J. Forest Res. https://doi.org/10.1093/forestry/cpae007.

Zink, M., Samaniego, L., Kumar, R., Thober, S., Mai, J., Sch¨

afer, D., Marx, A., 2016. The

German drought monitor. Environ. Res. Lett. 11, 074002 https://doi.org/10.1088/

1748-9326/11/7/074002.

C. Schulz et al.