[go: up one dir, main page]

Next Article in Journal
Drought and Flood Monitoring of the Liao River Basin in Northeast China Using Extended GRACE Data
Next Article in Special Issue
Mapping Mining Areas in the Brazilian Amazon Using MSI/Sentinel-2 Imagery (2017)
Previous Article in Journal
Inventory of Glaciers in the Shaksgam Valley of the Chinese Karakoram Mountains, 1970–2014
Previous Article in Special Issue
Multitemporal Cloud Masking in the Google Earth Engine
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Global Estimation of Biophysical Variables from Google Earth Engine Platform

by
Manuel Campos-Taberner
1,*,
Álvaro Moreno-Martínez
2,3,
Francisco Javier García-Haro
1,
Gustau Camps-Valls
3,
Nathaniel P. Robinson
2,
Jens Kattge
4 and
Steven W. Running
2
1
Department of Earth Physics and Thermodynamics, Faculty of Physics, Universitat de València, Dr. Moliner 50, 46100 Burjassot, València, Spain
2
Numerical Terradynamic Simulation Group, College of Forestry & Conservation, University of Montana, Missoula, MT 59812, USA
3
Image Processing Laboratory (IPL), Universitat de València, Catedrático José Beltrán 2, 46980 Paterna, València, Spain
4
Max-Planck-Institute for Biogeochemistry, Hans-Knöll Straβe 10, 07745 Jena, Germany
*
Author to whom correspondence should be addressed.
Remote Sens. 2018, 10(8), 1167; https://doi.org/10.3390/rs10081167
Submission received: 6 June 2018 / Revised: 19 July 2018 / Accepted: 20 July 2018 / Published: 24 July 2018
(This article belongs to the Collection Google Earth Engine Applications)
Graphical abstract
">
Figure 1
<p>Work flow of the proposed retrieval chain over GEE.</p> ">
Figure 2
<p>Constrained random samples of leaf chlorophyll (C<math display="inline"><semantics> <msub> <mrow/> <mrow> <mi>a</mi> <mi>b</mi> </mrow> </msub> </semantics></math>), leaf dry matter (C<math display="inline"><semantics> <msub> <mrow/> <mrow> <mi>d</mi> <mi>m</mi> </mrow> </msub> </semantics></math>), and leaf water (C<math display="inline"><semantics> <msub> <mrow/> <mi>w</mi> </msub> </semantics></math>) contents based upon prior knowledge of the TRY database, the MODIS land cover (MCD12Q1), kernel density estimators, and copulas.</p> ">
Figure 3
<p>Theoretical performance of the Random forest regression over PROSAIL simulations of LAI, FAPAR, FVC and CWC. The colorbar indicates density of points in the scatter plots.</p> ">
Figure 4
<p>LAI, FAPAR, FVC, and CWC global maps and latitudinal transects corresponding to the mean values estimated by the proposed retrieval chain for the period 2010–2015.</p> ">
Figure 5
<p>LAI and FAPAR global maps and latitudinal transects corresponding to the mean values of the GEE MODIS reference product (MOD15A3H) for the period 2010–2015.</p> ">
Figure 6
<p>Sites location of the BELMANIP-2.1 network used for intercomparison of LAI and FAPAR retrievals and MOD15A3H LAI/FAPAR product.</p> ">
Figure 7
<p>Biome-dependent scatter plots of the retrieved LAI over BELMANIP2.1 sites for the period 2002–2017. The colorbar indicates density of points in the scatter plots.</p> ">
Figure 8
<p>Biome-dependent scatter plots of the retrieved FAPAR over BELMANIP2.1 sites for the period 2002–2017. The colorbar indicates density of points in the scatter plots.</p> ">
Figure 9
<p>LAI and FAPAR global maps and latitudinal transects corresponding to the difference of mean values between derived estimates by the proposed retrieval chain and the GEE MODIS reference product for the period 2010–2015.</p> ">
Figure 10
<p>LAI (<b>left</b>) and FAPAR (<b>right</b>) latitudinal profiles over Africa (longitude 22°E) corresponding to the mean values of the GEE MODIS reference product and estimated by the proposed retrieval chain for the period 2010–2015.</p> ">
Versions Notes

Abstract

:
This paper proposes a processing chain for the derivation of global Leaf Area Index (LAI), Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), Fraction Vegetation Cover (FVC), and Canopy water content (CWC) maps from 15-years of MODIS data exploiting the capabilities of the Google Earth Engine (GEE) cloud platform. The retrieval chain is based on a hybrid method inverting the PROSAIL radiative transfer model (RTM) with Random forests (RF) regression. A major feature of this work is the implementation of a retrieval chain exploiting the GEE capabilities using global and climate data records (CDR) of both MODIS surface reflectance and LAI/FAPAR datasets allowing the global estimation of biophysical variables at unprecedented timeliness. We combine a massive global compilation of leaf trait measurements (TRY), which is the baseline for more realistic leaf parametrization for the considered RTM, with large amounts of remote sensing data ingested by GEE. Moreover, the proposed retrieval chain includes the estimation of both FVC and CWC, which are not operationally produced for the MODIS sensor. The derived global estimates are validated over the BELMANIP2.1 sites network by means of an inter-comparison with the MODIS LAI/FAPAR product available in GEE. Overall, the retrieval chain exhibits great consistency with the reference MODIS product (R 2 = 0.87, RMSE = 0.54 m 2 /m 2 and ME = 0.03 m 2 /m 2 in the case of LAI, and R 2 = 0.92, RMSE = 0.09 and ME = 0.05 in the case of FAPAR). The analysis of the results by land cover type shows the lowest correlations between our retrievals and the MODIS reference estimates (R 2 = 0.42 and R 2 = 0.41 for LAI and FAPAR, respectively) for evergreen broadleaf forests. These discrepancies could be attributed mainly to different product definitions according to the literature. The provided results proof that GEE is a suitable high performance processing tool for global biophysical variable retrieval for a wide range of applications.

Graphical Abstract">

Graphical Abstract

1. Introduction

Earth vegetation plays an essential role in the study of global climate change influencing terrestrial CO 2 flux exchange and variability through plant respiration and photosynthesis [1,2]. Vegetation monitoring can be achieved through the evaluation of biophysical variables such as LAI (Leaf Area Index), FVC (Fraction Vegetation Cover) and FAPAR (Fraction of Absorbed Photosynthetically Active Radiation) [3,4]. LAI accounts for the amount of green vegetation that absorbs or scatters solar radiation, FVC determines the partition between soil and vegetation contributions, while FAPAR is a vegetation health indicator related with ecosystems productivity. In addition, canopy water content (CWC) accounts for the amount of water content at canopy level, varies with vegetation water status, and is usually computed as the product of leaf water content (C w ) and LAI [5,6]. These essential variables can be estimated using remote sensing data and are key inputs in a wide range of ecological, meteorological and agricultural applications and models.
Biophysical variables can be derived from remote sensing data using statistical, physical and hybrid retrieval methods [7,8,9]. Statistical methods rely on models to relate spectral data with the biophysical variable of interest, usually through some form of regression. Statistical methods such as neural networks [10], random forests [11] or kernel methods [12], extract patterns and trends from a data set and approximate the underlying physical laws ruling the relationships between them from data. Physically-based retrieval methods are based on the physical knowledge describing the interactions between incoming radiation and vegetation through radiative transfer models (RTMs) [13,14]. In particular, the MODIS LAI/FAPAR product is based on a three-dimensional RTM which links surface spectral bi-directional reflectance factors (BRFs) to both canopy and soil spectral and structural parameters [15]. On the other hand, hybrid methods couple statistical with physically-based approaches inverting a database generated by an RTM [16,17,18]. For example, CYCLOPES global products [19] were derived inverting the PROSAIL radiative transfer model [20] using neural networks, while the global EUMETSAT LAI/FAPAR/FVC products are being produced inverting PROSAIL with multi-output Gaussian process regression from the EUMETSAT Polar System (EPS) [21]. The use of RTMs implies modeling leaf and canopy structural and biochemical parameters. The ranges and distribution of parameters used for running the simulations are usually based on field measurements that are very useful for simulating specific land covers [16,17]. Nevertheless, when the objective is to simulate a wide range of vegetation situations and land covers, ground data is often a limitation. In these scenarios, the scientific community uses distributions based on several experimental datasets such as the HAWAII, ANGERS, CALMIT-1/2 and LOPEX [22,23,24,25] which together embrace hundreds of observations [26]. Continuous update of ground measurements used for radiative transfer modeling are key in order to better constrain the RTM inversion process [27]. In this framework, the use of global plant traits database (TRY) [28] containing thousands of leaf data could alleviate this limitation.
From an operational standpoint, processing remote sensing data on an ongoing basis demands high storage capability and efficient computational power mainly when dealing with time series of long term global data sets. This situation also occurs because of the wide variety of free available remote sensing data disseminated by agencies such as the National Aeronautics and Space Administration (NASA) (e.g., MODIS), the United States Geological Survey (USGS) (e.g., Landsat), and the European Space Agency (ESA) (e.g., data from the Sentinel constellation) [29]. Recently, Google (Mountain View, Cal., USA) developed the Google Earth Engine (GEE) [30], a cloud computing platform specifically designed for geospatial analysis at the petabyte scale. The GEE data catalog is composed by widely used geospatial data sets. The catalog is continuously updated and data are ingested from different government-supported archives such as the Land Process Distributed Active Archive Center (LP DAAC), the USGS, and the ESA Copernicus Open Access Hub. The GEE data catalogue contains numerous remote sensing data sets such as top and bottom of atmosphere reflectance, as well as atmospheric and meteorological data. Data processing is performed in a parallel on Google’s computational infrastructure, dramatically improving processing efficiency, and opens up excellent prospects especially for multitemporal and global studies that include vegetation, temperature, carbon exchange, and hydrological processes [31,32,33,34,35].
The present study proposes a generic retrieval chain for the production of global LAI, FAPAR, FVC and CWC estimates from 15 years of MODIS data (MCD43A4) on the GEE platform. The methodology is based on a hybrid method inverting a PROSAIL radiative transfer model database with random forests (RFs) regression. Major contributions of the presented work are:
  • The development of a general methodology for global LAI/FAPAR estimation including FVC and CWC which are not provided by MODIS.
  • The use of a global plant traits database (composed of thousands of data) for probability density function (PDF) estimation with copulas to be used for radiative transfer modeling leaf parameterization.
  • The enforceability of biophysical parameter retrieval chain over GEE exploiting its capabilities to provide climate data records of global biophysical variables at computationally both affordable and efficient way.
Validation was performed by means of inter-comparison with the official MODIS LAI/FAPAR product available on GEE (MCD15A3H) over a network of globally distributed sites. The only process computed locally is the RTM simulation, while the inversion of the database, the derivation of the global maps, and the assessment of the retrievals have been performed into the GEE platform. Furthermore, the proposed methodology estimates both FVC and CWC variables which are not part of the official MODIS products, giving an added value to this work.
The remainder of the paper is structured as follows. Section 2 describes the data used in this work while Section 3 outlines the followed methodology. Section 4 exhibits the obtained results and the validation of the global estimates, and Section 6 discusses the main conclusions of this work.

2. Data Collection

2.1. MODIS Data

In this study, we used the MCD43A4 and the MCD15A3H MODIS products both available in GEE. Both the MCD (MODIS Combined Data) reflectance and LAI/FAPAR products are generated combining data from Terra and Aqua spacecrafts and are disseminated in a level-3 gridded data set. The MCD43A4 product provides a Bidirectional Reflectance Distribution Function (BRDF) from a nadir view in the 1–7 MODIS bands (i.e., red, near infrared (NIR), blue, green, short wave infrared-1 (SWIR-1), short wave infrared-2 (SWIR-2), and middle wave infrared (MWIR), see Table 1). MCD43A4 offers global surface reflectance data at 500 m spatial resolution with 8-day temporal frequency.
On the other hand, GEE also offers access to MODIS derived LAI and FAPAR estimates through the MCD15A3H collection 6 product. The temporal frequency of the biophysical estimates is every four days, and the retrieval algorithm chooses the “best” pixel available from all the acquisitions of both MODIS sensors from within the 4-day period. The MCD15A3H main retrieval algorithm uses a look-up-table (LUT) approach simulated from a 3D RTM. Basically, this method searches for plausible values of LAI and FAPAR for a specific set of angles (solar and view), observed bidirectional reflectance factors at certain spectral bands, and biome types [15]. In addition, the MCD15A3H employs a back-up algorithm (when the main one fails) that uses empirical relationships between NDVI (Normalized Difference Vegetation Index) and the biophysical parameters. Similarly to MCD43A4, the pixel’s spatial resolution is 500 m.

2.2. Global Plant Traits

The TRY database represents the biggest global effort to compile a massive global repository for plant trait data (6.9 million trait records for 148,000 plant taxa) at unprecedented spatial and climatological coverage [28,36]. So far, the TRY initiative has delivered to the scientific community around 390 million trait records which have resulted in more than 170 publications (https://www.try-db.org/). The applications of the database range from functional and community ecology, plant geography, species distribution, and vegetation models parameterizations [37,38,39,40].
We use a realistic representation of global leaf trait variability provided by the TRY to optimize a vegetation radiative transfer model (PROSAIL), commonly used by the remote sensing community [41]. Instead of using the common lookup tables available in the literature to parametrize the model [19,21,42], we exploit the potential of the TRY database to infer the distributions and correlations among some key leaf traits (leaf chlorophyll C a b , leaf dry matter C d m and water C w contents) required by PROSAIL. Table 2 shows some basic information about the considered traits extracted from the TRY.

3. Methodology

Physical approaches to retrieve biophysical variables rely on finding the best match between measured and simulated spectra. The solution can be achieved by means of numerical optimization or Monte Carlo approaches which are computationally expensive and do not guarantee the convergence to an optimal solution. Recently, new and more efficient algorithms relying on Machine Learning (ML) techniques have emerged and have become the preferred choice for most RTM inversion applications [16,17,18,19,21]. In this work, we have followed the latter hybrid approach, combining radiative transfer modeling and the parallelized machine learning RFs implementation available in GEE to retrieve the selected biophysical variables. Figure 1 shows a schema of the work flow.

3.1. Creation of Leaf Plant Traits’ Distributions

Recent research has highlighted the importance of exploiting a priori knowledge to constrain solutions of the ill-posed inversion problem in RTMs [27,43]. In this work, we used the TRY database and the available literature to extract prior knowledge and improve our results. Despite using the biggest plant trait database available, the representation of trait observations in a spatial and climate context in the TRY is still limited, and it shows significant deviations among observed and modelled distributions [28]. Trait measurements in the TRY represent the variation of single leaf measurements because they are not abundance-weighted with respect to natural occurrence [28]. Trait distributions are biased due to the availability of samples, which vary significantly due to technical difficulties for sampling (e.g., very dense forests and remote areas) or the availability of funds to carry out expensive field measurement campaigns in the different parts of the globe. To overcome these issues, we have computed our leaf traits’ univariate distribution functions by combining the plant trait database (TRY) with a global map of plant functional types (PFTs). The chosen global map of plant functional types was the official MODIS (MCD12Q1) land cover product [44], we used this product to compute global fractions of PFTs to weight more realistically species’ occurrence for the selected traits (leaf chlorophyll, leaf dry matter, and leaf water contents).
The categorical information available in TRY allowed us to group leaf trait measurements in the common PFT definitions. These grouped data were then used to compute individual normalized histograms for each PFT, whereas the final leaf trait histogram was calculated as the weighted sum of each PFT normalized histogram according to the global PFT’s spatial occurrence fractions. Repeating this process for all considered leaf traits, we obtained their final univariate distribution functions. These functions were inferred directly from the data by means of a non-parametric kernel density estimation (KDE, [45]). This approach allowed us to model leaf probability distributions without requiring any assumption regarding parametric families.
Some combinations of plant traits, like the ones considered in this work, exhibit significant correlations and tradeoffs as a result of different plant ecological strategies [46]. In order to capture these dependencies among traits, we created distributions that were also able to model correlated multivariate data by means of different copula functions. These functions separate marginal distributions from the dependency structure of a given multivariate distribution [47]. More precisely, we use a multivariate Gaussian copula function [48]. Using the calculated marginal univariate distributions for each trait and the Gaussian multivariate copula, computed from leaf measurements of the TRY database, we created a set of random training samples while preserving the correlation structure among them (see Figure 2).

3.2. Radiative Transfer Modeling

We used the PROSAIL RTM, which results from coupling the PROSPECT leaf optical model [49] with the SAIL canopy reflectance model [50]. Note that we used the PROSPECT-5B [26] for the coupling, which accounts for chlorophylls and carotenoids separately. PROSAIL was run in forward mode for building a database mimicking MODIS canopy reflectance. These data were then used for training the retrieval model assuming turbid medium canopies with randomly distributed leaves. PROSAIL simulates top of canopy bidirectional reflectance from 400 to 2500 nm with a 1 nm spectral resolution as a function of leaf biochemistry variables, canopy structure and background, as well as the sun-view geometry. Leaf optical properties are given by the mesophyll structural parameter (N), leaf chlorophyll (C a b ) and carotenoid (C a r ) contents, leaf brown pigment (C b p ) content, as well as leaf dry matter (C m ) and water (C w ) contents. The average leaf angle inclination (ALA), the LAI, and the hot-spot parameter (Hotspot) characterize the canopy structure. A multiplicative brightness parameter ( β s ) was used to represent different background reflectance types [19]. The system’s geometry was described by the solar zenith angle, the view zenith angle, and the relative azimuth angle between both angles, which in our case corresponded to illumination and observation zenith angles of 0°. Sub-pixel mixed conditions (i.e., spatial heterogeneity) were tackled assuming a linear spectral mixing model, which pixels are composed by a mixture of pure vegetation (vCover) and bare soil (1-vCover) fractions’ [21].
The leaf variables were randomly generated following the calculated kernel density distributions from the available leaf traits measurements, whereas distributions of the canopy variables as well as the soil brightness parameter, were similar to those adopted in other global studies [19,21]. Brown pigments were intentionally set to zero in order to account only for photosynthetic elements of the canopy (see Table 3). In addition, with the aim of accounting for different sources of noise (e.g., atmospheric correction, BRDF normalization or radiometric calibration) a wavelength dependent white Gaussian noise was added to the reflectances of the PROSAIL simulations. Specifically, a Gaussian noise with σ = 0.015 was added in the blue, green, and red channels, σ = 0.025 in the NIR, and σ = 0.03 in the SWIR-1, SWIR-2 and MWIR.

3.3. Random Forests Regression

The inversion of PROSAIL was done using standard regression. There is a wide variety of machine learning models for regression and function approximation. In this paper, we focus on the particular family of methods called random forests (RFs). An RF is essentially an ensemble method that constructs a multitude of decision trees (each of them trained with different subsets of features and examples), and yields the mean prediction of the individual trees [51]. RFs’ classification and regression have been applied in different areas of concern in forest ecology, such as modelling the gradient of coniferous species [52], the occurrence of fire in Mediterranean regions [53], the classification of species or land cover type [54,55], and the analysis of the relative importance of the proposed drivers [55] or the selection of drivers [54,56,57]. The selection of RFs in our study is not incidental, and we capitalize on several useful properties. The main advantage of using RFs over other traditional machine learning algorithms like neural networks (NNETs) is that they can cope with high dimensional problems very easily thanks to their pruning strategy. In addition, unlike kernel machines (KMs), RFs are more computationally efficient. The RF strategy is very beneficial by alleviating the often reported overfitting problem of simple decision trees. Moreover, this paper training data set has been split into train and an independent test set that was only used for the assessment of the RFs. In addition, RFs excel in the presence of missing entries, heterogeneous variables, and can be easily parallelized to tackle large scale problems, which is especially relevant in the application described in this work. This way, we can exploit large datasets and run predictions within Google Earth Engine easily. In this work, we predicted the considered biophysical variables (LAI, FAPAR, FVC, and CWC) using the full set of MODIS land bands shown in Table 1.

4. Results and Validation

4.1. Random Forests Theoretical Performance

In this section, we evaluate the RFs’ theoretical capabilities for LAI, FAPAR, FVC and CWC retrieval. The training database was composed of 14,700 cases of reflectances in the MODIS channels (Table 1) and the corresponding biophysical variables (i.e., LAI, FAPAR, FVC, and CWC) accounting for any combination of the PROSAIL parameters. We first trained RFs with 70% of the PROSAIL samples and then evaluated the estimation results over the remaining 30% of the samples (not in the training). Figure 3 shows the scatter plots of the RFs’ estimates of every biophysical parameter over the unseen test set. High correlations (R 2 = 0.84, 0.89, 0.88, and 0.80 for LAI, FAPAR, FVC and CWC, respectively) low Root-Mean-Squared Errors (RMSE = 0.91 m 2 /m 2 , 0.08, 0.06, and 0.27 kg/m 2 for LAI, FAPAR, FVC and CWC, respectively) and practically no biases were found in all cases (see Figure 3).

4.2. Obtained Estimates over GEE

After the RFs’ regression assessment undertaken in the previous section, we ran the retrieval chain in GEE and obtained 15 years of global biophysical parameters. Here, we show the global mean values of LAI, FAPAR, FVC and CWC derived from 2010 to 2015 (Figure 4). The spatial distribution of retrieved parameters is expected, reaching the highest mean values close to the Equatorial zones (Central Africa forests and Amazon basin) followed by the Northern latitudes (e.g., boreal forests). In addition, Figure 5 shows the mean LAI and FAPAR values for the same period computed from the GEE MODIS reference product (MOD15A3H) freely distributed from the Land Processes Distributed Active Archive Center (LP DAAC) portal https://lpdaac.usgs.gov/.

4.3. Validation

Validation of LAI and FAPAR retrievals was undertaken by means of inter-comparison with the available LAI/FAPAR product (MCD15A3H) on GEE. The inter-comparison was conducted over a network of sites named BELMANIP-2.1 (Benchmark Land Multisite Analysis and Intercomparison of Products). These sites were especially selected for representing the global variability of vegetation, making them suitable for global intercomparison of land biophysical products [58]. BELMANIP-2.1 is an updated version of the original BELMANIP sites which includes 445 sites located in relatively homogeneous areas all over the globe (see Figure 6). The sites are aimed to be representative of the different planet biomes over an 10 × 10 km 2 area, mostly flat, and with minimum fractions of urban area and permanent water bodies.
We selected the MODIS pixels for every BELMANIP-2.1 location and then we computed the mean value of the MODIS valid pixels within a 1 km surrounding area. We also considered the contribution of partial boundary pixels by weighting their contribution to the mean according to their fractions included within the selected area. Non-valid pixels (clouds, cloud shadows) and low-quality pixels (back-up algorithm or fill values) were excluded according to the pixel-based quality flag in the MCD products. In addition, since the MCD15A3H and MCD43A4 differ in temporal frequency, only the coincident dates between them were selected for comparison. Due to the large amount of data available in GEE, we were able to select only high-quality MODIS pixels, resulting in ~60,000 valid pixels from 2002–2017 for validation. Figure 7 shows per biome scatter plots between the estimates provided by the proposed retrieval chain and the reference MODIS LAI product over the BELMANIP2.1 sites from 2002 to 2012. Goodness of fit (R 2 ) ranging from 0.70 to 0.86 and low errors (RMSE) ranging from 0.23 to 0.57 m 2 /m 2 are found between estimates in all biomes except for evergreen broadleaf forest, where R 2 = 0.42 and RMSE = 1.13 m 2 /m 2 are reported.
Similarly, Figure 8 shows the obtained scatter plots for FAPAR. In this case, very good agreement (R 2 ranging from 0.89 to 0.92) and low errors (RMSE ranging from 0.06 to 0.08) are found between retrievals and the MODIS FAPAR product, over bare areas, shrublands, herbaceous, cultivated, and broadleaf deciduous forest biomes. For needle-leaf and evergreen broadleaf forests lower correlations (R 2 = 0.57 and 0.41) and higher errors (RMSE = 0.18 and 0.09) are obtained. It is worth mentioning that over bare areas, the MODIS FAPAR presents an unrealistic minimum value (~0.05) through the entire period.
Figure 9 shows the LAI and FAPAR difference maps computed from the mean estimates (2010–2015) provided by the proposed retrieval chain and the mean reference MODIS LAI/FAPAR product. Mean LAI map revealed that most of the pixels fall within the range of ±0.5 m 2 /m 2 , which highlights the consistency between products. However, for high LAI values, there is an underestimation of the provided estimates over very dense canopies that may reach up to 1.4 m 2 /m 2 . In the case of FAPAR, there is a constant negative bias of ≈0.05 which is also noticeable in the scatter plots shown in Figure 8. This is related with a documented systematic overestimation of MODIS FAPAR retrievals [59,60,61], which is partly corrected by the proposed retrieval approach. The spatial consistency of LAI/FAPAR estimates was also compared over the African continent (Figure 10). The latitudinal transects provided by Figure 10 clearly show an underestimation of LAI retrievals in equatorial forests, having a better agreement for the remaining biomes.

5. Discussion

The usefulness of GEE for providing global land surface variables related to vegetation status was demonstrated in this work. GEE offers some major advantages mainly related to storage capacity and processing speed. Despite the variety of algorithms implemented in GEE, its capabilities are constrained by the number of state-of-the-art algorithms (in this case, regression-based) which are currently implemented in GEE. However, this limitation is being overcome by the increasing number of users developing algorithms that may be potentially implemented in GEE for a wide range of applications. The functions in GEE utilize several built-in parallelization and data distribution models to achieve high performance [30]. The RFs’ implementation in GEE is not an exception to that. It allowed for the exploitation of large data sets and to obtain global estimates very efficiently. In GEE, the system handles and hides nearly every aspect of how a computation is managed, including resource allocation, parallelism, data distribution, and retries. These decisions are purely administrative; none of them can affect the result of a query, only the speed at which it is produced [30]. Under these circumstances, it is very difficult to give exact computation times because they vary in every run. As an example, in the present work, to compute the mean biophysical maps implied to process 230 (46 yearly images × 5 years) FAPAR images at 500 m spatial resolution (~440 million cells) and compute their annual mean, it took around 6 h.
The GEE data catalogue includes MODIS surface reflectance daily products, which can be advantageous to fully exploit the information contained in the reflectance signal of the surface. In this paper, we have preferred to use the normalized reflectance (MCD43A4) as an input. The BRDF normalization and temporal compositing steps assume: (1) perfectness of the linear kernel model inversion; (2) change of vegetation cover within temporal window is insignificant. These underlying assumptions are approximate but allow robust estimates of the BRDF kernel coefficients. A number of algorithms to retrieve satellite products (LAI, FAPAR and FVC) such as CYCLOPES (SPOT/VGT), Copernicus (SPOT/VGT and PROBA-V), and LSA-SAF (MSG and EPS) use as input top of canopy (TOC) normalized to a standard geometrical configuration [19,21,62]. This approach reduces considerably the requirements in terms of number of inputs and computational load. Despite the good speed potential of the proposed chain, the computational cost may be relevant when the aim is to generate time series of global products. In order to reduce input uncertainties, we used the quality flag provided by MCD43 products to filter non-valid pixels (persistent clouds and/or cloud shadows in the MCD43 composite) as well as for identifying zones with low-quality pixels. Because of the above-mentioned assumptions of the MCD43 product, further improvements of the proposed methodology will include the uncertainty propagation of the reflectance input data to our retrievals for operational use.
The comparison results between retrieved LAI/FAPAR and the reference MODIS product revealed good spatial consistency. However, there are differences in mean LAI values over dense forests (up to 1.4 m 2 /m 2 ). The underestimation in high LAI values could be partly explained by two factors: (1) differences in the algorithms used to estimate the LAI; and (2) the use of distinct LAI definitions for each product estimates. Namely, the LAI retrieved by the proposed chain is based on the inversion of a RTM assuming the canopy as a turbid medium. This approximation provides estimates closer to an effective LAI (LAI e f f ). In turn, the MODIS retrieval algorithm accounts for vegetation clumping at leaf and canopy scales through radiative transfer formulations, therefore estimated values should be closer to actual LAI (LAI a c t u a l ). The relationship between LAI e f f and LAI a c t u a l is given by L A I a c t u a l L A I e f f Ω being Ω the cumpling index. Similar underestimation behaviour was found by other studies when comparing MODIS LAI products and LAI retrievals from RTM inversion [59,63]. Yan et al. [59] found RMSE = 0.66 m 2 /m 2 and RMSE = 0.77 m 2 /m 2 when comparing MODIS C6 LAI estimates with ground LAI a c t u a l and LAI e f f measurements respectively, as well as larger uncertainties in high LAI values. Regarding FAPAR, an overall negative bias is found for all biomes. However, this bias could not be regarded as an issue in the estimations, since different studies have pointed out a systematic overestimation of MODIS retrievals in both C5 and C6 at low FAPAR values as a main drawback of the product [59,60,61]. For example, Xu et al. [64] assessed MODIS FAPAR through comparisons to ground measurements available from 2012–2016, obtaining a reasonable agreement (R 2 = 0.83, RMSE = 0.10) but with an overall overestimation tendency (bias = 0.08, scatters distributed within 0–0.2 difference). Similar results (R 2 = 0.74, RMSE = 0.15) were reported by Yan et al. [59] using globally distributed FAPAR measurements. The study evidenced a clear overestimation of FAPAR over sparsely-vegetated areas, as noted previously in other studies [60].
It is worth mentioning that neither the FVC nor the CWC products are available on GEE. Moreover, there is no global and reliable CWC product with which to compare the CWC estimates derived by the proposed retrieval chain. Regarding FVC, there are only a few global products that differ in retrieval approaches and spatiotemporal features. Since the main objective of the manuscript is to provide a generic biophysical retrieval chain, including the validation with the corresponding biophysical variables over GEE, the comparison of parameters not provided by GEE is out of the scope of the paper and could be addressed in future works.

6. Conclusions

This paper proposed a processing chain for the estimation of global biophysical variables (LAI, FAPAR, FVC, and CWC) from long-term (15-year) MODIS data in GEE. The approach takes the advantage of exploiting Earth observation data rapidly and efficiently through the GEE cloud storage and parallel computing capabilities. The retrieval methodology is based on a hybrid approach combining physically-based radiative transfer modelling (PROSAIL) and random forests regression.
The leaf parameter co-distributions employed during the radiative transfer modelling step were obtained by means of exploiting the TRY database. This allowed a better PROSAIL parametrization based on thousands of chlorophyll, water, and dry matter content ground measurements at leaf level. The increasing amount of available plant trait data in TRY (containing thousands of records) alleviates the need of a more realistic representation for some of the input parameters in radiative transfer models.
A validation exercise was undertaken over the BELMANIP2.1 network of sites by means of inter-comparison of the derived LAI and FAPAR with the MODIS reference LAI/FAPAR product available on GEE. The obtained results highlight the consistency of the estimates provided by the retrieval chain with the reference MODIS product. However, lower/poorer correlations were found for evergreen broadleaf forests when compared with the rest of biomes. These discrepancies could be mainly attributed to different retrieval approaches and variables definition, since derived LAI estimates are closer to LAI e f f rather than LAI a c t u a l derived by the MOD15A3H product. In addition, derived FAPAR stands only for photosynthetic elements of the canopy while FAPAR provided by MODIS also accounts for non-photosynthetic elements. The proposed retrieval chain also derived globally both FVC and CWC variables which are not provided by any GEE dataset.
The results demonstrated the usefulness of GEE for global biophysical parameter retrieval and opened the door to user self-provisioning of leaf and canopy parameters in GEE for a wide range of applications including data assimilation and sensor fusion.

Supplementary Materials

Author Contributions

All co-authors of this manuscript significantly contributed to all phases of the investigation. They contributed equally to the preparation, analysis, review and editing of this manuscript.

Funding

The research leading to these results was funded by the European Research Council under Consolidator Grant SEDAL ERC-2014-CoG 647423, the NASA Earth Observing System MODIS project (Grant NNX08AG87A), and supported by the LSA SAF CDOP3 project, and the Spanish Ministry of Economy and Competitiveness (MINECO) through the ESCENARIOS (CGL2016-75239-R) project.

Acknowledgments

The authors want to acknowledge the efforts of the TRY initiative on plant traits (http://www.try-db.org), hosted at the Max Planck Institute for Biogeochemistry, Jena, Germany.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Raich, J.W.; Schlesinger, W.H. The global carbon dioxide flux in soil respiration and its relationship to vegetation and climate. Tellus B 1992, 44, 81–99. [Google Scholar] [CrossRef]
  2. Beer, C.; Reichstein, M.; Tomelleri, E.; Ciais, P.; Jung, M.; Carvalhais, N.; Rödenbeck, C.; Arain, M.A.; Baldocchi, D.; Bonan, G.B.; et al. Terrestrial Gross Carbon Dioxide Uptake: Global Distribution and Covariation with Climate. Science 2010, 329, 834–838. [Google Scholar] [CrossRef] [PubMed]
  3. Huete, A.; Didan, K.; Miura, T.; Rodriguez, E.P.; Gao, X.; Ferreira, L.G. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 2002, 83, 195–213. [Google Scholar] [CrossRef]
  4. Fensholt, R. Earth observation of vegetation status in the Sahelian and Sudanian West Africa: Comparison of Terra MODIS and NOAA AVHRR satellite data. Int. J. Remote Sens. 2004, 25, 1641–1659. [Google Scholar] [CrossRef]
  5. Clevers, J.G.P.W.; Kooistra, L.; Schaepman, M.E. Estimating canopy water content using hyperspectral remote sensing data. Int. J. Appl. Earth Obs. Geoinf. 2010, 12, 119–125. [Google Scholar] [CrossRef]
  6. Yebra, D.P.; Chuvieco, E.; Riaño, D.; Zylstra, P.; Hunt, R.; Danson, F.M.; Qi, Y.; Jurdao, S. A global review of remote sensing of live fuel moisture content for fire danger assessment, moving towards operational products. Remote Sens. Environ. 2013, 136, 455–468. [Google Scholar] [CrossRef]
  7. Wulder, M. Optical remote-sensing techniques for the assessment of forest inventory and biophysical parameters. Prog. Phys. Geogr. 1998, 22, 449–476. [Google Scholar] [CrossRef]
  8. Zheng, G.; Monika, M. Retrieving leaf area index (LAI) using remote sensing: theories, methods and sensors. Sensors 1998, 9, 2719–2745. [Google Scholar] [CrossRef] [PubMed]
  9. Verrelst, J.; Camps-Valls, G.; Muñoz-Marí, J.; Rivera, J.P.; Veroustraete, F.; Clevers, J.G.; Moreno, J. Optical remote sensing and the retrieval of terrestrial vegetation bio-geophysical properties—A review. ISPRS J. Photogramm. Remote Sens. 2015, 108, 273–290. [Google Scholar] [CrossRef]
  10. Haykin, S. Neural Networks—A Comprehensive Foundation, 2nd ed.; Prentice Hall: Upper Saddle River, NJ, USA, 1999. [Google Scholar]
  11. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef]
  12. Camps-Valls, G.; Bruzzone, L. Kernel Methods for Remote Sensing Data Analysis; Wiley & Sons: Chichester, UK, 2009; p. 434. ISBN 978-0-470-72211-4. [Google Scholar]
  13. Myneni, R.B.; Ramakrishna, R.; Nemani, R.; Running, S.W. Estimation of global leaf area index and absorbed PAR using radiative transfer models. IEEE Trans. Geosci. Remote Sens. 1997, 35, 1380–1393. [Google Scholar] [CrossRef]
  14. Kimes, D.S.; Knyazikhin, Y.; Privette, J.L.; Abuelgasim, A.A.; Gao, F. Inversion methods for physically-based models. Remote Sens. Rev. 2000, 18, 381–439. [Google Scholar] [CrossRef]
  15. Knyazikhin, Y.; Glassy, J.; Privette, J.L.; Tian, Y.; Lotsch, A.; Zhang, Y.; Wang, Y.; Morisette, J.T.; Votava, P.; Myneni, R.B.; et al. MODIS Leaf Area Index (LAI) and Fraction of Photosynthetically Active Radiation Absorbed by Vegetation (FPAR) Product (MOD15) Algorithm Theoretical Basis Document; NASA Goddard Space Flight Center: Greenbelt, MD, USA, 1999; Volume 20771. [Google Scholar]
  16. Campos-Taberner, M.; García-Haro, F.J.; Camps-Valls, G.; Grau-Muedra, G.; Nutini, F.; Crema, A.; Boschetti, M. Multitemporal and multiresolution leaf area index retrieval for operational local rice crop monitoring. Remote Sens. Environ. 2016, 187, 102–118. [Google Scholar] [CrossRef]
  17. Campos-Taberner, M.; García-Haro, F.J.; Camps-Valls, G.; Grau-Muedra, G.; Nutini, F.; Busetto, L.; Katsantonis, D.; Stavrakoudis, D.; Minakou, C.; Gatti, L.; et al. Exploitation of SAR and Optical Sentinel Data to Detect Rice Crop and Estimate Seasonal Dynamics of Leaf Area Index. Remote Sens. 2017, 9, 248. [Google Scholar] [CrossRef]
  18. Svendsen, D.H.; Martino, L.; Campos-Taberner, M.; García-Haro, F.J.; Camps-Valls, G. Joint Gaussian Processes for Biophysical Parameter Retrieval. IEEE Trans. Geosci. Remote Sens. 2018, 56, 1718–1727. [Google Scholar] [CrossRef] [Green Version]
  19. Baret, F.; Hagolle, O.; Geiger, B.; Bicheron, P.; Miras, B.; Huc, M.; Berthelot, B.; Niño, F.; Weiss, M.; Samain, O.; et al. LAI, fAPAR and fCover CYCLOPES global products derived from VEGETATION: Part 1: Principles of the algorithm. Remote Sens. Environ. 2007, 110, 275–286. [Google Scholar] [CrossRef] [Green Version]
  20. Baret, F.; Jacquemoud, S.; Guyot, G.; Leprieur, C. Modeled analysis of the biophysical nature of spectral shifts and comparison with information content of broad bands. Remote Sens. Environ. 1992, 41, 133–142. [Google Scholar] [CrossRef]
  21. García-Haro, F.J.; Campos-Taberner, M.; Muñoz-Marí, J.; Laparra, V.; Camacho, F.; Sánchez-Zapero, J.; Camps-Valls, G. Derivation of global vegetation biophysical parameters from EUMETSAT Polar System. ISPRS J. Photogramm. Remote Sens. 2018, 139, 57–74. [Google Scholar] [CrossRef]
  22. Jacquemound, S.; Bidel, L.; Francois, C.; Pavan, G. ANGERS Leaf Optical Properties Database (2003). Data Set. Available online: http://ecosis.org (accessed on 5 June 2018).
  23. Gitelson, A.A.; Merzlyak, M.N. Remote sensing of chlorophyll concentration in higher plant leaves. Adv. Space Res. 1998, 22, 689–692. [Google Scholar] [CrossRef]
  24. Gitelson, A.A.; Buschmann, C.; Lichtenthaler, H.K. Leaf chlorophyll fluorescence corrected for re-absorption by means of absorption and reflectance measurements. J. Plant Physiol. 1998, 152, 283–296. [Google Scholar] [CrossRef]
  25. Hosgood, B.; Jacquemoud, S.; Andreoli, G.; Verdebout, J.; Pedrini, G.; Schmuck, G. Leaf Optical Properties Experiment 93 (LOPEX93); European Commission—Joint Research Centre: Ispra, Italy, 1994; p. 20. Available online: https://data.ecosis.org/dataset/13aef0ce-dd6f-4b35-91d9-28932e506c41/resource/4029b5d3-2b84-46e3-8fd8-c801d86cf6f1/download/leaf-optical-properties-experiment-93-lopex93.pdf (accessed on 5 June 2018).
  26. Feret, J.B.; François, C.; Asner, G.P.; Gitelson, A.A.; Martin, R.E.; Bidel, L.P.R.; Ustin, S.L.; Le Maire, G.; Jacquemoud, S. PROSPECT-4 and 5: Advances in the leaf optical properties model separating photosynthetic pigments. Remote Sens. Environ. 2008, 112, 3030–3043. [Google Scholar] [CrossRef]
  27. Combal, B.; Baret, F.; Weiss, M.; Trubuil, A.; Mace, D.; Pragnere, A.; Myneni, R.B.; Knyazikhin, Y.; Wang, L. Retrieval of canopy biophysical variables from bidirectional reflectance using prior information to solve the ill-posed inverse problem. Remote Sens. Environ. 2002, 84, 1–15. [Google Scholar] [CrossRef]
  28. Kattge, J.; Díaz, S.; Lavorel, S.; Prentice, I.C.; Leadley, P.; Bönisch, G.; Garnier, E.; Westoby, M.; Reich, P.B.; Wright, I.J.; et al. TRY—A global database of plant traits. Glob. Chang. Biol. 2011, 17, 2905–2935. [Google Scholar] [CrossRef] [Green Version]
  29. Wulder, M.A.; Coops, N.C. Make Earth observations open access: Freely available satellite imagery will improve science and environmental-monitoring products. Nature 2014, 513, 30–32. [Google Scholar] [CrossRef] [PubMed]
  30. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. [Google Scholar] [CrossRef]
  31. Robinson, N.P.; Allread, B.W.; Jones, M.O.; Moreno, A.; Kimball, J.S.; Naugle, D.E.; Erickson, T.A.; Richardson, A.D. A dynamic Landsat derived Normalized Difference Vegetation Index (NDVI) product for the Conterminous United States. Remote Sens. 2017, 9, 863. [Google Scholar] [CrossRef]
  32. Attermeyer, K.; Flury, S.; Jayakumar, R.; Fiener, P.; Steger, K.; Arya, V.; Wilken, F.; Van Geldern, R.; Premke, K. Invasive floating macrophytes reduce greenhouse gas emissions from a small tropical lake. Sci. Rep. 2016, 6, 20424. [Google Scholar] [CrossRef] [PubMed]
  33. Yu, M.; Gao, Q.; Gao, C.; Wang, C. Extent of night warming and spatially heterogeneous cloudiness differentiate temporal trend of greenness in mountainous tropics in the new century. Sci. Rep. 2017, 7, 41256. [Google Scholar] [CrossRef] [PubMed]
  34. He, M.; Kimball, J.S.; Maneta, M.P.; Maxwell, B.D.; Moreno, A.; Begueria, S.; Wu, X. Regional Crop Gross Primary Productivity and Yield Estimation Using Fused Landsat-MODIS Data. Remote Sens. 2018, 10, 372. [Google Scholar] [CrossRef]
  35. Kraaijenbrink, P.D.A.; Bierkens, M.F.P.; Lutz, A.F.; Immerzeel, W.W. Impact of a global temperature rise of 1.5 degrees Celsius on Asia’s glaciers. Nature 2017, 549, 257–260. [Google Scholar] [CrossRef] [PubMed]
  36. Reichstein, M.; Bahn, M.; Mahecha, M.D.; Kattge, J.; Baldocchi, D.D. Linking plant and ecosystem functional biogeography. Proc. Natl. Acad. Sci. USA 2014, 111, 13697–13702. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  37. Van Bodegom, P.M.; Douma, J.C.; Verheijen, L.M. A fully traits-based approach to modeling global vegetation distribution. Proc. Natl. Acad. Sci. USA 2014, 111, 13733–13738. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  38. Madani, N.; Kimball, J.S.; Ballantyne, A.P.; Affleck, D.L.R.; Bodegom, P.M.; Reich, P.B.; Kattge, J.; Sala, A.; Nazeri, M.; Jones, M.; et al. Future global productivity will be affected by plant trait response to climate. Sci. Rep. 2018, 8, 2870. [Google Scholar] [CrossRef] [PubMed]
  39. Wirth, C.; Lichstein, J.W. The Imprint of Species Turnover on Old-Growth Forest Carbon Balances-Insights From a Trait-Based Model of Forest Dynamics. In Old-Growth Forests; Wirth, C., Heimann, M., Gleixner, G., Eds.; Springer: Jena, Germany, 2009; pp. 81–113. ISBN 978-3-540-92705-1. [Google Scholar]
  40. Ziehn, T.; Kattge, J.; Knorr, W.; Scholze, M. Improving the predictability of global CO2 assimilation rates under climate change. Geophys. Res. Lett. 2011, 38, 10. [Google Scholar] [CrossRef]
  41. Berger, K.; Atzberger, C.; Danner, M.; D’Urso, G.; Mauser, W.; Vuolo, F.; Hank, T. Evaluation of the PROSAIL Model Capabilities for Future Hyperspectral Model Environments: A Review Study. Remote Sens. 2018, 10, 85. [Google Scholar] [CrossRef]
  42. Bacour, C.; Baret, F.; Béal, D.; Weiss, M.; Pavageau, K. Neural network estimation of LAI, fAPAR, fCover and LAIxCab, from top of canopy MERIS reflectance data: Principles and validation. Remote Sens. Environ. 2014, 105, 313–325. [Google Scholar] [CrossRef]
  43. Si, Y.; Schlerf, M.; Zurita-Milla, R.; Skidmore, A.; Wang, T. Mapping spatio-temporal variation of grassland quantity and quality using MERIS data and the PROSAIL model. Remote Sens. Environ. 2012, 121, 415–425. [Google Scholar] [CrossRef]
  44. Friedl, M.A.; Sulla-Menashe, D.; Tan, B.; Schneider, A.; Ramankutty, N.; Sibley, A.; Huang, X. MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sens. Environ. 2010, 114, 168–182. [Google Scholar] [CrossRef]
  45. Parzen, E. On estimation of a probability density function and mode. Ann. Math. Stat. 1962, 33, 1065–1076. [Google Scholar] [CrossRef]
  46. Reich, P.B. The world-wide ‘fast–slow’ plant economics spectrum: A traits manifesto. J. Ecol. 2014, 102, 275–301. [Google Scholar] [CrossRef]
  47. Nelsen, R.B. An Introduction to Copulass, 2nd ed.; Springer Science & Business Media: New York, NY, USA, 2009; ISBN 978-0387-28659-4. [Google Scholar]
  48. Žežula, I. On multivariate Gaussian copulas. J. Stat. Plan. Inference 2009, 111, 3942–3946. [Google Scholar] [CrossRef]
  49. Jacquemoud, S.; Baret, F. PROSPECT: A model of leaf optical properties spectra. Remote Sens. Environ. 1990, 34, 75–191. [Google Scholar] [CrossRef]
  50. Verhoef, W. Light scattering by leaf layers with application to canopy reflectance modeling: The SAIL model. Remote Sens. Environ. 1984, 16, 125–141. [Google Scholar] [CrossRef] [Green Version]
  51. Breiman, L.; Friedman, J.H. Estimating Optimal Transformations for Multiple Regression and Correlation. J. Am. Stat. Assoc. 1985, 391, 1580–1598. [Google Scholar]
  52. Evans, J.S.; Cushman, S.A. Gradient modeling of conifer species using random forests. Landsc. Ecol. 2009, 24, 673–683. [Google Scholar] [CrossRef]
  53. Oliveira, S.; Oehler, F.; San-Miguel-Ayanz, J.; Camia, A.; Pereira, J.M.C. Modeling spatial patterns of fire occurrence in Mediterranean Europe using Multiple Regression and Random Forest. For. Ecol. Manag. 2012, 275, 117–129. [Google Scholar] [CrossRef]
  54. Gislason, P.O.; Benediktsson, J.A.; Sveinsson, J.R. Random Forests for land cover classification. Pattern Recognit. Lett. 2006, 27, 294–300. [Google Scholar] [CrossRef]
  55. Cutler, D.R.; Edwards, T.C.; Beard, K.H.; Cutler, A.; Hess, K.T.; Gibson, J.; Lawler, J.J. Random Forests for classification in ecology. Ecology 2007, 88, 2783–2792. [Google Scholar] [CrossRef] [PubMed]
  56. Genuer, R.; Poggi, J.M.; Tuleau-Malot, C. Variable selection using random forests. Pattern Recognit. Lett. 2010, 31, 2225–2236. [Google Scholar] [CrossRef] [Green Version]
  57. Jung, M.; Zscheischler, J. A Guided Hybrid Genetic Algorithm for Feature Selection with Expensive Cost Functions. Procedia Comput. Sci. 2013, 18, 2337–2346. [Google Scholar] [CrossRef]
  58. Baret, F.; Morissette, J.T.; Fernandes, R.; Champeaux, J.L.; Myneni, R.B.; Chen, J.; Plummer, S.; Weiss, M.; Bacour, C.; Garrigues, S.; et al. Evaluation of the representativeness of networks of sites for the global validation and intercomparison of land biophysical products: proposition of the CEOS-BELMANIP. IEEE Trans. Geosci. Remote Sens. 2006, 44, 1794–1803. [Google Scholar] [CrossRef]
  59. Yan, K.; Park, T.; Yan, G.; Liu, Z.; Yang, B.; Chen, C.; Nemani, R.R.; Knyazikhin, Y.; Myneni, R.B. Evaluation of MODIS LAI/FPAR Product Collection 6. Part 2: Validation and Intercomparison. Remote Sens. 2016, 8, 460. [Google Scholar] [CrossRef]
  60. Camacho, F.; Cernicharo, J.; Lacaze, R.; Baret, F.; Weiss, M. GEOV1: LAI, FAPAR essential climate variables and FCOVER global time series capitalizing over existing products. Part 2: Validation and intercomparison with reference products. Remote Sens. Environ. 2013, 137, 310–329. [Google Scholar] [CrossRef]
  61. Nestola, E.; Sánchez-Zapero, J.; Latorre, C.; Mazzenga, F.; Matteucci, G.; Calfapietra, C.; Camacho, F. Validation of PROBA-V GEOV1 and MODIS C5 & C6 fAPAR Products in a Deciduous Beech Forest Site in Italy. Remote Sens. 2017, 9, 126. [Google Scholar]
  62. Baret, F.; Weiss, M.; Lacaze, R.; Camacho, F.; Makhmara, H.; Pacholcyzk, P.; Smets, B. GEOV1: LAI and FAPAR essential climate variables and FCOVER global time series capitalizing over existing products. Part 1: Principles of development and production. Remote Sens. Environ. 2013, 137, 299–309. [Google Scholar] [CrossRef]
  63. Campos-Taberner, M.; García-Haro, F.J.; Busetto, L.; Ranghetti, L.; Martínez, B.; Gilabert, M.A.; Camps-Valls, G.; Camacho, F.; Boschetti, M. A Critical Comparison of Remote Sensing Leaf Area Index Estimates over Rice-Cultivated Areas: From Sentinel-2 and Landsat-7/8 to MODIS, GEOV1 and EUMETSAT Polar System. Remote Sens. 2018, 10, 763. [Google Scholar] [CrossRef]
  64. Xu, B.; Park, T.; Yan, K.; Chen, C.; Zeng, Y.; Song, W.; Yin, G.; Li, J.; Liu, Q.; Knyazikhin, Y.; et al. Analysis of Global LAI/FPAR Products from VIIRS and MODIS Sensors for Spatio-Temporal Consistency and Uncertainty from 2012–2016. Forests 2018, 9, 73. [Google Scholar] [CrossRef]
Figure 1. Work flow of the proposed retrieval chain over GEE.
Figure 1. Work flow of the proposed retrieval chain over GEE.
Remotesensing 10 01167 g001
Figure 2. Constrained random samples of leaf chlorophyll (C a b ), leaf dry matter (C d m ), and leaf water (C w ) contents based upon prior knowledge of the TRY database, the MODIS land cover (MCD12Q1), kernel density estimators, and copulas.
Figure 2. Constrained random samples of leaf chlorophyll (C a b ), leaf dry matter (C d m ), and leaf water (C w ) contents based upon prior knowledge of the TRY database, the MODIS land cover (MCD12Q1), kernel density estimators, and copulas.
Remotesensing 10 01167 g002
Figure 3. Theoretical performance of the Random forest regression over PROSAIL simulations of LAI, FAPAR, FVC and CWC. The colorbar indicates density of points in the scatter plots.
Figure 3. Theoretical performance of the Random forest regression over PROSAIL simulations of LAI, FAPAR, FVC and CWC. The colorbar indicates density of points in the scatter plots.
Remotesensing 10 01167 g003
Figure 4. LAI, FAPAR, FVC, and CWC global maps and latitudinal transects corresponding to the mean values estimated by the proposed retrieval chain for the period 2010–2015.
Figure 4. LAI, FAPAR, FVC, and CWC global maps and latitudinal transects corresponding to the mean values estimated by the proposed retrieval chain for the period 2010–2015.
Remotesensing 10 01167 g004
Figure 5. LAI and FAPAR global maps and latitudinal transects corresponding to the mean values of the GEE MODIS reference product (MOD15A3H) for the period 2010–2015.
Figure 5. LAI and FAPAR global maps and latitudinal transects corresponding to the mean values of the GEE MODIS reference product (MOD15A3H) for the period 2010–2015.
Remotesensing 10 01167 g005
Figure 6. Sites location of the BELMANIP-2.1 network used for intercomparison of LAI and FAPAR retrievals and MOD15A3H LAI/FAPAR product.
Figure 6. Sites location of the BELMANIP-2.1 network used for intercomparison of LAI and FAPAR retrievals and MOD15A3H LAI/FAPAR product.
Remotesensing 10 01167 g006
Figure 7. Biome-dependent scatter plots of the retrieved LAI over BELMANIP2.1 sites for the period 2002–2017. The colorbar indicates density of points in the scatter plots.
Figure 7. Biome-dependent scatter plots of the retrieved LAI over BELMANIP2.1 sites for the period 2002–2017. The colorbar indicates density of points in the scatter plots.
Remotesensing 10 01167 g007
Figure 8. Biome-dependent scatter plots of the retrieved FAPAR over BELMANIP2.1 sites for the period 2002–2017. The colorbar indicates density of points in the scatter plots.
Figure 8. Biome-dependent scatter plots of the retrieved FAPAR over BELMANIP2.1 sites for the period 2002–2017. The colorbar indicates density of points in the scatter plots.
Remotesensing 10 01167 g008
Figure 9. LAI and FAPAR global maps and latitudinal transects corresponding to the difference of mean values between derived estimates by the proposed retrieval chain and the GEE MODIS reference product for the period 2010–2015.
Figure 9. LAI and FAPAR global maps and latitudinal transects corresponding to the difference of mean values between derived estimates by the proposed retrieval chain and the GEE MODIS reference product for the period 2010–2015.
Remotesensing 10 01167 g009
Figure 10. LAI (left) and FAPAR (right) latitudinal profiles over Africa (longitude 22°E) corresponding to the mean values of the GEE MODIS reference product and estimated by the proposed retrieval chain for the period 2010–2015.
Figure 10. LAI (left) and FAPAR (right) latitudinal profiles over Africa (longitude 22°E) corresponding to the mean values of the GEE MODIS reference product and estimated by the proposed retrieval chain for the period 2010–2015.
Remotesensing 10 01167 g010
Table 1. Spectral specifications of the MODIS MCD43A4 product.
Table 1. Spectral specifications of the MODIS MCD43A4 product.
MCD43A4 BandWavelength (nm)
Band 1 (red)620–670
Band 2 (NIR)841–876
Band 3 (blue)459–479
Band 4 (green)545–565
Band 5 (SWIR-1)1230–1250
Band 6 (SWIR-2)1628–1652
Band 7 (MWIR)2105–2155
Table 2. Information about the TRY data used in this work.
Table 2. Information about the TRY data used in this work.
Trait NameNumber of SamplesNumber of Species
C a b 19,222941
C d m 69,78311,908
C w 32,0204802
Table 3. Distributions of the parameters within the PROSAIL RTM at leaf (PROSPECT-5B) and canopy (SAIL) levels. * KDE refers to kernel density estimation method, which does not provide any parameters being a non parametric model of the marginal distributions.
Table 3. Distributions of the parameters within the PROSAIL RTM at leaf (PROSPECT-5B) and canopy (SAIL) levels. * KDE refers to kernel density estimation method, which does not provide any parameters being a non parametric model of the marginal distributions.
ParameterMinMaxModeStdType
LeafN1.22.21.60.3Gaussian
C a b ( μ g·cm 2 )----KDE *
C a r ( μ g·cm 2 )0.61657Gaussian
C d m (g·cm 2 )----KDE *
C w ----KDE *
C b p 0000-
CanopyLAI (m 2 /m 2 )083.54Gaussian
ALA (°)35806012Gaussian
Hotspot0.10.50.20.2Gaussian
vCover0.310.990.2Truncated Gaussian
Soil β s 0.110.80.6Gaussian

Share and Cite

MDPI and ACS Style

Campos-Taberner, M.; Moreno-Martínez, Á.; García-Haro, F.J.; Camps-Valls, G.; Robinson, N.P.; Kattge, J.; Running, S.W. Global Estimation of Biophysical Variables from Google Earth Engine Platform. Remote Sens. 2018, 10, 1167. https://doi.org/10.3390/rs10081167

AMA Style

Campos-Taberner M, Moreno-Martínez Á, García-Haro FJ, Camps-Valls G, Robinson NP, Kattge J, Running SW. Global Estimation of Biophysical Variables from Google Earth Engine Platform. Remote Sensing. 2018; 10(8):1167. https://doi.org/10.3390/rs10081167

Chicago/Turabian Style

Campos-Taberner, Manuel, Álvaro Moreno-Martínez, Francisco Javier García-Haro, Gustau Camps-Valls, Nathaniel P. Robinson, Jens Kattge, and Steven W. Running. 2018. "Global Estimation of Biophysical Variables from Google Earth Engine Platform" Remote Sensing 10, no. 8: 1167. https://doi.org/10.3390/rs10081167

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop