Elsevier

Geoderma

Volume 404, 15 December 2021, 115387
Geoderma

Estimating soil organic carbon of sown biodiverse permanent pastures in Portugal using near infrared spectral data and artificial neural networks

https://doi.org/10.1016/j.geoderma.2021.115387Get rights and content

Highlights

  • Soil organic matter is an important indicator to assess soil quality.

  • SOM content was estimated for Portuguese sown biodiverse pastures.

  • A combined approach of spectral data with artificial neural networks was used.

  • Cross-validation was performed with an 8-fold leave-one-out approach.

  • ANN model was able to estimate SOM contents with low error.

Abstract

Grasslands in Portugal are key managed ecosystems, supporting and providing a diverse number of ecosystem services. Here, we developed a procedure for rapid estimation of soil organic carbon (SOC) in soil samples of sown biodiverse permanent pastures rich in legumes (SBP) in Portugal. We combined laboratory NIR spectral data analysis with artificial neural networks (ANN) to estimate the SOC content of SBP soil samples. To train and test the ANN, we used more than 340 soil samples collected in the 0–20 cm topsoil layer from three farms in 2018 and 2019 and two other farms in 2019 only. The number of bands of the spectra (800–2778 nm) was reduced using two different approaches: (a) aggregation to Sentinel-2 (S2) bands using the average reflectance within each bandwidths; and (b) principal component analysis (PCA). For the S2 approach, we considered the six S2 bands that overlap with the spectral range of the instrument used. For the PCA approach, we considered the five first principal components. Additional covariates were used for prediction, including weather and terrain attributes, e.g. accumulated precipitation, average temperature, elevation, and slope. To test for transferability of the models to different farms, we used an eight-fold leave-one-out cross-validation approach to calculate estimation errors. Each fold is a unique combination of farm and year and is used to assess the model's performance calibrated from the seven other folds. The ANN was able to estimate both low and high SOC contents without systematic errors and with similar estimation errors for both full and reduced spectral data approaches. The average root mean squared error (RMSE) for the S2 approach was 1.95 g kg−1 (0.45 – 2.33 g kg−1 depending on the hold-out fold) and for the PCA approach was 1.81 g kg−1 (0.74 – 2.42 g kg−1) (compared to the average SOC content of 12 g kg−1). These RMSE values were similar to the RMSE obtained using the full spectra, suggesting that the original spectral resolution could be reduced without losing information. These results suggest the potential for using remotely sensed data to estimate the variation of SOC content for SBP. They are a first step towards developing algorithms that can alleviate the cost and time of soil sampling and chemical SOC laboratory analysis through indirect estimation.

Introduction

Grasslands are one of the most important terrestrial ecosystems for humans due to their contribution to food systems through the production of feed for grazing animals (Naidoo et al., 2008, Sloat et al., 2018). They also play a relevant role in providing ecosystem services (Egoh et al., 2016) and in some cases, they can be biodiversity hotspots (Davidson et al., 2012, Isbell et al., 2011, Soliveres et al., 2016). Nevertheless, inadequately managed grasslands or grasslands converted from natural land use (e.g., tropical forests) can also be responsible for environmental degradation, e.g. loss of climate regulation, biotic production and biodiversity (Abreu et al., 2017, Parton et al., 1993).

An example of enhancing the positive effects of grasslands, developed in Portugal since the 1960 s, is the system of sown biodiverse permanent pastures rich in legumes (SBP). SBP improve grass yield and allow increasing animal stocking rates in semi-natural pastures (Teixeira et al., 2011, Teixeira et al., 2008a, Valada et al., 2012). Their installation requires sowing a mixture of up to 20 species or cultivars of legumes and grasses that provide quality animal feed (Morais et al., 2018a, Teixeira, 2010, Teixeira et al., 2015, 2018). Currently, more than 90% of the total Portuguese SBP areas are located in the agricultural region of the Alentejo, mostly in areas of “montado/dehesa” agro-silvo pastoral ecosystems (Pereira et al., 2009, Teixeira et al., 2015). Nevertheless, those SBP areas represent less than 10% of the permanent grassland area in this region.

High grass yields in SBP lead to a substantial input of fresh soil organic carbon (SOC) (Teixeira et al., 2011), associated to CO2 removal from the atmosphere (Morais et al., 2018b, Teixeira et al., 2019) and soil carbon sequestration. Between 2008 and 2014, the Portuguese Carbon Fund (PCF – funded by the Portuguese Government), through the Terraprima – PCF project, supported the installation and maintenance of SBP using a system of payments proportional to carbon sequestration to assist Portugal in complying with the Kyoto Protocol goals under the Agriculture, Forestry and Other Land Uses activities (APA, 2018, Martins et al., 2015). More than 1,000 farmers established SBP in an area of more than 4% of the country’s agricultural land due to this project (Teixeira et al., 2015). However, it was not possible to monitor the real C sequestration in all these farms since conventional laboratory methods for SOC assessment were too costly and time-consuming. Farmers were paid on the basis of pre-determined sequestration factors established from data obtained in past studies (Teixeira et al., 2011) and not based on the carbon content increase measured on the farm. Indirect monitoring of a combined set of management practices was carried out to ensure that the sequestration factors were applicable. All farmers complying with the set practices received an equal payment. However, C sequestration, like most costs and benefits of grasslands, is highly site-specific and depends on variables such as soil properties, vegetation composition, climate, presence or absence of tree cover and grazing management practices (van Oijen et al., 2018). Therefore, the management of these ecosystems towards maximizing C sequestration requires better knowledge of grassland properties. Future initiatives to promote C sequestration, including subsidies based on the amount of C sequestered, should consider cost-effective methods for SOC quantification to survey large areas more effectively and differentiate performance between farmers.

Among cost-effective methods, visible (400–700 nm), near-infrared reflectance (NIR – 700–1300 nm) and shortwave-infrared reflectance (SWIR – 1300–2500 nm) spectroscopy has been used to indirectly estimate soil properties as an alternative to laboratory methods (Gomez et al., 2008, Srivastava et al., 2015). Various soil characteristics, e.g. SOC, clay content and soil moisture (Kuang et al., 2012, Stenberg et al., 2010), are well correlated with visible-NIR (VNIR), in particular with the NIR wavelengths (Qiao et al., 2017, Stevens et al., 2013).

Several methods can be used to correlate spectral data with SOC content, partial least-squares regression (PLSR) being the most widely used (Gogé et al., 2012, Gomez et al., 2008, Mouazen et al., 2010, Serrano et al., 2021, Shi et al., 2014). Recently, a wide range of alternative machine learning methods have been increasingly applied (Baligh et al., 2020, Fernandes et al., 2019, Morellos et al., 2016, Mouazen et al., 2010), artificial neural networks (ANN) being one of the most frequently used (Kuang et al., 2015). Advantages of ANN are that they can (i) estimate complex, high-dimensional and nonlinear relationships (Liu et al., 2016); and (ii) use both quantitative and qualitative data as inputs (Yang et al., 2018). Nevertheless, the use of these models for land use systems monitoring is still underexplored (Brown et al., 2006). ANN are increasingly used to estimate SOC content. Using Scopus® to search in the titles and abstract of published studies and with the following search string: “artificial neural network” AND (“soil organic matter” OR “soil organic carbon”)), we found 115 hits. In these, more than 75% were published since 2016.

The present study aimed to assist with expedited and efficient monitoring of SBP through the combination of spectral data with machine learning methods, namely ANN. For this, an ANN algorithm was calibrated for the estimation of SOC (at 20 cm depth) in soil samples from SBP soils using 800–2800 nm (laboratory) spectral data as an input. This work is intended as a first step in automating SOC estimation using progressively more remote sources of data. Testing the decreased spectral resolution through the use of Sentinel-2 (S2) aggregation enables us to prepare for future use of remotely sensed spectral data, as we explored how accurately data collected in those bands can estimate SOC for SBP. Here, we provide an ANN-based algorithm for estimating SOC spectroscopically and paves the way for future development of algorithms based on S2 that enable remote estimation of SOC. The calibration used conventional (laboratory) SOC analyses from the same soil samples. We tested two data reduction methods: (i) principal component analysis (PCA), and (ii) aggregation of spectral data based on S2 bandwidths. In both instances, we used an 8-fold leave-one-location-and-year-out approach to estimate the generalisation capacity of the model.

Section snippets

Study area and soil sampling design

This work used soil samples collected in five farms in south and central Portugal (farms 1, 2, 4 and 5 in the Alentejo and farm 3 in Beira Interior), across latitudes and longitudes ranging respectively between 38°10′ and 40°30′N and 7°40′ and 8°30′ W (Fig. 1). Farms 1 and 5 are located near Évora, farm 2 near Cabeço de Vide, farm 3 near Covilhã, and farm 4 near Grândola. All farms are in the hot-summer Mediterranean climate region, according to the Köppen climate classification system (IPMA,

Laboratory analysis of soil organic carbon

The main trend between folds that can be observed in the values of SOC determined with analytical techniques (combination of farm and year) is that, in general, SOC variation within the same fold tends to be higher than that between folds (Fig. 3). For each of the two years, the differences in SOC between farms are small. For example, Farm 3–2019, Farm 4–2019 and Farm 5–2019 have similar SOC. In Farm 3–2019 the median is 7.48 g kg−1 and interquartile range is 3.16 g kg−1, in Farm 4–2019 the

Discussion

In this paper, we have combined spectral images with ANN for measuring SOC in SBP systems in Portugal. We have shown that two different spectral dimensionality reduction approaches performed as well as the raw spectra (Table 4). PCA is a dimensionality reduction based on the statistical correlation between different wavelengths that uses the actual measurements, and we therefore expected it to perform better than S2. Further, the considered first five PC explains more than 99% of the variance

Conclusion

We provide an algorithm for estimating SOC in Portuguese SBP using spectroscopy. We concluded that spectroscopical methods are a viable solution for expediting SOC monitoring. The method applied laboratory analysis using high-resolution NIR spectral data. Moreover, it provided evidence that those spectroscopical methods can be replaced by different data products or data collection methods with lower spectral resolution. A comparison of ANN modelling results between two reduced spectra methods

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.

Acknowledgments

This work was supported by Programa de Desenvolvimento Rural (PDR2020) through “Viabilização de pastagens semeadas biodiversas através da otimização da fertilização fosfatada” (PDR2020-101-030690) and “GO SOLO: Avaliação da dinâmica da matéria orgânica em solos de pastagens semeadas biodiversas através do desenvolvimento de um método de monitorização expedito e a baixo custo” (PDR2020-101-031243). The work was also supported by Fundação para a Ciência e Tecnologia through projects “Animal

References (72)

  • A.M. Mouazen et al.

    Comparison among principal component, partial least squares and back propagation neural network analyses for accuracy of measurement of selected soil properties with visible and near infrared spectroscopy

    Geoderma

    (2010)
  • D. Rogge et al.

    Building an exposed soil composite processor (SCMaP) for mapping spatial and temporal characteristics of soils with Landsat imagery (1984–2014)

    Remote Sens. Environ.

    (2018)
  • R.F.M. Teixeira et al.

    Soil organic matter dynamics in Portuguese natural and sown rainfed grasslands

    Ecol. Modell.

    (2011)
  • R.F.M. Teixeira et al.

    A conceptual framework for the analysis of engineered biodiverse pastures

    Ecol. Eng.

    (2015)
  • S.Z. Wang et al.

    A multi-factor weighted regression approach for estimating the spatial distribution of soil organic carbon in grasslands

    Catena

    (2019)
  • M. Wiesmeier et al.

    Soil organic carbon storage as a key function of soils - A review of drivers and indicators at various scales

    Geoderma

    (2019)
  • S. Yang et al.

    Modeling grassland above-ground biomass based on artificial neural network and remote sensing in the Three-River Headwaters Region

    Remote Sens. Environ.

    (2018)
  • Hui Yu et al.

    A method to avoid spatial overfitting in estimation of grassland above-ground biomass on the Tibetan Plateau

    Ecol. Indic.

    (2021)
  • R.C.R. Abreu et al.

    The biodiversity cost of carbon sequestration in tropical savanna

    Sci. Adv.

    (2017)
  • I. Ali et al.

    Modeling Managed Grassland Biomass Estimation by Using Multitemporal Remote Sensing Data-A Machine Learning Approach. IEEE J

    Sel. Top. Appl. Earth Obs. Remote Sens.

    (2017)
  • APA

    Portuguese National Inventory Report on Greenhouse Gases, 1990–2018

    (2018)
  • S. Ayoubi et al.
  • P. Baligh et al.

    Predicting Soil Particulate Organic Mattter Using Artificial Neural Network with Wavelet Function

    Commun. Soil Sci. Plant Anal.

    (2020)
  • C.M. Bishop

    Neural networks for pattern recognition

    (1995)
  • Carneiro, J.P., Freixial, R.C., Pereira, J.S., Campos, A.C., Crespo, J.P., Carneiro, R., 2005. Relatório Final do...
  • R.N. Clark et al.

    High spectral resolution reflectance spectroscopy of minerals

    J. Geophys. Res.

    (1990)
  • R.C. Cornes et al.

    An Ensemble Version of the E-OBS Temperature and Precipitation Data Sets

    J. Geophys. Res. Atmos.

    (2018)
  • A.D. Davidson et al.

    Ecological roles and conservation challenges of social, burrowing, herbivorous mammals in the world’s grasslands

    Front. Ecol. Environ.

    (2012)
  • J.A.M. Demattê et al.

    Morphological Interpretation of Reflectance Spectrum (MIRS) using libraries looking towards soil classification

    Sci. Agric.

    (2014)
  • J.A.M. Demattê et al.

    Bare Earth’s Surface Spectra as a Proxy for Soil Resource Monitoring

    Sci. Rep.

    (2020)
  • H. Demuth et al.

    Neural Network Toolbox For Use with MATLAB

    Math Work

    (2013)
  • S. Diek et al.

    Barest Pixel Composite for Agricultural Areas Using Landsat Time Series

    Remote Sens.

    (2017)
  • EC, 2003. European Soil Database (distribution version v2.0). European Commission Joint Research...
  • Egoh, B.N., Bengtsson, J., Lindborg, R., Bullock, J.M., Dixon, A.P., Rouget, M., Bengtsson, J., Lindborg, R., Bullock,...
  • ESA, 2015. Sentinel-2 User Handbook, Sentinel-2 User Handbook....
  • Gedminas, L., Martin, S., 2019. Soil Organic Matter Mapping Using Hyperspectral Imagery and Elevation Data, in: IEEE...
  • Cited by (0)

    View full text