Synthetic aperture radar and optical satellite data for estimating the biomass of corn

https://doi.org/10.1016/j.jag.2019.101933Get rights and content

Highlights

  • Corn biomass estimation from RADARSAT-2 and RapidEye demonstrated high accuracies for both satellites.

  • Using crop phenology for the corn biomass estimation significantly improved the accuracies.

  • The biomass estimates from RADARSAT-2 were significantly improved after adjusted using a neural network algorithm trained by biomass estimates from RapidEye.

  • An integrated SAR-optical approach for biomass monitoring is feasible.

Abstract

Above ground biomass is an important crop biophysical parameter for monitoring crop condition and determining crop productivity, in particular if linked with phenological growth stage. Optical reflectance and Synthetic Aperture Radar (SAR) backscatter have been used to model above ground biomass for some crops. However, to date, direct comparisons of biomass retrievals from these two data sources, and more importantly an integration of biophysical products from optical and SAR satellite data, has not been fully explored. In this study, SAR and optical satellite data are assessed and compared for estimating the wet and dry biomass of corn. Wet and dry biomass are estimated from RADARSAT-2 using a calibrated Water Cloud Model (WCM) and from RapidEye using vegetation indices which include the Normalized Difference Vegetation Index (NDVI), Red-Edge Triangular Vegetation Index (RTVI), Simple Ratio (SR) and Red-Edge Simple Ratio (SRre). Data collected during the SMAP Validation Experiment 2012 (SMAPVEX12) are used for testing the model accuracies. The results show high accuracies for both SAR and optical sensors for dry biomass with a correlation coefficient (R) of 0.83, Root Mean Square Error (RMSE) of 0.16 kgm-2 and Mean Absolute Error (MAE) of 0.15 kgm-2 from RADARSAT-2 and R of 0.92, RMSE of 0.11 kgm-2 and MAE of 0.07 kgm-2 from RapidEye. For wet biomass, the accuracies are 0.78 (R), 1.25 kgm-2 (RMSE) and 1.00 kgm-2 (MAE) from RADARSAT-2 and for RapidEye, 0.92 (R), 0.71 kgm-2 (RMSE) and 0.49 kgm-2 (MAE). For dry biomass, the accuracies from RapidEye are slightly higher but the results from these two sensors are comparable. For wet biomass, the accuracies from RADARSAT-2 decreased, likely due to the effect of Vegetation Water Content (VWC) on backscatter intensity. This study also developed a neural network transfer function between the biomass derived from SAR and from optical data. The purpose of this function is to facilitate the integration of optical and SAR derived biomass in order to monitor corn throughout the growing season, regardless of source of satellite data. Integration of data from multiple sources is an important strategy to improve temporal coverage, in particular when clouds obscure optical observations. With this approach in mind, biomass estimates from RADARSAT-2 were cross-calibrated against those from RapidEye and their time-series biomass maps were generated.

Introduction

Total above ground biomass is an important crop biophysical parameter which can be useful for monitoring crop condition and its productivity (Asrar et al., 1985; Lobell, 2013). Many studies have demonstrated that optical satellite data are able to estimate biomass of various crops (Günlü et al., 2014; Kross et al., 2015; Liao et al., 2018; Persson et al., 2013; Punalekar et al., 2018; Zheng et al., 2004). These studies often rely on band ratios to track crop development. These indices exploit the increasing differential in amount of red light absorbed by leaf chlorophyll and infrared light reflected due to cell structure, as leaves accumulate (Gates, 1980). In several studies, the Normalized Difference Vegetation Index (NDVI) was proven highly sensitive to biomass especially from the early to mid-growing season (Jin et al., 2014; Zhao et al., 2014). Other researchers have advocated for the use of the Simple Ratio (SR) (Stenberg et al., 2004) or indices with near-infrared red-edge bands (red-edge simple ratio (SRre) and Red-Edge Triangular Vegetation Index (RTVI) (Chen et al., 2010) which have produced higher correlations with crop biomass (Kross et al., 2015). However, cloud cover impedes visible-infrared wavelengths making it particularly challenging to rely exclusively on this class of satellites for monitoring rapidly developing crops. Conversely, longer microwaves used by Synthetic Aperture Radars (SARs) permit collection of data even during cloudy conditions. Recent studies have demonstrated that SAR sensors are also able to estimate biomass (Betbeder et al., 2016; Hamdan et al., 2011; Hosseini and McNairn, 2017; Ndikumana et al., 2018; Reisi-Gahrouei et al., 2019; Sarker et al., 2013). Ndikumana et al. (2018) used the backscatter from a single polarization (VH) to model the biomass of rice using Sentinel-1 C-band data. Hosseini and McNairn (2017) demonstrated the potential of multi-polarization C-band and L-band SAR data for biomass estimation of wheat. Reisi-Gahrouei et al. (2019) extended this analysis to report on the sensitivity of 14 L-band polarimetric parameters derived from multiple decomposition algorithms to the biomass of corn, soybean and canola. Despite these findings, studies that assess SAR for agricultural applications have been few compared to optical sensors. Access to SAR data has been more limited and processing of these data can be more challenging.

SAR scattering changes as crop structure develops, with soil moisture also impacting response. Therefore, efforts to model backscatter should include both soil and crop contributions. The semi-empirical Water Cloud Model (WCM) (Attema and Ulaby, 1978) parameterizes the biophysical properties of vegetation. This model has been used successfully to estimate properties of a number of crops, using both C- and L-band SAR data (Bériaux et al., 2011, 2013 and 2015; Svoray and Shoshany, 2002; Hosseini et al., 2015 and 2017).

Building upon these documented results, the objectives of the research presented here are twofold.

  • 1

    Acquire near coincident SAR and optical sensor data to compare biomass estimates of corn from optical wavelengths with those from C-Band SAR backscatter intensities. Derive optical-based estimates from vegetation indices, and parameterize the WCM using field measures of corn biomass and soil moisture, with accuracies established using independent biomass measures.

  • 2

    Cross-validate optical-based and SAR-based biomass estimates for corn. Assess whether a neural network transfer function could be developed to relate optical-based and SAR-based biomass estimates. The purpose of such a function is to evaluate whether optical and SAR satellite data could be used interchangeable for estimation of corn. Such an integration is likely to improve the temporal density of satellite data needed to effectively monitor crop productivity.

Section snippets

Study area

For this study, ground and satellite data were acquired in southern Manitoba, an important region of agricultural production for Canada (Fig. 1). The study area stretches 70 km by 20 km and is characterized as flat with slopes from 0 to 2%. This region has fine textured clay soils which overlay a flat topography and as such, there are annual concerns of flooding due to spring snowmelt. These risks, along with the importance of this region for crop production, have led to a number of intensive

Description of the water cloud model (WCM)

The WCM was selected to estimate wet and dry biomass of corn from the backscatter intensity measured by RADARSAT-2. The WCM models total backscatter as the sum of scattering from vegetation and scattering from the soil, attenuated by the vegetation. The general formulation of the WCM is as follows (Eqs. (1), (2), (3) (Attema and Ulaby, 1978).σ0=σveg0+τ2σsoil0σveg0=AV1cosθ1-exp-2BV2cosθτ2=exp(-2BV2/cosθ)where σ0 is total backscatter in power unit, σveg0 is direct backscatter from the vegetation,

Forward modeling of the WCM

The calibrated WCM model is to be used to estimate biomass for a validation dataset, data not used to parameterize the WCM. However, before validating model retrievals on unseen data, the performance of the parameterized model was assessed in a forward modeling exercise to test the goodness of fit of modeled backscatter (from the WCM) to measured backscatter (from RADARSAT-2). The forward modeling accuracies for the WCM model are shown in Fig. 3. These accuracies are also documented in Table 3.

Conclusions

Direct comparisons of optical and SAR models for crop biomass estimation is lacking in the literature. In this study, near coincident RADARSAT-2 and RapidEye data were used to estimate biomass for corn fields. The semi-empirical WCM model was used for estimation of biomass from RADARSAT-2, and optical models were developed using four vegetation indices specifically Normalized Difference Vegetation Index (NDVI), Red-Edge Triangular Vegetation Index (RTVI), Simple Ratio (SR) and Red-Edge Simple

Acknowledgement

This study was funded by the Canadian Space Agency Government Related Initiatives Program (GRIP) and Agriculture and Agri-Food Canada. The authors would like to thank the field and aircraft crews who participated in the SMAPVEX12 experiment.

References (40)

  • J.R. Adams et al.

    Evaluating the Cloude-Pottier and Freeman-Durden scattering decompositions for distinguishing between unharvested and post-harvest agricultural fields

    Can. J. Remote Sensing

    (2013)
  • E.P.W. Attema et al.

    Vegetation modelled as a water cloud

    Radio Sci.

    (1978)
  • E. Bériaux et al.

    Estimating surface-soil moisture for retrieving maize leaf-area index from SAR data

    Can. J. Remote Sensing

    (2011)
  • E. Bériaux et al.

    Multiyear independent validation of the water cloud model for retrieving maize leaf area index from SAR time series

    Int. J. Remote Sens.

    (2013)
  • E. Bériaux et al.

    Maize leaf area index retrieval from synthetic quad pol SAR time series using the water cloud model

    Remote Sens. (Basel)

    (2015)
  • J. Betbeder et al.

    Assimilation of LAI and dry biomass data from optical and SAR images into an agro-meteorological model to estimate soybean yield

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

    (2016)
  • P. Chen et al.

    New index for crop canopy fresh biomass estimation

    Spectrosc. Spectr. Anal.

    (2010)
  • Y. Duguay et al.

    Potential of C and X b and SAR for shrub growth monitoring in sub-arctic environments

    Remote Sens. (Basel)

    (2015)
  • David M. Gates

    Biophysical Ecology

    (1980)
  • A. Günlü et al.

    Estimating aboveground biomass using Landsat TM imagery: a case study of Anatolian Crimean pine forests in Turkey

    Ann. For. Res.

    (2014)
  • Cited by (0)

    View full text