15arspc submission 44

8/8/2019 15arspc Submission 44

http://slidepdf.com/reader/full/15arspc-submission-44 1/9

ESTIMATION OF PASTURE BIOMASS AND SOIL-MOISTURE

USING DUAL-POLARIMETRIC X AND L BAND SAR –ACCURACY ASSESMENT WITH FIELD DATA

Tishampati Dhar [1],Carl Menges [1],John Douglas [1],Michael Schmidt [2],JohnArmston [2]

Author affiliation:[1]Apogee Imaging International12B, 1 Adelaide-Lobethal Rd.

Lobethal, SA 5241Australia+61-8-8389-5499,+61-8-8389-5488

[email protected][2]Remote Sensing Centre, Queensland

Department of Environment and Resource Management80 Meiers Road,

Indooroopilly, Queensland, 4068

Abstract

This paper presents the results of a study conducted to relate X and L bandpolarimetric SAR backscatter to pasture soil moisture and biomass as part of an environmental monitoring program. Extensive field data was collectedconcurrently with satellite SAR data acquisition – including dry/wet aboveground biomass, soil moisture, surface roughness profiles and EM-38electromagnetic sensor data. This data is used for both electromagneticmodelling of the surface to work out the theoretical backscatter as well asempirical fitting regression models to the recorded SAR data and validation of existing inversion models.

Introduction

The work in this paper was carried out in conjunction with QueenslandDepartment of Environmental and Resource Management. The department hasworked in the past with dual-polarimetric ALOS-PALSAR L-band to ascertainlandcover changes[1].

This study utilised a TerraSAR-X HH/VV dual-polarimetric stripmap image inconjunction with ALOS-PALSAR HH/HV and fieldwork. The aim is to ascertain

the utility of SAR for rangeland Feed-On-Offer (FOO) estimation.

1



The fieldwork collected provided information regarding the soil moisture androughness as well as live and dead biomass cover.

The field work was performed from 2009-09-07 to 2009-09-15 in central parts of Queensland, between Aramac and Barcaldine in the Mitchell Grass Downs,specifically the 33 field sites shown in the map in Figure 1.

Analysis Techniques

The study performed dual-polarimetric decomposition on the ALOS-PALSARdata and compared field soil moisture measurements with AMSR-E passive L-band radiance and ALOS-PALSAR backscatter. The dual-polarimetric entropy-

alpha decomposition was used to isolate areas in the SAR with low-entropy andalpha angles indicating surface scattering as the dominant backscatter mechanism.

The backscatter values at the survey points in both PALSAR and TerraSAR-Xare compared at the survey points and over distributed scatterers as indicatedby landcover maps.

Field work details

33 sites were surveyed to quantify the following observables:

2

Figure 1: Field site locations with respect to PALSAR and TSXscenes



1.Cover – measured as percentages in various types

2.Soil Moisture – measured using in-situ probes.

3.Soil Roughness - pinmeter is used to measure the soil surface roughness at

30mm spacing.4.Wet and Dry Biomass – measured using samples and recorded together witha visual estimate.

5.Electromagnetic Conductivity - using an EM38 Sensor.

These parameters adequately represent a 2 layer target model with shortvegetation on top of a slightly conductive rough soil medium. The samplingcaptures spatial variability of the soil moisture and roughness. Areas wherelarger vegetation such as trees are present is noted and points near power lineswhere the EM-38 sensor is adversely affected by other magnetic fields thereadings from this sensor are rejected.

Sampling Strategy

EM38 and Soil moisture measurements are performed in a pattern around acentre point as in Figure 2 in a 40m radius circle to the outermost points and25m radius circle to the inner group of points.

Soil Surface Roughness Parameterisation

A number of soil surface profiles were collected as illustrated below at the studysites using a pin-meter with 30mm spacing between the centres of the rod asillustrated below in Figure 2. The soil profile is converted to Gaussian standarddeviation and correlation length representation for use with established soilbackscatter simulation models which represent the surface as a randomGaussian with zero mean and exponential correlation[2].

3

Figure 2: Pin-Profiler used for estimating soil surface roughness



Variable Independence analysis

The set of field variables selected for inclusion in the study are examined for

linear independence using correlation statistics. It is established that the wetand dry biomass is linearly related hence all vegetation contains 9.6% moistureon average.

The surface roughness and EM38 readings are weakly correlated a few other parameters have correlation above 0.5, but all the values are unrelated andsuitable for use in multiple regression (Table 1).

Satellite data collection

ALOS-PALSAR and TerraSAR-X dual polarimetric data was concurrentlycollected. TerraSAR-X acquisition was on 2009-09-14 while the PALSAR

acquisition was on 2009-9-10. Both dates fall within the field work window.Variables which change rapidly such as soil moisture are expected to have ahigh degree of error between satellite and field observations, but due to thegenerally dry weather at this time the variation is not expected to be significant.

TerraSAR-X was collected in stripmap mode with 38.73° incidence angle atscene center. The TerraSAR-X Scenes have a swath width of 15km and alength of 50km.

ALOS-PALSAR was collected in Fine Beam Dual (FBD) – equivalent toTerraSAR-X dual-polarimetric stripmap mode with an incidence angle of 38.93°.PALSAR scenes have a nominal size of 70x70 km.

4

Biomass Wet 1 0.28 0.33 0 0.08 0.37 0.21Bare Soil 0.28 1 0.13 0.13 0.52 0.53 0.34

0.33 0.13 1 0.53 0.1 0.05 0.01

0 0.13 0.53 1 0.09 0.02 0.25

0.08 0.52 0.1 0.09 1 0.51 0.31Em38 mean 0.37 0.53 0.05 0.02 0.51 1 0.6Em38 std 0.21 0.34 0.01 0.25 0.31 0.6 1

Biomass

Wet

Bare

Soil

SoilMoisture

mean

SoilMoisture

std

SurfaceRoughness

std

Em38

mean

Em38

std

Soil MoisturemeanSoil MoisturestdSurfaceRoughnessstd

Table 1: Correlation Analysis of Measured Field variables



PALSAR and TerraSAR-X Data Analysis

The ALOS-PALSAR data is focused using the Gamma Interferometric SARprocessor and used in single-look complex form for further analysis. The data ismultilooked with 2 looks in range and 8 looks in azimuth to produce almostsquare pixels. This product (Figure 4) is then Lee filtered and processed withdual-interferometric techniques.

The HH/VV TerraSAR-X data (Figure 3) is suitable for both standard channel bychannel backscatter comparisons to observed field variables as well as dual-polarimetric techniques particularly analysis of the phase difference distributionand dual-polarimetric entropy-alpha decomposition.

Backscatter Analysis

The backscatter amplitude in the HH and HV channels of PALSAR is relatedempirically to the field values of soil surface rougness, soil moisture and wetbiomass measured. Ordinary least squares regression is performed at thisexploratory stage. Later model development and transformed explanatoryvariables will be used

TerraSAR-X is highly sensitive to small scale roughness and grassland

vegetation. TerraSAR-X has been successfully used to monitor rice crops in

5

Figure 4: HH/HV Backscatter Amplitude for ALOS-PALSAR

Figure 3: TerraSAR-X HH/VVCovariance Image



Spain[3] and wheat crops in Southern parts of Australia[4]. This makesTerraSAR-X a feasible option for pasture biomass measurement since pastureshave similar grass-like vegetation, though with a much lower biomass. In thiscase the pastures have very low moisture content and surface scattering forms

a significant part of the net backscatter.The R 2 regression statistic for individual explanatory variables is too lowindicating the vegetation as well as soil surface contribute to the netbackscatter. An exploratory linear regression was performed against allvariables. Multivariate regressions with more than one of the independentvariables provide better explanation of the scattering observed. The chosenindependent variables are:

• Vegetation Wet Biomass• Bare Soil Percentage

• Soil surface standard deviation• Soil moisture mean• Soil moisture standard deviation• Soil Conductivity mean• Soil Conductivity standard deviation

The standard deviations of the measured variables are included in theregression to account for the error in the measurements performed.

A multi-variate model is selected after statistical analysis and physicalscattering considerations[5]. From physical considerations the dominantexplanatory variables for scattering are the surface roughness and moisturecontent and the vegetation wet biomass. These factors are borne out by the R 2

analysis for a multivariate model selection. The R 2 values range from 0.6 to0.74 with 4 parameters. The discrepancy is expected due to the high spatialvariability observed in the soil moisture sampling and temporal differencesbetween the sampling and image acquisition.

The backscatter is affected by soil moisture and roughness and the principalcontribution of vegetation return is the moisture content. The soil parameters aswell as the backscatter values are used as independent variables in a linear

regression to determine moisture content in kg/ha.

6

Table 2: Regression coefficients for vegetation water content estimation

Variable Coeffi ci ent t-va lue

Intercept 585.08 4.17PALSAR HH 11222 0.9PALSAR HV -39318 -2.73TerraSAR-X HH -17425 -3.21TerraSAR-X VV 16045 2.8Soil Roughness -47.93 -1.41Soil Moisture 14.53 1.67



Polarimetric Decomposition

Dual-polarimetric Entropy/Alpha (H/a) decomposition is applied to the PALSARimage and the results classified according to the Cloude-Pottier scheme[6](Figure 5). Considering the shifts produced by dual-polarimetry[4] as opposedto full-polarimetry some of the pixels are reclassified. The results indicatesurface scattering in the areas shown in purple and volume scattering in theareas shown in blue.

The polarimetric classification results indicate heavily vegetated areas along theriver interspersed with pastures. When relating biomass observed in the field to

7

Figure 5: Linear Model fitted to estimate vegetation water content from SAR backscatter



the SAR backscatter the different scattering mechanisms (Figure 7 and 6) inthe different areas need to be accounted for.

Conclusions

This study performs a preliminary data exploration for a potential pasturebiomass estimation and investigates how different microwave frequencies cancomplement each other. The vegetation in this case had very low moisturecontent and pasture biomass cannot be ascertained with great certainty withoutknowledge of other affecting conditions such as soil surface roughness andmoisture content.

Future Work

Future work will explore electromagnetic modeling of the pastures at L and Xband leading to physically sound inversion techniques for dual-polarimetricmulti-band SAR imagery. The field variables will be used to test the theoreticalSAR simulation model provided by PolsarPRO for L-band bare ground andground + short vegetation. Fieldwork will need to be carried out at different

times of the season to assess the impact of vegetation and soil-moisturecontent on the total resultant backscatter.

8

Figure 6: Entropy/Alpha ClassificationResult for ALOS-PALSAR. Blue areas are

vegetated, Purple areas are bare

Figure 8: Dual-polarimetricentropy/alpha for PALSAR illustrating

surface scattering (entropy in the x-axis, mean-alpha in the y-axis, the

colour represents point density – blue islow density and red is high density)

Figure 7: Dual-polarimetricentropy/alpha for PALSAR illustrating

volume scatterin



References[1] Richard Lucas, John Armston, John Carreiras and Bunting Peter. RegionalCharacterisation and Mapping of Australian Forest Structural Types from ALOSPALSAR data. In Alos 2008 symposium . 2008.

[2] Oh Y, Sarabandi K & Ulaby F. An empirical model and an inversiontechnique for radar scattering from bare soil surfaces. Geoscience and RemoteSensing, IEEE Transactions on (1992) 30 : pp. 370-381.

[3] Lopez-Sanchez J, Ballester-Berman J & Hajnsek I. Rice monitoring in Spainby means of time series of TerraSAR-X Dual-pol images. TerraSAR-X ScienceTeam Meeting 2008 (2008)

[4] Dhar T, Gray D & Menges C. Agricultural performance monitoring with dual-polarimetric TerraSAR-X imagery. In Proceedings of iet china 2009 radar conference . 2009.

[5] Stiles J, Sarabandi K & Ulaby F. Electromagnetic scattering from grassland.II. Measurement and modeling results. IEEE Transactions on Geoscience and Remote Sensing (2000) 38 : pp. 349-356.

[6] Cloude S. The dual polarisation entropy/alpha decomposition: A palsar casestudy. European Space Agency, (Special Publication) ESA SP (2007) : p. 6 -

9

15arspc submission 44

Documents