ENVISAT ASAR POLARIMATRIC DATA FOR SOIL MOISTURE MAPPING

ENVISAT ASAR POLARIMATRIC DATA FOR SOIL MOISTURE MAPPING Y.S. Rao, Ashwani Kumar Singh, Sonika Sharma, G. Venkataraman Centre of Studies in Resources ...
Author: Lynn Williams
1 downloads 0 Views 4MB Size
ENVISAT ASAR POLARIMATRIC DATA FOR SOIL MOISTURE MAPPING Y.S. Rao, Ashwani Kumar Singh, Sonika Sharma, G. Venkataraman Centre of Studies in Resources Engineering, Indian Institute of Technology, Bombay, Mumbai-400076, India Email: - [email protected] ABSTRACT ENVISAT ASAR Data acquired over four test sites were analyzed for soil moisture mapping using various models. The polarimetric data covers dual polarized HH/VV, HH/HV and single polarized VV in swaths IS2, IS2, IS4, IS5 and IS6. SIR-C L- and C-band data were also used for the verification of models. Dubois et al. empirical and linear regression equations were used for soil moisture estimation. The test sites cover bare, rice, sugarcane, corn, etc. fields. Using the SIR-C data, we found that Dubois et al. model overestimates soil moisture at C-band compared to L-band. The difference is about 5%. The linear regressions equations developed by Baghdadi et al. predict soil moisture with reasonable accuracy for bare fields using ENVISAT ASAR data. However, these regression equations are site specific and do not take into account surface roughness and vegetation cover. More groundtruth data are needed for the verification of these relations. 1. INTRODUCTION Importance of soil moisture in several disciplines leads to the development of active and passive microwave sensor system for continuously mapping soil moisture at high and low spatial resolutions [1][2]. It has been demonstrated with SIR-C L-band data that soil moisture can be estimated with an error of 3 to 4% [3][4]. Until the launch of ALOS PALSAR in Jan 2006,, only data available through space based platform are ERS-1&2. Radarsat-1 and ENVISAT. Except ENVISAT ASAR, all others are operated/operating at single polarization. ENVISAT ASAR can acquire data at two polarizations and at different incidence angles. Several investigators [1][5-7]exploited the potential of ENVISAT ASAR data for soil moisture and derived relations with various combinations of ASAR polarizations and incidence angles. In our study, we used Dubois et al. semi-empirical model and linear regression relations derived by Baghdadi et al[1] for retrieving soil moisture using ENVISAT ASAR data. In the subsequent sections, test sites, data sets, models and results are discussed. 2. TEST SITES Test sites cover different parts of India. The test site at coastal Andhra Pradesh in the south-eastern part of India covers variety of agriculture fields. At the time of acquisition of scenes, most of the fields are bare-dry harvested rice crop. The other major crop is sugarcane. The river Krishna and canals from the river flow

through this test area. There are number of bore wells that extract underground water for irrigation. The annual rainfall in this area is about 1000 mm. The soil is black cotton soils and it texture varies from loamy to clay loam. Gurdaspur is a part of Punjab state with flat terrain and contains good network of canals. Soil is fine loamy and fertile. The average annual rainfall is 875 mm and major crops in the area are rice, wheat and sugarcane. Ahmedabad district in Gujarat and surrounding areas are taken for flood and soil moisture studies. The annual rainfall in this area is about 2000 mm during June – Sept. The maximum intensity of the rainfall is in July and August. Major crops in this area are rice, wheat and cotton. The soil texture is loamy to loamy clay. Mumbai is cosmopolitan city and most of the area covers with building. Other parts of city covers with vegetation (trees) and some patches are used for agriculture. Annual rainfall over this area is around 2170 mm. Most of the soil texture is loamy. 3. DATA SOURCES Coastal Andhra Pradesh : ENVISAT ASAR in swath IS5 with dual polarization HH/VV for Apr. 4, 2006 in descending pass was acquired. The swath 5 covers in the look angle of 350 – 390. Another scene of ASAR with swath IS4 with HH/VV polarization was acquired on 22 Apr. 2006. The IS4 covers over the look angle from 310-360. IRS-P6 LISS-3 optical data was also acquired on April 7, 2006 which is very close to ASAR data acquisition. Gujarat : SIR-C L- and C-band data of April and Oct. 5 in fully polarimetric mode at 360 incidence angle at 25m spatial resolution were acquired for this test site. The product is Multi-look complex data with 10 byte per pixel. ENVISAT ASAR scenes were acquired over this area for flood affected area mapping. But we tried to use the same data for soil moisture mapping. Unfortunately we do not have synchronous ground truth for this scene. The ENVISAT ASAR scenes for this area are as follows: ASAR-IS4(HH,HV) Look Angle: 310 360, 03 Aug. 2004 ASAR-IS4(HH,HV) - 22 Aug. 2004 ASAR-IS4(HH,HV) – 03 July, 2005 ASAR-IS1(HH,HV) Look angle 140-220- 29 Aug. 2005 Gurdaspur, Punjab : ENVISAT scenes for this area are as follows: ASAR-IS6(HH,VV) – Look Angle 390-430 17 July 2006 ASAR-IS4(HH,VV)-Look Angle -310-360, 20 July 2006

ASAR-IS2(HH,VV)-Look Angle -220-260, 22 July 2006 Groundtruth data in terms of soil moisture, GPS and crop pattern were collected synchronous with ENVISAT passes on 17 and 20 July 2006. Mumbai : The data sets acquired over Mumbai are : ASAR-IS2 (HH, VV) – May20, 2004 ASAR-IS2 (HH, VV) – June 13, 2004 ASAR-IS2 (HH, VV) – July 20, 2004 IS2 swath covers in the look angle of 220-260. All the data sets were acquired in SLC format so that they can be used for interferometric applications. 4. MODELS FOR SOIL MOISTURE MAPPING There are several models for soil moisture estimation using dual polarized data [1],[3][4],[8]. We used Dubois et al. [3] and Baghdadi et al. [1] models for soil moisture estimation. Dubois et al. model is an empirical model derived using ground scatterometer data and verified using SIR-C data for the retrieval of soil moisture. The model is good for kh≤2.5, mv≤35% and θ≥300, where k=2π/λ, h is RMS surface height, mv-soil moisture and θ is look angle. The effect of vegetation is taken into account using the ratio between HV and VV polarized data. If the ratio (σ0HV/σ0VV) is greater than 11db, the soil moisture estimation is not possible with this model. The Dubois et al. model gives dielectric constant and RMS surface height. The dielectric constant is converted into soil moisture using Pemlinski et al. [9] dielectric constant model. Baghdadi et al. [1] model is simply the regression equations obtained using groundtruth soil moisture and ENVISAT ASAR backscattering coefficient data at several polarizations and incidence angle. They found that dual polarization data does not contribute much to improve the accuracy of soil moisture estimation. They concluded that single polarized data acquired at lower incidence angle is good for soil moisture estimation. Their equations for the estimation of soil moisture using extensive data are as follows: 0 = α 0σ mn (θ ) +

mv β 0 for single polarization 0 0 mv = α1σ HH (θ ) + β1σ HV (θ ) + γ 1 forHH&VV mv

⎛ σ 0 (θ ) ⎞ 0 ⎟ + γ 2 for = α 2σ mn (θ ) + β 2 ⎜ HH 0 ⎜ σ (θ ) ⎟ ⎝ HV ⎠

HH/VV

4. DATA PROCESSING SIR-C L- and C-band data were processed using ENVI image processing software for the calculation of backscattering coefficient. The product was MLC with pixel resolution of 12.5m. ENVISAT ASAR SLC DN values were converted into backscattering coefficient using the calibrations equations provided by ESA and software developed by us. The SLC product is multilooked in azimuth direction 5 times and the range pixel is kept as it is. Later, the slant range is converted to ground range using ENVI software. The final spatial

resolution of the data is about 20mx20m and slightly varies with incidence angle. All the data were subjected to speckle filtering using enhanced Lee filter. Around 25 groundtruth soil moisture measurement were collected in coastal Andhra Pradesh synchronous with ENVISAT ASAR passes and obtained volumetric soil moisture by multiplying gravimetric soil moisture with bulkdensity. A few more ground locations were also taken in terms of GPS and crop condition, but without soil moisture information. IRS-LISS-3 data were registered to ASAR data for identification of test sites. Soil moisture was estimated using backscattering coefficient obtained using SIR-C and ENVISAT ASAR data using Dubois et al. and Baghdadi et al. models. 5. RESULTS AND DISCUSSIONS 5.1 SIR-C Data Analysis Soil moisture map produced using SIR-C L-band April 14, 1994 data is shown in Fig. 1. The actual groundtruth soil moisture is around 2.5% volumetric. The estimated soil moisture shows between 0-5% and 20-25%. Dark areas in the soil map are trees and settlements in the village. At the left side-middle of intensity image, black- buck national park (triangle dark part) with dry grass of 60 cm was noted during the observation. We retrieved the soil moisture using Shi et al. [3] model and found that the soil moisture over the entire area is between 0-5%. Similarly, we used C-band data to estimate soil moisture over the same area and the results are shown in Fig. 2. Dubois et al. model over estimated soil moisture at many places and it is more than 20% in most of the area. Using Baghdadi et al. equation for HH polarization, soil moisture is estimated and given in Fig. 2. Baghdadi et al. model also over estimates but the results are better than that of Dubois et al. model. We found that the combination HH and VV and their variation lead to big difference in the estimation of soil moisture using Dubois et al. model. 5.2 Coastal Andhra Pradesh Out of 25 ground-truth soil moisture values for this test area, we took only 12 samples after eliminating sugarcane fields and other areas which do not come into the scene. Fig. 3 shows soil moisture versus measured ASAR backscattering coefficient at HH and VV polarizations for April 4, 2006. Most of the points fall at low soil moisture range. The correlation coefficient is 0.65 at HH and 0.50 at VV polarization. Similar correlation was observed by Holah et al. [5] at 34-370 incidence angles. As the Dubois et al. model over estimates soil moisture, we used regression equations obtained by Baghdadi et al. [1] and Holah [5]. The equations may be ideal for their test sites, but we used them for verification of retrieval results for our test site. The estimated soil moisture versus measured soil moisture is shown in Fig. 4. The results are almost similar except Holah et al. gives slightly better correlation.

30

Fig. 1. SIR-C L-band (a) HV polarized intensity data April 14, 1994, (b) Soil moisture estimated using Dubois et al. [4] and (c) Soil moisture map produced using Shi et al. [3].

30

Fig. 2. SIR-C C-band (a) HV polarized intensity data April 14, 1994, (b) Soil moisture estimated using Dubois et al. [4] and (c) Soil moisture map produced using Shi et al. [3].

0 -2 -4 -6 -8 -10 -12 -14

0

y = 0.174x - 12.65 2 R = 0.4949

-2

σ0 ( VV )

σ0(H H )

y = 0.2819x - 14.873 R2 = 0.6501

-4 -6 -8 -10 -12

0

5

10

15

20

25

30

-14 0

Soil Moisture Vol.

5

10

15

20

25

30

Soil Moisture Vol.

35 y = 0.8447x + 7.5293 30 2 R = 0.5787 25 20 15 10 5 0 0 10 20 Estimated Soil Moisture

30 Soil Moisture Vol.

Soil Moisture Vol.

Fig. 3. ENVISAT ASAR (IS5) HH and VV backscattering coefficient versus soil moisture over coastal Andhra Pradesh.

25

y = 0.5074x + 6.3484 2

R = 0.6501

20 15 10 5 0

30

0

10 20 30 Estimated Soil Moisture

Fig. 4. Measured soil moisture versus estimated soil moisture using (a) Baghdadi et al.[1] and (b) Holah et al. [5]

40

Suggest Documents