Combination of Envisat, CryoSat-2 and SARAL/AltiKa … · 2016. 5. 30. · Jain M., Andersen O.B.,...
1
Franziska Göttl, Christian Schwatke and Denise Dettmering Deutsches Geodätisches Forschungsinstitut der Technischen Universität München (DGFI-TUM), München, Germany [email protected] Combination of Envisat, CryoSat-2 and SARAL/AltiKa measurements for estimating water level variations of lakes Combination of Envisat, CryoSat-2 and SARAL/AltiKa measurements for estimating water level variations of lakes Franziska Göttl, Christian Schwatke and Denise Dettmering Deutsches Geodätisches Forschungsinstitut der Technischen Universität München (DGFI-TUM), München, Germany [email protected] Motivation The estimation of lake water level variations with satellite altimetry is a challenging task because the majority of altimeter waveforms of smaller lakes are contaminated by land. The main goal of this study is the determination of improved long-term water level time series of lakes due to combination of Envisat, CryoSat-2 and SARAL/AltiKa altimeter measurements in an optimal way. While Envisat ant SARAL/AltiKa carry classical pulse-limited radar altimeters on board, CryoSat-2 carries a Delay-Doppler/SAR altimeter on board that allows to exploit the full Doppler bandwidth. Depending on the observation technique and surface conditions the altimeter waveforms vary from specular over land to multi peaks at the land/water transition and brown-like in the center of the lake. In a first step, we have developed a new processing strategy for CryoSat-2 SAR data over lakes based on waveform classification, retracking and outlier rejection. The results are validated with water level time series derived from classical altimeter missions and in-situ gauging data. Fig. 1: OCOG parameters for the waveforms of the three classes (a) water, (b) land/water transition and (c) land. The sub-waveforms which are identified within the retracker procedure are marked in red. Processing strategy for CryoSat-2 SAR data over lakes 1. Waveform classification: Three classes of CryoSat-2 SAR multi-looked waveforms: water, land/water transition and land can be identified by using the three parameters of the Offset Center of Gravity (OCOG) retracker: amplitude (A) and width (W) of the waveform as well as the gate position of the waveform center of gravity (COG), see Figure 1. The cluster centers of the three classes are determined by applying the k-means algorithm to the three OCOG parameters of the CryoSat-2 SAR waveforms in the training area of the lake Tonle Sap (see Fig. 2). While the cluster for water is well identified, the cluster for the land/water transition contains data points which should be referred to the cluster land. This problem can be solved by adopting the condition that waveforms at the land/water transition should contain one single peak before the main peak. In future we want to consider this feature within the clustering. We have tested if the cluster centers of the training area can be used to classify the waveforms of other lakes such as Vänern, Vättern, Okeechobee, Towuti, and Lough Neagh. We found that this classification strategy can be used for all other lakes, too. 2. Waveform retracking algorithms: Over water, the leading sub-waveform which has to be retracked is located around the maximum peak of the waveform (see Fig. 1). Therefore in our implementation of the Improved Threshold Retracker (ITR; Hwang et al., 2006) the sub-waveform around the maximum peak is chosen. We call it the Maximum Peak Threshold Retracker (MPTR). Waveforms near the land/water transition are contaminated by land signals and therefore they exhibit a smaller peak before the main peak (see Fig. 1). Here the leading sub-waveform is located around the peak before the maximum peak. Therefore these waveforms are retracked with the Narrow Primary Peak Threshold Retracker (NPPTR; Jain et al., 2015). To avoid offsets between the two different retrackers MPTR and NPPTR the same threshold level of 50% is used for both retrackers (see Fig. 3). 3. Outlier rejection: Inaccurate water levels must be rejected before the determination of the median of the water heights of each track. Therefore the following outlier criteria are applied: • latitude and longitude thresholds, • water height thresholds, • moving 5-point standard deviation [i-2 … i+2] smaller than 10 cm. References: Göttl F., Schwatke C., Dettmering D. & Müller F.: Cryosat-2 SAR mode retracking strategy for lakes: Waveform classification, sub- waveform retracking, outlier rejection. Remote Sens., in preparation. Hwang C., Gua J., Deng X., Hsu H.Y. & Liu Y. (2006): Coastal gravity anomalies from retracked Geosat/GM altimetry: improvement, limitation and the role of airborne gravity data. J. Geodesy, 80, 204-216. Jain M., Andersen O.B., Dall J. & Stenseng I. (2015): Sea surface height determination in the Arctic using CryoSat-2 data from primary peak empirical retrackers. Adv. Space Res., 55(1), 40-50. Envisat and SARAL/AltiKa retracking strategy Living Planet Symposium 2016, 9-13 May 2016, Prague, Czech Republic Fig. 2: Classification results for the training area Tonle Sap. a) Data points of each cluster: water (blue), land/water transition (red) and land (green). The cluster centers are marked in black. (b) CryoSat-2 tracks classified as water (blue), land/water transition (red) and land (green) by the k-means algorithm. (c) Revised classification results. For Envisat and SARAL/AltiKa the water level time series are determined with DAHITI (Database for Hydrological Time Series over Inland Waters; Schwatke et al., 2015). Hereby the waveform classification is based on statistical waveform parameters such as skewness, kurtosis, peakiness, signal to noise ration and maximum power of the altimeter waveform (Schwatke & Dettmering, 2015). The waveforms corresponding to the single peak classes 20 to 22 are retracked with the Improved Threshold Retracker with a threshold level of 10%. 13.2 12.6 latitude [°] 12.8 13.0 (b) 103.8 104 104.2 104.4 longitude [°] 13.2 12.6 latitude [°] 12.8 13.0 (c) 103.8 104 104.2 104.4 longitude [°] (a) 20 40 60 80 20 40 60 80 20 40 60 80 COG W Ax10 11 Conclusions / Outlook Fig. 3: Water heights determined with the MPTR (blue) and NPPTR (red) retrackers for a track crossing the lake Okeechobee. The water height thresholds are mark by green dashed lines and the moving standard deviations of the water heights are displayed by the gray errorbars. Valid results after the outlier rejection are marked in green. (a) water (b) land/water transition (c) land Data sources for the validation • Modeled water level time series derived from CryoSat-2 SAR observations by the Technical University of Denmark (DTU) are provided via the open web service AltWater (Altimetry for inland Water, http://www.altwater.dtu.space/index.html). • Multi-mission water level time series derived from classical altimeter missions such as Envisat, Jason-2 and SARAL/AltiKa are provided by the open web service DAHITI (Database for Hydrological Time Series over Inland Waters, http://dahiti.dgfi.tum.de/en/), which is run by the Deutsches Geodätisches Forschungsinstitut der Technischen Universität München (DGFI-TUM). • In-situ gauging data are provided by the Mekong River Commission (http://ffw.mrcmekong.org/), the National Water Information System (http://waterdata.usgs.gov/nwis/sw) and the Rivers Agency, Department of Agriculture and Rural Development (https://www.dardni.gov.uk/rivers-agency). Results Comparison of lake water level time series derived from CryoSat-2 SAR data (DGFI- TUM) with external data sets. 13.2 12.6 latitude [°] 12.8 13.0 (a) Tonle Sap: 2600-10400 km² 103.8 104 104.2 104.4 longitude [°] 27.1 26.8 latitude [°] 26.9 27.0 (b) Okeechobee: 1890 km² 279.1 279.2 279.3 longitude [°] 54.5 latitude [°] 54.6 54.7 (c) Lough Neagh: 392 km² 353.4 353.5 353.6 353.7 longitude [°] Conclusions / Outlook • The CryoSat-2 SAR mode waveform classification based on the three OCOG parameters is successfully applied over smaller and larger lakes. • DGFI-TUM water level time series based on CryoSat-2 SAR data show similar quality as the modeled CryoSat-2 water level time series of DTU. • Water heights especially of smaller lakes can be derived from CryoSat-2 SAR multi- looked waveforms with higher precision than from classical altimeter waveforms. • In the next step, the CryoSat-2 SAR data processing shall be integrated in DAHITI in order to determine improved long-term water level time series. • In future, this classification algorithm and class-dependent processing strategy shall be adopted to the new CryoSat-2 Baseline C and Sentinel-3 products. water height [m] 26.8 26.9 27 27.1 latitude [°] 27.2 0 1 2 3 4 5 Lake Okeechobee 14 12 10 8 6 4 water height [m] Wrt. gauge RMS [m] Corr. DGFI-TUM 0.94 0.94 DTU 0.98 0.94 DAHITI 0.78 0.96 2009 2010 2011 2012 2013 2014 2015 Time [decimal years] 5 4 2009 2010 2011 2012 2013 2014 2015 Time [decimal years] water height [m] Wrt. gauge RMS [m] Corr. DGFI-TUM 0.03 1.00 DTU 0.04 1.00 DAHITI 0.08 0.99 2010 2011 2012 2013 2014 2015 Time [decimal years] 13 12.4 water height [m] 12.2 12.6 12.8 13.2 2009 Wrt. gauge RMS [m] Corr. DGFI-TUM 0.04 0.96 DAHITI 0.08 0.92
Transcript of Combination of Envisat, CryoSat-2 and SARAL/AltiKa … · 2016. 5. 30. · Jain M., Andersen O.B.,...
Franziska Göttl, Christian Schwatke and Denise Dettmering
Deutsches Geodätisches Forschungsinstitut der Technischen Universität München (DGFI-TUM), München, Germany [email protected]
Combination of Envisat, CryoSat-2 and SARAL/AltiKa measurements for estimating water level variations of lakes
Combination of Envisat, CryoSat-2 and SARAL/AltiKa measurements for estimating water level variations of lakes
Franziska Göttl, Christian Schwatke and Denise Dettmering
Deutsches Geodätisches Forschungsinstitut der Technischen Universität München (DGFI-TUM), München, Germany [email protected]
Motivation
The estimation of lake water level variations with satellite altimetry is a challenging task because the majority of altimeter waveforms of smaller lakes are contaminated by land. The main goal of this study is the determination of improved long-term water level time series of lakes due to combination of Envisat, CryoSat-2 and SARAL/AltiKa altimeter measurements in an optimal way. While Envisat ant SARAL/AltiKa carry classical pulse-limited radar altimeters on board, CryoSat-2 carries a Delay-Doppler/SAR altimeter on board that allows to exploit the full Doppler bandwidth. Depending on the observation technique and surface conditions the altimeter waveforms vary from specular over land to multi peaks at the land/water transition and brown-like in the center of the lake. In a first step, we have developed a new processing strategy for CryoSat-2 SAR data over lakes based on waveform classification, retracking and outlier rejection. The results are validated with water level time series derived from classical altimeter missions and in-situ gauging data.
Fig. 1: OCOG parameters for the waveforms of the three classes (a) water, (b) land/water transition and (c)
land. The sub-waveforms which are identified within the retracker procedure are marked in red.
Processing strategy for CryoSat-2 SAR data over lakes
1. Waveform classification:
Three classes of CryoSat-2 SAR multi-looked waveforms: water, land/water transition and land can be identified by using the three parameters of the Offset Center of Gravity (OCOG) retracker: amplitude (A) and width (W) of the waveform as well as the gate position of the waveform center of gravity (COG), see Figure 1.
The cluster centers of the three classes are determined by applying the k-means algorithm to the three OCOG parameters of the CryoSat-2 SAR waveforms in the training area of the lake Tonle Sap (see Fig. 2). While the cluster for water is well identified, the cluster for the land/water transition contains data points which should be referred to the cluster land. This problem can be solved by adopting the condition that waveforms at the land/water transition should contain one single peak before the main peak. In future we want to consider this feature within the clustering.
We have tested if the cluster centers of the training area can be used to classify the waveforms of other lakes such as Vänern, Vättern, Okeechobee, Towuti, and Lough Neagh. We found that this classification strategy can be used for all other lakes, too.
2. Waveform retracking algorithms:
Over water, the leading sub-waveform which has to be retracked is located around the maximum peak of the waveform (see Fig. 1). Therefore in our implementation of the Improved Threshold Retracker (ITR; Hwang et al., 2006) the sub-waveform around the maximum peak is chosen. We call it the Maximum Peak Threshold Retracker (MPTR).
Waveforms near the land/water transition are contaminated by land signals and therefore they exhibit a smaller peak before the main peak (see Fig. 1). Here the leading sub-waveform is located around the peak before the maximum peak. Therefore these waveforms are retracked with the Narrow Primary Peak Threshold Retracker (NPPTR; Jain et al., 2015).
To avoid offsets between the two different retrackers MPTR and NPPTR the same threshold level of 50% is used for both retrackers (see Fig. 3).
3. Outlier rejection:
Inaccurate water levels must be rejected before the determination of the median of the water heights of each track. Therefore the following outlier criteria are applied:
• latitude and longitude thresholds, • water height thresholds, • moving 5-point standard deviation [i-2 … i+2] smaller than 10 cm.
References:
Göttl F., Schwatke C., Dettmering D. & Müller F.: Cryosat-2 SAR mode retracking strategy for lakes: Waveform classification, sub-waveform retracking, outlier rejection. Remote Sens., in preparation.
Hwang C., Gua J., Deng X., Hsu H.Y. & Liu Y. (2006): Coastal gravity anomalies from retracked Geosat/GM altimetry: improvement, limitation and the role of airborne gravity data. J. Geodesy, 80, 204-216.
Jain M., Andersen O.B., Dall J. & Stenseng I. (2015): Sea surface height determination in the Arctic using CryoSat-2 data from primary peak empirical retrackers. Adv. Space Res., 55(1), 40-50.
Envisat and SARAL/AltiKa retracking strategy
Living Planet Symposium 2016, 9-13 May 2016, Prague, Czech Republic
Fig. 2: Classification results for the training area Tonle Sap. a) Data points of each cluster: water (blue),
land/water transition (red) and land (green). The cluster centers are marked in black. (b) CryoSat-2 tracks
classified as water (blue), land/water transition (red) and land (green) by the k-means algorithm. (c) Revised
classification results.
For Envisat and SARAL/AltiKa the water level time series are determined with DAHITI (Database for Hydrological Time Series over Inland Waters; Schwatke et al., 2015). Hereby the waveform classification is based on statistical waveform parameters such as skewness, kurtosis, peakiness, signal to noise ration and maximum power of the altimeter waveform (Schwatke & Dettmering, 2015). The waveforms corresponding to the single peak classes 20 to 22 are retracked with the Improved Threshold Retracker with a threshold level of 10%.
13.2
12.6
latitu
de
[°]
12.8
13.0
(b)
103.8 104 104.2 104.4 longitude [°]
13.2
12.6
latitu
de
[°]
12.8
13.0
(c)
103.8 104 104.2 104.4 longitude [°]
(a)
20 40
60 80
20 40
60 80
20
40
60
80
COG W
Ax1
011
Conclusions / Outlook
Fig. 3: Water heights determined with
the MPTR (blue) and NPPTR (red)
retrackers for a track crossing the lake
Okeechobee. The water height
thresholds are mark by green dashed
lines and the moving standard
deviations of the water heights are
displayed by the gray errorbars. Valid
results after the outlier rejection are
marked in green.
(a) water (b) land/water transition (c) land
Data sources for the validation
• Modeled water level time series derived from CryoSat-2 SAR observations by the Technical University of Denmark (DTU) are provided via the open web service AltWater (Altimetry for inland Water, http://www.altwater.dtu.space/index.html).
• Multi-mission water level time series derived from classical altimeter missions such as Envisat, Jason-2 and SARAL/AltiKa are provided by the open web service DAHITI (Database for Hydrological Time Series over Inland Waters, http://dahiti.dgfi.tum.de/en/), which is run by the Deutsches Geodätisches Forschungsinstitut der Technischen Universität München (DGFI-TUM).
• In-situ gauging data are provided by the Mekong River Commission (http://ffw.mrcmekong.org/), the National Water Information System (http://waterdata.usgs.gov/nwis/sw) and the Rivers Agency, Department of Agriculture and Rural Development (https://www.dardni.gov.uk/rivers-agency).
Results
Comparison of lake water level time series derived from CryoSat-2 SAR data (DGFI-TUM) with external data sets.
13.2
12.6
latitu
de [°]
12.8
13.0
(a) Tonle Sap: 2600-10400 km²
103.8 104 104.2 104.4
longitude [°]
27.1
26.8
latitu
de
[°]
26.9
27.0
(b) Okeechobee: 1890 km²
279.1 279.2 279.3
longitude [°]
54.5
latitu
de
[°]
54.6
54.7
(c) Lough Neagh: 392 km²
353.4 353.5 353.6 353.7
longitude [°]
Conclusions / Outlook
• The CryoSat-2 SAR mode waveform classification based on the three OCOG parameters is successfully applied over smaller and larger lakes.
• DGFI-TUM water level time series based on CryoSat-2 SAR data show similar quality as the modeled CryoSat-2 water level time series of DTU.
• Water heights especially of smaller lakes can be derived from CryoSat-2 SAR multi-looked waveforms with higher precision than from classical altimeter waveforms.
• In the next step, the CryoSat-2 SAR data processing shall be integrated in DAHITI in order to determine improved long-term water level time series.
• In future, this classification algorithm and class-dependent processing strategy shall be adopted to the new CryoSat-2 Baseline C and Sentinel-3 products.
wa
ter
he
igh
t [m
]
26.8 26.9 27 27.1 latitude [°]
27.2 0
1
2
3
4
5
Lake Okeechobee
14
12
10
8
6
4
wa
ter
he
igh
t [m
]
Wrt. gauge RMS [m] Corr.
DGFI-TUM 0.94 0.94
DTU 0.98 0.94
DAHITI 0.78 0.96
2009 2010 2011 2012 2013 2014 2015
Time [decimal years]
5
4
2009 2010 2011 2012 2013 2014 2015
Time [decimal years]
wa
ter
he
igh
t [m
]
Wrt. gauge RMS [m] Corr.
DGFI-TUM 0.03 1.00
DTU 0.04 1.00
DAHITI 0.08 0.99
2010 2011 2012 2013 2014 2015
Time [decimal years]
13
12.4
wa
ter
he
igh
t [m
]
12.2
12.6
12.8
13.2
2009
Wrt. gauge RMS [m] Corr.
DGFI-TUM 0.04 0.96
DAHITI 0.08 0.92
