Similarities and Differences in the Earth’s Water Variations Signal Provided by Grace and AMSR-E Observations Using Maximum Covariance Analysis at Various Land Cover Data Backgrounds
Published Online: Jul 18, 2023
Page range: 63 - 87
Received: Mar 22, 2023
Accepted: Jun 16, 2023
DOI: https://doi.org/10.2478/arsa-2023-0006
Keywords
© 2023 Viktor Szabó et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
Soil moisture (SM) is a critical hydrologic state variable of the land that crosses the interfaces of several disciplines, of significant importance for numerous applications for meteorology, hydrology, climatology, and ecology (Robinson et al., 2008). Small changes in gravity measured from space also deduced water mass fluctuations. Launched in March 2002 twin-satellite system Gravity Recovery and Climate Experiment (GRACE) (Tapley et al., 2004b) and GRACE Follow-On (GRACE-FO) (Flechtner et al., 2016) provided unique information regarding gravity changes caused by the mass transport over the Earth’s surface. Changes in total water storage (ΔTWS) (Wahr et al., 1998) show the Earth’s mass change on a near-monthly timescale. The derivative of the TWS signal is TWS anomaly (TWSA), understood as a combined monthly averaged water storage change by removing the long-term average divided by standard deviation. TWSA corresponds to the sum of all above and below surface water storage, including SM, canopy water, lakes, rivers, and groundwater. The importance of SM and ΔTWS for understanding the Earth’s water cycle, and the factors affecting it over the years, has been considered in many studies individually.
The influence of estimating spatial and temporal variations of SM on climate changes was described in multiple studies (Betts et al., 1994, Engman, 1992, Entekhabi et al., 1994, Fast and McCorcle, 1991, Jackson et al., 1987, Petropoulos et al., 2014, Saha, 1995, Topp et al., 1980). Spatial and temporal variability of water was well documented in previous work for SM (Crow et al., 2012, Famiglietti et al., 2008, Vereecken et al., 2014) and ΔTWS (Landerer and Swenson, 2012, Tapley et al., 2004a, Zhao et al., 2017). From a hydrological point of view, analysis of spatiotemporal patterns of SM and ΔTWS observations is essential to understanding their behavior. In literature, existing methods describe variability only in the spatial domain (Haining et al., 2010, Khaki et al., 2017) or only in the temporal domain, based on time series analysis (Fu, 2011, Sprott and Sprott, 2003, Vishwakarma et al., 2021). Several methods can be found in the literature that analyzes ΔTWS and SM space and time domains together such as temporal stability analysis (TSA) (Martínez-Fernández and Ceballos, 2005, Wang et al., 2018), triple collocation (TC) (Crow et al., 2015, Gruber et al., 2017, Hasan and Tarhule, 2021, Yin and Park, 2021), and empirical orthogonal functions (EOFs) (Eom et al., 2017, Lei et al., 2012, Navarra and Simoncini, 2010, Schrama et al., 2007, Yoo and Kim, 2004). Whether the analysis is temporal or spatiotemporal, researchers in previous work have indicated the importance of SM as a component of the ΔTWS signal.
Water content in near-surface soil layers is a significant component of the ΔTWS signal observed by the GRACE mission. There have been many significant studies examining the relationship between SM and ΔTWS. A joint comparison of the remote sensing retrieval products’ metric entropy and fluctuation complexity was considered in (Kumar et al., 2018). The satellite products of Advanced Microwave Scanning Radiometer (AMSR-E), Advanced Scatterometer (ASCAT), Soil Moisture and Ocean Salinity (SMOS), and Advanced Microwave Scanning Radiometer 2 (AMSR2) show significant noise (high entropy, low complexity), except Soil Moisture Active Passive (SMAP) is slightly noisy and more informative. The correlation greater than 0.7 between TWSA and SM data was shown in previous work (Abelen and Seitz, 2013, Crow et al., 2017, Swenson et al., 2008b). Expanding the shallow groundwater variation under the SM root zone is an essential issue in scientific research. Since using microwave satellites may be a possible way to isolate groundwater storage (GWS) variations from the GRACE signal (Frappart and Ramillien, 2018, Yeh et al., 2006), a significant area of research is the possibility of using microwave observations to determine SM.
Microwave remote sensing observations have been applied for the determination of SM (Babaeian et al., 2019). Active and passive microwave remote sensing provides an observation of SM at global and regional scales (Bartalis et al., 2007, Chen et al., 2018, Jackson et al., 2010, Kerr et al., 2016, Koike et al., 2004, Ulaby, 1982, Vinnikov et al., 1999, Wagner et al., 2013). It helps in much scientific research in hydrology and climate studies and gives an opportunity to understand environmental changes (Njoku and Entekhabi, 1996). GRACE ΔTWS and remote sensing microwave SM observations have recently been used to improve SM and GWS simulations (Tangdamrongsub et al., 2022, Tian et al., 2017).
One of the essential microwave sensors providing SM data was the AMSR-E mission. Owing to the long joint period in orbit during the operation of GRACE and AMSR-E missions, numerous previous studies have considered comparing SM from AMSR-E and ΔTWS signals from these sensors. Comparisons between the AMSR-E surface wetness index (ASWI) and the GRACE drought severity index (DSI) were shown in the previous work (Du et al., 2019). The indicated comparisons showed robust correlations in regions in the United States (R higher than 0.7 for 29 percent of the area) during the summer months (June–August) from 2002 to 2017 for regions where a semiannual temporal lag between fast surface water changes and the slower GWST was considered. The study explores multivariate data assimilation (DA) using synthetic ΔTWS from GRACE and synthetic AMSR-E passive microwave brightness temperature spectral differences (dTb) in case estimation of snow water equivalent (SWE) over snow-covered terrain was presented by Wang et al. (2021) and Wang and Forman (2020). In a previous study (Seo et al., 2010), the authors propose methods to estimate solid precipitation accumulation in winter in the northern Arctic region. Based on the GRACE and AMSR-E, winter season solid precipitation accumulation was estimated. In the second step, estimated values was compared with the traditional estimations from the Global Precipitation Climatology Project (GPCP) and Climate Prediction Center’s Merged Analysis of Precipitation (CMAP). Correlation, time shift, and principal component analyses of SM from the WaterGAP Global Hydrology Model (WGHM) and the satellite sensors AMSR-E and ASCAT to total water storage variations from the satellite gravity mission GRACE in the area of the La Plata Basin in South America were provided by Abelen et al. (2015). Regional and global variations in SM from satellite sensor AMSR-E and GRACE was also considered in Abelen et al. (2011). Global Land Data Assimilation System (GLDAS) (Rodell et al., 2004) product was used to evaluate AMSR-E observations over central Tibetan Plateau (Chen et al., 2013). To effectively catch drought disasters in the Guangdong province of southern China in 2004–2005, 2007, and 2009 SM from AMSR-E was used (Chen et al., 2012). The highest SM variability in the surface soil layer can be observed because of meteorological and environmental interactions such as precipitation, temperature changes, porosity, topography, vegetation processes, and human factors.
Although many studies have been performed on evaluating extreme hydrological events using GRACE and AMSR-E, there is a gap in the published literature concerning ΔTWS and SM signal convergence considering land cover data described soil conditions, vegetation cover, urbanization status, and cultivated land. Since the information collected by gravimetric sensors has a lower temporal frequency and spatial resolution than microwave measurements, it is crucial to investigate the convergence of these signals. The key question posed in the article is: is it possible to use the information contained by sensors characterized by higher noise and signal variance, such as AMSR-E, in the global analysis of ΔTWS variability from GRACE satellites? In work, it was decided to present the similarities and differences in the Earth’s water resource measurements. This article analyzes the spatiotemporal variations of SM and ΔTWS in the context of the similarity pattern comparison. The study used maximum covariance analysis (MCA) to extract principal components for ΔTWS and SM signals.
GRACE data is available at
The Advanced Microwave Scanning Radiometer for the Earth Observing System is a passive multiband sensor of NASA’s Earth Observing System Aqua satellite. AMSR-E uses the X-band and C-band to measure the water cycle and SM content retrievals corresponding to the depth of (2.5–3.75cm) and (3.75–7.5cm), respectively. Owing to the fact that radio frequency interference (RFI) in the C-band (6.9 and 10.7 GHz), the X-band has been extensively used for SM retrieval (Njoku et al., 2005). AMSR-E dataset is available as daily files at
The intersection of the GRACE and AMSR-E sensors datasets was selected for analysis. The time range of the selected data for this study was chosen to cover the maximum part intersection of existing GRACE and AMSR-E datasets. The dataset in the analysis contains data from 2002 to 2012 from both missions.
Data preparation involved averaging with moving window data collected by the AMSR-E sensor over the GRACE epochs. As the compared sensors have different spatial resolutions, the data from AMSR-E were linearly interpolated on the GRACE resolution. The values for ΔTWS observed by GRACE and AMSR-E have different amplitudes. To be able to compare these results to each other, it was decided to normalize data for each season and then compare the normalized values for given seasons to minimize the effects of seasonality. Volumetric soil water content collected by AMSR-E sensor is the volume of water per unit volume of soil
To reveal the similarities and differences between the values, both sensor signals were grouped for the winter, spring, summer, and autumn months. Moreover, a complementary correlation analysis was performed to assess the level of agreement between different data sources:
Intense spatial averaging filters with a high radius of smoothing kernel can cause signal loss, known as ”leakage error” (Longuevergne et al., 2010, Swenson and Wahr, 2002). Filtering decreases the spatial resolution of the GRACE observation, making it challenging to identify the mass water signal of the main stem. The EOF analysis is a method for GRACE data to separate signals from signal noise. It is beneficial in cases such as problems with loss of geophysical signal with diminishing spatial resolution during filtration (Wouters and Schrama, 2007). The use of this method is justified in the case of comparison of microwave data with higher spatial resolution and greater time frequency of measurements than gravimetric satellite measurements. Concerning the EOF’s of standard MCA (Rieger et al., 2021), the spatial amplitude (As) provides a means to understand which regions contribute the most to the given mode. The spatial amplitude is easily computed via the complex EOF and the complex conjugate EOF*:
We can determine exactly how the individual regions are dynamically linked to each other. Phase shifts between these two cases are signals that can be combined into one mode with standard MCA by the following equation:
The surface soil layer commonly shows the most considerable SM variability due to the relations with meteorological, environmental, and anthropogenic factors such as porosity, topography, vegetation, precipitation, and temperature decreasing with depth. To analyze land cover conditions, the Harmonized World Soil Database was used (Fischer et al., 2008) from
Drainage characteristics of soils broadly define oxygen availability in soils. The determination of soil drainage classes is based on procedures developed at FAO. These procedures consider soil type, texture, terrain slope, and phases with mean proportion of water, air, and solids in soil. This publication contains characteristics of forest land, oxygen availability to roots, total cultivated land, and share of build-up land in Figure 1.
Figure 1.
Land cover data of forest land (a), oxygen availability to roots (b), total cultivated land (c), and share of build-up land (d) based on Harmonized World Soil Database
The SM and ΔTWS variables are characterized by high variability over time. The main components are related to seasonal factors included and the occurrence of dry and rainy seasons. This decline over the years is presented in Figure 2. The figure clearly shows the negative trend of ΔTWS value over the years. There are no similarities between the averaged SM and ΔTWS observations for a given month. Since the cyclic signal can be reset by cyclical phenomena occurring in a given area, the article presents averaged anomalies concerning time and latitude.
Figure 2.
Seasonal patterns of ΔTWS (a), SM from band X (b), and SM from band C (c) grouped by month over time
In order to characterize the values collected by gravimetric and microwave sensors, the averaged values of the observation epochs in the years 2002–2011 were determined concerning the latitude and longitude, respectively, as shown in Figure 3. Mean anomalies and standard deviation of anomalies in time over the latitude are presented in Figure 4. Figure 4 a) c) e) show an increase in the average values of TWSA and SMA in 2009–2011 for latitudes 0–20°S with a slight standard deviation for these latitudes in the given years. Both sensors picked up the same anomaly in these areas. Time series analyses in this area can be characterized by high convergence. For latitudes 20–40°N, we observe a significant TWSA anomaly that was not captured by the AMSR-E sensors. In the years 2003–2005, we observed a significant standard deviation of anomalies, which indicates a large scatter of observations and substantial variability, which was not captured when determining the average SMA values.
Figure 3.
Average SM from AMSR-E and ΔTWS from GRACE grouped by latitude (a) and longitude (b)
Figure 4.
GRACE (a,b) and AMSR-E (c,d,e,f) average anomaly (a,c,e) and standard deviation (b,d,f) grouped by latitude over time
As the data on the water content in the ground shows the cycle of seasonal changes in the groundwater level, the average values were compared separately for each season of the year. The analysis was divided into the C and X bands for the SM observation. After normalizations of ΔTWS, ΔTWS and SM signals were grouped for the winter, spring, summer, and autumn seasons. Where winter months are marked as December, January, February (DJF), spring as March, April, May (MAM), summer as June, July, August (JJA), and autumn as September, October, and November (SON).
Data from the C- and X-ranges are very similar. However, they are visible in the saturation of the SM parameter. In Figure 5, higher values of VSM in the areas of Amazonia can be noticed for the C- range and the latitude of 60–70 degrees. When comparing the percentages of GRACE and AMSR-E, there are apparent differences. Some of them may be due to data noise in GRACE. Theoretically, all observations from the Sahara area should be close to zero due to the near-zero water content in that area. However, variations in the water content around Lake Chad are observed (Boy et al., 2012), which partially explain this effect. More similarities can be seen between the C-band and the GRACE data, especially in the equatorial regions.
Figure 5.
GRACE ΔTWS (a,d,g,j) and AMSR-E band C (b,e,h,k) and band X (c,f,i,l) SM averaged and normalized values grouped by seasons DJF (a,b,c), MAM (g,h,i), JJA (j,k,l) and SON (j,k,l)
Often long-term microwave SM datasets, such as the Climate Change Initiative (CCI), based on C- and X-band observations, are typically masked over densely vegetated areas due to the soil signal’s strong attenuation by the vegetation signal canopy (Dorigo et al., 2011, Liu et al., 2011). It is worth emphasizing here that the X-band penetrates only the surface layer, the C-band a bit deeper, into with highly dense vegetation. Both bands cannot penetrate the soil in some cases (El Hajj et al., 2018). Pearson’s correlation coefficient for the tested signals is presented in Figure 6.
Figure 6.
Pearson correlation coefficient between SM from band X and C from AMSR-E (a), ΔTWS from GRACE and SM from band C from AMSR-E (b), ΔTWS from GRACE and SM from band X from AMSR-E (c)
Significant values of humidity in the X- and C-bands and low coefficients of correlation with GRACE data observed in the northern regions of the globe, are strongly related to the permafrost region. Data from this area deviates significantly in quality from other observations. No reduced correlations can be seen in forest areas during the comparison of the water content obtained from gravimetric and microwave sensors. The central part of Europe and the eastern regions of China are mainly urbanized areas. There we observe a negative correlation between GRACE and AMSR-E sensors. The anthropogenic factor related to the urbanization of space strongly influences the quality of observation (Ahmed et al., 2014, Chen et al., 2019, Wang et al., 2017). A high rate of urbanization also characterizes the Indian subcontinent. Moreover, over 60% of the area is arable land, which, due to the large number of people living in the region, is necessary to produce the right amount of food. Owing to the large open area and the lack of limitations in oxygen availability in the root zones, we can observe a significant amplitude of the SM signal. Phase compliance contributes to a high correlation in this area despite the progressive urbanization of the area, in particular in the X-band. The cultivated areas worldwide showed highly coherent GRACE and AMSR-E signals for GRACE and AMSR-E observations. The open areas do not have barriers or limitations for rainfall, which allows water to penetrate the root zone. The only exception is the eastern part of Europe, for which the overlapping of urbanization factors and soil constraints on oxygen content, and thus lower soil porosity, slows down water penetration into the soil. This causes a phase shift for the observed signals manifested by the negative correlation coefficient in this area.
EOF method is effective due to its capacity to find spatial correlation in spatiotemporal data. ΔTWS retrieved from the GRACE and SM retrieved from AMSR-E missions are decomposed using the EOF method to extract the signal, mainly describing the river discharge along the main gravity stream. Before determining the EOF, the linear trend was removed from the observations to eliminate the bias. Applying orthogonal decomposition MCA to geophysical datasets permits extracting common dominant patterns between two variables. Regions with the same color are in phase, that means their time series correlate with each other, while regions whose color is different are anticorrelated as shown in Figure 8 and Figure 9.
Figure 7.
Dominant spatial pattern of water variability presended by decomposition of signal using EOF for ΔTWS from GRACE (a,d,g) and SM from AMSR-E (b,c,e,f,h,j). The first spatial pattern (EOF1) (a,b,c), the second spatial pattern (EOF2) (d,e,f), and the third spatial pattern (EOF3) (g,h,j)
Figure 8.
EOF signal amplitude for ΔTWS form GRACE (a), SM from band X from AMSR-E (b), and SM from band C from AMSR-E (c)
Figure 9.
EOF signal phase shift for ΔTWS form GRACE (a), SM from band X from AMSR-E (b), and SM from band C from AMSR-E (c)
The preservation of the flow of subsurface waters is a significant regional issue, depending on the climate determining the amount of rainwater, the topography, the arrangement of permeable layers, and the presence of river sources. In this part of the article, regional studies were carried out for selected river basins with the most significant area by selecting cases for all continents. Regional analyses appear in earlier articles by Vishwakarma et al. (2021), where time series analysis was carried out for major river basins. In this article, scientists capture significant dips and identify constraints due to too short an observation period using the trend to variability ratio (TVR) metric. This section focuses on the reasons for similarities and differences in gravimetric and microwave signals in selected areas. The observations provided by the GRACE mission are characterized by a significantly lower spatial resolution than microwave observations. The application of grouping to the studied signals within rivers allows for finding patterns resulting from minimizing errors resulting from noise or artifacts of the filtration process. For each continent, a set of rivers with the largest area and different land cover features and different latitudes was selected, thus eliminating bias in the dataset sample. For selected river basins, Pearson’s correlation coefficients and cross-correlation, taking into account the phase shift calculated according to formula (2) and presented in Figure 10 and Figure 11, were determined.
Figure 10.
Pearson correlation over selected rivers basin between ΔTWS from GRACE and SM from band X from AMSR-E (a) and ΔTWS from GRACE and SM from band C from AMSR-E (b)
Figure 11.
Cross-correlation over selected rivers basin between ΔTWS from GRACE and SM from band X from AMSR-E (a) and ΔTWS from GRACE and SM from band C from AMSR-E (b)
Examples of ΔTWS and SM time series and TWSA and SMA anomalies are shown in Figure 12.
Figure 12.
River basin time series containing ΔTWS and SM (a,c,e,g,i,k,m,o,q,s,u,w), TWSA and SMA (b,d,f,h,j,l,n,p,r,t,v,x) for European (a,b,c,d), North America (e,f,g,h), South America (i,j,k,l), Asian (m,n,o,p), African (q,r,s,t,w,x), and Australian (u,v) rivers
The analysis shows that the size of the river basin is not directly related to the differences in GRACE and AMSR-E signals. Large European rivers, such as the Danube and the Vistula, show the mutual shift of hydrological signals for gravimetric and microwave remote sensors as can be seen in Figure 12 a), c). There is a more significant variance in the signal for observations from the X- and C-bands than in GRACE observations. Therefore, the determined anomalies are characterized by high noise for these ranges. Similar to the analysis performed in Kuczynska-Siehien et al. (2019), the GRACE and AMSR-E sensors pick up an anomaly related to the 2010 hydrological flood. However, contrary to the cited article, the SM determined from AMSR-E indicates the occurrence of anomalies in the years 2007–2009, which is not recorded in the GLDAS models. The snowfall in these regions during the months of DJF indicates a lower moisture content in the soil, while GRACE sensors capture the mass contained in the snow equivalent. This is explained by the method used to process AMSR-E data. Under frozen surface conditions, the dielectric properties of the water change dramatically. Therefore, the method assigns all pixels where the surface temperature is observed to be at or below 273 K with an appropriate data flag (Holmes et al., 2009).
The Nile basin shows a very high agreement between GRACE and AMSR-E signals in both X- and C-bands for the ΔTWS and SM values and their anomalies. The Pearson correlation coefficient between these variables is greater than 0.8 for this region. The lack of soil constraints and little human intervention in the form of agricultural or urban activities contribute to the consistency of observations (Gossel et al., 2004). The percentage of arable land with additional irrigation is less than 5% (Villholth, 2013). In the case of the Nile, a significant factor influencing the changes in ΔTWS is surface runoff. The weather extremes and climatic variances observed over the years using gravimetric observations indicate the high sensitivity of these sensors to extreme phenomena such as droughts (Scanlon et al., 2022, Seka et al., 2022a,b). The GRACE and AMSR-E sensors catch the 2005 and 2010 drought anomalies, shown in Figure 12 t). Similar results were also described by Seka et al. (2022b) using meteorological drought indicators and a water storage deficit index (WSDI) occurring at the source of the Nile in the Turkana, Victoria, and Tanganyika lakes.
The Congo River basin, known as Zaire, is over 60% covered by tropical forests. Crops account for only 10% of the area. The correlation of gravimetric and microwave signals is lower than at the same latitude for the Amazon basin. In this case, data collected by the AMSR-E mission detects two seasonal signal peaks, while GRACE usually has only one, as presented in Figure 12 q), r). The X-band observations for shallow soil layers do not detect a split between longer and shorter rainfall. At the same time, the C-band distinguishes subeasonal changes more like the GRACE observations. Despite the lack of soil constraints, such a large area affected by changes in precipitation caused by the movement of circulation cells poses a challenge for scientists in interpreting ΔTWS and SM observations.
The Zambezi River basin maintained an above-average consistency between GRACE and AMSR-E signals in both X- and C-bands for the ΔTWS and SM values and their anomalies also described in Thomas et al. (2014) and Hassan and Jin (2016). Similar to Thomas et al. (2014), a water deficit was observed in the area Zambezi River basin due to a hydrological drought event in April 2005. For the Zambezi and Zaire river basins, the highest amplitudes of ΔTWS and SM signals on the African continent can be observed. It can therefore be concluded, similarly to the publication of Hassan and Jin (2016), that the ΔTWS in these regions is dominated mainly by precipitation. Despite the relatively poorly urbanized area, the most important anthropogenic factors include that the Zambezi River is used to produce electricity for southern Africa. In the middle stretch of the river, there is a large artificial water reservoir called Kariba. Incremental storage of a large mass of water favors capturing this effect by GRACE sensors with a minor time frequency. Large uncovered agricultural areas and lack of factors contributing to noise in microwave observations contribute to a significant convergence of results with gravimetric sensors.
The large rivers of North America have different results for the studied similarity between gravimetric and microwave observations. The Mackenzie River basin has its source in Great Slave Lake. Located in the north of Canada in subpolar regions, the source is closely related to the snow equivalent variances visible in the GRACE observations but not included in the X- and C-bands. A similar situation will be seen in the subpolar regions of the Ob River. This result is visible in small correlations and low aggregation of ΔTWS and SM signals and their anomalies.
The Mississippi River basin has the opposite statistics compared to the Mackenzie River described previously. The high agreement of ΔTWS and SM observations, shown by the cross-correlation coefficient > 0.7, is due to the large area of agricultural crops. No limitations for soil conditions, and <10% afforestation of the area does not retain water in the vegetation and allows free seepage to groundwater. The main components of EOF3 show similar signal strength in terms of area. Observations in the X-range have a slightly more substantial phase shift than observations from the C microwave band. However, the difference is not significant in the context of the examined similarity to gravimetric observations. The high compatibility of TWSA and SMA allows both sensors to quickly monitor and predict natural disasters caused by droughts or floods (Foroumandi et al., 2022).
The Amazon basin, well described in the literature previously by Chen et al. (2009), Cui et al. (2020, 2022), Eom et al. (2017) and Wu et al. (2022), is exceptionally consistent for GRACE and AMSR-E signals despite being mostly forested. Observations in the X-band captured the 2004 anomaly, which is not visible in the C-band observations. Both bands indicated an anomaly in 2009 resulting from the exceptional flood in this area (Chen et al., 2010b) and droughts in 2010–2012 (Nie et al., 2015). As in the two previously mentioned publications, extreme hydrological phenomena from 2009 to 2012 were captured by the GRACE and AMSR-E sensors in Figure 12 j). A large area and one of the largest amplitudes of water fluctuations resulting from tropical rains occurring at equatorial latitudes cause, despite minor soil limitations, the studied signals to be characterized by considerable convergence.
The La Plata basin region is characterized by a significant anomaly in the GRACE and AMSR-E observations in Figure 4. Extremes occurring in this area require special attention during interpretation (Abelen et al., 2015, Chen et al., 2010a). The low topographic complexity facilitates penetration of the microwave signal. Secondly, a higher value of EOF3 shown in Figure 7 in the X-band AMSR-E indicates the occurrence of phenomena that in the literature can be found as the flood of winter 2009/2010. It was correlated with the occurrence of the El Niño effect and the droughts occurring in 2009. The analysis of the main components indicates that extreme hydrological phenomena have a significant effect both in gravimetric and microwave data. However, due to the differentiation in the La Plata river basin, these phenomena are characterized by a phase shift.
The Amur is the tenth longest river in the world, forming the border between the Russian Far East and Northeast China. From the north of the basin, the area covers permafrost and is covered with the boreal forest. The southern part of the area is intensively cultivated and distorted by human activity. As can be seen in Figure 11, the cross-correlation coefficient is at the level of 0.3, which proves the extremely poor compatibility of the ΔTWS and SM signals. This is also confirmed in Figure 12 m), n) presenting the time series for this area. Extreme droughts and wildfires in 2008 described previously in Semenov et al. (2017) are reflected in GRACE observations in this research, represented by an anomaly in this period, which is entirely absent in microwave observations.
The Ganges and Brahmaputra valley is intensively used for agriculture and densely populated area near the Himalayas. The monsoons in this area have become a permanent part of the landscape of the local population. ΔTWS observations in that area were previously described by Felfelani et al. (2017), Forootan et al. (2016) and Papa et al. (2015). Figure 12 o) shows a good agreement between the ΔTWS and SM signals. Cross-correlation over this basin is at a level of 0.75. However, Figure 12 p) shows the discrepancy between TWSA and SMA characterized by the opposite trend in the 2006–2012 period. A similar difference was noted in (Felfelani et al., 2017), described as a significant divergence between the SM natural and GRACE ΔTWS trend lines. As in the case of the Zaire River, the observations provided by the AMSR-E mission capture two annual waves and only one primary wave during GRACE. In this case, the differences between the X- and C-bands are smaller than in the case of the Congo Basin. Strong amplitudes of GRACE and AMSR-E signals, especially in the X-range, presented in Figure 8 a), b) indicate the intensity of SM changes in shallow layers for the largest river delta in the world.
The Murray-Darling basin is a large geographic area in the interior of south-eastern Australia with intensive farmland use around Adelaide. The area is characterized by one of the best TWSA and SMA signal correspondences observed in Figure 12 v). The decreasing trend of water content in the soil in this area was described by Heimhuber et al. (2019), Tregoning et al. (2012) and Yang et al. (2014). We see considerable agreement in the detected anomaly in 2010-2011 for gravimetric and microwave sensors. Similar conclusions as in the article by Heimhuber et al. (2019) can be obtained regarding the interpretation of the results from the period 2010–2012. La Nina Floods can be observed in higher TWSA and SMA in Figure 12 v). Unlike previous works, ASMR-E sensors did not show decreasing trend related to the 2000–2009 Millennium Drought. This is partly explained by the aggregation of data over a large river basin area and the different intensification of phenomena in the northern and southern parts of the river basin. Observations in the C-band compared to GRACE are similar in phase. Figure 9 shows a more significant shift in the observations of the X-band for its main components, which explains the shallow penetration into the soil layers for this band. However, the lack of significant global constraints, large open spaces, and small built-up areas create favorable conditions for the GRACE and AMSR-E satellites to detect the same groundwater characteristics and variances. The high Pearson correlation coefficient at the level of 0.8 and the cross-correlation of about 0.7 are visible in Figure 10 and Figure 11.
This article discusses the conditions under which the ΔTWS observations provided by the gravimetric GRACE mission are characterized by a greater or lesser signal convergence with the observations provided by the passive multiwavelength microwave sensors of the AMSR-E mission. The interplay of ΔTWS and SM can provide a better and high-resolution understanding of the Earth’s processes related to the water cycle. The complexity of land uses processes and conditions impacts the detection and mapping of natural hazards, such as droughts or floods, observed on a global or regional scale. Understanding the limitations affecting the speed of detection of changes and consistency in the observations provided using various methods and sensors has a tangible impact on the quality of the solutions provided for the prediction of geo-hazards. The main conclusions and observations from the conducted study worth emphasizing are the mutual relationship between the use of cultivated and forested areas in the ΔTWS and SM compliance analyses. Naturally forested areas and large open spaces used for agriculture support the compatibility between GRACE and AMSR-E observations. The discussion showed a high correlation for these areas, at the same time pointing to the importance of good oxygen conditions for root zones in the soil. Existing soil constraints such as permafrost significantly eliminate the usefulness of X- and C-range microwave observations. For this reason, analyses carried out in subpolar regions using gravimetric sensors have a significant advantage. The referenced examples in the subsection for Europe indicate differences between GRACE and AMSR-E in signals leading to the conclusion of unfavorable conditions resulting from soil constraints and significant urbanization of the area. Moreover, the study opens the question of spatial data leakage caused by filtering low-resolution GRACE data. Regions with high signal variance averaged over the area of the entire river basin may cause the loss of a part of the geophysical signal, which was observed and described for the example of the Zaire River. The use of mathematical methods and a combination of signals with different spatial and temporal resolutions, for areas with appropriate conditions and no soil and urban restrictions, will be the next direction of the research.