Anthropogenic CO₂ emissions assessment of Nile Delta using XCO₂ and SIF data from OCO-2 satellite

The Nile Delta is considered to be one of the oldest cultivated areas in the world, having been continuously cultivated since 3000 B.C. (Negm et al 2017). It is a triangular shaped, tide-dominated delta (figure 1(a)) occupying about 2.5% of Egypt's area (∼25 000 km²) and giving shelter to about 60% of its population (Zeydan 2005, Negm et al 2017). Moreover, 96% of Egypt's population is concentrated around Nile Delta (north of Cairo) along with Nile valley (south of Cairo) (Abd El-Kawy et al 2011). Since the construction of Aswan High Dam in 1970, the Nile Delta has received less input of nutrients and sediments from the river floods. This shift has resulted in intensive use of fertilizers for agriculture, thereby degrading the soil quality and potentially leading to loss of soil organic matter (Negm et al 2017). Furthermore, about 75 000 ha of fertile agricultural land was lost to urbanization within the Nile Delta between 1992 and 2015 (Radwan et al 2019). Nevertheless, the Nile Delta produces two-thirds of Egyptian crops (FAOSTAT 2019, www.fao.org/faostat/en/) with its still rich and fertile soil allowing two or more crops over the year (Osama et al 2017). Two main growing season in Nile Delta is the winter growing season (peak in January/February) and summer season (peak in July/August). Important winter crops are wheat, vegetable and clover crops, whereas in summer—maize, rice and cotton are extensively grown (El-Beheiry et al 2015). The Nile Delta today is mostly characterized by artificially irrigated agriculture and spreaded urban settlements (figure 1(a)), with minimal natural sources or sinks of CO₂.

Zoom In Zoom Out Reset image size

We used five years of XCO₂ data from NASA's OCO-2 satellite. The OCO-2 satellite was designed to accurately measure XCO₂ at kilometer-scale spatial resolution around 13:30 local time with a repeat cycle of 16 d (Crisp 2015) and 10 km swath (Bhattacharjee and Chen 2020). The satellite uses the Atmospheric CO₂ Observations from Space (ACOS) algorithm (Wunch et al 2011a, 2011b, Crisp et al 2012, O'Dell et al 2012). We used the high quality measurement soundings of the latest version of the OCO-2 Level 2 Version 9 Lite product from September 2014 to July 2019 provided by the NASA's Jet Propulsion Laboratory website (https://co2.jpl.nasa.gov/build/?dataset=OCO2L2Stdv8&product=FULL#download). The measurement soundings are obtained using an improved bias-correction approach and are pre- and post-filtered for potentially unreliably measurements (O'Dell et al 2018). The downloaded measurement data also contain surface wind velocity information at each measurement sounding taken from the NASA's GEOS5-FP-IT (Goddard Earth Observing System Version 5-Forward Processing for Instrument Teams). Each trajectory of OCO-2 measurement covers both the delta and desert area (figure S1 (available online at stacks.iop.org/ERL/15/095010/mmedia)), which enables us to quantify the spatial (delta-desert) XCO₂ gradient.

In addition to XCO₂ data, we also employed OCO-2's SIF (Solar-induced fluorescence) data. SIF is a signal emitted from plants/crops during photosynthesis, thus indicating photosynthetic activities (Sun et al 2017, Shekhar et al 2019, Castro et al 2020). OCO-2 measures SIF at two wavelengths, 757 nm and 771 nm (Frankenberg et al 2014, 2015), based on the infilling of the Fraunhofer lines (Frankenberg et al 2011). We used SIF (measured at 757 nm) to help understand how the CO₂ emissions from the Nile Delta were related to agricultural activities.

We computed the expected values of XCO₂ over the Delta, compared to surrounding desert areas, using two CO₂ emission inventories, EDGAR (Emissions Database for Global Atmospheric Research) and ODIAC (Open-source Data Inventory for Anthropogenic CO₂). Both are constructed based on the IPCC methodology. The latest version of EDGAR v5.0 (Crippa et al 2020) was recently made available. It provides global annual CO₂ emissions at 0.1° × 0.1° spatial resolution until 2018. EDGAR emissions are calculated using a bottom-up methodology: the inventory scales activity data, which is based on international annual statistics with the best-available emission factors. Then, it uses the monthly share and spatial proxies such as population and road density to downscale the national or state-level data to a finer spatial resolution. We used the CO2_excl_short-cycle_org_C variable of the EDGAR v5.0 which includes total CO₂ emissions (in kgCO₂ (m² s)⁻¹) from the considered fossil fuel sources such as fossil fuel combustion, various non-metallic mineral processes like cement production, various metallic and chemical production processes, urea productions and agricultural liming processes. The EDGAR v5.0 dataset was downloaded from the European Commission's Joint Research Centre-EDGAR website (https://edgar.jrc.ec.europa.eu/overview.php?v=50_GHG). Mean EDGAR emissions over the Nile Delta from 2014–2018 was 108 Mt CO₂/year.

The latest ODIAC v2019 provides monthly CO₂ emissions (in tons of Carbon/(km²-month)) from fossil fuel combustions at a much higher spatial resolution of 1 km × 1 km (GeoTIFF file format) till the end of 2018 (Oda et al 2018). A hybrid approach is implemented to construct the ODIAC emission inventory, which integrates national emission estimates produced by CDIAC (Carbon Dioxide Information Analysis Center, Andres et al 1996) and nightlight data with individual power plant emissions/location profiles (Oda and Maksyutov 2011). The ODIAC v2019 dataset was downloaded from the CGER-NIES (Center for Global Environmental Research, National Institute for Environmental Studies) website (http://db.cger.nies.go.jp/dataset/ODIAC/DL_odiac2019.html).

Since neither the EDGAR nor the ODIAC dataset was fully available for the time period of this study (September 2014–July 2019), we used the latest available data point for each inventory for the remaining time period, i.e. the EDGAR and ODIAC estimates of 2018 were used for the year of 2019. It is important to note that neither of these two emission inventories include CO₂ emission from agricultural soil degradation. Both EDGAR and ODIAC emission datasets are widely used by the international carbon cycle research community (Quéré et al 2018). Mean ODIAC emissions over the Nile Delta from September 2014–July 2019 was 136.5 Mt CO₂/year. Figure S3 shows the CO₂ emission map for Nile Delta as per EDGAR and ODIAC.

To simulate the enhancement of XCO₂ values from the local CO₂ emissions sources we employed a forward model based on the emission inventories and an atmospheric transport model.

For any location, the XCO₂ can be modelled as the sum of the initial background concentration and the enhancement due to the upstream sources and sinks (Gerbig et al 2003, Hu et al 2018a). Equations (1) and (2) describe our calculations:

$$\begin{equation}XC{O_{2.fm}} = XC{O_{2.bkg}} + XC{O_{2.enh}}\end{equation} \tag{ 1 }$$

$$\begin{align}XC{O_{2, enh}} = & \ \mathop \sum \limits_{i = 1}^{{N_x}} \mathop \sum \limits_{j = 1}^{{N_y}} \left[ footprint\left( {{x_i}, {y_j}} \right) \right. \nonumber\\ & \left.\times emission\ \left( {{x_i}, {y_j}} \right) \right] \end{align} \tag{ 2 }$$

where, XCO_2,fm is the forward modelled concentration for any location, XCO_2,bkg is an initial background XCO₂ value and XCO_2,enh is the enhancement due to upstream CO₂ emission sources. The 'footprint' is obtained from an atmospheric transport model and the 'emission' data are obtained from the emission inventories (more information below). Therefore, XCO_2,enh is the summation of the enhancements caused by all locations (x_i,y_j): i = 1,2...N_x; j = 1,2...N_y., in the study domain containing the Nile Delta and valley (Latitude:[26, 32]; Longitude:[29, 33]). For the case of simplicity here, we have neglected the effect of sinks on XCO_2,enh as the Nile Delta is characterized by minimal or no natural vegetation, and net emissions are plainly positive. To study the XCO₂ enhancement in the Nile Delta we applied a local regression smoothing (LOESS, Jacoby 2000) with loess parameter (span) of 0.15 to the XCO₂ concentrations along the OCO-2 trajectory (latitude-wise) for each measurement day, and these locally regressed XCO₂ (XCO_2,O) values were used for further calculation.

Since OCO-2 measures XCO₂ along a narrow swath of 10 km (figure 1), selection of background concentration (XCO_2,bkg) can be tedious. In this study, we leveraged the wind data from ECMWF's ERA5 dataset (Copernicus Climate Change Service (C3S) 2017) to select a particular section of one degree latitude range (background section) of XCO₂ values in the desert that did not pass over the Nile Delta as XCO_2,bkg. The selection of 'background section' was semi-automated for all the 32 dates. The semi-automated process involved visual inspection and automated selection of the background section. The detailed procedure is given in supplementary section S1. Figure 2 illustrates the background section for few dates. For example on 2014-09-23, the wind direction is north, contaminating the proximate observations with CO₂ from the Delta, so we selected the lower section of OCO-2 measurements (shown in green box in figure 2), where the wind was coming from the east side. Similarly, for 2015-05-19, the selected section does not have any influence from Nile Delta and minimal influence from the Nile valley. The mean of the locally smoothed XCO₂ values within this section was taken as a XCO_2,bkg for each day. The selected background section for all the other dates is shown in figure S3.

Zoom In Zoom Out Reset image size

Calculation of XCO_2,enh requires footprint information from the atmospheric transport model, and local emission data. The geographic distributions of emissions were adopted from EDGAR and ODIAC (section 2.2). Here, we simulated the footprints using the STILT (Stochastic Time-Inverted Lagrangian Transport) model that is based on HYSPLIT model (Lin et al 2003, Fasoli et al 2018, Wu et al 2018). The STILT model simulates convective and diffusive air transport patterns by releasing a certain amount of particles (N) from a receptor (i.e. an observation location in this case) and tracking the particles backwards in time (Hu et al 2018b). Thus, the STILT footprints are equivalent to the sensitivity of XCO₂ (unit of ppm·m²·s·µmol⁻¹) at the receptor location with respect to changes in surface flux (Gerbig et al 2003).

Our STILT simulation was based on ERA5 meteorological data and we chose a backward time of 32 h and N = 500 particles and 11 vertical levels for each receptor location. Figure 3 shows column averaged STILT footprints for 2014-09-23 and 2016-02-15 at 13:30 (local overpass time of OCO-2). Finally for each OCO-2 trajectory we chose 5 receptor locations within the Nile Delta at a distance of 10, 15, 20, 25 and 30 km along the OCO-2 trajectory to calculate XCO_2,fm (in ppm), which is the accumulated CO₂ enhancement occurring over the 32 h back trajectory at the receptor location due to emissions from the upstream sources. Figure 1 shows the receptor locations along the delta and within the background section for OCO-2 measurements on 2016-02-15. It is also evident from figure 3 that the footprints of the background receptor are not influenced by the Nile Delta area. Ultimately we compared the XCO_2,fm with the LOESS smoothed OCO-2's XCO₂ value at the receptor locations (XCO_2,O) for each OCO-2 trajectory over study period (2014–2019) to estimate CO₂ emissions based on the two inventory data.

Zoom In Zoom Out Reset image size

We studied 62 d of OCO-2 measurements over the Nile Delta and desert area spanning five years of observations. We observed the expected steady seasonal cycle in XCO₂ concentration over the delta (XCO_2,del) and the desert (XCO_2,des) from 2014 to 2019 (figure 4). About 80% of the measurements showed a mean XCO_2,del of 1.11 (0.83,1.21) (95% confidence interval) ppm higher than mean XCO_2,des with a maximum ΔXCO₂ (XCO_2,del—XCO_2,des), of 2.2 ppm. Expectedly, consistent higher SIF values in the Nile Delta indicated the agricultural production (figure S4), with high SIF values during the months of peak of winter growing season (DJF) and summer growing season (JJA), but the latter season showed significantly lower ΔXCO₂ values (figure 5). This seasonal pattern of ΔXCO₂ can be attributed to the river flow dynamics and crops grown in the winter and summer season. The delta receives sediment and carbon-rich water during the winter season from the Aswan high dam (El Gamal and Zaki 2017), which is recycled and also used in the summer season. The emissions from newly arrived carbon-rich soil (carbon source) could result in high ΔXCO₂. However in summer, high photosynthetic activity of well-irrigated maize and rice during the mid-day could be a significant carbon sink (Rana et al 2016, She et al 2017) with reduced CO₂ emission from soil, thereby resulting in lower XCO₂ values in the delta. A detailed study combined with cropping pattern and river flow dynamics could give insight that is more accurate. On average higher SIF values correlated with higher XCO₂, which might not be expected for forest, but because of all the fossil fuel and soil CO₂ emissions associated with the farming and irrigation in the Nile Delta (figure 5 and S5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The OCO-2 clearly captured the latitudinal variation of XCO₂ concentrations along its trajectory as depicted in figure 6. Transportation of CO₂ plumes from emission sources in the delta region including Cairo city and Nile valley by the wind resulted in comparatively high XCO₂ values in the downwind desert area. For example, on 2015-05-19 the CO₂ emissions from Cairo results in as much as 3 ppm higher XCO₂ values in the downwind desert area compared to the southern desert, with minimum influence from the Nile Delta or valley (figure 6). Furthermore, emissions from the Nile valley clearly resulted in elevated XCO₂ values in the downwind desert area, for example on 2018-11-03 (figure 6, latitude: ∼26.5 deg.). Such wind transport of CO₂ plume resulted in high XCO₂ values in the desert region. Thus, OCO-2's XCO₂ measurement clearly demonstrates the capability to measure local enhancement in XCO₂ values.

Zoom In Zoom Out Reset image size

Out of the 62 d of OCO-2 measurements, many days had sparse measurements over the Nile Delta and desert area, making it difficult to select background XCO₂ values. Therefore, before performing the forward modeling of emission inventories to calculate XCO_2,m, we applied a threshold criteria of minimum 100 and 200 measurements over the Nile Delta and desert area, respectively. Finally, 32 d satisfied the above criteria. Correspondingly, we performed forward modeling calculations on 5 locations at 32 different dates from 2014 to 2019. Two out of 32 d (2017-06-09 and 2018-07-07) showed consistently higher measurement values in the desert area resulting in higher XCO_2,bkg values compared to the XCO_2,O values at the receptor locations. These two dates occur during the peak of summer growing season and might indicate the agricultural crops grown (e.g. maize) in summer as a CO₂ sink.

For all 32 dates at 5 locations (total 160 receptor locations), the observed enhancements (XCO_2,O—XCO_2,bkg) averaged 1.21 (1.05,1.36) ppm. The calculated enhancements were much lower, 0.27 (0.23,0.32) ppm and 0.36 (0.31,0.43) ppm based on EDGAR and ODIAC, respectively (figure 7). Very prominent enhancements of as high as 1.7 ppm and 4 ppm by EDGAR and ODIAC, respectively were obtained for receptor location in the Cairo city (figure 7).

Zoom In Zoom Out Reset image size

Over the 32 d of study, 23 d showed underestimation of modelled enhancement (figure 8), which indicates high underestimation of CO₂ emission by the emission inventories. Our calculations indicate an underestimation by a mean emission factor (MEF) (95% confidence interval) of 4.53 (4.34, 4.68) and 3.36 (3.19, 3.45) (calculated as a ratio of the mean observed enhancement over all the days, divided by the mean modelled enhancement for all days, and bootstrapped) for EDGAR and ODIAC, respectively. Based on these calculated MEFs we also estimated the CO₂ emissions from the Nile Delta based on scaling up the two emission inventories. Our estimates of mean CO₂ emission are about 489 Mt CO₂/year and 457 Mt CO₂/year based on EDGAR and ODIAC inventory over the study period (September 2014–July 2019).

Zoom In Zoom Out Reset image size

Our estimate compares relatively well with recent CO₂ emissions of about 1500 Mt CO₂/year estimated from Yangtze River Delta in 2013 based on remote sensing (nighttime light) and statistical data (socioeconomic indicators) by Zhou et al (2019) as shown in table 1. The Yangtze River Delta is about 4 times larger than Nile Delta and is similarly characterized by intense urbanization and agricultural activity. However, our emissions estimates are quite high as compared to values obtained for the Pearl River Delta (table 1). Xu et al (2018) used urbanization indicators (like land, population and economic urbanization) to estimate carbon emissions in Pearl River Delta and also highlighted uncertainties in emission estimates due to selection of indicators (Peng et al 2017). None of the above studies included emissions from exploited/degrading carbon-rich agriculture soils, which may be a significant source in Nile Delta.

Uncertainties in CO₂ emission inventories have been reported for developing countries, e.g. India and China (Guan et al 2012, Xue and Ren 2012, Janardanan et al 2016), and have mostly been attributed to uncertainties in the national level energy statistics and human activity data used to construct inventory data sets. Furthermore, the plot of ΔXCO₂ and SIF (indication of agriculture activity) in figures 5 and S5, clearly shows seasonal differences, especially for the two main growing season (winter and summer) with lower XCO₂ for the Delta during the summer growing season. We also calculated the seasonal MEFs for each season and found that the factors for the JJA season (MEF ∼ 1.05) was significantly lower than MAM season (MEF ∼ 5–8), and SON season (MEF ∼ 6–11) (see table S1). However, for DJF season, only two OCO-2 measurements dates have sufficient observations over the Nile Delta and desert area for the emission assessments (2015-02-15 and 2017-02-03). There are very high differences between modelled and observed XCO₂ enhancements (>2 ppm in figure 8), which resulted in very high MEF (∼30 to 50, see table S1), which may be biased due to less measurement samples. Nevertheless, such strong signal of seasonal variation of XCO₂ can be attributed directly or indirectly to agricultural farming and irrigation activities associated with the sediment-rich river inflow that occurs during the winter season.

Direct agriculture emissions include tillage enhanced soil respiration and crop residue burning, whereas indirect emission comes from fossil fuel consumption by farm machineries for various intercultural activities like tilling, weeding, harvesting etc, and pumps for irrigation. In 2010, as per report from EAS (2014), irrigation pumps (mostly diesel pumps) were used in 75% of the cultivated area. El-Gafy and El-Bably (2016) measured an average emission of 6.55 tCO₂/ha from on-farm diesel pumps used for irritation.

Apart from agricultural activities, anthropogenic emissions from transportation, industrial and household use (combined energy statistics) of various distributed cities in the Nile Delta might also be under-reported in these global emission inventories (Guan et al 2012). Gately and Hutyra (2017) argued that global emission inventories (like ODIAC) appear to deflect from spatial patterns in source activity in comparison to national scale inventories, and thus the former might not accurately represent anthropogenic emissions at subnational or urban scales. Our calculated delta emissions are based on yearly (for EDGAR) and monthly (ODIAC) average of CO₂ emissions, whereas the OCO-2 XCO₂ measurements are instantaneous, at 13:30 local time. This can result in under-/over-estimation of calculated enhancements as CO₂ emissions vary diurnally, weekly and seasonally, especially for urban areas. Burri et al (2009) measured CO₂ fluxes using the eddy covariance method at the University of Cairo and showed that the CO₂ fluxes have a significant diurnal and weekly variation, with peak CO₂ flux occurring between 14:00 and 16:00 and minimum flux on Friday (rest day due to Muslim prayer day on Friday). In our study, ODIAC performed a little better compared to EDGAR, as the former gives monthly average estimates and thus has the capability to capture the seasonal variability of CO₂ emissions, which cannot be implied for the latter.