Evaluating irrigation status in the Mekong Delta through polarimetric L-band SAR data assimilation

Highlights

• A technique to detect soil-submersion covered by rice plants were introduced.

• Assimilation of fine-resolution SAR data based on spatio-temporal ergodicity.

• A technique was implemented to adaptatively optimize states with different scales.

• Stable assimilation even with sparse observation by using inundation probability.

• The estimated irrigation parameters showed a consistency with ground observed values.

Abstract

For sustainable food production in the Mekong Delta, local information on irrigation status is essential for allocating water resources efficiently at the community level. ALOS-2 PALSAR-2 L-band SAR can be used to detect submerged and nonsubmerged soil covered by rice plants at a spatial resolution compatible with field observation but a low temporal resolution. In this study, a new multiscale data assimilation technique is developed to estimate the spatiotemporal dynamics of field water levels at a temporal resolution appropriate to inform decision-making on when to initiate irrigation. The method employs an irrigation model based on parameters representing farmers' irrigation practices. ALOS-2 PALSAR-2 data are used to derive the hydrological parameters of the model, including an irrigation parameter representing how deep the field water level dropped until the next round of irrigation was initiated. We developed observation operators for the soil submersion status under vegetation and spatial submersion percentages instead of assimilating soil moisture products. The study uses ALOS-2 PALSAR-2 data (25–100 m spatial resolution, approximately every 42 days) over the Mekong Delta and experimental data collected in situ for model initialization and validation. The estimated irrigation parameter controlling the maximum depth of field water level showed spatio-temporal consistency with the ground-observed value (RMSE = 4.24 cm). The values of the parameters also showed spatial consistency with respect to DEM data; he paddies with low irrigation model parameter values were prone to be located in low-elevation zones (<2 m), whereas high model parameters values tended to be located in paddies in the high-elevation zones (>3 m).The results show promising applications using L-band SAR observations for monitoring paddy field water level, for irrigation practices and for estimation of the water consumption and of methane emissions.

Introduction

The objective of this study is to develop a SAR-based monitoring system that provides daily paddy field water level data for irrigation management, required in the simulation of GHG emissions and rice productivity. In Asia, rice is the most important staple food, providing an average 32% of total calorie uptake (Maclean et al., 2002). About 90% of the global paddy area and annual output of rice production are concentrated in monsoonal Asia (FAOSTAT, 2022). Approximately 75% of global rice production is produced in irrigated lowlands encompassing delta basins in Asia, such as the Mekong, Irrawaddy, Chao Phraya, and Bengal Rivers (Barker et al., 1985; Maclean et al., 2002). To meet the increasing food demand derived from global population growth, rice cropping in the region is becoming more intensive with the use of double/triple rice cropping techniques (FAO, 2020). However, rice production requires large amounts of water (3000–5000 L kg⁻¹ rice, IRRI, 2001) and has become a major source of the potent greenhouse gas methane, approximately 11% of anthropogenic methane emissions come from rice paddy submerged soils (IPCC, 2013). Therefore, water-saving irrigation practices that have the potential to mitigate methane emissions by oxidizing the soil environment, such as alternate wetting and drying (AWD), are desirable for dissemination in this region to ensure sustainable water demand while lowering greenhouse gas emissions.

AWD is an intermittent irrigation technique to lower the amount of water used for irrigation without a significant loss of rice productivity (Bouman et al., 2007; Lampayan et al., 2015). This approach consists of temporarily suspending irrigation until the paddy soil becomes dry/oxidized. Initially, irrigation water is applied to obtain flooded conditions (5–10 cm of water above the soil surface). Then, after a certain number of days have passed, the field water level drops below the soil surface. The next irrigation is scheduled before the soil becomes dry. However, the perception of soil dryness and consequently the number of days between 2 irrigations varies widely among farmers and sites (from 1 day to >10 days) depending on the soil texture (clay/silt/sand content), percolation rate, precipitation and evaporation (Bouman et al., 2007, Lampayan et al., 2015). Various soil drought indexes have been used for scheduling irrigation (Bouman et al., 2007); for example, the soil matric potential is used to define the value of the soil drought index as the irrigation criterion. Scientists have also advocated for the use of the farmer-determined water level has also been suggested (hereafter referred to as D_{before_irrigation}, i.e., how deep the field water level drops before the next round of irrigation is initiated). This specific water depth has been widely used by policy makers regarding irrigation activity, as rice farmers can easily monitor this value (Bouman et al., 2007; Rejesus et al., 2011). To define D_{before_irrigation}, several field experiments have been conducted in Asia (e.g., Arai et al., 2021; Belder et al., 2004; Bouman and Tuong, 2001; Lampayan et al., 2015). These studies report that increasing the interval between rounds of irrigations to allow the field water level to drop to deeper soil layers is a promising technique to save water, reduce methane emissions and increase rice yield in the Mekong Delta (Arai, 2022). ‘Irrigation when the field water level has leached 15–20 cm below the soil” is the criterion suggested by the International Rice Research Institute (IRRI). In Vietnam, in particular the Mekong delta, highly intensive cropping system was developed based on the construction and upgrading of a multilevel dike system, allowing double and triple rice crop per year. The AWD practices have also been adopted in certain regions as a water saving regulation.

However, despite the wide adoption of AWD, the values of D_{before_irrigation} differ by location (Ishido et al., 2016; Taminato and Matsubara, 2016). Most importantly, the adoption of AWD by farmers is constrained by several social/infrastructural factors. These factors include distance from the irrigation canal/drainage, density of subcanals inside the dikes, and location of pump ownership (Yamaguchi et al., 2017). At such community scale (i.e., 10-100 m spatial resolution with daily temporal resolution), heterogeneity is caused by nonsynchronized intermittent drainage in adjacent rice paddies, in land surface elevations and in cropping calendars (Dang et al., 2016; Evers and Benedikter, 2009, Miyashita et al., 2016). Other natural factors arise, such as irregular daily field water level dynamics of the irrigation canal due to ebbs and flows of tides (Wassmann et al., 2004).

These heterogeneous water management practices cause difficulties in monitoring/verifying mitigation measures operations (e.g., AWD). Such monitoring is required for irrigation management, such as allocation of irrigation/drainage canals and pumping apparatuses for irrigation/drainage schedule synchronization, opening/closing schedules of water gates attached to each dike surrounding rice paddies, dike heightening, selection of paddies that require field water level observations, etc.

In this context, there is a need to monitor the irrigation status [i.e., field water level dynamics and farmers' irrigation practices (e.g., D_{before_irrigation})] to evaluate the actual dissemination status of water-saving irrigation activities by each stakeholder and verify the longevity of already implemented mitigation measures. To address this issue, various studies have been conducted using remote sensing techniques to evaluate the soil submersion status. Using optical data, the rice paddy covered with water can be distinguished at the beginning of the season based on the reflectance of the water surface. The method often uses the normalized difference water index (NDWI) and the normalized difference vegetation index (NDVI) from remote sensing images to depict the differences between rice paddies and nonrice paddy areas (Zhang et al., 2020). However, as soon as the rice paddy is covered by vegetation, it is no longer possible to detect flooded and nonflooded paddy fields. Using SAR data with penetration capability, soil submersion/nonsubmersion is expected to be detectable in rice paddies for a longer period during the rice growth stage, depending on the SAR wavelength. Using C-band SAR data, Lam-Dao (2009) reported that soil submersion could be detected up to 20 days after sowing using the HH backscattering intensity of Advanced Synthetic Aperture Radar (ASAR). Using longer-wavelength SAR data with deeper penetration in the vegetation layer, Arai et al. (2018) demonstrated the performance of quadruple/dual-polarimetric L-band SAR (ALOS-2 PALSAR-2: Advanced Land Observing Satellite-2/Phased Array type L-band Synthetic Aperture Radar-2) data to distinguish submerged paddies and nonsubmerged paddies across all rice growth stages. However, the high resolution polarimetric images cover limited area (40 km swath) and irregular temporal frequency. For larger region, ScanSAR data are more adapted. However, the 42-day repeat cycle by which PALSAR-2 ScanSAR data are acquired is not sufficient to monitor the temporal dynamics of paddy field water levels. Future L-band SAR missions, such as ALOS-4, ROSE-L and NISAR, will have 14- and 12-day repeat cycles, respectively, will be still not sufficient for temporal monitoring. Many hydrological simulation studies (e.g., Al-Yaari et al., 2014; Ines et al., 2013; Montzka et al., 2011) applying data assimilation (hereafter, assimilation) have used high temporal resolution passive microwave radiometer data [e.g., Soil Moisture and Ocean Salinity (SMOS), in which the L-band (1.4 GHz) data are collected every 3 days]. However, their spatial resolution of 20–50 km is not adapted to the heterogeneity of the rice paddies in Asia. In this regard, it is very important to develop a technique using high spatial resolution SAR data at low temporal resolution to simulate the temporal dynamics of the irrigation status of rice paddies.

The objective of this study is to develop a system that provides daily field water level data for irrigation management at the community scale. The starting point is the use of ALOS-2 PALSAR-2 data which provide pixel-based information on submersion status of paddy fields every 42 days. We prepared a model simulating field water level dynamics, based on knowledge of the hysteresis of the soil water retention curve and our long-term ground observation data. and we use the SAR data to optimize model parameters. To estimate the values of D_{before_irrigation} at the community scale, we assimilate the backscattering coefficient of the ALOS-2 PALSAR-2 data by preparing a pixel-based observation operator to detect the submersion status of paddy fields. To solve the filter divergence issues that become critical due to the low temporal resolution of satellite observations, we designed a focal statistics-based observation operator to evaluate the temporal submersion percentage and the spatial submersion percentage in a focal analysis window.