HIGH RESOLUTION SURFACE SOIL MOISTURE FROM C- AND L-BAND SAR DATA TO DETECT IRRIGATION EVENTS

Irrigation is a crucial practice for agriculture, especially in arid and semi-arid regions, because it considerably influences the crop production and quality. In this respect, a detailed localization of irrigated croplands provides important information for scientific studies on the hydrologic cycle, water supply and climate change [e.g., 1]. Irrigation accounts for 85%-90% of global anthropogenic water consumption and, according to recent projections, global irrigated agriculture could double by 2050 due both to population growth and to climate change [2]. Therefore, having a periodically up-to-date estimate of irrigated areas could be important to pursue sustainable management of water resources. Earth Observations (EO)-based geo-information can largely contribute to more informed management of water resources. For instance, the space and time distribution of irrigated areas is a necessary input to simulate the water withdrawal which is vital information for effective water management [3]. This paper assesses the use of surface soil moisture (SSM) at high resolution (~100m) derived from Synthetic Aperture Radar (SAR) observations to identify in time and space and classify irrigated fields. A change detection approach implemented in the SMOSAR code [4] is used to obtain the SSM fields. The proposed methodology to identify irrigated areas is based on the comparison of the soil wetness level computed at two spatial scales. Indeed, a relative measure of SSM, i.e. the SSM contrast, C, is needed because the absolute SSM value is highly influenced by the characteristics of the site, e.g., climate, orography, soil properties, agriculture management, etc. For these reasons, the contrast between the SSM levels of irrigated and non-irrigated fields is a more robust metric, indicating whether or not a field is significantly dryer/wetter than the surrounding area. The criterion proposed for irrigation discrimination is that C must exceed a threshold level. The performance is tested over the irrigation district of Riaza in the Castilla y León region (Spain), and over two Italian sites, i.e. Apulian Tavoliere (Southern Italy) and Jolanda di Savoia (Northern Italy). Over the sites, detailed ground information is available, in particular, the crop maps, parcel borders and the list of events with timing information about irrigation, i.e., the date and time of irrigation. The SSM fields at high spatial resolution are derived from a time series of C-band Sentinel-1 images spanning from 2017 to 2022. Besides, time series of L-band SAOCOM data acquired over the Italian sites in 2022 are used in the classification. The output of the thresholding applied to the time series of SSM contrast consists of binary maps of irrigated/non-irrigated fields. The validation analysis consists of comparing the binary maps against the ground data available. The statistical scores are based on the computation of the confusion matrix and related accuracies of the irrigated and non-irrigated classes with respect to the available ground data. Figure 1 provides an example of an irrigated/non-irrigated field map on April, 22 2017 (middle panel) and the cumulative map (bottom panel) between April and June 2017 over the Riaza irrigation district where the number of the detected irrigation events per field is reported. The paper presents a comprehensive assessment of the classification algorithm over the experimental sites and discusses the main differences in the classification using C- and L-band data. The study demonstrates that SAR data are extremely valuable for the early detection of irrigated areas (well before vegetation indexes, derived from multispectral EO data, can even detect the presence of a crop). Besides, SSM can resolve the irrigation event in time, which remains unfeasible for vegetation indexes. This can be valuable information for irrigation management as it can, for example, support the detection of supplemental irrigation that is expected to significantly increase for rain-fed crops, such as wheat, to better cope with water scarcity [5]. ACKNOWLEDGMENTS This study was supported by the project "Use of multi-frequency SAR data for AGRIculture" (SARAGRI), funded by ASI under contract N. 2021-6-U.0. The authors also acknowledge ASI for providing the SAOCOM data under the ASI-CONAE SAOCOM License to Use Agreement.