Using SAR Data to Detect Wheat Irrigation Supply in an Irrigated Semi-arid Area

The objective of this study was to use SAR (sensitivity of Synthetic Aperture Radar) data to detect the supply of irrigation water during the anthesis and grain-filling phenological stages of wheat in the irrigated Tadla perimeter of Morocco. Backscattering coefficients were derived from four ERS-1 (European Remote-Sensing Satellite-1) images acquired between 31 March and 12 April 2011 and were compared with the irrigation water invoices database. The analysis showed that there were significant changes in backscattering values caused by irrigation, with average values ranging between 0.11 and 3.11 dB. A reference level of 0.52 dB was established for differentiating between (recently; up to 4 days) irrigated and non-irrigated plots. We also set an interval of 5 days for the acquisition of SAR images in order to ensure continuous monitoring of the irrigated wheat plots over time. The study showed that radar data contain important information for the assessment of irrigation supplies during the cropping season, which could help regional decision-support systems to monitor and control irrigation supplies over large areas.


Introduction
Irrigated areas throughout the world are facing increasing pressure due mainly to erratic precipitation regimes (Dore, 2005), long dry periods and rapidly growing population demands.In this context, the effective management and monitoring of irrigated areas require a good understanding of the spatial and temporal processes governing agricultural systems.
Managing and monitoring an irrigated area effectively can be done by analyzing these processes over an entire crop cycle and a large agricultural area where the surface is heterogeneous (various types of crops, classes of soil and management approaches) in order to assess the overall impact of crop management practices (Lionboui, Benabdelouahab, Hasib, Elame, & Boulli, 2018).
In Morocco, cereals are one of the major grain crops grown and they hold an important place in the agricultural production systems, occupying 75% of the cultivated area and accounting for 10-20% of the agricultural Gross Domestic Product (GDP).Nevertheless, yields remain low and fluctuate from one area to another and one season to another because of varying water management, field management and weather conditions (Benabdelouahab et al., 2016).
Given the importance of wheat production in semi-arid areas where water is the main limiting factor, large-scale irrigation of wheat is common practice.There is therefore a need for good management of irrigation supplies in order to improve irrigation scheduling and prevent water stress from adversely affecting yield (Lionboui, Benabdelouahab, Elame, Hasib, & Boulli, 2016).Remote sensing satellites can be used for monitoring land surface changes because of their extensive coverage capacity and frequent revisits (Benabdelouahab et al., 2015;Fieuzal et al., 2011;Kalluri, Gilruth, & Bergman, 2003;Ozdogan, Yang, Allez, & Cervantes, 2010).
The recent launch of Sentinel-1, which offers both high spatial resolution and frequent revisits, could be interesting (Aulard-Macler, 2011;Snoeij et al., 2008).It is still difficult, however, to acquire synchronous multi-sensor time series in order to analyze satellite data sensitivity over comparable surface conditions.C-band data are available from the ERS-1/2, EnviSat, Radarsat-1/2 and Sentinel-1 systems.
For wheat canopies and topsoil moisture, the sensitivity of radar backscattering coefficients was demonstrated by Mattia et al. (2003) and Picard, Le Toan, and Mattia (2003).Some attempts have been made to use simplified relationships between SAR backscattering coefficients and wheat canopy characteristics (Dente, Satalino, Mattia, & Rinaldi, 2008;Mattia et al., 2003).Few authors, however, have tried to apply the radar on a large scale and in a representative context (Fieuzal, Baup, & Marais-Sicre, 2013).
At C-band frequency, the temporal behavior of wheat in different studies (Mattia et al., 2003;Saich & Borgeaud, 2000) shows similar global trends for the same study site.Recent studies have proposed multi-sensor approaches for irrigation management purposes by combining optical and radar images (Fieuzal et al., 2011;Hadria et al., 2009;Hadria et al., 2010).
In irrigation monitoring, it is thought that the backscattering signal generated from SAR images reacts to changes in surface moisture (Hadria et al., 2009), which could be important indicator to detect irrigation inputs and monitoring surface moisture on a large scale and in a realistic context during the anthesis and grain-filling phenological stages of wheat.This information could be very useful in improving national grain yield forecast models that currently do not take into account production from irrigated areas, despite the fact that they occupy over 1million ha.Irrigation water supply data (time, duration and irrigated area) could be integrated into yield forecast models in order to improve their accuracy.
In this context, we conducted an analysis of a large number of agricultural plots using time series of SAR images in order to assess their sensitivity to surface moisture.This was done by evaluating the values of the backscatter signal compared with the variability of the surface moisture that is closely related to the irrigation supply program at wheat plot level (databases dates irrigation).

Methodology
Previous studies using SAR images have shown the potential of using backscattering signals to monitor vegetation water content and surface soil moisture via a simple linear relationship and one incidence-angle data (Fieuzal et al., 2011;Mattia et al., 2003;Zribi et al., 2005).Drawing on these studies, the methodology adopted in our study sought to detect irrigation water supplies to the wheat crop using SAR data.

Study Site
The irrigated Tadla perimeter (Figure 1) is in central Morocco, between the Atlantic coast in the north-west and the Atlas Mountains in the south-east (32°23′ N latitude; 6°31′ W longitude; 445 m above sea level).The studied area is characterized by a semi-arid climate; the annual average temperature is about 19 °C, with a large inter-seasonal variation (max = 38 °C in August and min = 3.5 °C in January).The average annual precipitation is about 300 mm (average over the 1970-2010 period), with significant inter-annual variation (from 130 mm to 600 mm).The processing level of the acquired images was (1B), which included radiometric and geometric corrections.

Supervised Classification
In order to define the cereal area, we used a maximum likelihood classification method that is a widely used supervised pixel-based method (Ouyang et al., 2011).
Training areas representative of the land cover classes were selected in order to develop class signature files.For each image, training areas were defined based on a field survey, expert field knowledge and ancillary data (tree crops mask and irrigation canals).Two thirds of the training areas were used in the classification process, remaining one third in the accuracy assessment.The main classes were: cereals; bare soil; industrial crops; perennial crops; and arboriculture.
For the wheat class, we selected a sufficient number of pixels representing 1.08% of the total pixels (500,123 pixels) (Yang, Everitt, & Murden, 2011).We performed the separability analysis, using the Jeffries-Matusita distance, for training samples in the final classification scheme with values of separability between 1.99 and 2, indicating good class separation.
The contingency matrix was used to evaluate the percentage of sampled pixels classified as expected.User accuracy and producer accuracy regarding the wheat class were 87.8% and 86.73%, respectively.The overall accuracy assessment and Kappa values were 85.7% and 0.84%, respectively, indicating good classification.

Integration and Intersection of Ground and Satellite Data
From the SAR data we obtained an amplitude time series.We also analyzed archival data of the irrigation supply schedule used by farmers in the irrigated Tadla perimeter at the plot level.The archival data was provided by ORMVAT.
The backscatter parameter from the ERS-2 images was averaged for each of the 341 training plots.This was followed by crossing all the information layers to monitor and analyze the spatiotemporal evolution of backscattering intensity, depending on irrigation water supplies used by farmers for their wheat plots.Figure 3 summarizes these steps. jas.ccsenet.

Data A
A referenc set an inte wheat plot regression more than vegetation water content and surface moisture at local and regional level in the irrigated Tadla perimeter.These results show that radar signal behavior can be generalized, especially where wheat plot conditions are fairly homogeneous.

Amplitude
SAR backscattering signal analysis shows potential for improving irrigation monitoring, detecting irrigation supplies and understanding surface water content changes at the field and regional levels in the study area.
Our findings need to be applied to other crops and other areas in order to test the validity of the proposed methodology.

Figure
Figure 1.black lin Satellite Images and Their ProcessingOne SPOT-5 HRV satellite image was acquired on 15 April 2011 (Table1), when the soil was completely covered by vegetation.It spanned the period between anthesis (March) and grain filling (April) in the 2010-2011 cropping season.
org Data: Time Se