**5. Discussion**

#### *5.1. Comparison with Other Studies*

Lawston et al. [33] proposed a method for detecting irrigation signals based on SMAP data. In this method, dates are first selected during the crop growing season and then the characteristics of SM are compared at irrigated and non-irrigated points. Finally, the method uses time integrated and SM normalized metrics of SM and precipitation to detect irrigation signals. According to the method, the precipitation and SM processing results are obtained, as shown in Figure 11a,b. Since this method does not deduct the e ffect of precipitation from the SM changes, in the southern part of the study area, su fficient precipitation a ffects the detection of irrigation signals. Simultaneously, the detection results of the proposed method are normalized, which is more conducive to the comparison of the two methods. In Figure 11, the amplitude change in (c) is more obvious than (d), and some obvious regions in the calculation results are marked. Region 1 contains two large reservoirs adjacent to the Taihang Mountains. Region 2 is the southern part of Beijing. Region 4 is a large wetland named Baiyangdian. The type of underlying surface may a ffect the monitoring of time series SM changes. Notably, region 3 is the main irrigation area in the southern part of the Hebei Province. However, the irrigation signal for this irrigated area is not significant in (d). Therefore, the method proposed in this paper is more suitable for irrigation signal detection in the study area.

The method proposed in this paper can acquire daily irrigation signal detection result, so the research can describe the irrigated information in the study area in more detail. In order to display the irrigation details more abundantly, the monthly irrigation signals were accumulated to acquire a monthly distribution of WW irrigation (as shown in Figure 12). At the end of February, the southern part of the study area warmed up, and the irrigated area of WW was mainly concentrated in the southern part. In March, a wide range of WW was irrigated, and irrigation in April and May was concentrated in the central and western regions. Compared with existing studies, Chen et al. statistically analyzed the climate distribution characteristics of WW growing season in the NCP for many years, which is consistent with the monthly spatial distribution of irrigation in this paper [50]. Yang et al. collected information on crop planting and irrigated area in the NCP for many years, and acquired crop and irrigation spatial distribution characteristics in this region [51]; the results of Yang's study are similar to the results acquired in this paper, but due to the change of crop pattern in the eastern region, inconsistencies have been caused. Overall, the results of this paper are consistent with existing research findings.

**Figure 11.** Comparison of the method proposed in this paper with the time-integrated and SM normalized irrigation signal detection methods. (**a**) Accumulated PRE and normalized result, (**b**) accumulated SM and normalized result, (**c**) irrigation intensity calculated by this paper proposed method, and (**d**) time-integrated and SM normalized irrigation signal detection methods. Both normalized results and irrigation intensity are dimensionless variables.

**Figure 12.** Spatial distribution of winter wheat irrigated area.

In the study of irrigated area extraction without considering SM changes, most of the research extraction methods are based on time series vegetation index changes and supervised classification to identify irrigated areas [13,34,41]. These methods for identifying irrigated areas through optical remote sensing datasets were based on identifying the type of crop to distinguish whether the area is irrigated [52]. The common advantage of these methods is that they can obtain a high resolution crop spatial distribution, and the accuracy can be increased as the spatial resolution of remote sensing images increases, and validated in many areas [16]. Based on the SMAP data extraction irrigation signal, the spatial distribution of crops with high spatial resolution is was introduced as irrigation reference area, which not only maintain irrigation time and frequency information, but also increases precision of SMAP data recognition irrigated area.

#### *5.2. A Rational Discussion of the Irrigation Signal Detection Model*

The establishment of the irrigation signal detection model in this paper was based on irrigation records and SMAP SM data. Since the SM change data in the irrigation record is measured every 10 days, the data does not express time-continuous SM variations, so the SMAP SM data is not compared with the in-situ data. However, validation of SMAP SM data and irrigation-induced SM increase researches can demonstrate that variations in SMAP SM data can be used to establish irrigation signal detection models. A validation study of SMAP SM data has been described in the first section, and this section will discuss the relationship between irrigation and SM variation.

Chen et al. [23] analyzed the continuous variation of SM before using the MODIS Greenness Index to detect irrigation signals in Gansu Province. Combined with irrigation and precipitation records, it was found that the sudden increase of SM generally originated from irrigation and e ffective precipitation. At the same time, the irrigation time was estimated using the continuous SM variation data in year 2016. Under the condition of no in-situ SM data, Lawston et al. [33] obtained the SMAP SM variation of di fferent crop types based on the location information of irrigation and rainfed crops, and according this, they extracted the irrigated area of many regions in the United States. Is the phenomenon of SM sudden increase caused by irrigation also obvious in the NCP region? Some studies based on the e ffects of di fferent irrigation patterns on WW yield provide a reliable basis. Wang et al. [53] collected SM variation in di fferent irrigation patterns of winter wheat. The data show that although the SM (soil depth 0–80 cm) covered by WW in drip irrigation is slightly lower than level-basin, there is obvious SM increase after WW irrigation. Zia et al. [54] collected more detailed time series SM variation data (soil depth 10 cm and 40 cm); at the soil depth of 10 cm, irrigation will cause significant SM increase, while at 40cm, irrigation will maintain a higher level of SM, and the sudden increase is not significant. In this study, when the in-situ SM data is insu fficient, the SMAP SM data can be used to analyze the SM variation characteristics of WW and rainfed crops. Referring to number of studies on the relationship between irrigation and SM response, this paper suggests that irrigation records and SM increase can be used to detect irrigation signals in agricultural areas.

It should be noted that the thresholds in the irrigation signal detection model proposed in this paper are not universal. For example, in the study area of this paper, there are significant di fferences in SM increase caused by di fferent irrigation patterns. In areas with more complicated irrigation patterns, the irrigation pattern of sample points needs to be considered. In addition, the SMAP SM data of 9 km resolution is acquired by 36 km data downscaling, and the uncertainty of scale conversion may also a ffect the application of the model. If necessary, consider using multiple filtering methods for data optimization.
