*3.2. Irrigated Lysimeter ET Estimation*

Figure 5a,b presents the northeast irrigated lysimeter daily measured ET, Landsat ET, and seven-day running average from 2001 to 2010. The 1:1 graph between the daily measured and calculated Landsat ET is shown in Figure 6, and the daily and monthly summary statistics are summarized in Table 6. The growing and non-growing season summary statistics are reported in Table 7. The daily 1:1 graph is shown in Figure 6, and the monthly ET is shown in Figure 7.


**Table 6.** Daily and monthly summary statistics for the NE irrigated lysimeter with Landsat ET.

The years 2004 and 2007 were omitted from the statistical analysis due to limited clear image availability.

**Figure 5.** Daily and seven-day running average measured and calculated Landsat ET from (**a**) 2001–2005, and (**b**) 2006–2010 for the northeast (NE) irrigated lysimeter. The years 2004 and 2007 had limited clear remote sensing observations during the growing season.


**Table 7.** Seasonal summary statistics for the NE irrigated lysimeter with Landsat ET.



GS: growing season, NG: non-growing season. The years 2004 and 2007 were omitted from the analysis due to limited clear image availability.

**Figure 6.** Daily 1:1 graph of measured and Landsat ET for the NE irrigated lysimeter. The years 2004 and 2007 were omitted from the analysis due to limited clear image availability.

**Figure 7.** Monthly average measured and Landsat estimated ET for the NE irrigated lysimeter. The years 2004 and 2007 had limited clear remote sensing observations during the growing season.

The daily mean ET was 2.4 mm d−<sup>1</sup> and 2.4 mm d−<sup>1</sup> for the Landsat and measured, respectively. The summary statistics improved with the irrigated field and provided a weak correlation with an *R2* value of 0.38, NSE of 0.37, RMSE of 2.1, and RMSE ~86.4%.

The growing season irrigated measured LAI was plotted versus Landsat estimates for the days where Landsat images were available during the year, as shown in Figure 8. Landsat better estimated LAI under irrigated conditions compared to the dryland conditions. Consequently, higher NDVI values were obtained [38], producing higher LAI values for irrigated fields, and resulted in better ET estimates.

**Figure 8.** Measured and Landsat leaf area index (LAI) for the NE irrigated lysimeter. The year 2004 graph omitted because Landsat LAI values were not available.

#### **4. Discussion**

### *4.1. Dryland Daily ET Comparison*

The relationship between measured and Landsat ET for the dryland lysimeter showed significant deviation with periods of both over and underestimation of ET throughout the year for the entire study period. The satellite-based LAI was assessed versus the measured LAI (Figure 4), and the LAI assessment summary is summarized in Hashem [38]. The daily time series ET deviations were related to errors in LAI estimation [38,41,47], where Landsat LAI estimates were significantly lower than measured LAI during the growing season for the dryland lysimeter. The higher the NDVI values, the more the LAI values increase, resulting in greater ET values.

In 2002, 2005, and 2009, the lysimeter field was fallow, and Landsat overestimated ET in each of the three years (Figure 1), and these results agree with Allen et al. [46]. Cotton was cultivated in 2001 and 2008, and Landsat estimates of ET closely matched the measured ET at the beginning of each year. However, towards the end of 2001, Landsat significantly underestimated ET due to low NDVI values and, consequently, underpredicting LAI [38]. ET data in 2004 and 2007 were omitted from the analysis, as the Landsat data overestimated the ET compared to the measured ET due to the large gap period in Landsat data and linear interpolation method used to fill the gap. In 2003, when sorghum was cultivated, the satellite-based-ET overestimated measure ET in both the beginning and towards the end of the year, and underestimated towards the middle of the growing season.

A detailed statistical analysis was performed for the growing and non-growing seasons (Table 5), where the growing season was defined as the days between planting and harvest. Monthly statistics showed better statistical performance, with monthly RMSE and NSE of −0.19 and 1.2 compared to values of −1.38 and 1.8 mm for the daily assessment [38,46,50]. The RMSE during the growing season was greater compared to the non-growing season (Table 5), with values almost double for the growing season compared to the non-growing season due to low measured ET values during the non-growing season. Hence, there was less variation between the measured and satellite-based ET values. However, the %RMSE error was higher during the non-growing season than the growing season, and these results agree with Allen et al. [46].

The satellite was able to distinguish between bare soil and vegetation in the field, providing useful information on when the field was fallow versus when a crop was growing. However, the overall LAI estimation from Landsat was lower than the measured LAI for all cultivated crops during this study under dryland conditions. Potential reasons for the LAI undercalculations are the water stress during the growing season producing low NDVI values under dryland conditions, uncertainties with aerodynamic resistance surface roughness length [36], long gap periods, and using the linear interpolation method to generate daily ET time series [38].
