**3. Results**

The maximum and minimum air temperatures from Jan to mid-May from 2008 to 2018 are shown in Figure 2. Overall, there is inter-annual variability in amplitudes and patterns of maximum and minimum air temperature. For example, in 2011 the minimum air temperatures were the lowest recorded during the period of study. The spatial patterns of the Landsat GNDVI images for a period of 10 years, each around the time of flowering, are shown in Figures 3 and 4. Overall, GNDVI was generally more variable for the earlier years, with values ranging between 0.1 and 0.8 for the 2008–2011 period (Landsat 5) and the majority of values over 0.6 from 2013 to 2018 (Landsat 8). The years 2010 and 2011 showed higher spatial variability than other years.

**Figure 2.** Maximum (full black line) and minimum (dotted black line) air temperature recorded at the Wangzhaung village from 2008 to 2018 from January to mid-May. The horizontal dotted line represents the 0 ◦C threshold while the vertical grey box represents the time when satellite images were acquired.

**Figure 3.** Calculated Green Normalized Vegetation Index (GNDVI) for LANDSAT 5 scenes for the growing season (**a**) 2008; (**b**) 2009; and (**c**) 2011. LANDSAT 5 scenes were obtained for late April/early May every year from 2008 to 2011.

The soil brightness resampled to 30 m resolution to match the Landsat data is shown in Figure 5a and the original soil brightness data at a 3 m resolution is shown in Figure 5b. The mean soil brightness was lower in the North-East portion of the village where the values ranged between 3 and 7. Areas with the highest soil brightness values were in the north-western and south-western portions of the village.

The spatial distributions of measured soil N and OC throughout the village are shown in Figure 6a,b respectively. Soil N varied between 0.68 and 1.37 g/kg (Figure 6a). Low values of soil N were located mainly on the North-West portion of the village and the highest values were located across the village without any clear spatial patterns (Figure 6a). The soil OC varied between 0.52 and 1.2% and its spatial variability matched the soil N content (Figure 6b). Therefore, there was not any clear spatial clustering of its value across the village.

The spatial and temporal decomposition of the GNDVI is shown in Figure 7. No obvious spatial clustering was found in the spatial variability metric (Figure 7a). The temporal variability of the GNDVI was higher in the eastern half of the village and also tended to coincide with areas of low GNDVI (Figure 7b).

**Figure 4.** Calculated GNDVI for LANDSAT 8 scenes for growing season (**a**) 2013; (**b**) 2015; and (**c**) 2018. LANDSAT 8 scenes were obtained for late April/early May from 2013 to 2018. There were no scenes available from 2012.

**Figure 5.** Soil brightness image (**a**) for 2017 resampled to match the LandSat8 spatial resolution (30 m) and (**b**) the original soil brightness image (3 m resolution).

**Figure 6.** Measured values of (**a**) soil nitrogen content (g/kg) and (**b**) soil organic carbon (%) at the field scale.

**Figure 7.** Ten-year time-series of GNDVI decomposed into (**a**) spatial variance and (**b**) temporal variance.

Using soil brightness to define the zones resulted in the three and four zones shown in Figure 8a,b. Using the GNDVI variability metrics to define the zones resulted in the three and four zones shown in Figure 8c,d. Using a combination of soil brightness and GNDVI variability metrics to define the zones resulted in the three and four zones shown in Figure 8e,f.

The RV of measured soil variables total N and OC within these clusters compared to the village mean is shown in Table 1. Soil brightness alone was a fairly poor predictor of measured N and OC in this study, explaining only up to 9% of the variability. The GNDVI variability metrics were a reasonable predictor (up to 39%) of variability in measured N and OC. The three-cluster solution in the combined model (soil brightness + GNDVI) was the best predictor of N and OC variability at up to 45%.

**Figure 8.** Results of clustering at field scale using (**<sup>a</sup>**,**b**) soil brightness alone; (**<sup>c</sup>**,**d**) the GNDVI variability metrics and (**e,f**) both combined, testing three and four cluster solutions.


**Table 1.** Relative variance of measured soil variables total N and organic carbon at the field scale for three and four cluster solutions using (i) soil brightness; (ii) the spatial and temporal variability in GNDVI and (iii) a combination of both.
