Study area

The research was carried out during the autumn-winter growing seasons 2016-2017 and 2017-2018 in the dairy farm E. and C. Asociados, located in Ejido Purísima, Matamoros, Coahuila, Mexico (Figure 1). Field preparation consisted of fallow, laser tracking and leveling. Three safflower cultivars were evaluated; one with spine (Guayalejo) and two without spines (Selkino and CD-868). Fertilization was carried out on fallow ground using 150-50-00 units of N, P₂O₅ and K₂O, respectively. A nitrogen fertilizer with nitrification inhibitor (3-4 Dimethylpyrazole phosphate) was used as a source of N and monoamonic phosphate was used as a source of P₂O₅. Fertilizers were incorporated with a step dray, later to make ground leveling. The experimental plot was two ha.

Figure 1 Location of study area. The plot is represented in the internal box. 

Crop establishment

Sowing was done with a small grain seeder in dry soil at 10 and 12 December, 2016 and 2017, respectively, using a seeding density of 50 kg ha^-1; Three days later, irrigation was applied. In the 2016-2017 growing season, three irrigations were applied at 20, 37 and 58 days after sowing (DAS) and in 2017-2018 two irrigations were applied at 30 and 48 DAS. Crop management was carried out by the farmer.

NDVI Measurement

NDVI began to measure at 37 DAS in 2016-2017 growing season and at 25 DAS in 2017-2018, at a height of 0.6 m above crop canopy every eight days, which is the frequency that satellites Landsat 7 and 8 passes through the study area. Five measurements per day were performed by cultivar in 10 linear meters in the plot center. Measurements were made at 10:00 a.m. East-West oriented with GreenSeeker device, which was designed by Oklahoma State University and marketed by NTech Industries (Ukiah, CA, USA). GreenSeeker works by directing a red light (0.660 μm), as well as a near-infrared (0.780 μm) to the crop canopy. The field of view is an oval ranging from 25 to 50 cm depending on the height at which is located the GreenSeeker respect to crop canopy. The measurement height that was used was 0.6 m, having a field of view of 40 cm. The NDVI_field obtained from GreenSeeker for three cultivars of safflower was compared with NDVI_satellite estimated by remote sensing.

Satellite images

Five satellite images were used in 2016-2017 and seven in 2017-2018 growing season to estimate NDVI. Satellite images were collected from Landsat 7 Enhanced Thematic Mapper plus (ETM +) and Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) (path = 30, row = 42) (Table 1). Each image was treated by radiometric, atmospheric correction and illuminated surface (^{Kjaersgaard & Allen, 2010}). The images were selected based on temporary coverage and cloudless conditions. Images with clouds located at 10 km from area of interest were considered acceptable. The images were processed following the recommendations of ^{Allen, Tasumi and Trezza (2007)}, and ^{Kjaersgaard and Allen (2010)} using the METRIC model, which is executed with ERDAS Imagine software version 14.0.

NDVI was estimated with reflectance value of the wavelength red and near-infrared. For Landsat 7 ETM + bands 3 (0.63-0.69 μm) and 4 (0.77-0.90 μm) were used and for Landsat 8 OLI bands 4 (0.636-0.673 μm) and 5 (0.851-0.879 μm). To avoid mixing NDVI_satellite pixels, a pixel located in the center of plot was used, as the same that measurements with GreenSeeker. Finally, a model of assimilated data was obtained between NDVI_satellite and NDVI_field.

Plant height

It was measured with a tape measure, from ground surface to last crop branch, taking 10 readings per crop per day and averaged. The readings were taken on the same date on which the NDVI field was taken (eight days), coinciding with satellite overpass dates.

Relationship between NDVI_field and Kc FAO-56

The relationship between NDVI_field and actual Kc taken from the FAO-56 manual (^{Allen et al., 1998}) for both growing seasons was established. The general equation obtained from two growing seasons was used to propose a Kc value as a function of NDVI. In the end, new Kc values were multiplied by ETo to generate actual crop evapotranspiration (ETc) and compare it with the reference evapotranspiration, which was taken from meteorological station of INIFAP La Laguna, as well as precipitation and heat units.

Precipitation and evapotranspiration

Cumulative rainfall was 25.6 mm for 2016-2017 and 38.4 mm for 2017-2018. Precipitation had better distribution in the first year of study, since the greatest events were at the end of growing season after irrigation events (Figure 2). Daily values of reference evapotranspiration (ET) ranged from 1.0 to 6.8 mm dia^-1 in 2016 to 2017 and from 0.4 to 5.5 mm dia^-1 in 2017-2018 growing season. However, the accumulated values during the growing season (from sowing to harvest) were similar (267 mm for 2016-2017 and 266 mm for 2017-2018). As expected, the minimum value of ETo was recorded at the beginning of each growing season and maximum at the end of it. These values coincided with average temperatures in the Region Lagunera, where cold temperatures (10 °C) are recorded in January and temperate (20 °C) in March. On the other hand, Figure 3 shows heat unit accumulation in two growing season, recording 1075 heat units in the first and 1050 in the second.

Figure 2 Precipitation, irrigation and ETo in two growing seasons of forage safflower in Matamoros, Coahuila. 

Figure 3 Heat units accumulation in forage safflower during two growing seasons in Matamoros, Coahuila.  

NDVI curves

NDVI curves were observed according to phenological stages of crop and only at the end of growing season were noted divergent. This was due to last irrigation for 2016-2017 growing season was applied on February 9, 2017 (58 DAS), in addition to receiving a precipitation of 14.6 mm before harvest, while in 2017-2018 last irrigation event was February 1, 2018 (51 DAS) and received 7 mm of rain before harvest. Therefore, NDVI values in 2016-2017 remained at 0.8 and values in 2017-2018 decreased to 0.6 at the end of the growing season (Figure 4). However, ^{Tasumi, Allen, Trezza and Wright (2005)} reported that NDVI values are unaffected by soil moisture in a study conducted on 383 agricultural fields in Southern Idaho, USA. In general, NDVI values obtained in field with GreenSeeker for both growing seasons were averaged 0.2, 0.8 and 0.6 for initial, maturity and senescence stages, respectively. Similar values of NDVI were reported by ^{Hajare, Patil and Verma (2008)} for two growing seasons in safflower sown in three types of soil in Nagpur, India, although, at the end of season the values fall (0.2) when safflower was dry.

Figure 4 NDVI curves taken with GreenSeeker for three safflower cultivars. Red circles represent NDVI derived from satellite in two growing seasons. 

The discrepancy between NDVI values derived from satellite and values obtained in the field with GreenSeeker were higher at the beginning and at the end of growing season (Figure 4). This difference was due to field measurements are directly to canopy covering an area of 0.4 m, while satellite estimates an area of 30 x 30 m canopy and bare soil. Field measurements were made at the same day that satellite overpass dates through the study area. The images used were in cloudless days.

Similar NDVI curves in wheat, sugarcane, cotton and corn were reported by ^{Garatuza-Payán et al. (2003)}; ^{Zhang, Anderson, and Wang (2015)}; ^{French, Hunsaker and Thorp (2015)}, and ^{Reyes-González et al. (2019)}, respectively. Those authors found discrepancy in the first and last stage of the crop between satellite images and field measurements. Results suggest that period of greatest difference between satellite and GreenSeeker occurs when NDVI values are less than 0.6. Vegetation Index Adjusted Soil (SAVI) (^{Huete, 1988}), is recommended for satellite, since this minimizes soil reflectance in the first and last stages of the crop when vegetation cover is limited. On the other hand, ^{Reyes-González et al. (2018)} documented that the beginning and end of growing season corn ET value obtained from NDVI was less accurate. However, for irrigation calendar purposes this method may be acceptable. In another study, ^{Anderson, Allen, Morse and Kustas (2012)} reported that ETc calculated from vegetation indices was underestimated in irrigated areas in Idaho, USA.

Figure 5 shows vegetation cover development of safflower variety CD-868 during the 2017-2018 growing season, which began with low values, subsequently increases and at the end decreases slightly.

Figure 5 Development of vegetation cover of safflower variety CD-868 in Dairy farm E and C Asociados during the 2017-2018 growing season. 

Relationship between NDVI_field and NDVI_satellite

With satellite images, five NDVI values were obtained in 2017-2018 and seven in 2017-2018. Figure 6 shows the relationship between NDVI obtained from field and NDVI obtained with images. A good relationship was found between measured and estimated with high coefficient of determination (r² = 0.96). This reliability degree shows that satellite is a useful tool not only for estimating NDVI at regional scale but also at field scale. High correlations (r² = 90) of NDVI between measured in the field with radiometer (MSR87) and estimated with remotely sensors were observed by ^{Duchemin et al. (2006)} in 13 agricultural fields in Morocco, Africa.

Figure 6 Linear relationships between NDVI_field and NDVI_satellite in two growing seasons of forage safflower. 

Plant height

Figure 7 shows growth dynamics in plant height of three safflower cultivars evaluated during two growing seasons. At 25 DAS crop showed a height of ~5 cm and in the rosette stage a height of 10 cm (40 DAS). Subsequently, at 44 DAS stem elongation began until reaching reproductive stage with the beginning of capitulum formation at 80 DAS, with plant height was between 90 and 100 cm. The stem and branches continue growing and differentiation and development of capitulum parts became distinguishable at 93 DAS, although capitulum was not yet open. On this date safflower plants reached between 111 and 151 cm. The Guayalejo and CD-868 cultivars had the highest plant height and the Selkino cultivar showed the lowest in both growing seasons. In previous work, it was reported that crop had a growing season between 82 and 95 DAS reaching heights between 82 and 93 cm (^{Reta et al., 2014}; ^{Reta et al., 2017}). This comparison indicates that plant heights were higher than those observed in previous studies in the Region. They were also higher than heights reported by ^{Tayebi et al. (2012)} who obtained maximum heights (80 cm) in safflower established with different levels of evapotranspiration in Tabriz, Iran. However, similar heights (from 104 to 149 cm) were reported by ^{Kizil, Cakmak, Kirici and Inan (2008)} in safflower established in Eskishehir, Turkey. ^{Bagheri, Saeedi and Zaeh (2006)} mention that available soil moisture affects plant height in a study in Esfahan, Iran.

Figure 7 Plant height of three forage safflower in the Region Lagunera. 

On the other hand, Figure 8 shows a polynomial second-order relationship between plant height and NDVI of safflower, which accounts between 64 and 92% total observed variation NDVI in the first growing season (2016-2017) and between 83 and 87% in the second growing season (2017-2018).

Figure 8 Relationship between plant height and NDVI of three forage safflower in the Region Lagunera. 

Relationship between NDVI_field and Kc FAO-56

NDVI values were taken from three safflower cultivars evaluated for two growing season, while Kc values were obtained from the FAO-56 manual. Figure 9 shows the relationship between NDVI_field of three safflower cultivars and tabulated Kc for two growing seasons. In the first growing season the relationship between NDVI_field and Kc showed an R ² = 0.93 and for the second season of 0.89, showing a good relationship between NDVI_field and Kc. Nevertheless, in the second season the “r” value decreased slightly, this was due to crop presented stress conditions due to bad programming irrigation schedule because only two irrigations were applied and bad precipitation distribution during the growing season. Hence, the relationship between NDVI_field and Kc was slightly low, since Kc from FAO is developed for well-irrigated crops and without moisture and nutrient restrictions (^{Allen et al., 1998}). Even though, FAO-56 Kc values after 50 DAS were 1.0, while the NDVI_field values ranged between 0.65 and 0.8.

Figure 9 Relationship between NDVI_field of three forage safflower cultivars and Kc FAO-56 in two growing seasons. 

High correlations have been reported between NDVI_field and Kc taken from FAO tables for many crops. ^{Rocha, Perdigão, Melo and Henriques (2012)} and ^{Reyes-González et al. (2016)} reported high determination coefficients (r² = 0.99), for corn and alfalfa. ^{Consoli, Russo and Snyder (2006)}; ^{Singh and Irmak (2009)}; ^{Gontia and Tiwari (2010}); ^{Kukal, Irmak and Kilic (2017)}, and ^{Palacios-Vélez, Palacios-Sánchez and Espinosa-Espinosa (2018)} reported good relationships between NDVI and Kc for orange (r ² = 0.80), sorghum (r ² = 0.87), wheat (r ² = 0.80), soybean (r ² = 0.81) and grape (r ² = 0.83), respectively. The differences between high and low ratios occur mainly by crop management.

Development of general equation of Kc

NDVI taken in field and Kc obtained from FAO-56 was used to develop linear regression equations for two growing seasons (Figure 9). The equations obtained for both seasons were compared using the "t" test method to test statistical difference between two independent regressions. After the test, the calculated “t” value was lower (0.76) than tabulated “t” value (1.68), indicating that there was no significant statistical difference between two equations. Based on this result, all data (two growing seasons) were pooled and a new equation was generated for safflower in the Region Lagunera (Figure 10).

Figure 10 Relationship between NDVI_field and Kc obtained from FAO-56 for two years. 

Several equations have been generated from vegetation indices for different crops. For instance, ^{Consoli et al. (2006)} developed an equation using NDVI in orange in Italy (Kc = 1.262 * NDVI + 0.287); ^{Singh and Irmak (2009)}, for soybeans in USA (Kc = 1.217 * NDVI-0.034); ^{Campos et al. (2010)} for grapes in Spain (Kc = 1.44 * NDVI - 0.10); ^{Gontia and Tiwari (2010)}, for wheat in India (Kc = 2.7109 * NDVI - 0.424); ^{Rocha et al. (2012)} for corn in Portugal (Kc = 1.37 * NDVI - 0.017), and ^{Reyes-González, Hay, Kjaersgaard, and Neale (2015)} for alfalfa in Mexico (Kc = 2.112 * NDVI - 0.4989). These equations were generated for developing new Kc values in local conditions as recommended ^{Doorenbos and Pruitt (1977)} and ^{Allen et al. (1998)}, to improve water management in agriculture.

The generated equation (Kc = 1.0644 * NDVI + 0.2128 ) in Figure 10 was used to develop a new Kc curve for forage safflower, taking days after sowing (Figure 11). Similar Kc values for safflower were reported by ^{Kar et al. (2007)} in Dhenkanal, India, who reported values of 0.37 in the initial stage, 0.94 in development, 1.29 in maturity and 0.35 at the end of growing season. However, in last two stages (maturity and final) there was a difference with those found in this study, being more remarkable at the end of season, since safflower evaluated at the Region Lagunera is used for forage and is harvested in green, so it does not reach physiological maturity as safflower that is used for seed.

Figure 11 New Kc curve compared to Kc FAO-56. 

Comparison between ETo and ETc

The ETo values were taken from weather station located at La Laguna Experimental station, which is 15 km from the study area, while ETc values are the result of multiplying new Kc values by ETo. Figure 12 shows comparison between ETo and ETc for 2017-2018 growing season. At beginning of growing season (1-20 DAS) ETo values were around 2 mm day^-1, while ETc were 0.23 mm day^-1. In development stage (21-57 DAS) ETo values oscillated in 2.5 mm day^-1 and ETc values ranged from 0.3 to 2.5 mm day^-1. In maturity stage (58-85 DAS) values between ETo and ETc were similar (3.5 mm day^-1). At the end of season (86-95 DAS) ETo remained at 4.5 mm day^-1 and ETc decreased to 3.0 mm day^-1. The figure shows that at beginning of season daily ETo was higher than ETc and in middle of season values were similar. This indicates that with ETc crop water demand is consistent with crop phenological stages. Higher ET values (6.0 mm day^-1) than our values were reported by ^{Lovelli et al. (2007)} and ^{Kar et al. (2007)}. However, different ETo values in safflower varieties under different weather and soil conditions can vary from 200 to 1 000 mm per season (^{Istanbulluoglu, 2009}).

Figure 12 Comparison of ETo and ETc during 2017-2018 growing season. 

Understanding different development stages of crop and the exact amount of water applied; producers may improve irrigation scheduling, water management and increase sustainability of water resources in crops. ^{Reyes-González et al. (2016)} reported that farmers should use ETc instead of ETo for scheduling irrigation in arid and semi-arid regions where water is scarce.