Eight sites in the provinces of Catamarca and La Rioja in the Northwest of Argentina were selected at latitudes of 28-30° S and an elevation gradient of 350-1200 m above mean sea level (AMSL) (Fig. 1; Table 1). Phytogeographically, the area is considered Arid Chaco with native vegetation being mainly scrubland. At each site, one olive orchard was used for the flowering observations. Most of the orchards were located in modern, commercial olive farms with densities of 250-300 trees/ha and were drip-irrigated. Trees in these orchards were 7-8 years-old at the start of the observations. The exceptions were the Sumalao and Tinogasta sites, which were located in older, traditional orchards with a plant density of 100 trees/ha and were flood-irrigated. Trees in these orchards were approximately 40 years-old. The orchard at the Sumalao site was part of the germplasm collection of the Instituto Nacional de Tecnología Agropecuaria (INTA). Trees were kept well irrigated at all sites during the years this study was conducted with 600-1000 mm of irrigation being applied per year depending on the amount of rainfall (70-450 mm/yr). Because rainfall events are infrequent in the winter months, all orchards were irrigated during this period to avoid water deficit conditions, which have been shown to reduce or delay the flowering of olive trees (Orlandi et al., 2010a; Oteros et al., 2013; Pierantozzi et al., 2014). Evapotranspiration reference values have been estimated to be 1600 mm/yr in the study area (Searles et al., 2011).

Flowering data for three Mediterranean cultivars were collected from 2004-2008 at the sites (Table 1). ‘Arbequina’ is a Spanish cultivar that had been previously observed to flower consistently in the region, while ‘Frantoio’ and ‘Leccino’ are Italian cultivars with less consistent flowering under our climatic conditions. The flowering of all three cultivars were evaluated at the Chañarito, Sumalao, Aimogasta, and Pajonal sites, while only the flowering of ‘Arbequina’ was assessed at the remaining four sites due to lack of availability of the Italian cultivars. To corroborate the genetic identity of the study material, leaf buds were collected from each site and cultivar. The identity of each cultivar was confirmed by microsatellite molecular markers at the Universidad Nacional de Cuyo. ‘Arbequina’ and ‘Frantoio’ material corresponded to the World Olive Germplasm Bank of Córdoba (Spain) patterns (Caballero & Del Río, 2008; Trujillo et al., 2014), while ‘Leccino’ material coincided with that cultivar as reported in Tuscany, Italy (Bandelj et al., 2004).

Hourly air temperature data was collected near each site using automatic weather stations located in El Infiernillo, Aimogasta, Pajonal, Chilecito, and Copacabana. For the remaining three sites, maximum and minimum temperatures were recorded daily with mercury thermometers. To determine the consistency and accuracy of both the automatic stations and the thermometers, two mercury maximum and minimum thermometers were rotated between the different sites. To provide an approximate comparison of temperature data from our Northwest Argentina sites with the Mediterranean Basin, daily temperature values for a period of years similar to those of our study were obtained for Córdoba, Spain and Florence, Italy. The Córdoba data were from the World Olive Germplasm Bank of Córdoba, while the Florence data is publically available on-line (http://clima.tiempo.com/clima-en-firenze+peretola-161700.html).

At each study site, 10-12 trees per cultivar were marked for phenological observations that were conducted every 3 days from mid-winter (August) until the end of the flowering period (late October) using the procedure of Fernández-Escobar & Rallo (1981). The date of full flowering was determined using these observations, and defined as the average of the dates at which at least 50% of the flowers of the observed trees were open. This is the same criteria used by the DMA model, and is equivalent to a BBCH scale value of 65 for olive growth stage phenology (Sanz-Cortés et al., 2002). Under our climatic conditions, flowering usually occurred between day of year (DOY) 270-290 (i.e., Sept. 27-Oct. 17). This would be equivalent to late March and April in the Northern Hemisphere. Similar to Ramírez-Santa Pau et al. (2002), flowering intensity per tree was estimated visually in the field by assigning numerical values ranging from 0 to 5 to the percentage of the tree crown that was flowering. At the low end, zero represented no flowering (0%), while five represented 80-100% flowering of the tree crown. Flowering was considered to be normal when more than half of the trees had flowering intensities equal to or greater than 2 (i.e., 20-40% flowering of the tree crown).

In addition to the flowering data (2004-2008), we also included yield data when available from previous years at the same sites to increase the number of observations. In such cases, moderate and high yields were considered to reflect normal flowering, while very low yield values were considered to be due to abnormal flowering. This was done because not all cultivars were present at each of the eight sites as mentioned earlier. To reduce the possibility of errors, alternative bearing behavior and crop management information at each site were considered when assessing yield data. Nevertheless, it is recognized that unidentified factors affecting fruit set could result in low yields even though flowering was normal.

In order to estimate the flowering date, Model-1 of De Melo-Abreu et al. (2004) was used to calculate: 1) the number of chilling units (CU) accumulated before bud dormancy release and 2) the thermal time (TT) accumulated in degree days between the end of bud dormancy and flowering. This model also indicates the likelihood of abnormal flowering due to insufficient CU or TT. For the Southern Hemisphere, the accumulation of CU was considered to start on April 1 (DOY 91) in the early Fall to be consistent with the start date (October 1; DOY 275) proposed by the DMA model for the Northern Hemisphere. The last possible day of CU accumulation for flowering was defined as October 27 (DOY 300; mid-Spring) because no CU accumulated after this date in our region. The model requires only maximum and minimum daily temperature and geographic location (latitude, longitude) as inputs. Using these data, the model simulates hourly temperatures and thus the accumulation of CU or TT, depending on the model phase. The model, which is written in Visual Basic for Applications, is available in Excel format on the Agrometeorology, Agriculture and Environment Tools webpage at the Universidade de Lisboa (http://home.isa.utl.pt/~jpabreu/PaginaTecnica.htm). It should be noted that neither Model-2 or Model-3 of De Melo-Abreu et al. (2004) were tested in this study because Model-2 lacks a mechanism to reduce the CU accumulated when high temperatures occur and Model-3 only considers TT accumulation after a certain date and not CU accumulation. Thus, these models were considered a priori to be less appropriate for our warm region than Model-1.

The cultivar-specific number of CU and TT required for flowering in each cultivar were obtained for ‘Arbequina’ from the De Melo-Abreu et al. (2004) study (Table 2). For ‘Leccino’ and ‘Frantoio’, 15 years (1991-2005) of air temperature and flowering data from the World Olive Germplasm Bank of Córdoba (Spain) were used to make these estimations with an optimization algorithm that employed the downhill simplex method (Nelder & Mead, 1965). The estimated number of CU required were very different among cultivars (‘Arbequina’=339, ‘Leccino’=612, ‘Frantoio’=671 units), but estimated TT was similar (468-490°C days). ‘Arbequina’ is considered an early flowering cultivar that was first widely cultivated under fairly mild temperature conditions in several regions of Spain such as Catalonia and Andalusia, while ‘Leccino’ and ‘Frantoio’ are late flowering cultivars from colder Tuscany in northern Italy.

The calculation of CU accumulation until the end of dormancy by the DMA model utilizes several parameters related to temperature including the optimum temperature for chilling (T _o=7.3°C) and the breakpoint temperature (T _x=20.7°C) above which a constant number of accumulated chilling units (a=–0.56) are nullified (Table 2). The parameter a, represents the maximum number of chilling units that are lost for each hour of high temperature. Hourly temperatures equal to or less than zero (T _h≤0) are not included as chilling units in the DMA model, but no CU are subtracted under freezing temperatures. After endodormancy release, a base temperature (T _b) of 9.1 is used to calculate thermal time until full flowering. In addition to these standard parameter values for olive, the De Melo-Abreu et al. (2004) study proposed alternative temperature coefficients for CU accumulation (T _x=21.7; a=–0.50) and TT (T _b=8.75) for early flowering cultivars such as ‘Arbequina’ to account for winter temperatures being above normal in some years when the model was tested in Córdoba, Spain.

Potential relationships between elevation and temperature variables as well as with the number of days with frosts were evaluated with GraphPad Prism 5 (San Diego, CA). The ability of the DMA model to accurately predict whether flowering occurred normally or not was assessed by comparing the number of successful predictions relative to the total number of observations. Prediction errors were categorized as type 1 (when the model predicted normal flowering, but it was not observed) or type 2 (when the model did not predict normal flowering, but it was observed). Flowering observations were excluded from the analysis if there were heavy winter frosts at a particular site in a given year.

To evaluate the model’s ability to predict full flowering date, a simple linear regression analysis between predicted and observed values was performed and the root mean square error (RMSE) was determined (Wilmott et al., 1985). The differences between predicted and observed flowering dates were also evaluated based on a chi-square distribution (Freese, 1960). For this analysis, it was tested whether the number of predicted flowering dates that fell within 7 days of the observed values was significant.

During the accumulation of CU and TT (early Fall-mid Spring) for our Northwest Argentina sites, the highest daily average maximum temperature was 25.2°C at 345 m AMSL. Although a statistically significant decrease in maximum temperature with increasing elevation was not observed, the lowest maximum was 21.6°C at 900 m (Fig. 2A). Daily minimum temperature values showed a negative linear relationship with elevation (R ²=0.54; p<0.05), and ranged from 10.1°C at 391 m to 2.4°C at 1100 m. The number of days with frosts (<0°C) was greater than 40 d above 1000 m with a large amount of variability at lower elevations (Fig. 2B). This variability was likely related to differences in cold air drainage associated with the microtopography in and around each study site.

To provide a comparison, average maximum temperature for two selected Mediterranean sites, Córdoba (Spain) and Florence (Italy), was approximately 6°C and 11°C lower; respectively, than in Northwest Argentina for the same dates (early Fall-mid Spring) (Fig. 2A). In contrast, daily minimum temperature values and the number of days with frosts were within the same range as those found at our sites.

Flowering at the eight sites was assessed based on 42 observations (Sites × Years) in ‘Arbequina’, 18 in ‘Frantoio’, and 18 in ‘Leccino’ (Table 3). Several observations for each cultivar (i.e., 5 or 6) were discarded because of frost damage to reproductive buds. In ‘Arbequina’, 94% of the remaining observations were consistent with normal flowering, and percentage flowering intensity was fairly high (i.e., 60-80% of the tree crown). Abnormal flowering only occurred in ‘Arbequina’ at two low elevation sites, Sumalao and Chañarito. Over the seven years analyzed, 30% of the years had abnormally low flowering intensities at Sumalao and the percentage was 15% at Chañarito. At these sites, low flowering intensity did not seem to be related to alternate bearing behavior.

The flowering behavior of ‘Leccino’ and ‘Frantoio’ was primarily examined at low elevation sites (391-859 m AMSL). Normal flowering occurred in some instances at one of the higher sites (Aimogasta; 856 m AMSL), but the overall percentages of observations with normal flowering were still low (i.e., 14% in ‘Leccino’ and 31% in ‘Frantoio’). Even when flowering was considered normal, the percentage flowering intensity was most often between 20-40% and was never greater than 50% of the tree crown. These observations are in agreement with the number of CU values required for these cultivars based on the data of the World Olive Germplasm Bank of Córdoba (Spain) (Table 2).

The model accurately predicted whether flowering was normal or abnormal in ‘Arbequina’ in 92% of the cases (Table 4) for our eight sites when using the alternative temperature coefficients proposed by De Melo-Abreu et al. (2004) for early flowering cultivars under warmer than average winter conditions in Córdoba, Spain (Table 2). The errors that did occur for ‘Arbequina’ in 3 of the 36 observations evaluated (8%) were type 1 (i.e., the model predicted normal flowering but it was not observed). The use of the standard temperature coefficients resulted in only slightly fewer successful predictions (84%), while the errors were divided into both type 1 and type 2 (data now shown). A type 2 error indicates that the model did not predict normal flowering, but it was observed.

In ‘Frantoio’ and ‘Leccino’, the model successfully predicted whether flowering was normal or not in 83% and 61% of the cases; respectively (Table 4). Similar to ‘Arbequina’ with the alternative coefficients, the errors that did occur for ‘Leccino’ were type 1 (5 of 13 cases evaluated). Thus, in both cultivars, the model sometimes predicted normal flowering when it did not occur. In ‘Frantoio’, only two prediction errors occurred with one error being type 1 and one error being type 2. In this case, the type 2 error occurred because insufficient CU accumulation was predicted, but normal flowering was observed.

The ability of the model to predict flowering dates was analyzed in ‘Arbequina’ and ‘Frantoio’ for the cases when normal flowering was successfully predicted (Fig. 3). Of 21 cases, the flowering date for these two cultivars was predicted within ±7 days in 71% of the cases using a chi-square analysis (p < 0.05), and the RMSE was just slightly higher (8 days). In ‘Arbequina’, the predicted flowering date was 10-15 days earlier than the observed date at some sites and in some years. In contrast, the predicted flowering date in ‘Frantoio’ was later than the observed date in some cases. ‘Leccino’ was not included in the analysis because normal flowering could only be estimated from previous production data at the Aimogasta site in two cases rather than from actual flowering data.