The first study was conducted in the lowland system (Hacienda Lucerna farm), located in the municipality of Bugalagrande, Valle del Cauca department (Colombia), at an altitude of 941 m a.s.l. The region has an average annual temperature of 26.3 °C and a relative humidity of 66.5%, under the characteristic conditions of the Tropical Dry Forest (b-ST; Holdridge, 1967(Holdridge, 1967)). The evaluated system was an intensive silvopastoral system, characterized by a high density of Leucaena leucocephala Lam. Cv Cunningham (>8000 shrubs per hectare), associated with “star grass” (Cynodon plectostachyus K. Schum. Pilg). Forty-four Lucerna cows were managed under a rotational grazing system, utilizing daily strips of approximately 1200 m², with a rotation cycle of 43 days and a one-day occupation period, providing each animal with approximately 60 m² daily. Nutritional management incorporated supplementation during milking, conducted mechanically twice daily, with each cow receiving 2.19 kg of rice bran and 2.25 kg of cassava bran per day.

The second evaluation was conducted in the highland system (La Jacoba farm), located in the municipality of La Unión, Antioquia department (Colombia), at an altitude of 2,500 m a.s.l., with an average annual temperature of 17 °C and a relative humidity of 80%, under the characteristic conditions of the Lower Montane Humid Forest (bmh-MB; Holdridge, 1967(Holdridge, 1967)). The rotational grazing system consisted of a “kikuyu” grass (Cenchrus clandestinus) monoculture, with 48 Holstein cows managed in daily strips of approximately 4,358 m² (90.8 m²/animal/d) using an electric mobile fence. The rotation cycle was 33 days, with a one-day occupation per strip. Nutritional management included 5 kg of commercial concentrate per cow, provided during milking twice a day.

Temperature and relative humidity were recorded every 20 minutes for 48 hours using thermohygrometers placed in the paddocks. The temperature-humidity index (THI) was calculated using the Kibler (1964)(Kibler, 1964) equation, and thermal stress levels were classified according to Wiersama (2005)(Wiersama, 2005).

THI= (1.8 x T° + 32) – (0.55 – 0.55 x RH / 100) x (1.8 x T° – 26)

Where: THI denotes the temperature-humidity index, T is temperature in degrees Celsius, and RH is relative humidity in percentage.

Daily forage offered (kg DM/animal or kg DM/100 kg BW) was estimated using the comparative yield method(Haydock & Shaw, 1975), which relies on a visual scale with three forage availability levels (low, medium, and high). Forage was harvested 8 cm above ground level, and nine out of 25 randomly placed 0.25 m² frames per strip were weighed. For shrub biomass, the method was modified using a linear meter and cutting three shrubs 10 cm above the ground level. Available forage biomass was calculated as total biomass minus dead material. Daily forage intake (kg DM/animal or kg DM/100 kg BW) was estimated from the difference between forage offered and the remainder after grazing. Water intake was measured using flow meters at drinking stations, and water from feed was estimated based on dry matter intake and forage moisture content.

A representative sample of the forages and feed supplements offered to animals was collected from each strip assigned for grazing. Samples were stored under refrigeration until transport to the laboratory, where they were dried in a forced-air oven at 65 °C for 72 h to determine dry matter (AOAC, 2019, Method 934.01)(International, 2019). Neutral detergent fiber (NDF) and acid detergent fiber (ADF) were determined following the methodology described by Van Soest et al. (1991)(Soest et al., 1991). Crude protein (CP) was determined by the Kjeldahl method (AOAC, 2019, Method 984.13)(International, 2019). Ether extract (EE) was quantified by Soxhlet extraction (AOAC, 2019, Method 920.39)(International, 2019). Ash was obtained by direct incineration in a muffle furnace (AOAC, 2019, Method 942.05)(International, 2019). Finally, calcium and phosphorus concentrations were determined according to the official AOAC methods (2019, Methods 968.08 and 965.17, respectively)(International, 2019). The nutritional composition of all feed sources is presented in Table 1.

Nutritional balance was estimated using the Cornell Net Carbohydrate and Protein System(Amburgh et al., 2015), based on field data on forage supply, feed composition, and animal productive and reproductive parameters. Two simulations were performed per system: one considering only baseline nutritional requirements and another including the energy cost of walking. To estimate walking distance, one animal per system was equipped with a GPS device (Oregon 650 Garmin) and a mobile phone running the STRAVA app, both mounted on a head collar. Georeferenced data were processed in QGIS (version 3.36.3) to calculate the total distance walked during the observation period.

Animal behavior was continuously monitored over 48 hours under grazing conditions. Observations were carried out by four trained researchers organized in alternating 12-hour shifts, using the scan sampling technique described by Sed (1991)(Setz, 1991). Behavioral records were taken every 20 minutes, resulting in a total of 144 observations per farm. The recorded behaviors included forage intake (FI), standing rumination (RuS), lying rumination (RuL), standing rest (ReS), and lying rest (ReL). This information allowed for the quantification of time allocation to each activity and the identification of behavioral patterns associated with the animals’ circadian rhythms.

Since the initial conditions of each farm differed from the beginning, no direct comparisons were made between them. Although the results are presented together, the description and statistical analysis were carried out independently for each farm. Variables related to forage supply, forage intake, and water intake were analyzed using descriptive statistics (mean and standard deviation). To explore the association between behavioral activities and time of day, a chi-square test was applied, followed by a multiple correspondence analysis (MCA). Additionally, the Friedman test was used to assess significant differences among the recorded behaviors. This non-parametric alternative to two-way ANOVA is suitable for randomized complete block designs when parametric assumptions are not met. All analyses were performed in R(Team, 2022).

During the observation period, the THI in the lowland system exceeded 85 units between 13:00 and 17:00, reaching the emergency threshold. In contrast, in the highland system, THI values remained below 75 units (Figure 1), within the alert range. Cows in the lowland system showed no clinical signs of heat stress (respiratory rate <60 breaths/min, no panting or salivation), indicating a high level of thermal adaptation. These findings suggest that the risk thresholds proposed by Wiersama (2005)(Wiersama, 2005) may require adjustment for this breed.

Data on forage supply and intake, nutritional composition, and water consumption are presented in Table 2. In the lowland system, dry matter intake as a percentage of body weight was higher, and forage composition reflected a greater fiber content. In the highland system, dry matter intake, feed efficiency, and available grazing area per animal were consistent with the agroecological conditions of the region. Forage in this system showed higher CP and lower NDF. Total water intake included both drinking water and water from feed and was associated with the productive characteristics observed in each system.

Average milk yield was 12.4 ± 2.03 kg/animal/day in the lowland system and 25.33 ± 7.19 kg/animal/day in the highland system. When expressed per unit of metabolic body weight (BW^0.75), milk production was 0.119 kg/kg BW^0.75 in the lowland system and 0.206 kg/kg BW^0.75 in the highland system. Feed conversion efficiency, expressed as milk yield per kilogram of dry matter intake, was 0.78 in the lowland system and 1.68 in the highland system.

Table 3 and Table 4 show the estimated nutritional balance according to the CNCPS model for farms located in lowland and highland systems, respectively. The nutritional evaluation indicated a metabolizable energy (ME) deficit in lowland and highland systems. In the lowland system, cows walked an average of 1,300 m per day (Figure 2B), resulting in an additional energy expenditure of 0.49 Mcal/day. This increased the ME deficit from −3.96 to −4.45 Mcal/day (Table 3), representing approximately 1.87% of the total ME requirement. In the highland system, cows walked an average of 3,440 m per day (Figure 3B), with an associated energy cost of 1.66 Mcal/day. This raised the ME deficit from −4.79 to −6.45 Mcal/day (Figure 3), equivalent to 3.81% of daily ME requirements. Detailed values for protein, amino acids, and minerals are presented in Table 3 and Table 4 .

Multiple correspondence analysis revealed distinct behavioral patterns based on time of day and animal posture. In the lowland system, Dimension 1 explained 63.8% of the variability and distinguished between standing and lying behaviors, while Dimension 2 (28.7%) was associated with periods of elevated THI, during which cows spent more time standing (Figure 2C). In the highland system, Dimension 1 (44.5%) also reflected postural differences, and Dimension 2 (38.6%) separated activities occurring in the paddock from those associated with milking (Figure 3C).

In both farms, forage intake followed a bimodal distribution, with peaks in the early morning (06:00–08:00;Figure 2A) and late afternoon (16:00–18:00;Figure 3A). Lying rumination and lying rest predominated during nighttime hours (19:00–03:00), while standing rest was more frequent around midday and during milking periods. Standing rumination was the least frequent activity and showed a more uniform distribution throughout the day.

The temporal distribution and daily duration of activities are summarized in Table 5. In the lowland system, cows spent an average of 7.4 hours on forage intake, followed by lying rumination (5.9 h) and standing rest (5.5 h). Lying rest and standing rumination were less frequent, with 3.0 and 2.2 hours, respectively. In the highland system, forage intake was slightly higher (7.9 h), with similar durations for lying rumination (5.4 h) and standing rest (4.2 h). Lying rest and standing rumination were recorded at 3.9 and 2.6 hours, respectively.

The Friedman test revealed significant differences in time allocation among behavioral categories of cows grazing in the lowland system (p = 0.019), with forage intake, lying rumination, and standing rest showing the highest average ranks. In the highland system, differences were not statistically significant (p = 0.06), although similar behavioral trends were observed.