NASA POWER satellite meteorological system is a good tool for obtaining estimates of the temperature-humidity index under Brazilian conditions compared to INMET weather stations data

Abstract

Heat stress negatively affects livestock, with undesirable effects on animals’ production and reproduction. Temperature and humidity index (THI) is a climatic variable used worldwide to study the effect of heat stress on farm animals. Temperature and humidity data can be obtained in Brazil through the National Institute of Meteorology (INMET), but complete data may not be available due to temporary failures on weather stations. An alternative to obtaining meteorological data is the National Aeronautics and Space Administration Prediction of Worldwide Energy Resources (NASA POWER) satellite-based weather system. We aimed to compare THI estimates obtained from INMET weather stations and NASA POWER meteorological information sources using Pearson correlation and linear regression. After quality check, data from 489 INMET weather stations were used. The hourly, average daily and maximum daily THI were evaluated. We found greater correlations and better regression evaluation metrics when average daily THI values were considered, followed by maximum daily THI, and hourly THI. NASA POWER satellite-based weather system is a suitable tool for obtaining the average and maximum THI values using information collected from Brazil, showing high correlations with THI estimates from INMET and good regression evaluation metrics, and can assist studies that aim to analyze the impact of heat stress on livestock production in Brazil, providing additional data to complement the existing information available in the INMET database.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00484-023-02493-5.

Introduction

Brazil is one of the largest producers and exporters of animal products in the world (USDA 2023), standing out in the global food trade (Ferraz and Felício 2010). Livestock production in Brazil and in the world is highly affected by climatic conditions. If greenhouse gas emissions are not curbed in the coming years, Brazil may expect a general temperature rise exceeding 1.5 ºC by 2030 (IPCC 2021). In this sense, animal responses to these climatic changes, as heat stress, may cause serious physiological changes, along with negative effects on animals’ production and reproduction (West 2003; Renaudeau et al. 2012). In studies on heat stress in livestock, the temperature and humidity index (THI), usually calculated using daily temperature and humidity data from weather station databases, is mostly used as a climatic indicator.

In Brazil, the National Institute of Meteorology (INMET) offers the possibility of immediate download of meteorological information from the year 2000 onwards at https://bdmep.inmet.gov.br. Several studies have used information from the INMET database to evaluate the effects of heat stress on livestock in Brazil (Santana et al. 2016, 2020; Cordeiro et al. 2020; Negri et al. 2021; Stefani et al. 2022). Unfortunately, the weather coverage is sparse in some regions and complete information is not always available for a given day or time period, since weather stations may have power outages. For example, between 2020 and 2022, around 20% of the stations experienced a power outage, and up to 42% had a blackout in April 2021, a negative impact of the COVID-19 pandemic (INMET 2022). Moreover, the properties that have production animals are not always located near INMET weather stations, which can reduce the precision of THI estimates for these locations.

An alternative to obtaining meteorological data is the National Aeronautics and Space Administration Prediction of Worldwide Energy Resources (NASA POWER) satellite-based weather system. Data can be downloaded from https://power.larc.nasa.gov by entering the latitude and longitude of the desired location, with single point, regional, or global coverage, and can be exported to a user-friendly output (e.g., in ASCII format). Recent studies have used meteorological data from the NASA POWER project to calculate THI and assess the effects of heat stress on livestock (Carrara et al. 2021; Manica et al. 2022; Salvian et al. 2022; Dauria et al. 2022; Rockett et al. 2023).

However, Brazil has a great climatic diversity and the production animals, mainly cattle, are mostly pasture-raised. In this way, would NASA POWER be a good provider of temperature and humidity data to calculate THI considering Brazilian climatic conditions compared to INMET data?

Monteiro et al. (2018) compared meteorological data from 1997 to 2016 belonging to 203 INMET weather stations and the authors concluded that NASA POWER satisfactorily estimates daily temperature. Duarte and Sentelhas (2020) compared data from 10 INMET weather stations with data from NASA POWER collected between 1990 to 2013 and found correlations greater than 0.70 for minimum and maximum daily temperatures and average daily relative humidity. Similarly, Aguiar and Lobo (2020) evaluated 30 stations from 2004 to 2014 and reported correlations ranging from 0.21 to 0.99, and 0.50 to 0.96 for minimum and maximum daily temperatures, respectively, depending on the location. Therefore, the degree of agreement of data obtained via NASA POWER and INMET depends on where the data are collected.

In Canada, Rockett et al. (2023) demonstrated that the use of NASA POWER weather parameter estimates could replace weather station data in the calculation of THI. However, for Brazilian climatic conditions, such studies are still scarce. Therefore, we aimed to compare THI estimates obtained from INMET and NASA POWER meteorological information sources using linear regression and Pearson correlation.

Material and methods

Meteorological information from 2002 to 2021 (last 20 years) were gathered from 609 Brazilian INMET weather stations. Some filtering parameters was applied to guarantee data quality and were kept: i) weather stations with a minimum of five years of recorded data; ii) days with at least 19 daily hours recorded; iii) days with at least two measurements between hours 00:00–06:00, 06:00–12:00, 12:00–18:00, and 18:00–23:00 considering the Coordinated Universal Time (UTC); iv) minimum relative humidity levels of 12%; and v) data from automatic stations, as they provide hourly information. After the quality check, 489 weather stations with valid measurements remained for further analyses. The spatial distribution of the stations, classified according to the Köppen's climate classification map for Brazil (Alvares et al. 2013), is presented in Fig. 1.

Fig. 1.

Spatial distribution of weather stations, separated according to the Köppen's climate classification map for Brazil (Alvares et al. 2013). Köppen's climate classification: Af = Tropical without dry season; Am = Tropical monsoon; As = Tropical with dry summer; Aw = Tropical with dry winter; BSh = Dry semi-arid low latitude and altitude; Cfa = Humid subtropical oceanic climate, without dry season and with hot summer; Cfb = Humid subtropical oceanic climate, without dry season and with temperate summer; Cwa = Humid subtropical with dry winter and hot summer; Cwb = Humid subtropical with dry winter and temperate summer

THI was calculated using the following equation (NRC 1971):

$$\mathrm{THI}=\left(1.8,\times ,\mathrm{T},+,32\right)-\left[\left(0.55,-,0.0055,\times ,\mathrm{RH}\right),\times ,\left(1.8,\times ,\mathrm{T},-,26\right)\right],$$

where T is the air temperature in ºC, and RH is the relative humidity in %.

Hourly information on air temperature and relative humidity were selected from the INMET database considering the Coordinated Universal Time (UTC). Latitude and longitude data of the 489 weather stations (single point) were used to obtain the 2-m air temperature and 2-m relative humidity data in NASA POWER platform.

The hourly data from NASA POWER is provided in Local Solar Time format by default. Hence, the data in UTC format was obtained through the Application Programming Interface (https://power.larc.nasa.gov/docs/services/api/) using the following simple script in Python (Pandas interface): 

curl -X GET "https://power.larc.nasa.gov/api/temporal/hourly/point?parameters=T2M,RH2M&community=AG&longitude=AAA&latitude=BBB&start=YYYYMMDD&end=YYYYMMDD&format=ASCII&time-standard=UTC " -H "accept: application/json" > FILE_CITY_01.txt,

where AAA is the longitude and BBB is the latitude of the respective INMET weather station, in decimal format, with start date (YYYYMMDD) equal to 20,020,101 and end date equal to 20,211,231 (i.e. period from 01/01/2002 to 12/31/2021).

The variables evaluated were:

• i)Hourly temperature (ºC, T_HOURLY);
• ii)Average daily temperature (ºC, T_AVERAGE);
• iii)Maximum daily temperature (ºC, T_MAX);
• iv)Hourly relative humidity (%, RH_HOURLY);
• v)Average daily relative humidity (%, RH_AVERAGE);
• vi)Minimum daily relative humidity (%, RH_MIN);
• vii)Hourly THI (THI_HOURLY);
• viii)Average THI (THI_AVERAGE), using the T_AVERAGE and RH_AVERAGE; and
• ix)Maximum THI (THI_MAX), using the T_MAX and RH_MIN.

Descriptive statistics of the variables are reported in Supplementary Table 1. Although the variables obtained from INMET and NASA POWER displayed similar mean and standard deviation values, the amplitude of the INMET variables was slightly greater than that of the NASA POWER variables. That is, overall, the variables derived from INMET exhibited a lower minimum value and a higher maximum value.

The relationship between INMET and NASA POWER data was evaluated by Pearson’s correlation coefficient (ρ), slope of linear regression (β), and regression evaluation metrics (mean absolute error, MAE; and adjusted coefficient of determination, R²_adj). The β provided the pattern of the relationship between the variables, where lower MAE and higher R²_adj indicate better adjustment. The ρ provided the degree of the relationship between the variables.

Three points of view were considered: the hourly THI (T_HOURLY, RH_HOURLY, and THI_HOURLY), the average THI of the day (T_AVERAGE, RH_AVERAGE, and THI_AVERAGE), and the maximum THI of the day (T_MAX, RH_MIN, and THI_MAX). For each THI calculation type (hourly, average daily and maximum daily), THI information was considered for the whole Brazil (THI general) and within regions according to the Köppen's climate classification map for Brazil (Alvares et al. 2013).

Results and discussion

The β and ρ parameters indicated a positive correlation between all INMET and NASA POWER variables. A better fit and higher correlation were verified when evaluated with the average daily values, i.e., T_AVERAGE, RH_AVERAGE, and THI_AVERAGE (general), with higher ρ (0.92, 0.82, and 0.94, respectively), lower error, i.e., lower MAE (1.29, 6.61, and 1.60, respectively), and better regression fit, i.e., higher R²_adj (0.85, 0.67, and 0.89, respectively) compared with the hourly and maximum daily THI parameters (Table 1). The hourly information (T_HOURLY, RH_HOURLY, and THI_HOURLY general) showed the lowest ρ (0.69, 0.56, and 0.79, respectively), highest MAE (3.46, 15.35, and 3.80, respectively) and lowest R²_adj (0.49, 0.31, and 0.62, respectively) (Table 1).

Table 1.

Linear regression parameters, regression evaluation metrics and Pearson´s correlation coefficients comparing the weather parameters and temperature-humidity index (THI) values from the National Aeronautics and Space Administration Prediction of Worldwide Energy Resources (NASA POWER) with the corresponding values from the National Institute of Meteorology (INMET) weather stations, according to the THI calculation type (hourly, average daily or maximum daily)^a

	Variable	β* ± SE	R2adj	MAE	ρ*
Hourly THI	Temperature (hourly)	0.67 ± 0.0001	0.49	3.46	0.69
	Humidity (hourly)	0.52 ± 0.0001	0.31	15.35	0.56
	THI general (whole Brazil)	0.77 ± 0.0001	0.62	3.80	0.79
	THI Köppenb Af	0.78 ± 0.0004	0.49	2.95	0.70
	THI Köppen Am	0.71 ± 0.0003	0.47	3.15	0.69
	THI Köppen As	0.59 ± 0.0003	0.41	3.23	0.64
	THI Köppen Aw	0.66 ± 0.0002	0.45	3.57	0.67
	THI Köppen BSh	0.53 ± 0.0004	0.40	3.33	0.63
	THI Köppen Cfa	0.72 ± 0.0002	0.61	4.60	0.78
	THI Köppen Cfb	0.72 ± 0.0003	0.55	4.99	0.74
	THI Köppen Cwa	0.64 ± 0.0004	0.42	4.31	0.65
	THI Köppen Cwb	0.63 ± 0.0005	0.42	4.48	0.65
Average daily THI	Temperature (daily average)	0.92 ± 0.0003	0.85	1.29	0.92
	Humidity (daily average)	0.76 ± 0.0004	0.67	6.61	0.82
	THI general (whole Brazil)	0.95 ± 0.0002	0.89	1.60	0.94
	THI Köppen Af	0.97 ± 0.0012	0.82	1.30	0.91
	THI Köppen Am	0.90 ± 0.0008	0.82	1.38	0.91
	THI Köppen As	0.82 ± 0.0014	0.65	1.61	0.80
	THI Köppen Aw	0.91 ± 0.0005	0.82	1.38	0.91
	THI Köppen BSh	0.91 ± 0.0012	0.82	0.92	0.90
	THI Köppen Cfa	0.90 ± 0.0005	0.91	1.68	0.96
	THI Köppen Cfb	0.89 ± 0.0013	0.76	2.76	0.87
	THI Köppen Cwa	0.87 ± 0.0014	0.76	1.84	0.87
	THI Köppen Cwb	0.92 ± 0.0017	0.73	2.37	0.86
Maximum daily THI	Temperature (daily maximum)	0.87 ± 0.0004	0.74	2.04	0.86
	Humidity (daily minimum)	0.75 ± 0.0004	0.61	9.09	0.78
	THI general (whole Brazil)	0.99 ± 0.0003	0.81	1.82	0.90
	THI Köppen Af	0.96 ± 0.0016	0.72	1.56	0.85
	THI Köppen Am	0.92 ± 0.0013	0.67	1.55	0.82
	THI Köppen As	0.82 ± 0.0020	0.48	1.85	0.69
	THI Köppen Aw	0.96 ± 0.0008	0.71	1.52	0.84
	THI Köppen BSh	0.88 ± 0.0018	0.63	1.56	0.79
	THI Köppen Cfa	0.93 ± 0.0006	0.86	1.88	0.93
	THI Köppen Cfb	0.92 ± 0.0016	0.67	3.10	0.82
	THI Köppen Cwa	0.90 ± 0.0018	0.69	1.79	0.83
	THI Köppen Cwb	0.93 ± 0.0022	0.62	2.77	0.79

^aβ = slope of the linear regression of weather parameters and THI values from NASA POWER on the corresponding INMET values; SE = standard error; R²_adj = adjusted coefficient of determination; MAE = mean absolute error; ρ = Pearson's correlation coefficient

^bKöppen’s climate classification: Af = Tropical without dry season; Am = Tropical monsoon; As = Tropical with dry summer; Aw = Tropical with dry winter; BSh = Dry semi-arid low latitude and altitude; Cfa = Humid subtropical oceanic climate, without dry season with hot summer; Cfb = Humid subtropical oceanic climate, without dry season and with temperate summer; Cwa = Humid subtropical with dry winter and hot summer; Cwb = Humid subtropical with dry winter and temperate summer

* All β and ρ were significant (p-value < 0.0001)

Similar behavior was observed for THI analyzed according to Köppen's climate classification, i.e., the ρ were higher, regressions showed better fit and MAEs were lower for average daily THI, followed by maximum daily and hourly THI, respectively (Table 1). Within Köppen's climate classification, ρ coefficients among the hourly, average daily and maximum daily THI ranged from 0.63 (BSh) to 0.78 (Cfa); 0.86 (Cwb) to 0.96 (Cfa); and 0.69 (As) to 0.93 (Cfa), respectively (Table 1).

Additionally, an hour-by-hour comparison was performed. In this instance, the variables were averaged based on UTC, incorporating data from all weather stations. This involved computing the mean hourly values of temperature, relative humidity, and temperature-humidity index considering all stations at each hour (mean of all 00:00 UTC, all 01:00 UTC, all 02:00 UTC, and so forth). This was conducted to verify if there exists a variation in the correlation between the different times of the day for the two information sources. In this case, correlations ranged from 0.73 to 0.89 for temperature values (Supplementary Table 2); 0.54 to 0.75 for relative humidity values (Supplementary Table 3); and 0.84 to 0.91 for temperature-humidity index values (Supplementary Table 4). There was no specific time of day found to have a higher correlation than others, nor was there a pattern of correlation based on the time of day.

Conclusions

NASA POWER satellite-based weather system was shown to be a suitable tool for obtaining the average and maximum THI values using information collected from Brazil, showing high correlations with THI estimates from INMET and good regression evaluation metrics. We recommend caution when using the NASA POWER data to calculate the hourly THI, since the correlation with INMET was slightly lower, although moderate to high (0.79).

Therefore, NASA POWER satellite-based weather system can assist studies that aim to analyze the impact of heat stress on livestock production in Brazil, providing additional data to complement the existing information available in the INMET database. It is particularly important for regions with low coverage by INMET.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

This research received a financial support from “Conselho Nacional de Desenvolvimento Científico e Tecnológico” (CNPq, grant number 150156/2022–7) and EMBRAPA (SEG number 10.19.03.064.00.00).

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

We certify that there is no conflict of interest with any financial organization regarding the material discussed in the manuscript. In addition, the authors have non-financial interests to disclose.