Determination of the Correlation between the Air Temperature Measured in Situ and Remotely Sensed Data from MODIS and SEVIRI in Congo-Brazzaville

This study compared data from the MODerate-resolution Imaging Spectroradiometer (MODIS) onboard NASA’s Terra satellite and the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on EUMETSAT’s Meterosat Second Generation (MSG) satellite with in situ data obtained from ground observation stations in Congo-Brazzaville. Remote sensing instruments can be used to estimate air temperature, which has an important role in monitoring the effects of climate change. Congo-Brazzaville is located in equatorial forest, which is difficult to access, and has a limited number of ground meteorological stations measuring air temperature. This study used MODIS and MSG data for the period 2009-2014 to assess the performance of land surface temperature data from satellites against in situ data from ground-based stations in Congo-Brazzaville using a linear regression model. This work has allowed us to determine which satellite is best adapted for use in Central Africa.


Introduction
Central Africa has a rich biodiversity, but there have been few studies of the dynamics, mass balance and regional climatology of this part of the world.Land surface temperature (LST) data from remote sensing satellites have been used to study the atmospheric processes in this region, either as a substitute for, or in combination with, more spatially limited ground measurements of the nearsurface air temperature.However, there is a need to improve the assessment and reconstruction of the spatial and temporal variability of the measurements of air temperature at the ground surface recorded by instruments onboard satellites.
Remote sensing technology is a powerful tool to regularly monitor and evaluate the Earth's surface.A major challenge for the scientific community is to ensure that these sensors are correctly calibrated.Post-launch, the onboard instruments are radiometrically calibrated by simulating signals from the surface and the atmosphere based on a single site, which will cause errors between different satellite sensors.The lack of good quality remotely sensed data from different countries and different satellites with multi-source consistency has the potential to limit the scope of remote sensing.
It is therefore important to assess the quality of remote sensing data in Central Africa and to develop a feasible strategy for the quality control of these data (e.g. the AMESD and MESA projects) to ensure the performance of satellites, the quality of the data obtained from them and to guarantee accuracy.This paper reports a comparison of remote sensing data with temperature data measured at ground level to determine whether data from satellites can be used in research work in Central Africa and in the Congo region in particular.
The LST is an important measurement in the global system of data collection by specialized international organizations [1].LST data have previously been used to improve the quality of prediction in global weather models [2] [3].The LST is an important parameter in interdisciplinary biological studies, in monitoring ecosystems [4] [5] [6] [7] [8] and in determining the energy balance between the Earth's surface and the atmosphere.The LST is observed by a number of different satellites, each of which is equipped with several different instruments.NASA's Terra satellite is equipped with MODIS (MODerat-resolution Imaging Spectroradiometer), which measures the reflectance of the Earth in 36 spectral bands at a medium spatial resolution and has been designed to monitor medium-and large-scale processes.EUMETSAT's Meterosat Second Generation (MSG) satellite is equipped with the SEVIRI (Spinning Enhanced Visible and Infrared Imager) sensor and transmits a scene recorded in 12 channels of visible, mid-infrared and thermal infrared every 15 minutes with increased spatial resolution [9].This study used LST data from the MSG and MODIS Terra satellites.
A number of studies have validated the algorithms for the MODIS LST products, taking into account the influence of meteorological parameters such as wind speed and air temperature, the zenith angle of the sensor view [10] [11], the radiometric resolution and the calibration in several thermal infrared bands.
The LST products have been used to analyze land use [12] [13] and land cover [14] [15] [16] and the thermal imager has been used to forecast freezing conditions [17].This study first determined the number of days on which these two satellites made observations over Central Africa, which corresponds to the number of days with clear skies.The daily and annual temporal distributions of the LST were compared using statistical analysis.The LST behavior was then analyzed with the inclusion of climatic factors.
This paper is structured as follows.Section 2 describes the study area, data and methodology.Section 3 presents the results and their interpretation based on our analysis.Conclusions and recommendations drawn from our findings are outlined in Section 4. The MSG is a geostationary satellite covering Europe, Africa, and parts of the Atlantic and Indian oceans.It provides images at 15-min intervals [22].The LST MSG data were recovered in the 10.8 and 12.0 μm thermal infrared spectral bands using the generalized split-window algorithm, which corrects for atmospheric effects on the basis of differential absorption in two adjacent infrared bands [23] [24] [25] [26] [27].The dataset is also available in HDF at https://landsaf.ipma.pt.

Study Area
Congo-Brazzaville (Figure 1) covers 342,000 km 2 and is located in Central Africa between (3˚30'N and 5˚S) and (11 and 18˚E).It is bordered to the west by Gabon, to the northeast by Cameroon, to the north by the Central African Republic, to the east and southeast by the Democratic Republic of Congo, and to

Methodology
Data derived from the satellite LST products were downloaded from Landsaf.ipma.ptfor the MSG data and lpdaac.usgs.govfor the MODIS Terra data.
The data for each day at the time of passage of the satellite over Congo-Brazzaville were stored in two separate data folders for the analysis period 2009-2014.The data files were extracted to provide time series by hour, day and year [31].To reduce the time required to read and analyze these data, Matlab and ENVI were used to write processing codes.
The codes were intended: 1) to read HDF files to re-projection; 2) to resize the data according to the study area; and 3) to extract the values of each pixel based on the geographical coordinates of each station to generate a file of the time series for the data.The time series were limited to the available observations of the satellites, which were limited by the presence of clouds.Another code was written to match the dates of the LST time series with the in situ data for each station.We numerized the in situ data and used a regression method to determine the correlation coefficient between the MODIS and MSG data and the in situ data.The number of cloudless days is defined by Equation ( 1): Percentage number of data cloudless days *100% where Cd N is the number of data points for cloudless days and Y N is the number of days per year.
To determine the relationship between the MODIS and MSG LST data and the in situ data, we used the Pearson correlation coefficient (r) and the root mean square error (RMSE).These values show the strength of the link (positive or negative) between the variables.According to the regression model, the r correlation determines a dimensionless scale in the range 0 -1 and can be expressed as a percentage.The RMSE statistic (Equation ( 2)) [32] was used to assess the level of agreement between the observed and estimated temperatures for each site.Statistical analysis was conducted using the Originlab statistical software package.
( )  influence is locally accentuated by the extensive forest cover and the large number of lakes, rivers and wetlands [36].Toward the southeast (Brazzaville) at 700 hPa, the relative humidity varies between 60 and 80% and extends to the southwest (Pointe-Noire).This corresponds to the dry season in the southern hemisphere in June-September.At 850 hPa in the northern hemisphere, the average relative humidity varies between 70 and 80% from November to July.When the relative humidity at the surface increases, the clouds change from shallow to deep convective types as a result of an increase in the available potential energy [37] [38].

Determination of the Number of Cloudless Days
The results shown in Figure 3 and Figure 4 confirm that the presence of cloud cover over the Congo-Brazzaville is responsible for the low number of daily satellites observations shown in Figure 2.  3).
The data from Ouesso, Brazzaville, Dajambala and Gamboma stations show In general, the values of the MSG LSTs overestimated the observed temperature, whereas the MODIS LST values underestimated the observed temperature (Table 3).These results are in agreement with previously reported work [41].

Conclusions
We carried out a comparative study between the data obtained from the MSG and MODIS LST products and the air temperature data measured at ground level to determine whether these satellite data are suitable for use in research work in Central Africa.There is currently little published climate data available for this part of Africa.
The MSG LSTs showed a moderate correlation with air temperature measurements at 2 m above ground level at meteorological ground stations in the equatorial climate zone and good accuracy at some of the stations in the area.
The MODIS LST data also showed good precision at some stations, especially those located in zones characterized by a subequatorial climate or humid tropical climate.These results corroborate those obtained in previous studies on the validation of satellite data [43] [44] [45].
The air temperature measured by these satellites can be used in future studies of climate change and in high-resolution regional climate models.The small amount of data from the two types of temperature measurement are a result of a number of factors, including temperature inversion, heterogeneities in surface emissivity and self-limiting satellites.
Future work should be carried out to evaluate LST products from other Earth observation satellites and to compare these results with a numerical model for this region.

Figure 1 .
Figure 1.Map of the Congo-Brazzaville showing the types of climate and the locations of weather stations (reproduced from the Atlas du Congo (2001), 2 edn).

Figure 2 Figure 2 .
Figure2shows the percentage of LST measurements obtained from the MODIS and MSG satellites each year from 2009 to 2014.These values correspond to the days when the sky was clear and the satellites were able to make measurements.The satellite data were not generated continuously because the cloud cover in this region may last for several consecutive days.Congo-Brazzaville is located in an area where there may be persistent cloud coverage of 100%-for example, for the Pointe-Noire station, over the six years studied, LST measurements were

Figure 4 Figure 3 .
Figure 4 shows that the average relative humidity at 700 hPa in the period 2009-2014 was >70% throughout the year.The humidity at 850 hPa varied between 80% and 100%.This high relative humidity is the consequence of the regular humid trade winds and Atlantic monsoon in this part of the world.Their

Figure 5 Figure 4 .
Figure 5 shows the annual cycle in the variation in temperature between the air temperature measured at the ground stations and the LST product from the MSG and MODIS satellites.During the period June-September, corresponding to the dry season, the temperatures generally decreased (e.g. at the Gamboma Brazzaville and Pointe-Noire stations).By contrast, the seasonal variations were
53˚C and those from MAE range from 1.94˚C to 5.02˚C.The MSG LST data did not give better accuracy, with the exception of Kelle station, where the correlation coefficient r = 0.73, RMSE = 2.36˚C and MAE = 3.68˚C in 2010.A better accuracy was obtained with the MODIS LST data, with RMSE = 2.28˚C, MAE = 1.94˚C and r = 0.4 in 2009.The stations located in the area with a humid tropical climate were identical to those in the subtropical zone, with the exception of Brazzaville station, which had a correlation coefficient > 0.5 between 2009 and 2014, Pointe-Noire station with a correlation coefficient of 0.51 in 2009, and Makabana station with a correlation coefficient of 0.663, RMSE = 2.5˚C and MAE = 1.97˚C in 2014.A better accuracy was observed between the MSG LST and the air temperature measured at ground level.There was a better accuracy between the MODIS LST and the air temperature measured at ground level in 2014 than in other years.
Satellite technology based on imaging in the thermal infrared region presents an opportunity to measure ground temperatures at different spatial and temporal scales for use in in-depth studies of biological, hydrological and climatological ecological systems and in the identification of surface-atmosphere interactions M. O. C. Kambi et al.DOI: 10.4236/acs.2018.82013194 Atmospheric and Climate Sciences The aim of this study was to assess the performance of the MODIS LST product aboard NASA's Terra satellite and the LSF edge onboard the MSG satellite to characterize the spatio-temporal variations in the LST over a six-year period from 2009 to 2014.To achieve this, the MODIS and MSG data were compared with the surface temperature measured at meteorological stations in Congo-Brazzaville.

Table 1 .
Satellite image products used in the study.

Table 2 .
Meteorological stations by climate type in the Congo-Brazzaville.

Table 3 .
Mean difference between the MSG LST air temperature and the MODIS LST air temperatures.
moderate correlation coefficients of 0.23 for the MSG LST and 0.64 for air temperature.The RMSEs vary between 1.7˚C and 2.7˚C and MAEs ranged from 1.6˚C to 4.66˚C from 2009 to 2014.There is a moderate correlation coefficient

Table 4 .
Correlation coefficient (r) and RSME between the LST data and the air temperature in regions with an equatorial climate.

Table 5 .
Correlation coefficient (r) and RSME between the LST data and the air temperature in regions with a subequatorial climate.

Table 6 .
[43]elation coefficient (r) and RSME between the LST data and the air temperature in regions with a tropical humid climate.The stations located in the area of subequatorial climate showed moderate correlation coefficients between the MSG LST and the air temperature measured at ground level, varying between 0.25 and 0.487, with the exception of the Ewo and Djambala stations with correlation coefficients between 0.062 and 0.20.The correlation coefficients between MODIS LST and the air temperature measured at ground level are very low (between 0.05 and 0.28), with the exception of Kelle station, which showed moderate values between 0.4 and 0.58.The RMSE between the MSG LST and the air temperature measured at ground level varied between 2˚C and 3.88˚C and those of MAE between 2.64˚C and 9.26˚C.Similar results for MAE in West Africa have been reported previously[43].The RMSE values between the MODIS LST data and the air temperature measured at ground level range from 2.49˚C to 4.