Air temperature is an important meteorological variable for understanding the physics of land surface processes. This variable is commonly used to monitor vegetation water stress, assess the surface energy balance, estimate precipitation and crop yields and to derive daily evaporation and soil moisture distribution over large areas (Lagouarde 1991; Rugege 2002; Taconet et al. 1986). Air temperature is also extensively used to weather forecast (Smith et al. 1988) and to study climate change on a global to local scale (Aikawa et al. 2008; Jones et al. 1986a; Jones et al. 1986b; Söderström and Magnusson 1995). In addition, air temperature influences crop growth and livestock well-being, both in terms of production capacity and spatial distribution. Maximum, average and minimum are the three components of air temperature commonly used in the energy and water cycle of the Earth–Atmosphere system. They are the key input elements for hydrology and land surface models. Air temperature is one of the most commonly recorded parameters at government weather stations (Lozada García and Sentelhes 2008), it is generally measured by temperature sensors of ground weather stations in a standardized way (Unger et al. 2001). The sensors must be well protected from solar radiation and adequately ventilated and air temperature must be measured at 1.5m to 2m above the ground, to assure quality and comparability of measurements. When air temperature is measured at different height levels, its values are corrected to ensure the inter-comparability.
The sparse availability of in situ measurements of meteorological variables is the main problem encountered when realizing agrometeorological studies over large geographic areas. This is mainly due to the low density of meteorological stations especially in rural areas. In addition, they are often located near urban or built-up areas like airports or experimental sites and are, as such, far from agricultural areas. Depending on the topography of each region, the maximum distance between two weather stations should not exceed 60km to characterize the seasonal patterns in spatial mean air temperature. The same distance should not exceed 30km to characterize the seasonal patterns in spatial minimal air temperature (Hubbard 1994). Data collection and their analysis and the maintenance of sensors constitute other significant factors affecting the accuracy and quality of data.
In Morocco, the Direction of National Meteorology (DNM) is the responsible agency for collecting, analyzing and interpolating weather data as well as weather forecasting at the national level. It manages a network of 46 synoptic weather stations throughout the country. Among these stations, 18 are explored for aeronautical purposes. The density of weather stations is relatively low in Morocco and is better in coastal plains than in mountainous regions, occidental and desert regions (Balaghi et al. 2012).
The accuracy of the interpolation methods is strongly affected by many local parameters like the density of weather stations, altitude of the station, consistency of measurements at a particular station, sun exposure of the sensors, the topography of the region and the distance from the coast (Hudson and Wackernagel 1994; Marquı́nez et al. 2003; Söderström and Magnusson 1995). The accuracy of the interpolated grids represents a major limitation for specific operational applications for agriculture monitoring and crop yield modeling.
Satellite sensors measure temperature fluxes and can be used to estimate air temperature consistently over large geographic areas. These estimations could contribute to the quality and accuracy of the interpolated grids, in particular in areas where weather stations are scarce or less representative for vast regions. Satellite derived air temperature estimations are specifically useful for Africans countries in general and for Morocco in particular given the limited number of available meteorological stations, large north-south gradient (between 21°N and 36°N north), as well as the topographic and environmental diversity of the country.
The Moderate Resolution Imaging Spectroradiometer (MODIS), aboard the Terra and Aqua satellites (http://modis.gsfc.nasa.gov), and the Advanced Very High Resolution Radiometer (AVHRR), carried on NOAA Satellites (http://noaasis.noaa.gov/NOAASIS/ml/avhrr.html; http://www.noaa.gov) are the two most used radiometers for estimating surface temperature, called also surface skin temperature in the literature. Air temperature and surface temperature are significantly different in terms of magnitude and diurnal cycle due to the different heat capacity between air and land, and weather conditions (Jin and Dickinson 2010). However, both temperatures are correlated due to heat exchange between air and earth surface. In the literature, many studies evaluated the accuracy of air temperature, principally the maximum air temperature component, calculated from satellite derived surface temperature in various parts of the world: northern China and the former USSR (Shen and Leptoukh 2011), Chile (Fabiola Flores and Mario Lillo 2010), Northern Québec and Alaska (Hachem et al. 2012), Croatia (Hengl et al. 2012), Southern Spain (Vogt et al. 1997), Spain (de Wit et al. 2004), over Africa (Vancutsem et al. 2010) and worldwide (Mildrexler et al. 2011). All these studies showed high correlation between air temperature and surface temperature derived from satellite data, but reported variability in accuracy depending on topography and land cover (vegetation, water, snow). It is thus important to conduct local studies and propose the best suited methods to each country context.
In Morocco, satellite based air temperature estimations over agricultural areas have been little studied. This paper evaluates local correlations to derive average and maximum air temperature from surface temperature measured by NOAA-AVHRR sensor in Morocco. The methodology adopted here consists in a first step to compare surface temperature, derived from 349 NOAA-AVHRR cloud free images and the corresponding maximum and average air temperatures measured between 1995 and 2012 by nine meteorological stations well distributed throughout the study area. The location of each station (latitude and longitude) was used to extract the surface temperature of corresponding pixel.
In a second step, remotely surface temperature was compared to interpolated average and maximum air temperature of agricultural zones of 46 provinces in Morocco, covering about 328311km², between 1995 and 2012. The stability of obtained correlations were then evaluated using K-Fold cross validation approach (Cassel 2007).The accuracy of obtained regression models were finally discussed trough their spatial analysis.