Skip to main content
NCBI home page
UKPMC Funders Author Manuscripts logoLink to UKPMC Funders Author Manuscripts
. Author manuscript; available in PMC: 2012 May 11.
Published in final edited form as: Remote Sens Environ. 2002 Feb;79(2-3):166–175. doi: 10.1016/S0034-4257(01)00270-X

The potential of Pathfinder AVHRR data for providing surrogate climatic variables across Africa and Europe for epidemiological applications

Robert M Green 1, Simon I Hay 2
PMCID: PMC3350066  EMSID: UKMS47533  PMID: 22581983

Abstract

Surface climatic conditions are key determinants of arthropod vector distribution and abundance and consequently affect transmission rates of any diseases they may carry. Remotely sensed observations by satellite sensors are the only feasible means of obtaining regional and continental scale measurements of climate at regular intervals for real-time epidemiological applications such as disease early warning systems. The potential of Pathfinder AVHRR Land (PAL) data to provide surrogate variables for near-surface air temperature and vapour pressure deficit (VPD) over Africa and Europe were assessed in this context. For the years 1988-1990 and 1992, correlations were examined between meteorological ground measurements (monthly mean air temperature and VPDgrd) and variables derived from Advanced Very High Resolution Radiometer (AVHRR) data (LST and VPDsat). The AVHRR indices were derived from both daily and composite PAL data so that their relative performance could be determined. Furthermore, the ground observations were divided into African and European subsets, so that the relative performance of the satellite data at tropical/sub-tropical and temperate latitudes could be assessed.

Significant correlations were shown between air temperature and LST in all months. Temporal variability existed in the strength of correlations throughout any twelve-month period, with the pattern of variability consistent between years. The adjusted r2 values increased when elevation and the Normalised Difference Vegetation Index (NDVI) were included, in addition to LST, as predictor variables of air temperature. Attempts to derive monthly estimates of atmospheric moisture availability resulted in an over-estimation of VPDsat compared to ground observations, VPDgrd. The use of daily PAL data to derive monthly mean climatic indices was shown to be more accurate than those obtained using monthly maximum values from 10-day composite data. A subset of the 1992 data was then used to build linear regression models for the direct retrieval of monthly mean air temperature from PAL data. The accuracy of retrieved estimates was greatest when NDVI was included with LST as predictor variables, with root mean square errors varying from 1.83°C to 3.18 °C with a mean of 2.38 °C over the twelve months.

Keywords: AVHRR, land surface temperature, vapour pressure deficit, epidemiology, Africa, Europe

INTRODUCTION

Abiotic climatic conditions are critical factors in determining the geographical distribution of arthropod vectors and the incidence of infections they can transmit (Hay et al., 1997). For example, environmental variables derived from meteorological satellite sensors have been used to explain the distribution and seasonal abundance of economically and medically important tick species (Randolph, 1993) or tick-borne diseases at both temperate and tropical latitudes (Randolph, 2000), as well as the distribution and abundance of a range of tsetse fly species in West Africa (Rogers et al., 1996; Rogers, 2000). Spatial and temporal fluctuations in climatic conditions also influence disease risk, such as the seasonality of malaria in human populations (Hay et al., 1998a; 1998b), as well as providing the information to drive proposed epidemic early warning systems for malaria (Connor et al., 1999; Hay et al., 2000a). Reliable assessments of the changing risk to human populations requires accurate, timely and extensive (geographically and temporally) information on climate (Hay, 1997) and is therefore of considerable interest to epidemiologists and public health workers alike (Hay, 2000; Hay et al. 2000b).

Studies that require meteorological data for large geographical areas depend on the regional infrastructure for meteorological data collection. These data are typically collected as point samples from weather stations, whose distribution is rarely designed to capture the range of climate within a region. Dissemination periods of these climatic data are also variable, limiting their use for real-time predictions. The paucity in spatial resolution of meteorological records may be compensated for by various methods of interpolation between known sites. This has proved very successful in predicting temperatures distant from meteorological stations, but measurements need to be of relatively high density and uniformly distributed (Lennon and Turner, 1995). Remotely sensed observations by satellite sensors, by contrast, allow quantitative measurements of key climatic variables at regular intervals over regional and global scales (Hay et al., 1996; Goetz et al., 2000).

This paper reports on the potential of Pathfinder AVHRR Land (PAL) data to provide continental-scale surrogate variables of near-surface air temperature and vapour pressure deficit, two climatic variables important in understanding arthropod vector ecology and disease distribution. The PAL data (James and Kalluri, 1994) are used to calculate monthly indices of land surface temperature (LST) and vapour pressure deficit (VPDsat), which are then compared with monthly ground observations of mean air temperature (Ta) and (VPDgrd), respectively, in Africa and Europe. In addition, the AVHRR observations are composited into monthly values using different methods, and their relative performances as surrogates of climatic conditions are assessed using an independent data set (1992 daily data for Europe). Finally, estimates of monthly mean air temperature are retrieved from empirical regression models and compared to surface observations.

This study recognises that the relationship between surface temperature and air temperature is rarely 1:1 and that other studies have concentrated on estimating air temperature directly from AVHRR data by utilising the negative relationship between LST and the Normalised Difference Vegetation Index (NDVI) (e.g. Prince et al., 1998; Czajkowski et al., 1997; Prihodko and Goward, 1997; Prince and Goward, 1995). This has become known as the TVX (temperature/vegetation index) method (Goward et al. 1994; Prince and Goward, 1995). Epidemiological applications often need extended temporal climatic data sets, spanning years rather than days or months. The TVX method is computationally demanding so that the simplicity of calculating LST using the Price (1984) using Pathfinder AVHRR data was favoured. The suitability of the TVX method in epidemiological mapping will be assessed at a future date.

This investigation is an extension of previous work (Hay and Lennon, 1999), which assessed the accuracy of both spatial interpolation and remote sensing methods for providing land surface temperature, vapour pressure deficit and rainfall surfaces for Africa during 1990. The present study uses a longer period of observations (1988-1990 inclusive) to compare the accuracy of remotely sensed data for providing LST and VPDsat surfaces in tropical Africa and temperate Europe.

MATERIALS AND METHODS

The analyses consisted of comparisons between meteorological variables derived from ground observations (monthly mean Ta and VPDgrd) and variables calculated using thermal channels 4 and 5 from PAL data (LST and VPDsat). Monthly meteorological observations are published by the NOAA National Climatic Data Center (NCDC) for selected meteorological stations that adhere to World Meteorological Organisation (WMO) instrumentation and data collection standards (NOAA, 1990). Monthly mean air temperature, Ta (°C), and monthly mean vapour pressure, Vp (mb), observations were collated from 250 and 178 meteorological stations in Africa (1988-1990 inclusive) and Europe (1992 only) respectively (Figure 1).

Figure 1.

Figure 1

Open in a new tab

Distribution of the NOAA-NCDC meteorological stations used in this study.

VPDgrd (mb) was calculated from Ta in Kelvin, and Vp from formulae provided by Unwin (1980)

VPDgrd=Svp−Vp (1)

where the saturation vapour pressure, Svp (mb) was given by:

log10Svp=9.24349−(230∕Ta)−(500∕Ta2)−(100000∕Ta3) (2)

The VPD is a measure of the lack of moisture equilibrium between an object and the surrounding atmosphere; the higher the VPD the more rapid the potential rate of desiccation.

PAL data (Agbu and James, 1994) were obtained from the Goddard Space Flight Center Distributed Active Archive Center (GSFC-DAAC) for the same duration as the meteorological data. PAL data are distributed with 12 data layers per observation, including five bands of spectral AVHRR data, the NDVI, and six bands of ancillary information, including date, sun/sensor geometry and data quality flags. Channel 4 (Ch4) and channel 5 (Ch5) brightness temperatures were used to calculate LST using the equation of Price (1984),

LST=Ch4+3.33(Ch4−Ch5) (3)

and total precipitable water content in the atmospheric column, U, (kgm−2) (Eck and Holben, 1994) where;

U=A+B(Ch4−Ch5) (4)

and, A and B are constants, 1.337 and 0.837 respectively.

The estimated precipitable water content, U (in cm), was then converted to a near surface dew point temperature, Td (°F), the temperature to which a sample of air must be cooled for it to become saturated and condense, using the following relationship (Smith, 1966);

Td=lnU−((0.113−ln(λ+1)0.0393 (5)

where l is a variable that is a function of the latitude and the time of the year. In this analysis, mean values of l = 2.99 and 2.74 were calculated from the annual mean l presented by Smith (1966) for latitudes representing Africa and Europe, respectively.

The dew point temperature values were then converted to Kelvin and used to calculate VPDsat (KPa), using the equation provided in Prince and Goward (1995):

VPDsat=0.611[exp(17.27×Tair−273Tair−36)−exp(17.27×Td−273Td−36)] (6)

in which Tair represents air temperature. VPDsat was converted to units of mb for direct comparison with VPDgrd. In equation 6, remotely sensed estimates of LST were substituted for the Tair variable assuming that LST is an adequate surrogate. The validity of this assumption is assessed in the following analyses.

The relationship between LST and ambient air temperature is influenced, among other things, by the amount of vegetation cover at the Earth’s surface. In this analysis vegetation cover is represented by the NDVI:

NDVI=Ch2−Ch1Ch2+Ch1 (7)

where Ch2 and Ch1 are, respectively, reflectance measured by the infrared and red bands of the AVHRR. The NDVI is positively related to both green vegetation amount and its level of photosynthetic activity (Myneni, et al., 1995).

PAL data are supplied as daily images or decade composites (two 10-day and one variable compositing period per month). Each pixel in a ten-day composite stores data from a single date within a consecutive 10-day period when atmospheric contamination is considered to be at a minimum i.e. the date of maximum NDVI (Holben, 1986). For the period 1988-90 inclusive, daily images were used to calculate climatic indices for Africa, but the composite data were used for Europe, since the two continental data sets were originally derived separately. To assess the influence of the different PAL data types on the accuracy of the climatic surrogates, indices were calculated for Europe in 1992 using both daily and composite data.

All images were subjected to a quality control process whereby cloudy pixels as determined by the Clouds from AVHRR (CLAVR) algorithm (Stowe et al. 1991), were masked, as were pixels with extreme sun/sensor viewing geometry (see methodology in Hay and Lennon, 1999). The daily and 10-day LST images were condensed into single monthly images by taking the maximum value of LST over the monthly period, thereby reducing the occurrence of cloud-contaminated pixels still further. The rationale is that clouds are generally colder than the land so that the highest thermal value in the series will probably be the least cloud contaminated, especially at tropical latitudes (Lambin and Ehrlich, 1995). The VPD data were composited into monthly images by using the VPD from the date in the month or decade when the NDVI was highest. Very high latitude stations in Europe tended to return negative values of VPD in the winter months. These were excluded from the subsequent analyses.

LST and VPDsat were extracted from each monthly image for those pixels corresponding to the latitude and longitude of each meteorological station. Regression analyses were performed on the meteorological and satellite variables for each month, and for each year of data (1988-1990) to assess inter-annual consistency. It was hypothesised that Ta would be strongly correlated with LST at monthly temporal resolution. Since Ta can deviate from LST significantly depending on surface conditions (Goetz 1997), such as vegetation cover and topography, NDVI and elevation are added to the multiple regressions as dependent variables to test how they affected the amount of explained variance. Another hypothesis was that some of the unexplained variance may be related to the method used to composite the AVHRR data. More regression analyses are performed using an independent dataset (daily PAL data for Europe in 1992). Climatic indices extracted from the maximum monthly AVHRR value are termed MMAX variables and those using the satellite observation from the date of the maximum NDVI are termed MVC (Maximum Value Composite). In addition, the daily observations were used to derive a mean value from all the cloud-free observations recorded for the month and are termed MMEAN variables. The optimum combination of predictor variables was then used to build regression models to retrieve estimates of monthly mean near-surface temperature directly, and to assess their accuracy using root mean square errors (RMSEs). For this, the 1992 meteorological station data set for Europe are divided randomly in two; one half was used to build the regression models and the other to test the accuracy of retrieved estimates.

RESULTS AND DISCUSSION

Relationships between Remotely Sensed LST and Monthly Mean Ta Observations

There are consistent significant correlations between remotely sensed LST estimates and monthly mean Ta observations for both the African and European continents. Linear regressions show positive and significant relationships in all months (P << 0.01, minimum n = 105, maximum n = 164 observations). RMSEs ranged from 2.8 to 7.5 K with a mean of 4.8 K for Europe, and 3.2 to 4.7 K with a mean of 3.9 K for Africa. The adjusted r2, of the monthly data are plotted as a temporal profile (Figure 2a). The proportion of explained variance accounted for by LST suggests that this variable can be used with confidence as a reliable surrogate indicator of Ta over wide geographical areas with certain caveats.

Figure 2.

Figure 2

Open in a new tab

Coefficients of determination, r2, from regressions between monthly mean air temperature Ta and (a) monthly maximum LST (b) LST, elevation and monthly maximum NDVI, for each month from January 1988 to December 1990. NDVI values were unavailable for May 1989, Africa.

The important observations in figure 2a are the seasonal trends in the strength of the relationships and their consistency for all three years of data. If LST is to be used as a predictor of Ta, knowledge of the annual variability in the strength of the correlation and its cause is essential. Spatial and temporal variations in surface conditions, e.g. vegetation cover and elevation, are likely to be responsible for a large degree of residual variance in the regressions and the observed temporal variability in correlations. When NDVI and elevation, in addition to LST, are used as predictor variables of Ta in multiple regressions, r2 values are higher than those obtained using LST alone (Figure 2b and Table 1). The inclusion of NDVI as a predictor variable helps reduce temporal fluctuations in r2, probably because the vegetation phenology and its effects on surface thermal properties are represented. Elevation has not been explicitly parameterised in any equations to estimate LST from satellite data, but should be considered where application is required at the broad spatial scale or where the range in local elevation is large. At an altitude of 2000 m, for example, atmospheric attenuation of thermal IR radiation will be significantly less than for a site at sea level, since the radiation must travel through 2 km less atmosphere before reaching the satellite sensor. Such inaccuracies in the original remotely sensed signal will obviously reduce the accuracy of the LST estimate derived from Price (1984). Topographic variability also has a strong influence on ambient air temperature and, as with vegetation, can determine the difference between surface and air temperatures. The inclusion of elevation in the multiple regressions therefore improves r2 values in a similar fashion to the inclusion of NDVI. Fortunately, global elevation data, such as the United States Geological Survey Global Land Information System (USGS GLIS) exist at sufficient resolution to be included in continental scale analyses.

Table 1.

The mean of the January to December (1988-90) adjusted coefficients of determination (r2) obtained from linear regressions between monthly mean air temperature, Ta, and predictor variables (i) monthly maximum LST (ii) LST and elevation (iii) LST, elevation and NDVI. 1.96 standard errors are given in brackets.

LST LST + Elevation LST + Elevation + NDVI
Year Africa Europe Africa Europe Africa Europe
1988 0.58 (0.07) 0.52 (0.06) 0.69 (0.04) 0.63 (0.08) 0.70 (0.05) 0.71 (0.05)
1989 0.53 (0.09) 0.53 (0.06) 0.69 (0.05) 0.66 (0.07) 0.72 (0.06) 0.73 (0.07)
1990 0.52 (0.07) 0.55 (0.06) 0.66 (0.05) 0.70 (0.04) 0.70 (0.05) 0.77 (0.02)
Open in a new tab

Although there is little difference in the yearly mean of the monthly r2 values between the continental data sets (Table 1), the pattern of monthly variability in the LST/Ta relationship is not consistent between Africa and Europe (Figure 2a or 2b). Differences between the two continents exist in the timing of the maximum and minimum r2. For Africa, the maximum r2 values occurred around the months of May and June, whereas for Europe maximum r2 occurred from August to October. Within any given month, there is a broad latitudinal range of meteorological stations covering a wide diversity of climatic regions and biomes. Future analyses will stratify the meteorological stations with respect to latitude, to determine the significance of the differences in temporal patterns of r2 between Europe and Africa. The temporal consistency of regression slope coefficients and intercepts also needs to be examined.

This study has assumed that the empirical relationship between LST and Ta is sufficiently strong for practical applications within epidemiological mapping and the results support this to some extent. It is recognised, however, that the relationship between LST and Ta is not 1:1, as exemplified in the scatterplots (Figure 3), and that LST can deviate significantly from air temperature over areas varying in surface characteristics such as vegetation density, surface albedo and wetness. The influence of vegetation helps explain the increase in r2 when NDVI is introduced into the multiple regressions. As mentioned previously, other studies have concentrated on retrieving air temperature estimates directly from remotely sensed data by exploiting the negative relationship between Ts and NDVI (Prince et al., 1998; Czajkowski et al., 1997; Prihodko and Goward, 1997; Prince and Goward, 1995). Due to the low thermal inertia properties of vegetation canopies, ambient air temperature does not deviate greatly from canopy temperature, even when the vegetation is actively transpiring. The observed NDVI/LST relation is extrapolated to an NDVI of an infinitely thick vegetation canopy to provide an estimate of air temperature. In site-specific tests Goward et al., 1994 report accuracy of predicted air temperature typically within ±2°C of actual air temperature. More recent studies report an accuracy of 3.9°C over a range of 36°C (Prince et al., 1998), 5 K (Czajkowski et al., 1997) and 2.92°C (Prihodko and Goward, 1997). The aim of this study, however, was not to estimate air temperature directly, but to test whether Ts can be useful as a surrogate variable of Ta across continental scales and over extended periods of time.

Figure 3.

Figure 3

Open in a new tab

Comparison of Ta with LST for the months with highest correlations (a) sites across Europe for October 1989 (b) sites across Africa for December 1990.

The type and quality of the data used in this study may have significant influence on the observed relationships. Ideally, satellite data used for environmental applications should originate from a consistent source. A factor which complicates the comparability of the 1988-90 data series for Africa and Europe is that the African monthly composites were derived by processing daily Pathfinder data, whereas the European monthly images were derived from three 10-day composite images per month using the decade with the maximum NDVI value. It would be unwise to derive a monthly mean value from a maximum of only three satellite observations. Hence, comparisons have been made between remotely sensed maximum value composites and the monthly mean surface observations. This was not an ideal comparison but, as expected, maximum and mean values were significantly correlated.

The results show the benefit of investing more processing time in the PAL daily data to calculate a mean LST value (MMEAN index), using AVHRR channels 4 and 5 data from all the cloud-free dates left after CLAVR cloud-screening (Figure 4). The MMEAN index produces the highest r2 value in nearly all months of 1992 for Europe. Both indices (MMEAN, MMAX) derived from the PAL daily data out-perform the MVC index, derived from the Pathfinder composite data set based on the remotely sensed responses from the decade of maximum NDVI. Compositing using the date of maximum NDVI is popular for removing atmospheric contamination from the time-series, but minimum atmospheric attenuation of short-wave (optical bands) and long-wave (thermal bands) radiation do not necessarily coincide (Lambin and Ehrlich, 1996). The results show that to characterise near surface conditions of temperature accurately for a monthly period, it is best to composite a maximum value or calculate the mean value independent of the date of maximum NDVI, and only the daily PAL data set can provide the data for this.

Figure 4.

Figure 4

Open in a new tab

Coefficients of determination r2 from regressions between Ta and three LST indices derived from AVHRR data (MMAX = maximum per month of daily data; MMEAN = Mean per month of daily data; MVC = maximum value composite using LST from decade of maximum NDVI.

Having identified the significant empirical relationship between LST and Ta, and the remotely sensed index yielding the strongest correlations (MMEAN), the ability to retrieve Ta estimates empirically was assessed. Linear regression models were built using a randomly generated training data set of 1992 Europe meteorological observations. For each month, three separate regression models were derived using as predictor variables LST, LST+NDVI, LST+NDVI+elevation, respectively. Estimates of Ta were derived using the randomly generated testing data set of meteorological observations (independent from the training data set) as input to the regression models. All r2 values are significant at P<0.01, except December LST+NDVI+elevation (not significant). The r2 values derived using LST+NDVI are consistently higher than LST alone, except in July and August (Figure 5). When elevation is included as a predictor variable, the range of r2 values increases. The highest r2 values are obtained when elevation is included, but relatively low values in January, March, April and especially December (r2 =0.08) reduce the mean r2 to 0.58 (Table 3). RMSEs of retrieved estimates of monthly mean air temperatures were at best 1.77°C (Table 3) and, when averaged over 1992, ranged from 2.38 °C to 2.97 °C depending on the combination of predictor variables.

Figure 5.

Figure 5

Open in a new tab

The strength of correlations, r2, between observed and predicted Ta at selected meteorological stations in Europe 1992 using predictor variables LST; LST and NDVI; LST, NDVI and elevation.

Table 3.

The correlations (r2) between predicted and observed monthly mean air temperatures and the accuracy of retrieved estimates (RMSE) for 1992.

LST LST + NDVI LST + NDVI + elevation
Month r2 rmse r2 rmse r2 rmse n
Jan 0.46 2.44 0.50 2.33 0.39 3.46 23
Feb 0.24 2.60 0.34 2.67 0.37 2.91 47
Mar 0.50 2.07 0.60 1.83 0.65 2.27 53
Apr 0.75 2.12 0.78 2.01 0.62 4.00 58
May 0.67 2.04 0.70 1.96 0.77 1.77 64
Jun 0.50 2.23 0.51 2.20 0.52 2.21 65
Jul 0.79 1.99 0.79 2.00 0.86 1.64 65
Aug 0.72 2.78 0.70 2.88 0.45 5.30 78
Sep 0.73 2.72 0.73 2.72 0.87 2.17 75
Oct 0.78 3.35 0.80 3.18 0.80 3.63 67
Nov 0.63 1.99 0.65 1.97 0.61 2.62 53
Dec 0.51 2.87 0.57 2.79 0.08 3.66 15
Mean 0.61 2.43 0.64 2.38 0.58 2.97
Open in a new tab

Relationships between Remotely Sensed VPD and Monthly Mean Ground Observations

Significant positive linear relationships (P < 0.01) exist between VPDgrd and VPDsat for the African data set (Figure 6a), with RMSEs ranging from 4.7 to 7.1 mb with a mean of 5.8 mb. In contrast, r2 values for the European data set are in most cases lower (Figure 5 and Table 2). RMSE for Europe range from 0.7 to 3.5 mb with a mean of 1.6 mb, which are smaller than for Africa, but reflect the narrower dynamic range and smaller magnitudes of data values for Europe, as well as the problem of these techniques in dealing with low temperature VPD retrieval. The inclusion of elevation, or elevation and NDVI, as additional predictor variables only marginally improved the fit to the ground data (Figure 6b).

Figure 6.

Figure 6

Open in a new tab

Coefficients of determination, r2, from regressions between VPDgrd and (a) VPDsat (b) VPDsat, elevation and monthly maximum NDVI, for each month from January 1988 to December 1990. NDVI values were unavailable for May 1989, Africa.

Table 2.

The mean of the January to December (1988-90) adjusted coefficients of determination (r2) obtained from linear regressions between monthly mean saturation deficit and predictor variables (i) monthly maximum AVHRR derived vapour pressure deficit VPDsat (ii) VPDsat and elevation (iii) VPDsat, elevation and NDVI. 1.96 standard errors are given in brackets.

VPD VPD + Elevation VPD + Elevation + NDVI
Year Africa Europe Africa Europe Africa Europe
1988 0.63 (0.07) 0.35 (0.11) 0.66 (0.06) 0.37 (0.10) 0.68 (0.05) 0.38 (0.10)
1989 0.63 (0.07) 0.48 (0.10) 0.67 (0.05) 0.50 (0.10) 0.68 (0.05) 0.50 (0.10)
1990 0.63 (0.04) 0.49 (0.11) 0.66 (0.03) 0.51 (0.12) 0.68 (0.03) 0.52 (0.10)
Open in a new tab

Scatterplots (Figure 7) reveal that remotely sensed estimates of VPD are significantly larger than the ground observations. VPDsat can exceed 200 mb which, compared to expected maxima approximating 70 to 80 mb, are unrealistic. The calculation of VPDsat involves a number of stages in which errors can arise and propagate. Atmospheric water vapour (U) can be problematic (Czajkowski et al., this issue) and in this instance is estimated from an empirical algorithm developed using meteorological data from Mali (Eck and Holben, 1994), but its applicability to other geographical regions needs to be assessed. Estimating U from algorithms derived from established physical models may be a preferable alternative to estimates derived from purely empirical observations. For example, Prince and Goward (1995) used an atmospheric radiative transfer model, LOWTRAN-7 (Kneizys et al., 1988), to derive an empirical relationship between U and the function of LST and AVHRR channel 4/channel 5 radiant temperature differences.

Figure 7.

Figure 7

Open in a new tab

Comparison of VPDgrd and VPDsat for the months with highest correlations (a) sites across Europe - July 1990 (b) sites across Africa - April 1989.

The major source of error in the VPDsat, however, is likely to arise from the use of LST as a straight substitution for air temperature, in the final stages of the calculation. Future work will concentrate on using inferred air temperature estimates, derived using the multiple regression methods performed on the 1992 Europe data, rather than LST as input to the VPD equation (eq. 6). Despite the strong correlations, directly retrieving VPD from remotely sensed data using LST as a substitute for Ta will not produce accurate estimates.

SUMMARY AND CONCLUSIONS

Epidemiologists require readily available and/or easily calculated remotely sensed indicators of near-surface climatic variables affecting disease vector biology. In particular, air temperature and atmospheric moisture conditions are key determinants of arthropod vector ecology and consequent parasite/pathogen transmission. This paper has examined the potential to derive proxy measures of these variables from 8 × 8 km PAL data using a three-year time series of satellite observations covering two continents, Africa and Europe. Techniques to retrieve air temperature directly from AVHRR data, i.e. the TVX method, are not yet validated at the spatial (typically continental) and temporal (multi-year) scales required for immediate use by epidemiologists. Hence, this study has not used the TVX method but investigated LST as a surrogate variable. The simplicity of the calculation of LST from Price (1984) will ensure that it continues to be readily used by researchers of many disciplines (not just epidemiologists) eager to incorporate remotely sensed data into their work. This study has assessed the extent, spatially and temporally, to which LST can be used as a proxy measure of air temperature before breaking down as a surrogate variable.

Correlations between remotely sensed monthly LST observations and monthly ground-based air temperature, Ta, data are high in most cases especially for the tropical and sub-tropical latitudes of Africa. There is temporal variability in the monthly correlations indicating a seasonal influence on the relationships. This is probably due, in part, to changing surface characteristics, especially vegetation cover, between different months of the year. The strength of correlations increases significantly when NDVI and elevation data are included in the multiple regression analyses. Importantly, there was consistency in relationships between the years of study (1988 - 1990).

LST indices derived from the daily AVHRR Pathfinder data are more strongly correlated with monthly mean Ta than those derived from the 10-day composite data. Mean values of LST for each month in 1992 were derived from all the cloud-free daily observations for Europe. These indices are more highly correlated with surface observations than monthly maximum values and observations from the date of maximum NDVI. The use of daily data to derive monthly mean LSTs is, however, dependent on accurate cloud-screening and expensive in terms of computer processing time.

NDVI is an important variable for increasing the strength of relationships between LST and Ta and increasing the accuracy of empirically derived Ta estimates. The accuracy of retrieved Ta over Europe (1992) range from 1.83 °C to 3.18° C, with a mean over twelve months of 2.38°C when NDVI and LST are used as predictor variables in a multiple regression.

Attempts to derive monthly estimates of atmospheric moisture availability from AVHRR Pathfinder data result in an over-estimation of VPDsat compared to ground observations, VPDgrd. This study substituted LST for air temperature in the calculation of VPDsat, which is likely to be the main reason for the over-estimation. Hence, continental-scale mapping of VPD is dependent on accurate estimation of Ta and will be the topic of future investigations.

The relationships obtained using the Africa and Europe data emphasise the potential of NOAA AVHRR data to provide useful surrogates of climatic variables, on a consistent temporal basis at continental scales. Remotely sensed meteorological data present a significant advance in both temporal frequency and spatial coverage over sparsely distributed meteorological station records. This paper has been a first step in showing significant relationships, with consistency over a relatively long time period (three years), between satellite observations and climatic variables of key importance to epidemiology.

ACKNOWLEDGEMENTS

We are grateful to Sarah Randolph, David Rogers and Scott Goetz for their comments on this manuscript and the support and interest in this from the NASA based Interagency Research Partnership for Infectious Diseases (INTREPID). The following formal acknowledgement is requested of those who use PAL data: “Data used by the authors in this study include data produced through funding from the Earth Observing System Pathfinder Program of NASA’s Mission to Planet Earth in co-operation with the National Oceanic and Atmospheric Administration. The data were provided by the Earth Observing System Data and Information System (EOSDIS) Distributed Active Archive Center at Goddard Space Flight Center, which archives, manages and distributes this data set”. This work was carried out with funds from the Wellcome Trust. SIH was funded by a grant from the Department for International Development (DFID), U.K. (Livestock Production Programme: ZC0012) and is currently funded through an Advanced Training Fellowship with the Wellcome Trust (#056642).

REFERENCES

  1. Agbu PA, James ME. The NOAA/NASA Pathfinder AVHRR Land Data Set User’s Manual. Version 3.1 NASA, Goddard Distributed Active Archive Centre, Maryland; Maryland, USA: 1994. [Google Scholar]
  2. Connor SJ, Thomson MC, Molyneux DH. Forecasting and prevention of epidemic malaria: new perspectives on an old problem. Parassitologia. 1999;41:439–448. [PubMed] [Google Scholar]
  3. Cracknell AP. The Advanced Very High Resolution radiometer (AVHRR) Taylor and Francis; London: 1997. [Google Scholar]
  4. Czajkowski KP, Goward SN, Shirey D, Walz A. Thermal remote sensing of near-surface water vapor. Remote Sensing of Environment. 2000 in press. [Google Scholar]
  5. Czajkowski KP, Mulhern T, Goward SN, Cihlar J, Dubayah RO, Prince SD. Biospheric environmental monitoring at BOREAS with AVHRR observations. Journal of geophysical Research. 1997;102:29651–29662. [Google Scholar]
  6. Eck TF, Holben BN. AVHRR split window temperature differences and total precipitable water over land surfaces. International Journal of Remote Sensing. 1994;15:567–582. [Google Scholar]
  7. Goetz SJ. Multi-sensor analysis of NDVI, surface temperature, and biophysical variables at a mixed grassland site. International Journal of Remote Sensing. 1997;18:71–94. [Google Scholar]
  8. Goetz SJ, Prince SD, Small J. Advances in satellite remote sensing of environmental variables for epidemiological applications. Advances in Parasitology. 2000;47 doi: 10.1016/s0065-308x(00)47012-0. in press. [DOI] [PubMed] [Google Scholar]
  9. Goward SN, Waring RH, Dye DG, Yang J. Ecological remote sensing at OTTER: satellite macroscale observations. Ecological Applications. 1994;42:322–343. [Google Scholar]
  10. Hay SI. Remote sensing and disease control: past, present and future. Transactions of the Royal Society of Tropical Medicine and Hygiene. 1997;91:105–106. doi: 10.1016/s0035-9203(97)90186-3. [DOI] [PubMed] [Google Scholar]
  11. Hay SI, Packer MJ, Rogers DJ. The impact of remote sensing on the study and control of invertebrate intermediate hosts and vectors for disease. International Journal of Remote Sensing. 1997;18:2899–2930. [Google Scholar]
  12. Hay SI, Tucker CJ, Rogers DJ, Packer MJ. Remotely sensed surrogates of meteorological data for the study of the distribution and abundance of arthropod vectors of disease. Annals of Tropical Medicine and Parasitology. 1996;90:1–19. doi: 10.1080/00034983.1996.11813021. [DOI] [PubMed] [Google Scholar]
  13. Hay SI, Snow RW, Rogers DJ. Prediction of malaria seasons in Kenya using multi-temporal meteorological satellite sensor data. Transactions of the Royal Society of Tropical Medicine and Hygiene. 1998a;92:12–20. doi: 10.1016/s0035-9203(98)90936-1. [DOI] [PubMed] [Google Scholar]
  14. Hay SI, Snow RW, Rogers DJ. From predicting mosquito habitat to malaria seasons using remotely sensed data: practice, problems and perspectives. Parasitology Today. 1998b;14:306–313. doi: 10.1016/s0169-4758(98)01285-x. [DOI] [PubMed] [Google Scholar]
  15. Hay SI, Lennon JJ. Deriving meteorological variables across Africa for the study and control of vector-borne disease: a comparison of remote sensing and spatial interpolation of climate. Tropical Medicine and International Health. 1999;4:58–71. doi: 10.1046/j.1365-3156.1999.00355.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Hay SI. An overview of remote sensing and geodesy for epidemiology and public health application. Advances in Parasitology. 2000;47 doi: 10.1016/s0065-308x(00)47005-3. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Hay SI, Rogers DJ, Myers MF, Snow RW. Malaria early warning in Kenya. Parasitology Today. 2000a doi: 10.1016/s1471-4922(00)01763-3. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hay SI, Omumbo JA, Craig MH, Snow RW. Earth observation, geographic information systems and Plasmodium falciparum malaria in sub-Saharan Africa. Advances in Parasitology. 2000b;47 doi: 10.1016/s0065-308x(00)47009-0. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Holben BN. Characteristics of maximum-value composite images from temporal AVHRR data. International Journal of Remote Sensing. 1986;7:1417–1434. [Google Scholar]
  20. James ME, Kalluri SNV. The pathfinder AVHRR land data set - an improved coarse resolution data set for terrestrial monitoring. International Journal of Remote Sensing. 1994;15:3347–3363. [Google Scholar]
  21. Kneizys FX, Shettle EP, Abreu LW, Chetwood JH, Anderson GP, Gallery WO, Selby JEA, Clough SA. Users Guide to LOWTRAN7. Report No. AFGL-TR-88-0177. Air Force Geophysics Laboratory; Bedford, Massachusetts, USA: 1988. [Google Scholar]
  22. Lambin EF, Ehrlich D. Combining vegetation indices and surface temperature for land-cover mapping at broad spatial scales. International Journal of Remote Sensing. 1995;16:573–579. [Google Scholar]
  23. Lambin EF, Ehrlich D. The surface temperature-vegetation index space for land cover and land-cover changes analysis. International Journal of Remote Sensing. 1996;17:463–487. [Google Scholar]
  24. Lennon JJ, Turner JRG. Predicting the spatial distribution of climate: temperature in Great Britain. Journal of Animal Ecology. 1995;64:370–392. [Google Scholar]
  25. Myneni RB, Maggion S, Iaquinto J, Privette JL, Gobron N, Pinty B, Kimes DS, Verstraete MM, Williams DL. Optical remote sensing of vegetation: modeling, caveats, and algorithms. Remote Sensing of Environment. 1995;51:169–188. [Google Scholar]
  26. National Oceanic and Atmospheric Administration . Monthly climate data for the world. Volumes 41-43. National Climate Data Centre; NOAA, North Carolina, USA: 1990. [Google Scholar]
  27. Price JC. Land surface temperature measurement for the split window channels of the NOAA 7 advanced very high resolution radiometer. Journal of Geophysical Research. 1984;89:7231–7237. [Google Scholar]
  28. Prihodko L, Goward SN. Estimation of air temperature from remotely sensed observations. Remote Sensing of Environment. 1997;60:335–346. [Google Scholar]
  29. Prince SD, Goward SN. Global primary production - a remote-sensing approach. Journal of Biogeography. 1995;22:815–835. [Google Scholar]
  30. Prince SD, Goetz SJ, Czajkowski K, Dubayah R, Goward SN. Inference of surface and air temperature, atmospheric precipitable water and vapor pressure deficit using AVHRR satellite observations: validation of algorithms. Journal of Hydrology. 1998;212/213:231–250. [Google Scholar]
  31. Randolph SE. Climate, satellite imagery and the seasonal abundance of the tick Rhipicephalus appendiculatus in southern Africa: a new perspective. Medical and Veterinary Entomology. 1993;7:243–258. doi: 10.1111/j.1365-2915.1993.tb00684.x. [DOI] [PubMed] [Google Scholar]
  32. Randolph SE. Ticks and tick-borne disease systems in space and from space. Advances in Parasitology. 2000;47 doi: 10.1016/s0065-308x(00)47010-7. in press. [DOI] [PubMed] [Google Scholar]
  33. Rogers DJ, Hay SI, Packer MJ. Predicting the distribution of tsetse-flies in West-Africa using temporal Fourier processed meteorological satellite data. Annals of Tropical Medicine and Parasitology. 1996;90:225–241. doi: 10.1080/00034983.1996.11813049. [DOI] [PubMed] [Google Scholar]
  34. Rogers DJ. Satellites, space, time and the African trypanosomiases. Advances in Parasitology. 2000;47 doi: 10.1016/s0065-308x(00)47008-9. in press. [DOI] [PubMed] [Google Scholar]
  35. Smith WL. Note on the relationship between total precipitable water and the surface dew point. Journal of Applied Meteorology. 1966;5:726–727. [Google Scholar]
  36. Stowe LL, McClain EP, Carey R, Pellegrino P, Gutman GG, Davis P, Long C, Hart S. Global distribution of cloud cover derived from NOAA AVHRR operational satellite data. Advances in Space Research. 1991;3:51–54. [Google Scholar]
  37. Unwin DM. Microclimate measurement for ecologists. Academic Press; London: 1980. (Biological techniques series 3). [Google Scholar]

RESOURCES