Schistosomiasis, caused by Schistosoma haematobium and S. mansoni, is a disease that mainly affects under-resourced communities and is often not prioritized in national budgets in sub-Saharan Africa (WHO, 2014). Approximately 600 million people are at risk worldwide and over 200 million people are infected (Chitsulo et al., 2000; King, 2009). The Schistosomiasis Control Initiative (SCI) works with national governments in sub-Saharan Africa to control and eliminate schistosomiasis. A recent national schistosomiasis survey contributed to the development of the schistosomiasis and soil transmitted helminths (STH) National Control Program in Zimbabwe that involves mass drug administration (MDA) (Midzi et al., 2014). Despite the efforts to control the disease in sub-Saharan Africa, 80-95% of the global, total number of schistosomiasis infected individuals still live in Africa (WHO, 2002; Steinmann et al., 2006; Utzinger et al., 2009; Hurlimann et al., 2011). It is estimated that about 200 000 deaths per year are caused by schistosomiasis in sub-Sahara Africa (WHO, 2014). The distribution of schistosomiasis is reflected by the geographical distribution of the intermediate host snail species (Despommier et al., 1994), which is a well-known phenomenon but difficult to predict and monitor. It is envisaged that knowing the drivers of the current distribution of schistosomiasis could improve our understanding on how climate and environmental changes may influence the distribution of schistosomiasis in the future. This is critical to effectively and efficiently manage surveillance, control and prevention of the disease (Stensgaard et al., 2005). King et al. (2006) emphasize that the next generation of schistosomiasis control will be optimized using new monitoring tools and effective transmission containment. Enhanced by the advancements in statistical ecological modelling, spatial technologies which provide spatial data and tools for spatial analysis and predictive modelling have opened a new way for developing such monitoring tools. The development of geo-spatial technology such as geographical information systems (GIS) and global positioning system (GPS) have facilitated the integration of Earth observation (EO) driven environmental parameters with health data for the development of disease surveillance and control models (Beck et al., 2000). GIS applications in public health include the estimation of spatial variation of disease, determination of risk factors of disease, and improved delivery of health services (Tanser et al., 2003). This review looks at the progress and challenges in the application of spatial technologies in mapping and modelling schistosomiasis in Africa. It was inspired by the works of Brooker et al. (2002a) and Brooker (2007) and Simoonga et al. (2009), who published detailed reviews of the past developments and use of GIS and remote sensing in schistosomiasis mapping and modelling in Africa. The potential future research priorities with emphasis on application of spatial technology in schistosomiasis modelling at local levels in Africa are discussed.
Selection criteria for the literature search undertaken
This review is based on a systematic search for relevant literature in the PubMed electronic search engine (http://www.ncbi.nlm.nih.gov/pubmed) following the method used by Simoonga et al. (2009). This search considered the studies using geo-spatial technologies for schistosomiasis prediction, modelling and mapping in Africa based on the following combination of terms and Boolean operations: geo-spatial analysis OR remote sensing OR earth observation OR geographic information systems OR prediction OR mapping OR modelling AND schistosomiasis AND Africa. Any literature which did not satisfy these criteria was excluded. The snowballing technique was used to obtain more literature based on the bibliography or reference list of previous reviews obtained by the search strategy described above. Literature on schistosomiasis without the element of geo-spatial analysis were used to strengthen the discussion and understanding why geo-spatial technology has been and/or can be used to understand the schistosomiasis transmission. The relevant literature was used to determine the levels of appreciation and use of geospatial technology in schistosomiasis modelling and/or mapping by extending the yearly publication graph by Simoonga et al. (2009). These authors accessed the PubMed database in early 2009 and gave a snapshot of the number of publications on GIS and remote sensing applications used with reference to schistosomiasis in Africa between 1996 and 2008. In this review, we extended the snapshot up to 2013 using the methodology mentioned above.
The search for literature conducted for the period 2009-2013 revealed 59 hits of which 36 were considered relevant. The relevant publications (36) were further categorized by year of publication and combined with Simoonga et al. (2009) results for the period 1996-2008 as presented in Figure 1. The annual number of publications on remote sensing and GIS with application to schistosomiasis in Africa has generally increased over the years as shown in Figure 1. This indicates the increased appreciation and usefulness of geospatial technology for schistosomiasis control and management through mapping, modelling or prediction. The key publications from 2009 to 2013 and 2 from 2008 which were not captured by Simoonga et al. (2009) as well as one for 2014 were categorized into the following three groups: modelling intermediate snail hosts (Table 1), modelling schistosomiasis (Table 2) and those modelling co-infection or co-endemicity of schistosomiasis and STHs (Table 3). The subsequent sections give a detailed discussion based on these publications as well as those mentioned elsewhere for example by Simoonga et al. (2009).
Progress in geo-spatial technology application in schistosomiasis modelling
Spatial technologies have provided an invaluable analytical tool to better understand the determinants and distribution of schistosome infections in Africa (Figure 1 and Tables 1-3). It is well known that climate and environmental factors determine the distribution of schistosomiasis (Appleton, 1978; Brown, 1994; Brooker, 2002) hence it is restricted in space and time by environmental factors (Rollinson et al., 2001; Malone, 2005). In this regard, spatial technologies are useful in understanding the distribution of parasites and their hosts as depicted in the maps showing the interplay with spatial and temporal features of the environment. The general objective of the models is to link the variables related to schistosomiasis transmission or snails with spatial parameters (Simoonga et al., 2009). Most studies use spatial modelling techniques such as logistic regression (Brooker et al., 2001, 2002b), Maxent (Stensgaard et al., 2013; Pedersen et al., 2014), genetic algorithm for rule-set prediction (GARP) (Stensgaard et al., 2006), generalized linear models (GLMs) and generalised additive models (GAMs) (Pfukenyi et al., 2006). There are also non-regression models such as the Bayesian geostatistical approach for modelling intermediate host snails distribution and prevalence of schistosomiasis (Raso et al., 2005; Vaunatsou et al., 2009; Schur et al., 2013). The use of GIS and remote sensing in these models has contributed towards optimized schistosomiasis control efforts at different spatial scales through identifying vulnerable populations for mass treatment (Brooker, 2007) and permitting more rational allocation of resources for cost-effective control (Beck et al., 1997, 2000). However, there are limitations with regards the applicability and robustness of these models that compromise their effectiveness in promoting community public health especially at local levels.
Geo-spatial technology in schistosomiasis modelling: pros and cons
Earth observation has been providing spatial data (and will continue to do so) for developing GIS models for predicting and mapping the risk of schistosomiasis, mostly in inaccessible regions of Africa and in scenarios involving environmental or climate change. However, there are still challenges which warrant further research and refinements/improvements (Herbreteau et al., 2007; Simoonga et al., 2009). In the early stages of the development of GIS and remote sensing technology, the main challenge was the possible resistance to the uptake of EO technology mainly attributed to costs of image processing equipment, expertise and subsequent ground validation (Hay, 1997). To date, it might be true that remote sensing has not become the wonder tool as it was expected to be, to echo Herbreteau et al. (2007), mainly because of limited capacity in processing and use of remote sensing data especially in Africa. EO data requires processing and understanding of the purpose for which it is intended without which the output may be as meaningless as the raw data. Therefore the existing schistosomiasis predictive models are weakened by several factors, including statistical uncertainty in variable selection criteria and methods used, low spatial resolution, failure to utilize the temporal aspect of EO data for spatiotemporal prediction of schistosomiasis, limited application of the developed models in different areas as well as uncertainty and lack of vigorous validation as discussed below.
Unjustified variable selection criteria
The role of non-climatic factors such as topography, distance to water and soil types have been considered in large-scale studies (Tables 1-3). However they have not been given considerable attention compared to climatic factors as determinants of the spatial distribution of schistosomiasis particularly at the local level. These factors could easily be mapped using GIS to determine their influence on schistosomiasis transmission. For example, the distance to water determines the human water-contact behaviour, which has a strong influence on the prevalence and intensity of schistosomiasis (Stensgaard et al., 2013; Chimbari et al., 2003). Furthermore, studies have shown that Biomphalaria pfeifferi can tolerate a maximum flow speed of up to 0.3 m/s (Appleton, 1978; Kloos et al., 2001) and soil types determine the water-soluble calcium (Ca) and pH levels. Together, these factors have an effect on the presence and density of snails that ultimately decides the risk for infection. Further, Raso et al. (2005) observed that distance to permanent rivers was significantly associated with S. mansoni infection in the non-spatial logistic regression but showed no significant association in spatially explicit models. Saathof et al. (2002) also concluded that most of the environmental factors (slope, distance to water and vegetation cover) inadequately explain the spatial pattern of schistosomiasis infection at the sub-district level except with regard to altitude. Although this is probably true, this assertion still requires further investigation. Moreover, rainfall and temperature were not used in this particular study as the necessary data for that spatial resolution were not available and it was anticipated that the variation within such a small area would have been small (Moodley et al., 2003). Rainfall and normalized difference vegetation index (NDVI) have been widely used as proxies for water availability. In their study, Standley et al. (2012) noted that water availability may be insignificant in the context of permanent large water bodies. However, the significance of these environmental factors may still need to be investigated as they may be proxies for water availability in temporal and spatial modelling of schistosomiasis at the local level.
In their large review, Simoonga et al. (2009) noted that a few researchers have attempted to relate disease prevalence and socio-economic profiles of the local population using GIS applications. Such analyses may be linked to intermediate host snails distribution datasets explained by EO environmental and climatic factors. Schistosomiasis is heterogeneous, so there is need to undertake localised studies to establish exposure risk factors and link water contact patterns with malacological surveys (Simoonga et al., 2008). Schur et al. (2013) used Bayesian modelling and they believe that their model could have performed better if they had included intermediate host snail data. Sturrock et al. (2013) highlighted the need to consider locations of transmission sites and not just distance to water bodies (a river in their case) to improve the performance of their prediction model. Simoonga et al. (2009) reported that socio-economic factors, such as availability of sanitary facilities and safe water supply may better explain different levels of transmission at local-scale than for instance, poverty quintiles as seen in a the micro-level study in Côte d’Ivoire by Raso et al. (2005). Chimbari et al. (2003) attributed differences in schistosomiasis distribution and intensity between Lake Kariba (Zimbabwe) and Siavonga (Zambia) (10 km apart) to different sanitation (better in Lake Kariba) and access to water from the lake (easy access for Siavonga compared to Kariba). Seto et al. (2012) used wearable GPS data-loggers for mapping and assessing the exposure of women and children to risk factors such as access to water. This, however may have influenced behaviour and affecting the performance of the prediction model. This opens avenues for further research as the inclusion of these factors may improve the performance of the model.
The use of simple threshold analysis (Malone et al., 2001) or logistic regression (Brooker et al., 2001, 2002a; Ekpo et al., 2008) modelling to predict infection risk are both limited by the inability of these methods to consider spatial correlation of infection and environmental variables. This leads to underestimation of standard errors of covariate coefficients resulting in erroneous inference and justifies the need for assessment of uncertainties inherent in data and modelling techniques (Brooker, 2007). Variable selection in predictive modelling is a major challenge due to analytical problems caused by over-fitting, confounding and non-independence of data (Craig et al., 2007). Schistosomiasis transmission predictive models mostly derived from regression models are conservative, because they use only a few (two or three) climate variables to model the disease. While this is simple to understand as required in most models, it could be more reasonable and realistic to include more factors such as the effects of sanitation, water-related activities (Simoonga et al., 2009; Stensgaard et al., 2013), snail presence and absence data (Moodley et al., 2003; Schur et al., 2013). The mapping of infected snails could also help clarify the complex interaction between snail, parasites and the environmental factors that are usually used to predict the distributions (Simoonga et al., 2009). However, the approach requires systematic and repeatable staged variable selection procedures, including spatial analysis to achieve a parsimonious model with the desired level of internal and external validity (Craig et al., 2007). This could be resolved by using the machine-learning algorithms such as Maxent (Stensgaard et al., 2013; Pedersen et al., 2014) and GARP (Stensgaard et al., 2006), which can handle larger number of variables (continuous and categorical) than regression models. Bayesian geostatistical approach has been widely used in schistosomiasis modelling over the seven latest years (Raso et al., 2005; Clements et al., 2006; Vaunatsou et al., 2009; Schur et al., 2011a, 2011b, 2011c; 2013). This approach is regarded as a flexible and robust (Brooker, 2007) that takes into account spatial variability of epidemiological and environmental data as it uses the semivariogram of the spatial process (Chiles and Definer, 1999). However, only a few aspects of geostatistical methods have been explored in schistosomiasis modelling. For example isotropy is more often used than anisotropy (Simoonga et al., 2008; Vaunatsou et al., 2009). Isotropy is independent of location and direction and assumes that spatial correlation is only a function of distance (Gosoniu et al., 2009). This is contrary to anisotropy, which assumes that spatial correlation is a function of distance in relation to both location and direction. It is likely that there is high correlation of intensity and prevalence of schsistosomiasis towards transmission sites (Vaunatsou et al., 2009) and linked to the main flow direction of the river (Beck-Wörner et al., 2007). Schur et al. (2013) applied anisotropy at the regional scale in East Africa and recommend the application of this method at the local scale. These authors emphasise that ignoring anisotropy could influence the strength of association and thus also the spatial range parameter estimates, which might reduce model ability, especially in the presence of strong anisotropy. Most of the studies on geospatial modeling and/or mapping of schistosomiasis at the local scale do not consider co-indemicity or co-infection of S. mansoni and S. haematobium (Gryseels, 1996; Raso et al., 2005; Brooker and Clements, 2009). Advanced geostatistical capabilities, such as shared component modelling (Schur, 2011a) and Bayesian geostatistical multinomial regression modelling (Magalhães et al., 2011), which can model co-infections, have not been fully exploited with reagrd to schistosomiasis co-infections and/or co-endemicity modelling. These techniques allow simultaneous modelling of co-endemicity or co-infection of the two common schistosome species in Africa (S. mansoni and S. haematobium) and investigate the independence between the two and how they respond to different climatic and environmental factors. However, the results are affected by inconsistency in data that can be due to different samples, different sample sizes and time of sampling, which may compromise the quality of model outputs (Hodges et al., 2012). This type of inconsistencies can be handled well using geostatistical methods such as shared component modelling as compared to Bayesian geostatistical multinomial regression modeling which can only use data from surveys screening for multiple infections simultaneously (Schur et al., 2011a). Generally, there is paucity of information on the geographical distribution of both species within co-endemic regions and knowledge of micro-geographical variation of single and mixed schistosoma infections and morbidity. Thus, modelling the two parasites could provide important insights into the drivers of infection and disease, which could help tailoring schistosomiasis control and elimination efforts (Meurs et al., 2013). It is believed that chronic infections cause adverse morbidity-related effects that are exacerbated by infections by multiple species and high parasite loads (Pullan and Brooker, 2008). Modelling schistosome co-infections could make it easier to determine whether these infections impact species-specific morbidity compared to single species infection (Gouvras et al., 2013). To date, the effects of schistosome co-infection on morbidity are not clear (Sang et al., 2014; Meurs et al., 2012) and modelling these co-infections could help targeting specific micro-geographical locations for further research on design of superior intervention strategies.
Low spatial resolution – remote sensing data
The increasing use of spatial low-resolution imagery (500 m or less) has provided an opportunity to explore the distribution of schistosomiasis at broad scales (country, regional and continental) (Brooker and Michael, 2000; Brooker, 2002; Clennon et al., 2006). Still, however, ineffective schistosomiasis prediction remains a public health concern in geographically restricted areas as low-resolution investigations do not consider local heterogeneity of snails and schistosomiasis (Kitron et al., 2006). The small-scale focality of schistosomiasis is well recognised and the causes of heterogeneity are varied and reflect many human and ecological factors (Kloos et al., 1997, 1998; Woolhouse and Chandiwana, 1989). This makes the small-scale distribution difficult to predict (Brooker, 2002) as the effects of local heterogeneity are averaged out at broader scales and ecological patterns often appear more regular (Wiens, 1989; Levin, 1992). Brooker (2002), emphasized that the use of remote sensing has not been very successful in capturing the well-known local variation of schistosomiasis transmission, for example, the 8-km advanced very high resolution radiometer (AVHRR) used by Brooker et al. (2002a, 2002b) may generalize the spatial variation of schistosomiasis. Although satellite instruments can deliver higher resultion imagery, such as SPOT 5 and 6 (http://www.geo-airbusds.com/en/147-spot-6-7-satellite-imagery) and Landsat 8 (http://landsat.usgs.gov/landsat8.php) even most recent African studies are based on low-resolution satellite products from the moderate resolution imaging spectroradiometer (MODIS) (http://modis.gsfc.nasa.gov/) and American National Oceanic and Atmospheric Administration (NOAA) (http://www.noaa.gov/) AVHRR (Tables 1-3). There is need for the use of high-resolution instruments to capture the local ecological spatial variation of intermediate host snails or prevalence of schistosomiasis (Brooker et al., 2001, 2002a). Indeed, high-resultion imagery is available but has only been used occasionly due to the current high cost. For example, the very recent study of De Roeck et al. (2014) on Fasciola hepatica in Belgium using drones and very high resolution (VHR) imagery from the commercial WorldView2 staellite (https://www.digitalglobe.com/sites/default/files/DG_WorldView2_DS_PROD.pdf) is a case in point. Fine-scale monitoring is of key importance to refine currently existing broad-scale infection risk models, and costs might dimish with time alowing a more widely use of the technology. Generally, however, there is no single natural scale, at which ecological patterns are studied (Levin, 1992). In each case, the appropriate scale is dictated by the goals of the study, system and available data. Hence, there is need for developing remote sensing predictive models for targeting schistosomiasis control at local levels. There is also need to focus on the household as spatial points in high endemic areas instead of schools, as this will help to avoid spatial aggregation and allow appropriate finite scale spatial mapping and give insight into the micro-epidemiology of schistosomiasis (Simoonga et al., 2008).
Failure to utilize the temporal domain of remote sensing
The temporal characteristic of remote sensing has not been fully utilized in modelling the temporal variation of schistosomiasis in Africa. Brooker (2002) has highlighted that the temporal variation of schistosomiasis has received far too little attention despite its relevance in understanding the spatial distribution of infection. Seasonal and intraseasonal modelling of schistosomiasis may capture the variation of snail density and occurrence as some snails maybe washed away during the rainy season (Appleton, 1978; Kloos et al., 2001) and some may die due to desiccation during the dry season (Rollinson et al., 2001). This may also explain the spatial and temporal variability of point prevalence, infection rate and intensity of schistosomiasis especially at local levels. This emphasizes the relevance of intra-seasonal and seasonal modelling of schistosomiasis in timely allocation of resources as well as targeting of control programs. Most of the schistosomiasis predictive models consider annual distribution of schistosomiasis (Stensgaard et al., 2013; Pedersen et al., 2014) but not the component of seasonality. Considering the temporal resolution of satellite data could help to syncronise the temporal differences between data collection of variables and disease or parasitological data as highlighted by Sturrock et al. (2013). There is need for ecological niche modelling of seasonal vector population dynamics combining ecological niche models with purpose-built, temporal high-resolution satellite remote sensing data (Kulkarni et al., 2010). On the other hand, remote sensing technology has been available for more than 30 years, but only a few studies such as that by Pedersen et al. (2014) have taken up the challenge to model the changes in snail habitats over the past years to estimate the possible distribution of schistosomiasis in relation to climate and environmental changes.
Brooker (2002) emphasized the need to develop separate models for each snail-schistosome system due to their different habitat types and environmental suitability. Thus, these models are not transferable to other regions or places. This is evident from the work of Malone et al. (2001), who developed a region-specific schistosomiasis predictive model with limited application elsewhere where different ecological conditions and snail species prevail. This compromises the validity of the continental- and regional-level models, advocating the local scale. Until 2009, there was no schisotosmiasis model at a continental level (Simoonga et al., 2009) and, to our knowledge, Stensgaard et al. (2013) is the only study investigating schistosomiasis distribution in Africa as a whole. Simoonga et al. (2009) highlight several challenges that must be overcome in order to further improve the GIS-based mapping of intermediate host snails at the continental scale. Consequently, there is a lack of large-scale, geo-referenced quality data on the presence/absence of snails as well as parasite-snail compatibilities. There is also need for a more complete understanding of snail species identities and their efficiencies as intermediate hosts as prescribed by Stothard et al. (2002). The challenge is to develop a composite risk map of schistosomiasis (Brooker, 2002), which could be achieved through shared component modelling technique as highlighted above.
Lack of vigorous validation or accuracy assessment
An important but often difficult part of a disease model is the assessment of applicability and validity, especially if outputs are to be used for disease control. The quality of predictive (presence or absence) models of a species is normally judged by the number of prediction errors or its accuracy (Fielding and Bell, 1997). Accuracy refers to the correctness of remotely sensed data or model outputs, which measures the agreement between a standard assumed to be the correct and classified image or the result of a model of unknown quality (Foody, 2001; Campbell, 2006). Despite the wide use of predictive models, most applications do not give sufficient consideration to model error and uncertainty (Barry and Elith, 2006). Datasets used to statistically develop the models are often of uncertain accuracy and are not always easily reproducible as the results vary with training data and methods used (Tanser et al., 2003). These generic disadvantages vary from worldwide or continent-wide (Stensgaard et al., 2013) to regional (Schur et al., 2011a, 2011b, 2011c) statistically-driven models; the models are too coarse to guide intervention efforts and their capacity to predict prevalence remains uncertain (Kulkarni et al., 2010).
Evaluating the predictive models is a crucial step for determining its suitability for specific applications (Guisan and Hofer, 2003; Allouche et al., 2006). In this case, the focus is on schistosomiasis control and the possibility of comparing with other models and classification techniques (Powell et al., 2004). Any approach to ecological modelling has little merit if predictions cannot be assessed (Verbyla and Litvaitis, 1989) and any maps or satellite products or models without associated accuracy remain untested hypotheses (Strahler et al., 2006). The most commonly used statistical measures of error of predictive models include error matrix (Morisette et al., 2005), the Cohen Kappa statistics, the threshold-independent receiver operating characteristic (ROC) approach (Fielding and Bell, 1997) achieved by calculating the area under the ROC curve (AUC), a Gini coefficient AUC (Copas, 1999), the true skill statistic (TSS) (McPherson et al., 2004), Cohen’s Kappa z-test and MacNemar’s test. Most of the studies use EO and disease data for schistosomiasis predictive modelling from electronic databases and only a few models for example Schur et al. (2011a, 2011b, 2011c, 2013) have been adequately evaluated and provided with proper statistical quantification of error. The main constraint in validation is lack of updated, comprehensive, good quality empirical data (Moodley et al., 2003; Tanser et al., 2003). This has compromised the quality, applicability and reliability of the developed models. The maps representing the world-wide or country-wide burden of schistosomiasis generally reflect the reported distribution of clinical episodes of this disease. However, the scope and accuracy of such reports are limited by the extent of health care coverage, the efficacy of surveillance and also by the quality of the reporting systems (Kiszewski et al., 2004). This paucity of epidemiological data hinders large-scale quantification of the burden of a disease (Brooker et al., 2002a, 2002b).
Potential future research priorities
Availability of affordable treatment of schistosomiasis (praziquantel) has led to increased interest and commitment to effective and efficient control of this disease. However, control resources are inevitably limited, necessitating predictive models that can rapidly and accurately identify and map high-risk communities so that interventions can be targeted in a spatially-explicit and cost-effective manner (Brooker, 2009). Geospatial technologies are promising with respect to meeting this objective. However, disease data, climatic and environmental data must not only be reliable, but also be possible to collect at suitable spatial and temporal resolutions. Figure 2 illustrates the proposed theoretical framework for schistosomiasis predictive modeling considering four phases; data collection, snail prediction modeling, schistosomiasis predictive modeling and validation which are based on data quality and availability as well as model accuracy. The whole system is sustained by health research institutional capacity in terms of skills, funds and equipment to generate high quality data and achieve desirable predictive model accuracy (Figure 2).
Although remote sensing has proved to be a reliable source of climatic and environmental data, there is need to consider satellites with higher spatial resolution such as SPOT-6 as opposed to low spatial resolution imagery of 1 and 8 km for MODIS and AVHRR, respectively, while ultrahigh-resultion imagry will have to wait until cost dimishes. The EO technology could offer higher spatial and spectral, more frequent coverage and lower cost data as suggested by Lleo et al. (2008), however, specialized skills and expertise are a pre-requisite so as to realize the full advantages of these developments. Mostly, EO data are developed for a wide range of applications and epidemiologists have to develop specific products or applications for specific purposes rather than relying on already processed or off-shelf products such as the normalized difference vegetation index (NDVI) and land surface temperature (LST) with no metadata. These products may have been developed for different purposes at different scales and may not serve the same purpose with same level of accuracy required in schistosomiasis modelling. This indicates the need for investing more time and resources in the development, application and use of the space technology in epidemiology.
Field studies will still be needed to generate high quality data including climatic and environmental factors for calibration of EO data as shown in Figure 1. The output predictive models should also be validated against field observations as argued by Bergquist et al. (2009) to realize their usefulness in community health and climate change decision making process especially at the local level in Africa. This will help to capture the local focality of schistosomiaisis as the use of high quality and reliable data could help to refine the geostatistical techniques and adoption of ecological tools such as Maxent (Phillips et al., 2006), which are promising to produce highly performing schistosomiasis predictive models (Stensgaard et al., 2013). The field measured data could also be complemented by laboratory experiments as the behavior of snails and schistosome parasites could change due to climatic and environmental changes. For example Brown (1994) studied the distribution of freshwater host snails in Africa and divided them into two groups (tropical and temperate species) based on the climate where they occur. However the snail species tolerance ranges for temperature might have changed over time, which requires more field studies for verification and comparison. On the other hand, systematic field studies on the relationship between densities and infection rates in snails and those in humans would still be highly useful (Gryseels, 1996). The rational use of remote sensing data is dependent of the quality of information from the field (De la Rocque et al., 2005), which requires sophisticated geo-spatial statistical methods for analysis and predictive modeling and dedicated fieldwork to validate the observations (Herbreteau et al., 2005). Hence, the need for interdisciplinary approach in which epidemiologists collaborate with software programmers, geographers and spatial analysts to create robust techniques and products for use with epidemiological data (Jacquez, 2000; Graham et al., 2004; Herbreteau et al., 2007).
Geo-spatial technologies are invaluable for schistosomiasis mapping and transmission prediction, particularly in Africa. However, more extensive applications of these tools have been hampered by lack of training, gaps in data (quality and quantity, particularly climatic, environmental, epidemiologic and parasitologic data) and inadequate tools for data gathering (Hay, 2000). These are clear indications of possible sources of errors and uncertainties that have propagated the schistosomiasis transmission modelling in Africa. Therefore, Herbreteau et al. (2007) is correct in saying that major elements of geospatial technologies have not yet met our current needs. However, in China, Yang et al. (2005) viewed future prospects of GIS and remote sensing application in disease mapping as bright and promising; hence, it might be too early to blame the technology. Instead, there is need to take advantage and sharpen ideas and skills to develop more and better methods through further research and refinement of the schistosomiasis predictive models to meet community needs. This could be achieved through collaboration between epidemiologists, geographers and software programmers, use of high-quality remote sensing and ground measured data and thorough validation protocols. It would also allow the adoption and use of some tools developed in ecology such as Maxent and GARP machine-learning algorithms and refining geostatistical techniques for predictive modelling of schistosomiasis at the local level.