**3. The roles of GIS**

GIS is the ideal tool for conducting wildlife habitat assessment as it involves geospatial data. Besides providing an integrated environment for managing and manipulating environmental data layers and georeferenced wildlife data, GIS can also offer capabilities to remedy or address some of the data quality issues associated with citizen contribute wildlife data. Firstly, geovisualization can be used to facilitate wildlife data elicitation from citizen participants and improve positional accuracy. Secondly, based on the cause of spatial bias, spatial analysis can be used to compensate for the biased coverage in observation efforts. Lastly, geospatial computation techniques can be employed to address the computational challenges arising from analyzing very large volumes of citizen-contributed wildlife data.

### **3.1 Geovisualization to improve positional accuracy**

In general, positional accuracy of wildlife data largely depends on the availability of positioning technology. Wildlife sightings can be accurately georeferenced with the aid of high-accuracy positioning techniques. For example, smartphones equipped with high-accuracy GPS units ensure generated data record is associated with accurate geographic coordinates. Nevertheless, the above observations hold only for citizen observers who are reporting or recording data at the time of sighting wildlife occurrences in the field. In many cases, local residents (e.g., farmers) do not keep records of daily wildlife sightings or they simply do not have access to GPS units or smartphones. Most often, wildlife data are elicited from their memories long after the time of sighting [17, 18, 23, 40].

Wildlife data (e.g., sightings) collected from local citizens through interviews or questionnaire surveys often have position information with unsatisfactory accuracy [14, 15]. Descriptions of the locations of wildlife sightings are often imprecise or vague, particularly if a long time has lapsed since the actual sightings. Such incapability partly results from the absence of an effective interviewing media (e.g., an intuitive and interactive representation of the natural environment where local citizens are active) that facilitates local citizens to recall and locate their sightings of wildlife. Ref. [17] collected distribution and abundance data of terrestrial tortoises from local

**25**

*Integrating Citizen Science and GIS for Wildlife Habitat Assessment*

shepherds over 1 km × 1 km grid cells with the aid of topographic maps. However, it is difficult to accurately locate wildlife sightings on topographic maps for the local

and improves visual search efficiency and navigation performance [53].

ecology to increase the accuracy of habitat suitability mapping.

**3.2 Geospatial analysis to tackle spatial bias**

determine the distance threshold properly.

Geospatially enabled and user-friendly geovisualization interfaces could help improve positional accuracy of the wildlife data elicited from local residents [50, 51]. Geovisualization, particularly 3D geovisualization techniques, can be adopted to help local residents to recall and locate their sightings of wildlife and obtain wildlife data with more accurate positional information [40]. Given the flat 2D topographic maps, the local residents need relief interpretation skills to re-construct the 3D topography of the landscape; local residents can then orientate themselves and locate places on the landscape. But they often do not have much training in basic map reading skills, not to mention relief interpretation. 3D geovisualization can facilitate relief interpretation by producing a realistic and intuitive terrain representation [52]

Geovisualization techniques as discussed above help improve positional accuracy of wildlife data at the very beginning of data generation. Sometimes, in cases where positional uncertainty does exist in wildlife data and is indeed of concern for wildlife habitat assessment, GIS-based methods have been developed to reduce its impact on the accuracy of wildlife habitat assessment. As an example, [54] proposed a spatial sampling method for deriving probable wildlife occurrence locations from patrol records using heuristics based on data recording context and species

Many geospatial analytical methods have been proposed to account for the spatial bias in wildlife data. An *AdaSTEM* approach that exploits variation in the density of wildlife observations was proposed to accommodate spatial bias in citizen-contributed wildlife observations [22, 55]. The continent- or hemispherewide study area is partitioned into rectangular spatial units (i.e., sub-areas) of size dependent upon density of wildlife observations. Environmental niches are modeled with only observations in each sub-area. By training local models in sub-areas, instead of training a global model using observations over the whole area, this approach mitigates spatial bias in the overall data set to a certain degree.

Filtering samples in the geographic or environmental space (i.e., remove observations that are within certain distance of one another) is also applied to reduce spatial bias [56, 57]. This method is based on the heuristic that removing localities (i.e., field samples) that are within certain distance of one another would somehow balance the unequal sampling or observation effort. The key of this method is to

If detailed information observation effort is available, such information can then be incorporated to correct for spatial bias. Spatial bias in wildlife observations was compensated for by weighting the observations with weights inversely proportional to the cumulative visibility at the observation sites, given that cumulative visibility is a good proxy of the underlying observation effort [23]. Here, cumulative visibility is the frequency at which a given location can be seen by observers from the routes the observers take. It can be computed based on a digital elevation model (DEM) representing the terrain and the routes using viewshed analysis, a common GIS function. A FactorBiasOut method was developed to correct for spatial bias in species presence-only data for species distribution modeling with MAXENT [58]. This method first estimates an empirical distribution to approximate the underlying but usually unknown sampling distribution that generated the presenceonly data. This approximate sampling distribution is then used to factor out

*DOI: http://dx.doi.org/10.5772/intechopen.83681*

residents who had no training in map reading.

#### *Integrating Citizen Science and GIS for Wildlife Habitat Assessment DOI: http://dx.doi.org/10.5772/intechopen.83681*

*Wildlife Population Monitoring*

**3. The roles of GIS**

by citizens are opportunistic in nature [23]. Unlike well-designed sampling or survey schemes which allocate observation sites in a way such that the geographic space and/or the environmental space are well covered by the observation sites, spatial distribution of the observation efforts of citizen volunteers would be considered neither random nor regular in the sense of sampling or survey design. One example to demonstrate this is wildlife sightings elicited from local residents. Local residents are not intentionally tracking wildlife of interest. Instead, they typically spot the wildlife en route to doing something else. The routes on which local citizens spot wildlife would be considered neither random nor regular but "ad hoc" [23]. As a result, wildlife sightings elicited from local residents are usually

Such spatial bias in wildlife data has a significant impact on environmental niche modeling and habitat suitability mapping for wildlife habitat assessment. Due to the spatial bias, citizen-contributed wildlife data might not be representative of the actual wildlife habitat use. The relationship derived based on the wildlife data thus might not well represent the underlying environmental niche. Spatial bias in citizen-contributed wildlife data, if not appropriately accounted for, would adversely affect the accuracy of wildlife habitat suitability mapping [47–49].

GIS is the ideal tool for conducting wildlife habitat assessment as it involves geospatial data. Besides providing an integrated environment for managing and manipulating environmental data layers and georeferenced wildlife data, GIS can also offer capabilities to remedy or address some of the data quality issues associated with citizen contribute wildlife data. Firstly, geovisualization can be used to facilitate wildlife data elicitation from citizen participants and improve positional accuracy. Secondly, based on the cause of spatial bias, spatial analysis can be used to compensate for the biased coverage in observation efforts. Lastly, geospatial computation techniques can be employed to address the computational challenges arising from analyzing very large volumes of citizen-contributed wildlife data.

In general, positional accuracy of wildlife data largely depends on the availability of positioning technology. Wildlife sightings can be accurately georeferenced with the aid of high-accuracy positioning techniques. For example, smartphones equipped with high-accuracy GPS units ensure generated data record is associated with accurate geographic coordinates. Nevertheless, the above observations hold only for citizen observers who are reporting or recording data at the time of sighting wildlife occurrences in the field. In many cases, local residents (e.g., farmers) do not keep records of daily wildlife sightings or they simply do not have access to GPS units or smartphones. Most often, wildlife data are elicited from their memories

Wildlife data (e.g., sightings) collected from local citizens through interviews or questionnaire surveys often have position information with unsatisfactory accuracy [14, 15]. Descriptions of the locations of wildlife sightings are often imprecise or vague, particularly if a long time has lapsed since the actual sightings. Such incapability partly results from the absence of an effective interviewing media (e.g., an intuitive and interactive representation of the natural environment where local citizens are active) that facilitates local citizens to recall and locate their sightings of wildlife. Ref. [17] collected distribution and abundance data of terrestrial tortoises from local

concentrated in areas with higher route accessibility.

**3.1 Geovisualization to improve positional accuracy**

long after the time of sighting [17, 18, 23, 40].

**24**

shepherds over 1 km × 1 km grid cells with the aid of topographic maps. However, it is difficult to accurately locate wildlife sightings on topographic maps for the local residents who had no training in map reading.

Geospatially enabled and user-friendly geovisualization interfaces could help improve positional accuracy of the wildlife data elicited from local residents [50, 51]. Geovisualization, particularly 3D geovisualization techniques, can be adopted to help local residents to recall and locate their sightings of wildlife and obtain wildlife data with more accurate positional information [40]. Given the flat 2D topographic maps, the local residents need relief interpretation skills to re-construct the 3D topography of the landscape; local residents can then orientate themselves and locate places on the landscape. But they often do not have much training in basic map reading skills, not to mention relief interpretation. 3D geovisualization can facilitate relief interpretation by producing a realistic and intuitive terrain representation [52] and improves visual search efficiency and navigation performance [53].

Geovisualization techniques as discussed above help improve positional accuracy of wildlife data at the very beginning of data generation. Sometimes, in cases where positional uncertainty does exist in wildlife data and is indeed of concern for wildlife habitat assessment, GIS-based methods have been developed to reduce its impact on the accuracy of wildlife habitat assessment. As an example, [54] proposed a spatial sampling method for deriving probable wildlife occurrence locations from patrol records using heuristics based on data recording context and species ecology to increase the accuracy of habitat suitability mapping.

#### **3.2 Geospatial analysis to tackle spatial bias**

Many geospatial analytical methods have been proposed to account for the spatial bias in wildlife data. An *AdaSTEM* approach that exploits variation in the density of wildlife observations was proposed to accommodate spatial bias in citizen-contributed wildlife observations [22, 55]. The continent- or hemispherewide study area is partitioned into rectangular spatial units (i.e., sub-areas) of size dependent upon density of wildlife observations. Environmental niches are modeled with only observations in each sub-area. By training local models in sub-areas, instead of training a global model using observations over the whole area, this approach mitigates spatial bias in the overall data set to a certain degree.

Filtering samples in the geographic or environmental space (i.e., remove observations that are within certain distance of one another) is also applied to reduce spatial bias [56, 57]. This method is based on the heuristic that removing localities (i.e., field samples) that are within certain distance of one another would somehow balance the unequal sampling or observation effort. The key of this method is to determine the distance threshold properly.

If detailed information observation effort is available, such information can then be incorporated to correct for spatial bias. Spatial bias in wildlife observations was compensated for by weighting the observations with weights inversely proportional to the cumulative visibility at the observation sites, given that cumulative visibility is a good proxy of the underlying observation effort [23]. Here, cumulative visibility is the frequency at which a given location can be seen by observers from the routes the observers take. It can be computed based on a digital elevation model (DEM) representing the terrain and the routes using viewshed analysis, a common GIS function.

A FactorBiasOut method was developed to correct for spatial bias in species presence-only data for species distribution modeling with MAXENT [58]. This method first estimates an empirical distribution to approximate the underlying but usually unknown sampling distribution that generated the presenceonly data. This approximate sampling distribution is then used to factor out

#### *Wildlife Population Monitoring*

the spatial bias in presence-only data. This is achieved by feeding MAXENT with background data that have the same spatial bias as the presence data. For example, occurrence data of a target group of species that are observed by similar methods can be taken as the estimate of the effort information and thus are used as the background data.

Recently, a general representativeness-directed approach was proposed to spatial bias mitigation in citizen-contributed wildlife observations (i.e., samples) for habitat suitability mapping [59]. The key idea is to define and quantify the representativeness of samples and then properly weigh the samples to improve representativeness. Sample representativeness is defined as the "goodness-of-coverage" of the samples in the environmental covariate space, which in turn is measured by the similarity between the probability distribution of the samples in the covariate space and the probability distribution of all mapping units (e.g., pixels) within the study area. Spatial bias is then mitigated by weighting the samples toward increasing sample representativeness. The optimal weights that maximize sample representativeness are determined through an optimization procedure using a genetic algorithm.

#### **3.3 Geocomputation to enable big data analysis**

Citizen-contributed wildlife data are an important source of geospatial big data. In spatial analysis or modeling of such large volume of data (e.g., point pattern analysis, wildlife habitat assessment, and species distribution modeling), it is urgent to address the associated computational challenges. Geocomputation technologies could be utilized to address such computational challenges.

For example, over 100 million bird observations were added to the eBird database each year. Point pattern analysis is commonly used to discover patterns from such data. Existing point pattern analysis software tools are not able to handle geospatial big data efficiently. Cutting-edge geocomputation technologies such as cloud computing and GPU (graphics processing units) computing can be leveraged to accelerate point pattern analysis algorithms. The massively parallel computing powers of cloud computing and GPU computing effectively sped up point pattern analysis tasks on big data by a factor of hundreds [60, 61]. Given the significant acceleration brought by the geocomputation technologies, geospatial big data analysis tasks that once were computationally prohibitive can now be conducted in a timely manner.
