(SAM), and artificial neural network (ANN) classifiers can be used for LULC mapping. To have a better understanding, the accuracy of classification can be improved by cloud patching and pan‐sharpening. The remotely sensed images can be used to classify area into the open, scrub and dense forest, delineation of water bodies, settlements and other important land use classes. The ANN‐based approach is reported to produce maps of high accuracy (Shaharum et al. 2018). In a study by Shaharum et al. (2018), it was observed that remote sensing tools can be used to assess socio‐economic activities that play a significant role in disturbing the natural environment of the study area.
2.3 LiDAR Technology
LiDAR is a remote sensing method in which a pulse of light is used to measure distances. The sensor emits a pulse of light to the earth's surface from an airborne or space‐borne laser for measurement. The technique provides a direct means to measure vegetation canopies' structure (Dubayah and Drake 2000). The pulse bounces off the tree canopy materials such as leaves and branches. The reflected energy is collected back at the instrument. Time taken for the pulse between emission, reflection, and recapture by the instrument is recorded. Various structure metrics are computed, analyzed, or modeled. Different LiDAR systems measure vegetation characteristics, mostly high pulse rate, small‐footprint, first‐ or last‐return‐only airborne systems which fly in the lower altitude region. Other systems are large footprint and full‐waveform digitizing that deliver superior vertical details about the vegetation canopy. Dubayah and Drake (2000) and Lefsky et al. (2002) provided a thorough overview of LiDAR application for land surface characterization and forest studies.
LiDAR systems have successfully recovered forest structure characteristics for different vegetation types quickly and directly. The technology has become an indispensable remote sensing tool for mapping forest inventory and structure. It has become popular for making informed decision‐making in forest management practices. LiDAR's ability to measure vertical as well as horizontal canopy structure can provide essential details for fuel estimation and fire behavior modeling. The flow chart in Figure 2.6 displays fusion of LiDAR and satellite data for improved image classification and feature extraction.
2.4 Artificial Intelligence and Remote Sensing
Remote sensing coupled with artificial intelligence (AI) provides essential technical supports to natural resource monitoring using various applications, including target detection, quantitative extraction of information, change detection and analysis, as well as multi‐source remote sensing information processing (SuperMap 2019).
Land use control is the primary means of developing and protecting land space. It protects the land by ensuring that all‐natural resources are utilized strictly according to the established plan. On the basis of target detection tools and technologies, the targets and scenes of prime interest can be precisely detected in the raster image and their size‐position can be instantly confirmed to identify different natural resources accurately. By recognizing the location of natural resources and their mutual relationships, the technology can aid the land space utilization control along with geological disaster prevention, ecological restoration and provide valuable information to law enforcement inspectors. AI image segmentation and classification technology can be used for quick high‐precision image classification. The quantitative indicators and boundaries of different natural resources can be automatically acquired, thereby assisting in the monitoring and evaluating of these natural resources. The supervision of natural resources and protection of farmland require automatic comparison, self‐inspection, self‐reporting, and regular verification of large land‐use areas. Earlier, this supervision was mainly dependent on the analysis of remote sensing images conducted visually by monitoring experts. With the integration of AI technology, changes in different natural resource categories can be quickly detected for a particular area and time.
Figure 2.6 Satellite and LiDAR data fusion for natural resource management.
The technology involving data processing and its optimization from different remote sensing sources can enrich the overall data quality. This technology can further improve the overall strength of the natural resources management process and may include:
1 LiDAR 3D point cloud processing techniques based on AI can enhance the monitoring data like buildings and terrains and improve natural resource management accuracy.
2 Optimization of remote sensing image quality using AI can improve the accuracy of image interpretation. Super‐resolution reconstruction and de‐clouding techniques can enhance the image quality and add more value to its use.
3 Hyperspectral image processing using AI can highlight various natural resources' subtle features and improve the natural resource classification results.
2.5 Machine Learning Tools for Natural Resource Management
In addition to biodiversity management, environmental researchers are also interested in Species Distribution Models (SDMs), which are widely used for predicting suitable habitat in space‐time for the choice of species (Bolliger et al. 2000; Raxworthy et al. 2003; Phillips et al. 2006; Baldwin 2009; Robinson et al. 2011). SDMs are beneficial in generating maps and results that identify suitable habitat areas and determine key environmental factors for driving species occurrence. These tools also report the threshold for suitable habitat demarcation and accuracy assessment. They are a valuable asset for many ecological studies and for species with narrow ranges it may also direct direction for future field surveys (Phillips et al. 2006). Integration of machine learning and geospatial techniques can easily handle such complex modeling problems.
Maximum Entropy (Maxent) is a presence‐only species distribution modeling technique used in the identification of a species' niche in environmental space. The prediction is based on the relation between observed occurrences to a set of climatic and non‐climatic environmental variables (Phillips et al. 2006; Pearson et al. 2007). Maxent is a free to download machine learning tool that provides a platform to integrate the species occurrence data with the bioclimatic variables using remote sensing and Geographical Information System and provides species habitat suitability and predicting future occurrence scenarios (Phillips et al. 2004; Phillips et al. 2006). Maxent has several advantages:
1 It can characterize probability distributions from incomplete information.
2 Presence data alone is sufficient and no absence data is required.
3 It uses both continuous and categorical environmental variables.
4 It produces an output with continuous prediction ranging from zero to one, where a higher value pixel indicates greater suitability for a given species at that pixel.
2.6 Applications of Unmanned Aerial Systems in Natural Resource Management
Most environmental monitoring and data collection systems are currently based on ground‐based measurements, satellite observations, and manned airborne sensors. These processes have spatiotemporal constraints that limit the current monitoring platforms. UAS provides an excellent opportunity to bridge this gap (Figure 2.1) by providing great spatial detail and enhanced temporal retrieval cost‐effectively (Manfreda et al. 2018). UAS systems provide very high spatial and temporal resolution images which are unmatched by satellite alternatives at a fraction of the satellite acquisition charge. The UAS‐mounted sensors have multiple additional
