Forestry & Forest Sciences · Mensuration, Inventory & Geospatial Analysis · also Geographic Information Systems, Geospatial Technology in Forestry, Precision Forestry
Spatial data and analysis for forest management: GIS data models, spatial analysis, GNSS, airborne and terrestrial LiDAR, and the digital tools of precision forestry. General remote sensing and GIS for land and water is in the Agricultural Engineering track (remote-sensing-gis).
In the exams: ICAR AICE-JRF 59 Unit 4 (GIS and remote sensing concept and scope); GISP and ASPRS certifications; growing share of SAF and FPBC professional practice
A forest is a spatial object. Every stand has a boundary, every road a route, every stream a buffer, every inventory plot a coordinate, and almost every management decision depends on where things are relative to each other. Geographic information systems (GIS) store, analyse and display this spatial information; global navigation satellite systems (GNSS) put field observations on the map; and light detection and ranging (LiDAR) measures the three-dimensional structure of the forest itself, tree by tree or hectare by hectare. Together with satellite remote sensing they form what is often called precision forestry: management based on measured, mapped and regularly updated information rather than on averages for large blocks.
This manual starts with the foundations, GIS data models, coordinate systems, map projections and datums, and moves through spatial analysis (overlay, buffers and terrain analysis), the stand and compartment databases that hold a forest's management record, and GNSS data collection under canopy. It then covers airborne LiDAR, from the physics of a laser pulse to canopy height models; the two main approaches to LiDAR-assisted inventory, area-based and individual-tree; terrestrial and mobile laser scanning; spatial decision support and multi-criteria analysis; open data and cloud platforms for forest monitoring; and map production and cartographic standards. Satellite image interpretation and classification are taught in the remote sensing subject of this track, and the statistics of inventory design in the forest inventory subject; this manual uses them where needed.
Geospatial skills are tested in doctoral and postgraduate forestry examinations (concept and scope of GIS and remote sensing, data models, applications), in professional geospatial certifications (data models, coordinate systems, accuracy standards, LiDAR processing), and increasingly in the practice standards of forest professional registration bodies, which expect a registered forester to commission, check and rely on spatial data. Short numericals are common: LiDAR range and footprint, point spacing, canopy height, slope, map scale, raster area, positional accuracy and weighted overlay scores.
A geographic information system is a set of hardware, software, data, methods and people for capturing, storing, checking, analysing and displaying information that is referenced to locations on the Earth. Its defining feature is that every record carries a location, so that layers of information about the same place can be combined. Two data models dominate. The vector model represents discrete features as points (an inventory plot, a nest tree, a well), lines (a road, a stream, a fire break) or polygons (a stand, a compartment, a protected area), each made of coordinate pairs and linked to a row in an attribute table. The raster model divides space into a regular grid of cells, each holding one value: elevation in a digital elevation model, a land-cover class in a classified image, canopy height in a LiDAR product. The cell size, or spatial resolution, controls detail and file size: halving the cell size quadruples the number of cells.
Vector data are compact and precise for boundaries and networks and suit legal and administrative records; raster data suit continuous surfaces and the analysis of images, and make overlay and map algebra simple. Most forest GIS work uses both and converts between them: rasterising stand polygons for a model, or vectorising a classified image into polygons. Topology is the explicit recording of spatial relationships, such as which polygons share a boundary or which lines meet at a node. Topological rules (polygons must not overlap, there must be no gaps between compartments, lines must connect at junctions) are used to find digitising errors. A triangulated irregular network (TIN) is a third surface model that joins irregularly spaced elevation points into triangles, and is often used as an intermediate step in building terrain models from LiDAR ground points.
A location is defined by coordinates relative to a reference. A geodetic datum defines the size and shape of the reference ellipsoid and how it is fixed to the Earth. The World Geodetic System 1984 (WGS 84) is the global datum used by GPS; many national and regional datums exist, some fixed to a continental plate so that coordinates of ground features do not drift as the plate moves. The same physical point can have coordinates differing by tens to hundreds of metres in different datums, so data must be transformed to a common datum before overlay. Geographic coordinates (latitude and longitude, in degrees) describe positions on the ellipsoid but are unsuitable for measuring distances and areas directly, because a degree of longitude shrinks towards the poles.
A map projection transforms the curved surface onto a plane, and every projection distorts something: shape, area, distance or direction. Conformal projections, such as Transverse Mercator and Lambert Conformal Conic, preserve local angles and shapes; equal-area projections, such as Albers Equal Area Conic and Lambert Azimuthal Equal Area, preserve area and must be used when forest area is calculated over large regions; equidistant projections preserve distance from one or two points. The Universal Transverse Mercator (UTM) system divides the world into 60 zones, each 6° of longitude wide, numbered eastward from 180° W. Zone n has a central meridian at 6n − 183 degrees (east positive), a scale factor of 0.9996 on that meridian, and a false easting of 500,000 m so that eastings are always positive; in the southern hemisphere a false northing of 10,000,000 m is added. UTM is widely used for forest maps of moderate extent because distortion within a zone is small. The spherical Web Mercator projection used by many online base maps is neither equal-area nor suited to area measurement; it greatly exaggerates areas at high latitudes. Coordinate reference systems are identified by standard codes in public registers, and recording the full reference system in the metadata of every dataset avoids the most common source of mismatched layers.
Map scale is the ratio of map distance to ground distance, written as a representative fraction such as 1:25,000, meaning one unit on the map is 25,000 of the same units on the ground. A large-scale map (1:5,000) shows a small area in detail; a small-scale map (1:1,000,000) shows a large area with little detail. In a GIS the data can be zoomed freely, but their accuracy remains that of the source, so a boundary digitised from a 1:50,000 sheet does not become more accurate when shown at 1:5,000. Data quality is described by positional accuracy, attribute accuracy, logical consistency (such as topological correctness), completeness, temporal validity and lineage (where the data came from and how they were processed). International metadata standards, notably ISO 19115, define how these properties are documented. A forester relying on a spatial dataset should always ask its source, date, scale or resolution, coordinate reference system and stated accuracy.
Spatial analysis derives new information from the location and attributes of features. Overlay combines two or more layers covering the same area. In vector overlay, intersect keeps only the area common to both inputs, union keeps all areas from both and splits them where they cross, identity keeps the whole of the first input and attaches attributes of the second where they overlap, and clip cuts one layer to the outline of another without adding attributes. A classic forest example is intersecting stand polygons with soil polygons and a slope-class layer to find stands on erodible soils on steep slopes. Overlay of independently digitised layers produces sliver polygons, tiny areas created where two versions of the same boundary do not coincide exactly; they are removed by setting a cluster tolerance or by merging small polygons into neighbours. Dissolve merges adjacent polygons that share an attribute value, such as all compartments of the same working circle.
A buffer is a zone of specified distance around a feature. Buffers around streams define riparian management zones, around roads define disturbance zones, and around nest trees define protection zones. The area of a buffer along a line is roughly twice the buffer width times the line length, plus the rounded ends; for a 5 km stream with a 30 m buffer on each side this is about 2 × 30 × 5,000 = 300,000 m², or 30 ha. Variable-width buffers, such as wider zones on steeper banks or larger streams, are created by taking the width from an attribute. Proximity analysis also includes distance surfaces (the distance of every raster cell from the nearest road), nearest-neighbour searches and cost-distance analysis, in which the effort of moving across each cell, such as skidding cost on steep or wet ground, is accumulated along the cheapest path.
In raster analysis each layer is a grid with the same cell size and extent, and new layers are computed cell by cell with map algebra. Local operations use only the value of each cell in the input layers, such as reclassifying elevation into bands or multiplying a suitability layer by a weight. Focal or neighbourhood operations use a moving window around each cell, such as smoothing a canopy height model with a 3 × 3 mean filter or finding the local maximum used to detect treetops. Zonal operations summarise one layer within the zones of another, such as the mean canopy height or total biomass of each stand. Global operations, such as Euclidean distance or cost distance, use the whole grid. Area is computed by counting cells: with 30 m cells each cell covers 900 m² (0.09 ha), so 12,000 forest cells represent 1,080 ha. Raster resolution must suit the question; a 30 m grid cannot resolve individual tree crowns, and a 0.5 m grid of a whole region may be impractically large.
A digital elevation model (DEM) is a raster of ground elevations. Terminology varies, but a digital terrain model (DTM) usually means the bare-earth surface and a digital surface model (DSM) the top surface including trees and buildings. From a DTM a GIS computes slope, the maximum rate of change of elevation, expressed in degrees (the angle whose tangent is rise over run) or in per cent (100 × rise ÷ run): a rise of 15 m over 60 m horizontal is a 25% slope, or about 14°. Aspect is the compass direction a slope faces, which strongly affects radiation, moisture and fire behaviour. Hillshade simulates illumination for display, and curvature distinguishes ridges from hollows. Hydrological analysis uses flow direction, often by the D8 method in which each cell drains to the steepest of its eight neighbours, to compute flow accumulation, stream networks and watershed boundaries; small depressions in the DEM are filled first so that flow is continuous. Topographic wetness indices combine upslope contributing area and slope to predict wet soils and are used to plan machine traffic and riparian protection. Viewshed analysis finds the area visible from a point, used for fire lookout siting and for assessing the visual impact of harvesting. Terrain derivatives depend on the source and resolution of the DEM: slopes computed from a coarse satellite DEM are smoother than those from a 1 m LiDAR DTM, which can reveal old skid trails, gullies and road prisms under canopy.
Every managed forest needs an authoritative record of its land units and their condition. Forests are usually divided hierarchically: a management unit or forest is divided into blocks or working circles, these into compartments with relatively permanent boundaries (often following ridges, streams and roads), and compartments into stands or sub-compartments, which are areas of similar species, age, structure and site that are treated alike and whose boundaries change as stands are harvested and regenerated. In a GIS the compartments and stands are polygon layers whose attribute tables hold the management record: identifiers, area, species composition, age or year of establishment, site class, stocking, volume, treatment history, planned treatments and constraints such as protection status.
Good stand databases follow relational database principles. Each table has a primary key, a field whose value uniquely identifies each record, such as a stand identifier. Related tables, such as treatment history, inventory plots or harvest records, carry the stand identifier as a foreign key and are joined to the stand layer when needed. Normalisation removes repeated data: rather than holding ten columns for ten past treatments, the treatment table holds one row per treatment per stand. A one-to-many relationship (one stand, many treatments) is handled by a related table, not by duplicating polygons. Domains and validation rules restrict values (a species code must exist in the species list; an age cannot be negative), and controlled vocabularies make data comparable across districts.
A stand database is only as useful as it is current. Changes come from harvesting, planting, fire, storms, pests, land purchases and boundary surveys, and from new inventory. Good practice records each change with its date and source, keeps a history of former boundaries rather than overwriting them, and uses versioning so that several editors can work at once and changes can be checked before they are accepted. Area is recalculated from the polygon geometry in an equal-area or suitable projected coordinate system, not typed in, and differences between GIS area and legal or historical area are documented rather than silently reconciled. Topology rules ensure that stands nest within compartments, compartments within blocks, and that there are no gaps or overlaps.
Stand attributes are increasingly derived from remote sensing and LiDAR rather than estimated by eye. A LiDAR canopy height model can be summarised by stand (a zonal operation) to give mean and dominant height, and an area-based inventory model can give volume and basal area per hectare for every grid cell, which are then aggregated by stand. Satellite change detection flags stands where canopy loss has occurred since the last update. The stand database then becomes the integrating layer that links inventory, growth models, harvest scheduling, certification reporting and operational planning. Data sharing within an organisation is easier when the database follows documented data standards, and many forest agencies and companies publish data dictionaries that define every field.
Typical problems in forest databases include stand areas that do not add up to the compartment area, duplicated or missing identifiers, attributes that refer to a stand that has since been split, polygons in the wrong coordinate system, and slivers from repeated edits. These errors propagate into harvest volumes, stumpage revenue, carbon estimates and certification audits. Regular quality checks, such as comparing the sum of stand areas with the legal area, checking for null values in required fields and running topology validation, are part of professional practice. When a forester relies on a stand database for a management plan, the plan should state the version and date of the data used.
Global navigation satellite systems include the United States' GPS, Russia's GLONASS, the European Union's Galileo and China's BeiDou, with regional systems such as Japan's QZSS and India's NavIC. Each satellite broadcasts signals carrying its orbit and precise time. A receiver measures the travel time of signals from several satellites and converts each to a range, called a pseudorange because it includes the receiver's clock error. Four unknowns must be solved, the three coordinates of the receiver and its clock offset, so at least four satellites are needed for a three-dimensional position. Multi-constellation receivers that use several systems see more satellites, which helps greatly under forest canopy and in steep terrain.
Errors arise from satellite orbit and clock errors, delays in the ionosphere and troposphere, multipath (signals reflected from trunks, branches, rocks and terrain before reaching the antenna), signal attenuation by wet foliage, and the geometry of the visible satellites. Dilution of precision (DOP) expresses the effect of geometry: when satellites are bunched in one part of the sky, DOP is high and errors are magnified; when they are well spread, DOP is low. Position dilution of precision (PDOP) multiplies the range error, so a PDOP of 2.5 with a range error of 4 m gives a position error of the order of 10 m. Field crews set a PDOP mask and an elevation mask, rejecting low satellites whose signals pass through more atmosphere and canopy.
Autonomous positioning by a single consumer receiver typically gives errors of a few metres in the open and considerably more under dense canopy. Differential correction compares observations with a base station at a known position. Satellite-based augmentation systems (SBAS), such as WAAS over North America and EGNOS over Europe, broadcast corrections that improve real-time accuracy to around a metre or two in the open. Post-processing against a nearby base station's data, using code or carrier-phase observations, can reach decimetre accuracy with suitable receivers. Real-time kinematic (RTK) positioning uses carrier-phase measurements and a correction stream from a base or a network of reference stations to reach centimetre-level accuracy in open conditions, though under canopy the receiver often cannot fix carrier-phase ambiguities. Precise point positioning (PPP) uses precise orbit and clock products without a local base, reaching decimetre to centimetre accuracy after a convergence period. The achievable accuracy under canopy depends on receiver grade, constellation use, occupation time and forest type, and must be tested locally rather than assumed from open-sky specifications.
In forestry, GNSS is used to locate inventory plots, map stand and harvest boundaries, record roads and skid trails, navigate to sample points, guide machines and record positions of observations such as wildlife signs or pest outbreaks. Good practice is to average many fixes at a point (logging for one or more minutes), use an external antenna raised on a pole above understorey, collect when satellite geometry is good, and record the datum, the correction method and the estimated accuracy with each feature. Plot centres that will be used to train LiDAR or satellite models need sub-metre to metre accuracy, because a plot located several metres off its true position is matched to the wrong canopy.
Accuracy is reported against independent check points of higher accuracy. The root mean square error (RMSE) is the square root of the mean squared difference between measured and true positions; for horizontal errors of 1, 2, 2 and 3 m the RMSE is √(18 ÷ 4) ≈ 2.12 m, larger than the mean error of 2.0 m because large errors are squared. The United States National Standard for Spatial Data Accuracy (NSSDA), widely copied elsewhere, reports horizontal accuracy at 95% confidence as 1.7308 × RMSEr when the x and y errors are similar, and vertical accuracy as 1.9600 × RMSEz. Other reporting conventions include circular error probable (CEP, the radius containing 50% of fixes); a forester comparing receivers must check which measure is quoted.
Light detection and ranging (LiDAR), also called airborne laser scanning (ALS), measures distance by timing laser pulses. A scanner on an aircraft or drone emits short pulses at a high pulse repetition frequency, commonly hundreds of thousands per second in modern systems, and a mirror or prism sweeps them across the flight line. The range to a target is half the round-trip travel time multiplied by the speed of light: a return after 6.67 microseconds comes from about 3 × 10⁸ × 6.67 × 10⁻⁶ ÷ 2 ≈ 1,000 m. Combining each range with the scanner's position from GNSS and its orientation from an inertial measurement unit (IMU) gives the three-dimensional coordinates of each reflection. Topographic systems usually use near-infrared wavelengths around 1,064 nm or 1,550 nm; bathymetric systems use green light around 532 nm, which penetrates water.
A laser beam diverges, so its footprint on the canopy grows with flying height: footprint diameter ≈ beam divergence (in radians) × range, so a 0.3 mrad beam at 1,500 m gives a footprint of about 0.45 m. Because the footprint is larger than a leaf, one pulse can hit several surfaces at different heights. Discrete-return systems record several returns per pulse, typically the first, intermediate and last; full-waveform systems digitise the whole returned energy profile. First returns come mostly from the upper canopy, and last returns include many from the ground in gaps between crowns, which is why LiDAR, unlike optical imagery, can map the terrain beneath a forest. Point density, in points or pulses per square metre, depends on pulse rate, flying height, speed, scan angle and overlap; mean point spacing is about 1 ÷ √density, so 4 points per m² gives a spacing of about 0.5 m. Area-based forest inventory works with densities of a few points per m², while individual-tree work and drone LiDAR use tens to hundreds.
Raw data are delivered as point clouds, commonly in the LAS format maintained by the American Society for Photogrammetry and Remote Sensing (ASPRS), or its compressed LAZ form. Each point carries coordinates, intensity, return number, number of returns, scan angle, time and a classification code; in the ASPRS standard class 2 is ground, classes 3, 4 and 5 are low, medium and high vegetation, and class 6 is buildings. Processing begins with strip alignment and checks against ground control, then classification of ground points by filtering algorithms that find the lowest surface consistent with plausible terrain slopes. Ground points are interpolated, often through a TIN, into a digital terrain model (DTM). A digital surface model (DSM) is built from the highest or first returns. The canopy height model (CHM) is the DSM minus the DTM, giving the height of the canopy above the ground: a DSM of 412.6 m above a DTM of 385.1 m gives a canopy height of 27.5 m. Alternatively, each point's height above ground is computed by subtracting the DTM elevation beneath it, producing a normalised point cloud from which height statistics are calculated.
Accuracy of the DTM is usually best on open, gentle ground and worse under dense, multi-layered canopy and on steep slopes, where fewer pulses reach the ground and classification errors are more likely; vertical accuracy is reported against surveyed check points, often as RMSEz. CHMs can contain pits, cells that are too low because a pulse penetrated deep into a crown, which are removed by pit-free algorithms or smoothing. LiDAR intensity, the strength of the return, is affected by range, angle and target and needs calibration before use.
Airborne LiDAR has become one of the most valuable data sources for forestry where it is available. Bare-earth terrain models support road and harvest planning, stream mapping under canopy, and identification of wet areas and erosion features. Canopy height models give stand height, gaps and structure. Statistical summaries of the point cloud predict timber volume, basal area, biomass and stem density, and density of returns in height layers describes understorey and fuel structure for fire models and habitat studies. Repeat surveys measure height growth and change from harvesting, storms and fire. Some countries and regions have flown national or regional LiDAR coverage and released it openly, while elsewhere surveys are commissioned for particular forests. The main limitations are cost per area, the need for field plots to calibrate models, and the gap between survey dates.
The area-based approach (ABA), developed in Scandinavia around the turn of the century and now used operationally in many countries, predicts forest attributes for grid cells from statistics of the LiDAR point cloud. Field plots, usually a few hundred spread across the range of forest conditions, are measured on the ground and located accurately by GNSS. For each plot the normalised point cloud is clipped and metrics are computed: height percentiles (such as the 90th percentile height, P90), mean and maximum height, the proportion of first returns above a height threshold such as 2 m (a measure of canopy cover), and density of returns in height layers. Regression or machine-learning models (linear and log-linear regression, k-nearest neighbour imputation, random forests) relate plot measurements of volume, basal area, biomass, mean height or stem number to these metrics. The models are then applied to a wall-to-wall grid whose cell area matches the plot area, because metrics change with the area over which they are computed: a circular plot of 12.62 m radius covers 500 m², matched by a grid cell of about 22.4 m side. Cell predictions are summed or averaged by stand.
For example, a model of the form volume = −45 + 14.2 × P90 (m³ per ha, P90 in m) would predict 267.4 m³ per ha for a cell with a P90 of 22 m; real models are fitted for each forest and sensor, and their coefficients cannot be transferred elsewhere without testing. Canopy cover from LiDAR is simply the share of first returns above the threshold: 1,800 of 2,400 first returns above 2 m gives 75% cover. Accuracy of ABA predictions at stand level is often better than that of traditional stand-level inventory by eye or few plots, and the results can be used in a model-assisted or model-based statistical framework to give unbiased estimates with known precision, which is how LiDAR is integrated into forest inventory design.
The individual-tree approach (ITD) finds and measures single trees in high-density point clouds or CHMs. Treetops are detected as local maxima in a smoothed CHM, using a window whose size may vary with height, and crowns are delineated by watershed segmentation or region growing around each top; point-cloud methods cluster points directly into trees. For each detected tree the height and crown dimensions are measured, and stem diameter and volume are predicted from allometric models relating diameter to height and crown size, often calibrated with field-measured trees. Species can be assigned from LiDAR structure, intensity and, more reliably, fusion with multispectral or hyperspectral imagery.
ITD works best for dominant and co-dominant trees in open or single-layered stands, especially conifers with distinct crown apexes. Suppressed and understorey trees are hidden under the upper canopy and are often missed (omission errors), while large broadleaved crowns with several tops may be split into several trees (commission errors). Results are therefore biased unless calibrated with field data. Accuracy is assessed by matching detected trees to field-mapped trees and reporting detection rates, omission and commission. In practice many operational systems combine approaches, using ABA for stand totals and ITD for tree lists and size distributions where point density allows.
Successful LiDAR inventory depends on matching field and LiDAR data in space and time: plot positions must be accurate, field measurements should be close in date to the flight, and plot size should be large enough that edge effects are small. Plots should cover the full range of conditions in the forest, including rare types, or predictions will be extrapolated outside the data. Leaf-on and leaf-off surveys give different metrics in broadleaved forests, so models should not mix seasons without testing. Because stand height and density drive volume strongly, LiDAR models usually explain much of the variation in volume and biomass, but species composition and wood density, which affect biomass, are less well captured by height alone. Costs per hectare fall with large survey areas, and many organisations share acquisitions with water, road and planning agencies.
Terrestrial laser scanning (TLS) places a scanner on a tripod inside the forest, where it records millions of points from the stems and crowns around it with millimetre to centimetre precision at close range. Scanners measure range either by time of flight, timing each pulse, which suits long ranges, or by phase shift, comparing the phase of a modulated continuous beam, which is fast and precise at shorter ranges. Because the scanner sees only what is in its line of sight, stems behind other stems are hidden; this occlusion is reduced by multi-scan designs in which several scans around and within a plot are registered into one point cloud, using artificial targets such as spheres or by matching shapes in overlapping clouds. Single-scan designs are faster but miss trees and parts of stems.
From TLS point clouds, stem diameter at breast height is estimated by fitting circles or cylinders to the points in a thin slice around 1.3 m above the ground: a fitted circle of 0.18 m radius gives a diameter of 36 cm. Stem curves (diameter at every height), tree height for trees whose tops are visible, crown dimensions and plot-level basal area can all be derived. Quantitative structure models (QSMs) reconstruct each tree as a set of cylinders representing stem and branches, giving volume of wood above ground; with wood density, this gives biomass without felling trees, which has made TLS an important tool for calibrating biomass allometry, particularly for large tropical trees that are rarely felled for weighing. Tree height from TLS can be underestimated in tall, dense canopies where the top is not visible from below.
Mobile laser scanning (MLS) mounts scanners on vehicles, forest machines, backpacks or hand-held units, and builds the point cloud while moving. Under canopy, where GNSS is unreliable, these systems use simultaneous localisation and mapping (SLAM), which estimates the scanner's trajectory by matching the geometry of successive scans while building the map. Personal laser scanning with backpack or hand-held units allows a plot or a strip of forest to be captured in minutes, giving stem positions and diameters for many trees. Accuracy is generally lower than multi-scan TLS but coverage per hour is much greater, and drift in the SLAM trajectory over long walks must be controlled by closing loops or using control points.
Drone or unmanned aerial vehicle laser scanning (ULS) flies a small LiDAR at low altitude, giving very high point densities from above, often hundreds of points per m², with some penetration to stems. It bridges airborne and terrestrial scanning for individual-tree mapping of small areas, research plots and operational blocks. Regulations on drone flights, such as line-of-sight rules, maximum heights and operator certification, vary by country and must be checked before survey. Harvesters equipped with scanners and positioning can also record the positions and dimensions of every tree felled, linking operations directly to the stand database.
Close-range laser scanning produces very large datasets that require processing pipelines for registration, ground classification, stem detection, diameter fitting and quality control. Results should be validated against tape or calliper measurements of a sample of trees, and the conditions of scanning (wind moves crowns and blurs branches; dense understorey hides stems) recorded. TLS and MLS are increasingly used to provide reference data for training airborne models, to measure plots in national forest inventories, to study canopy structure and light, and to create detailed records of permanent plots that can be re-measured digitally in later years.
A spatial decision support system combines spatial data, analytical models and an interface that lets decision makers explore alternatives. In forestry, such systems support harvest scheduling, road planning, fire management, habitat conservation, restoration prioritisation and land-use planning. They usually contain a database (stands, roads, constraints), models (growth and yield, cost, risk, habitat suitability), a solver or evaluation method, and tools to map and compare scenarios. Their value lies in making assumptions explicit and allowing stakeholders to see the consequences of different priorities, not in replacing judgement.
Harvest scheduling models, often formulated as linear or integer programming problems, choose which stands to treat in which period to maximise value or volume subject to constraints such as even flow of timber, minimum harvest age, and spatial constraints. Adjacency or green-up constraints prevent neighbouring stands from being cut within a set number of years, and maximum opening sizes limit clearfell areas; these constraints make the problem much harder to solve, and heuristic methods such as simulated annealing or tabu search are often used. Habitat and connectivity requirements can be included as constraints or objectives.
Multi-criteria analysis (MCA) ranks locations or alternatives against several criteria at once. In GIS-based MCA, each criterion is a layer: factors are continuous measures of suitability (such as distance to road, slope, soil quality), and constraints are Boolean masks that exclude areas outright (such as protected areas or slopes above a legal limit). Factors in different units are standardised to a common scale, usually 0–1 or 0–255, by linear rescaling or fuzzy membership functions. In weighted linear combination (WLC), the suitability of each cell is the sum of the standardised factor scores multiplied by their weights, which sum to 1, and the result is multiplied by the constraints. For a cell with standardised scores 0.8 for slope, 0.4 for access and 0.6 for soil, and weights 0.5, 0.3 and 0.2, the suitability is 0.5 × 0.8 + 0.3 × 0.4 + 0.2 × 0.6 = 0.64. WLC allows full trade-off between factors: a very high score on one can compensate for a low score on another. Ordered weighted averaging extends WLC to control the degree of trade-off and risk.
Weights are often derived by the analytic hierarchy process (AHP), developed by Saaty, in which experts compare criteria in pairs on a scale from 1 (equal importance) to 9 (extreme importance of one over the other). The weights are taken from the principal eigenvector of the comparison matrix. The consistency of the judgements is checked by the consistency index CI = (λmax − n) ÷ (n − 1), where λmax is the principal eigenvalue and n the number of criteria, and the consistency ratio CR = CI ÷ RI, where RI is a random index tabulated for each n (0.58 for three criteria). A CR below 0.10 is conventionally accepted; above that the comparisons are revised. For three criteria with λmax = 3.08, CI = 0.04 and CR ≈ 0.069, which is acceptable.
Every MCA result depends on the choice of criteria, standardisation and weights, so results must be tested by sensitivity analysis: changing weights within plausible ranges and checking whether the ranking of the top areas changes. Results that are stable under different weightings are robust; those that are not need further discussion. Participatory GIS brings community members and stakeholders into the definition of criteria and weights and into mapping local knowledge, such as sacred sites, grazing areas and resource use that do not appear in official data. Documenting who chose the weights and why is as important as the map itself, because decision support outputs are often questioned in public consultation and in court.
Free and open data have changed forest monitoring. The Landsat archive, with 30 m multispectral imagery from the 1980s onward (and coarser data from 1972), became free in 2008, enabling global analyses of forest change; each Landsat satellite revisits a location every 16 days. The European Copernicus programme provides Sentinel-2 optical imagery at 10–60 m resolution, with a revisit of about five days at the equator using two satellites, and Sentinel-1 C-band radar that sees through cloud. Moderate-resolution sensors such as MODIS give daily coverage at 250 m–1 km, used for fire detection and phenology. Spaceborne LiDAR adds structure: the GEDI instrument on the International Space Station, operating since 2018, samples canopy height and vertical structure in footprints of about 25 m diameter between about 51.6° N and S, and ICESat-2 uses photon-counting green laser to profile elevation and canopy. These are samples, not continuous maps, and are combined with imagery to produce wall-to-wall canopy height and biomass maps.
Derived open products include global forest cover and loss maps produced from Landsat time series, near-real-time deforestation alerts used by governments, companies and civil society, global land-cover maps at 10 m, canopy height maps and biomass maps. Many national forest agencies and statistical offices release forest maps, inventory plot summaries and, increasingly, LiDAR data under open licences. Users must check definitions: a global tree-cover loss map records loss of tree canopy, including plantation harvest and fire, and is not the same as deforestation in the sense of permanent conversion to another land use.
Analysing decades of imagery for a whole country once required downloading terabytes of data. Cloud geospatial platforms now hold petabyte-scale archives next to computing power, so analysts write scripts that run where the data are stored. Analysis-ready data are imagery already corrected to surface reflectance, cloud-masked and gridded, so that time series can be analysed without repeating pre-processing. Earth observation data cubes organise such data as stacks of images through time over a common grid, enabling per-pixel time-series analysis for change detection and phenology. Open standards make data findable and interoperable: the SpatioTemporal Asset Catalog (STAC) specification describes imagery collections, cloud-optimised GeoTIFF and point-cloud formats allow reading only the needed part of a large file over the internet, and Open Geospatial Consortium (OGC) web services serve maps and features.
Open-source desktop GIS and programming libraries in languages such as Python and R give access to these tools without licence costs, which matters for forest agencies and universities with small budgets. Data management principles summarised as FAIR (findable, accessible, interoperable, reusable) guide how forest data should be published. Sensitive data, such as locations of rare species, archaeological sites and community resources, may need to be generalised or withheld, so open data policies include exceptions.
National forest monitoring systems, built to report to the UN Framework Convention on Climate Change and to the FAO Global Forest Resources Assessment, combine satellite land-monitoring systems with field inventory, and increasingly use open data and cloud processing. Companies use the same tools to show that supply chains are free of deforestation, as required by some importing markets, and certification bodies use alerts to target audits. The forester's role is to understand what each product measures, its accuracy and its date, to validate it against field knowledge and higher-resolution imagery, and to report area estimates with proper statistical methods rather than simply counting pixels in a map that contains errors.
A map communicates; it is not just a display of data. A complete forest map carries a clear title stating subject, place and date; a legend explaining every symbol and colour; a scale, shown as a scale bar (which stays correct if the map is enlarged or reduced) and often as a representative fraction; a north arrow or graticule; the coordinate reference system; the data sources and their dates; the name of the author or organisation; and, where relevant, a statement of accuracy or a disclaimer. Inset or locator maps show where the area lies in a wider region. Operational maps for field crews also carry grid coordinates, access routes, hazards and contact information.
Visual hierarchy guides the eye to what matters: the subject of the map (for example, the harvest block) is shown strongly, and base information (contours, roads, streams) is lighter. Point, line and area symbols should follow conventions that readers know: blue for water, green for vegetation, brown contours, red or bold lines for boundaries of operations. Colour schemes depend on the data: sequential schemes (light to dark) for ordered values such as volume per hectare, diverging schemes for values above and below a meaningful midpoint such as change in canopy height, and qualitative schemes with distinct hues for categories such as species groups. Schemes that remain distinguishable for readers with common colour-vision deficiencies should be chosen, avoiding red–green contrasts as the only distinction.
Scale arithmetic is a frequent field and exam task. At 1:25,000, 4 cm on the map represents 4 × 25,000 = 100,000 cm, or 1 km. At 1:10,000, 1 cm represents 100 m, so 1 cm² represents 100 m × 100 m = 10,000 m², or 1 ha; areas scale with the square of the linear scale. Cartographic generalisation simplifies features for the map scale by selecting, simplifying, smoothing, aggregating and displacing them, so that a map at 1:250,000 does not attempt to show every stand boundary. Contour interval is chosen to suit the scale and the relief.
Choropleth maps colour areas according to a value, and should show rates or densities (volume per hectare, percentage cover) rather than raw totals, because large areas otherwise dominate visually. Values are grouped into classes by methods such as equal interval (classes of equal width), quantile (equal numbers of features per class, so 200 stands in 5 classes gives 40 per class), natural breaks (Jenks, which minimises variance within classes), or defined intervals that match management thresholds. The choice can change the impression a map gives, so the method and class limits should be stated.
Organisations adopt mapping standards so that maps are consistent and legally reliable: standard symbols for compartments, roads and protected features; standard scales for working plans and operational plans; and requirements for metadata. Metadata standards such as ISO 19115 record the content, quality, lineage, reference system and distribution of each dataset, and map products should reference them. Maps used for legal purposes, such as boundaries of concessions or protected areas and harvest plans submitted for approval, must meet the positional accuracy and survey requirements set by the relevant authority, which may require a licensed surveyor rather than a GIS analyst. A forester who signs a map or plan takes responsibility for its accuracy, so checking the sources, coordinate reference system and currency of the data is part of professional duty, as is showing uncertainty honestly, for example by marking boundaries mapped from imagery rather than ground survey.
A student uses this subject in GIS laboratories and field camps: digitising compartment boundaries, setting the correct coordinate reference system, buffering streams, computing slope and aspect from a DEM, logging plot centres with a GNSS receiver and checking their accuracy, and making a properly finished map with title, legend, scale bar and sources. Exams ask for definitions, the distinctions between DTM, DSM and CHM or between overlay operations, and short calculations of LiDAR range, footprint, point spacing, map scale and raster area. The commonest mistakes are measuring area in geographic coordinates or Web Mercator, mixing datums, and forgetting to halve the round-trip time.
A forester or analyst five years into practice maintains the stand database after harvests and fires, plans inventory plots and checks that GNSS positions are good enough to train LiDAR models, commissions or processes LiDAR surveys and checks the delivered DTM against control points, runs area-based models to produce stand volume and height, prepares harvest and road maps for approval, and uses deforestation alerts and open imagery to watch large estates. Daily decisions include which correction method to use under canopy, how dense a LiDAR survey must be for the intended products, and how to report the accuracy of a map honestly.
A senior professional decides whether an organisation invests in airborne LiDAR, drone scanning or open satellite products, sets data standards and versioning rules for the corporate stand database, and relies on spatial decision support for long-term harvest scheduling, conservation set-asides and road networks. They must judge whether model-based volume estimates are reliable enough for sale contracts and valuations, negotiate data sharing with other agencies, answer to auditors and regulators who check maps, and build teams that combine forestry and geospatial skills rather than leaving maps to specialists who do not know the forest.
A researcher with a PhD and long experience develops and tests methods: new point-cloud metrics and machine-learning models for volume, biomass and species, individual-tree algorithms for complex tropical canopies, fusion of spaceborne LiDAR with radar and optical data, and statistical frameworks that turn maps into unbiased estimates with stated precision. Such researchers advise national monitoring systems on methods for international reporting, design calibration plot networks, review accuracy claims in published products, and examine doctoral candidates on both the physics and the statistics behind the maps.
This manual gives principles and typical figures; sensor specifications, point densities, GNSS accuracies under canopy and model coefficients vary with equipment, forest type and terrain and must be checked for each project. It does not teach any particular software, gives only an outline of harvest-scheduling optimisation, does not cover radar interferometry or photogrammetric point clouds from imagery in depth, and does not list national LiDAR coverage or drone regulations, which differ by country and change often. Satellite image classification and inventory sampling statistics are covered in other subjects of this track.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.