UEF//eRepository
DSpace https://erepo.uef.fi
Rinnakkaistallenteet Luonnontieteiden ja metsätieteiden tiedekunta
2021
From comprehensive field inventories to remotely sensed wall-to-wall stand attribute data - a brief history of
management inventories in the Nordic countries
Maltamo, M
Canadian Science Publishing
Tieteelliset aikakauslehtiartikkelit
© 2021 Authors All rights reserved
http://dx.doi.org/10.1139/cjfr-2020-0322
https://erepo.uef.fi/handle/123456789/26274
Downloaded from University of Eastern Finland's eRepository
2
From comprehensive field inventories to remotely sensed wall-to-wall stand attribute data - a brief 1
history of management inventories in the Nordic countries 2
3
Maltamo, M.1, matti.maltamo@uef.fi 4
Packalen, P.1, petteri.packalen@uef.fi 5
Kangas, A.2, annika.kangas@luke.fi 6
1. University of Eastern Finland, School of Forest Sciences, P.O. Box 111, 80101 Joensuu, 7
Finland, 8
2. Natural Resources Institute Finland, Bioeconomy and Environment, P.O. Box 68, 80100 9
Joensuu, Finland 10
11
ABSTRACT 12
13
Forest Management Inventories (FMIs) provide critical information, usually at the stand 14
level, for forest management planning. A typical FMI includes i) the delineation of the 15
inventory area to stands by applying auxiliary information, ii) the classification of the stands 16
according to categorical attributes, such as age, site fertility, main tree species, stand 17
development, and iii) measurement, modelling and prediction of stand attributes of 18
interest. The emergence of wall-to-wall remote-sensing data has enabled a paradigm change 19
in FMIs from highly subjective, visual assessments to objective, model-based inferences.
20
Previously, optical remote-sensing data were used to complement visual assessments, 21
especially in stand delineation and height measurements. The evolution of airborne laser 22
scanning (ALS) has made objective estimation of forest characteristics with known accuracy 23
possible. New optical and Lidar-based sensors and platforms will allow further 24
3
improvements of accuracy. However, there are still bottlenecks related to species-specific 25
stand attribute information in mixed stands and assessments of tree quality. Here we 26
concentrate on approaches and methods that have been applied in the Nordic countries in 27
particular.
28 29
Keywords: airborne laser scanning, forest management planning, inventory by 30
compartments, relascope, tree species 31
32 33
4 INTRODUCTION
34 35
The National Forest Inventory (NFI) and the forest management inventory (FMI) are the two 36
main types of forest inventories. The main aim of the NFI is to provide sample-based large- 37
area forest resource information for national and regional policy-making (Table 1). However, 38
as forest management decisions are made at the stand level, i.e., the treatment unit level, 39
there has been a need for separate management inventories (Burkhart et al. 1978), which 40
have traditionally been based on field visits to each stand. Thus an FMI produces a wall-to- 41
wall map that covers each individual stand and estate. FMI data have several potential uses, 42
depending on the user’s needs and the extent of the areas of interest (AOIs) (Table 1).
43 44
The availability of wall-to-wall auxiliary information has brought a revolutionary change in 45
FMIs from highly subjective and often visual assessments towards objective, statistical 46
estimates with known accuracy. This is important, for the value of the data for decision- 47
making correlates strongly with the accuracy of the assessments (e.g., Kangas et al. 2018b).
48
The use of auxiliary remote-sensing data enables a localization of sample plot data to the 49
single stand level, which has led to the development of model-based small-area estimation 50
in management inventories (Breidenbach et al. 2016). This means that the differences 51
between a management inventory and a sample-based inventory have diminished. Sample- 52
based forest inventories can be regarded as a continuum from the stand level to the national 53
or even global level (Kangas et al. 2019b).
54 55
Whether the inventory is based on field visits or remote sensing, the basic steps are i) the 56
delineation of the inventory area to stands by the use of wall-to-wall auxiliary information, ii) 57
5
classification of the stands according to categorical attributes, such as age, site fertility, main 58
tree species, stand development class and iii) measurement, modelling and prediction of 59
stand attributes of interest, possibly by tree species. In the years following the inventory, the 60
information must be updated in regard to the treatments carried out in the forests as well as 61
the growth of the forests to meet the requirements of the users. Predictions of the future 62
development of the forest stock are needed for forest planning. A typical inventory cycle is 63
approximately 10 years, depending on the conditions, but the increased use of remote- 64
sensing data is expected to shorten it in the future.
65 66
In this paper we describe the evolution of FMIs and the impact of the information obtained 67
on the management of forests. We focus on the Nordic countries and the paradigm shift 68
from subjective field inventories to objective wall-to-wall inventories, based on remote 69
sensing, and their developments in near future.
70 71
HISTORY OF MANAGEMENT INVENTORIES 72
The era of field assessments 73
74
The basic principles of the stand-level management inventory were developed in central 75
Europe during the 19th century (Loetsch et al. 1973). The first management inventories were 76
conducted in special areas (Næsset 2014), such as Koli National Park in Finland (see e.g.
77
Vehmas et al. 2009) where the first inventory of slash-and-burn areas, fields, meadows and 78
forest lands was carried out between 1835–1845, and the first forest inventory was 79
performed in 1910 (Vehmas et al. 2009).
80 81
6
The process followed in the management inventory based on field measurements is 82
described in Fig. 1. In the earliest applications, the inventory areas were delineated 83
according to base maps or, if no map was available, the inventory would also include a land 84
survey (see, e.g., Koivuniemi and Korhonen 2006). The use of aerial photographs as auxiliary 85
data for stand delineation was initiated in the 1940s. Historically, the tree stock 86
characterization was based on visual stand density assessments and height measurements, 87
for example, which were used to predict the stand volume (Ilvessalo 1965). The introduction 88
of the relascope principle and the corresponding measurement device improved the stand 89
density measurements (Bitterlich 1984), enabling accurate and rapid measurements of the 90
stand basal area. When this information was combined with mean height data, the 91
prediction of stand volume (through the application of volume tariff models or tables) was 92
easy, even in the field scale.
93 94
In many cases, the fieldwork-based management inventory is still based on relascope and 95
height measurements or different modifications of this method. This is also true outside the 96
Nordic countries. In Russia, for example, 60 % of the state forests have been inventoried 97
with this type of approach (Kinnunen et al. 2007). In many countries, however, the field- 98
based approach has already been replaced with remote-sensing applications.
99 100
The delineation of stands is typically based on differences in tree species, site fertility and 101
stand development class (or stand age). Field plots are subjectively placed at 102
“representative” locations within the stand. The field work is thus based on purposive rather 103
than design-based sampling. This is because any systematic placement of the sample plots 104
within a stand would make their measurement too slow and too expensive (Koivuniemi and 105
7
Korhonen 2006). Within each plot, the few measurements made include measurements of 106
the basal area and the height (possibly by tree species) plus a visual recording of the age, soil 107
type, site fertility, stand development class, and other categorical stand attributes. In some 108
countries, the mean diameter in the field is also assessed, and the final stand-level estimates 109
are averaged from the plot assessments. It is also possible to visually assess the tree quality 110
and characterize the needs of silvicultural operations in the field. In North America, a stand- 111
level forest inventory is typically known as cruising, but this characterization may refer either 112
to the management inventory or the pre-harvest inventory.
113 114
When stand-level volume models are used, the prediction of stand attributes not assessed in 115
the field is straightforward, provided that the characterization of the tree stock is kept at the 116
stand level. When stand-level models are used, a compatible system of equations may be 117
used to both predict the current state of the tree stock and project future development.
118
Systems of this type may include equations for the basal area and the dominant height and 119
volume (e.g., Hyink and Moser 1983). However, if information on the mean diameter is also 120
recorded, it is possible to back-transform the tree stock to the tree level by the applying 121
theoretical diameter distribution models (Kilkki et al. 1989). This means using field estimates 122
of stand attributes to predict or recover parameters of some probability density function, 123
possibly by tree species, which are then used as a surrogate for actual diameter distribution.
124
Tree heights and volumes can also be predicted at the tree level in the same way. This 125
approach allows the characterization of the horizontal forest structure and the prediction of 126
timber assortments, which is very important information for management planning despite 127
the considerable uncertainty that usually attaches to the predictions. This approach also 128
allows for predicting the development of the tree stock using single-tree growth models.
129
8 130
As field plots are typically located subjectively, the accuracy of a stand-level inventory 131
cannot be calculated within the design-based approach. Though a model-based approach 132
could be applied, it is usually not. However, validation inventory applications for the stand 133
level have been developed (Jonsson and Lindgren 1978, Ståhl 1992). They are usually based 134
on a systematic layout of validation plots in sampled stands. In validation inventories 135
conducted in Finland and Russia, the relative root mean square error (RMSE) values 136
associated with stand-level tree volume predictions have been over 20 % and 30 %, 137
respectively, and species-specific estimates have been considerably more inaccurate (Haara 138
and Korhonen 2004, Kinnunen et al 2007). The validation inventories also show that stand 139
attributes are often underestimated, which may be due to visibility problems with the basal 140
area measurements and to the precautionary principle followed in the assessments, 141
especially in densely stocked stands.
142 143
The need for up-to-date data has lead to the use of growth models in updating the stand- 144
level data. The forest companies have been updating their data from the 1980s, and 145
digitalization has brought the updating of data within reach of private forest owners, too.
146
The accuracy of the updated data has been evaluated by means of separate validation 147
studies like the original data. Hyvönen and Korhonen (2003), for instance, found 5-year-old 148
updated data to be as accurate as a new inventory. However, the reliability of the updated 149
information decreases over time (Haara and Leskinen 2009), necessitating a new inventory 150
by and by. Remote sensing has enabled the updating of the treatments by means of satellite 151
images (e.g., Pitkänen et al. 2020).
152 153
9
In plantation forestry, the main emphasis may be on the prediction of site quality (Rombauts 154
et al. 2010, Tesfamichael et al. 2010), usually based on the dominant height or volume 155
besides stand age and initial tree density. Another specific use of the management inventory 156
is timber sale inventories, i.e., pre-harvest inventories, in which the AOI may be just one 157
marked stand (Siipilehto et al. 2016). In that case the primary interest lies in the description 158
of diameter distribution and the quality attributes of individual stems (Kankare et al. 2014).
159
Another example is the inventory that seeks to map some specific tree species, often related 160
protecting commercial trees of high value (Sverdrup-Thygeson et al. 2016, Kivinen et al.
161
2020).
162 163
The use of optical remote-sensing data as auxiliary information 164
165
The use of aerial image data in forest inventory work started after 1945, though some forest- 166
related experiments had been conducted earlier (e.g., Seely 1929, Spurr 1955). By the 1950s, 167
studies to characterize the growing stock by visual interpretation of aerial images were 168
ongoing (e.g., Nyyssönen 1955, Willingham 1957), although the common practical 169
applications that we know had not yet emerged.
170 171
The situation changed in the 1970s with the development of stereo photogrammetry. In 172
Norway, for instance, stand delineation, stand height and crown closure were determined by 173
means of analogue images and stereo plotters, and stand volumes were then predicted by 174
means of the existing volume models (see Næsset 2014) (also see Fig. 1). Other parameters, 175
such as tree species and site quality, were also interpreted visually. Field visits were 176
performed only occasionally. Inventory approaches of this type were also common in many 177
10
other countries, such as Sweden, and are still used in Canada, for example (White et al.
178
2013).
179 180
The next developments in remote-sensing materials were the availability of digitized images 181
in the 1990s and images taken with digital aerial cameras in the early 2000s. The digitized 182
and digital images were often orthorectified, so that they could be treated similarly to 183
georeferenced maps (Mikhail et al. 2001). At the same time, advances in satellite positioning 184
systems improved the positioning accuracy of field plots. This led to multiple approaches to 185
utilizing aerial and satellite images (Holmgren et al. 2000, Hyvönen et al. 2005). Most of 186
these were computer-aided automated methods operating at the stand, plot, or tree level.
187
The plot and the stand-level approaches typically used tone and texture metrics to predict 188
stand characteristics. Several algorithms were also developed for individual tree detection 189
(ITD) from images (see, e.g., Packalen 2009). Despite active research, most of these 190
approaches never got to be used operationally, with the exception of some aerial image- 191
based 2D systems that were put to practice for a short time (Anttila and Lehikoinen 2002).
192
The reason for their lack of use in operational inventories lies in the limited information 193
contained in the images: in most cases the RMSE values associated with volume predictions 194
were clearly over 30 %, thereby making field inventory information more preferable.
195 196
THE EVOLUTION OF ALS-BASED MANAGEMENT INVENTORIES 197
Basic principles 198
In the past 20 years, the emergence of accurate wall-to-wall remote-sensing data from 199
airborne laser scanning (ALS) has had a major impact on stand-level management 200
inventories (White et al. 2013, Maltamo and Packalen 2014, Næsset 2014); the first studies 201
11
in this field were initiated as early as the 1990s (Næsset 1997, also see review by Nelson 202
2013). The usability of ALS data is based on 3D point clouds that describe the tree canopy in 203
detail. While the attributes of the canopy are usually not of primary interest in a 204
management inventory, the metrics derived from point clouds have a very strong statistical 205
relationship to volume and height, for example. Their correlation to tree species information 206
is markedly weaker, which is why stand-level management inventories are often carried out 207
by also using digital optical images (Kukkonen et al. 2018).
208 209
In ALS studies, there has been a strong emphasis on management inventories from the start, 210
while research related to sampling-based large-area inventories and biomass and ecological 211
applications have started later (Vauhkonen et al. 2014a, Saarela et al. 2015, Kangas et al.
212
2018a). To date, ALS-based commercial management inventories have been applied in 213
practical forestry in the Nordic countries (Norway, Sweden, Denmark, and Finland), Estonia, 214
USA, Brazil, South Africa, New Zealand, Spain, Australia, and Canada (see Turunen et al.
215
2012, McRoberts et al. 2014, Nord-Larsen et al. 2017, Nilsson et al. 2017), and associated 216
research has been conducted in numerous other countries. It is notable that even after a 217
short period of research, the use of ALS has become one of the main methods of stand-level 218
management inventories in many countries. In Norway, ALS-based commercial inventories 219
started in 2002, and in Finland, approximately 3 000 000 hectares are interpreted by means 220
of ALS annually (Maltamo and Packalen 2014).
221 222
There are two basic ways of using ALS information in forest inventories. They are the area- 223
based approach (ABA) and ITD (Vauhkonen et al. 2014a). The ABA model employs the 224
relationship between ALS metrics and the stand attributes. The training for the ABA model 225
12
uses a set of sample plots, from which the attributes are predicted to grid cells covering the 226
entire inventory area. In ITD, individual trees are first identified and the attributes of interest 227
are then predicted by means of ALS data obtained from the area of the tree crown. Most of 228
the practical management inventory applications have been conducted with ABA, but ITD or 229
a combination of ITD and ABA has been used in several projects in Sweden (e.g., Lindberg et 230
al. 2010, Lindberg and Holmgren 2017). In general, low pulse-density (< 1 pulse per m2) ALS 231
data have been used, but the pulse density has increased markedly in recent years; In 232
Finland, for example, the density is currently >5 points per m2. Below, we describe the 233
principles of the ABA-based inventory system applied in the Nordic countries (Næsset 2002, 234
2004, Packalén and Maltamo 2007, Nilsson et al. 2017, Nord-Larsen et al. 2017).
235 236
An inventory project starts with delineation of the AOIs. Larger areas are preferred for the 237
inventory, for they make for better cost-efficiency in the ALS data acquisition. Typically, the 238
area is over 100 000 hectares. The next phases involve the acquisition of remote-sensing 239
data and the planning of the placement of field data plots on the basis of the available 240
auxiliary data. Usually, a degree of pre-stratification is performed and non-forest areas 241
omitted. In the pre-stratification phase, older inventory data updated with growth models 242
may be utilized. If remote-sensing data are available, a more detailed pre-stratification can 243
be carried out. Besides, stratification allows concentrating the field plots in the strata of 244
interest, so that sparsely stocked areas, such as peatlands and mountain forests, can be 245
omitted from the inventory.
246 247
In Norway, the stand delineation and stratification are carried out by means of 248
orthophotography and digital stereo photogrammetry (Næsset 2014). If no auxiliary data are 249
13
used and the selection is either simple random sampling or systematic sampling, there is a 250
risk of the number of field plots, usually small, not capturing all the variation present in the 251
inventory area. If auxiliary data, such as older inventory data or remote-sensing data, are 252
available, they can be used in the selection of the field plots by means of stratified sampling 253
(McRoberts et al. 2014), balanced sampling (Grafström et al. 2014), or other similar methods 254
to enhance the selection efficiency (e.g., Maltamo et al. 2011). If a model-based approach is 255
used, it is also possible to optimize the sample with respect to a given model.
256 257
The acquisition of field data requires highly accurate plot geo-referencing. Studies to date 258
have shown that positioning errors in forested areas are approximately 1 m (e.g., Gobakken 259
and Næsset 2009). Usually, diameter at breast height (DBH) is measured and the species is 260
recorded for all trees in the sample plot, while tree height is usually measured from a subset 261
of trees (Maltamo and Packalen 2014). Furthermore, stand age and certain categorical 262
variables are recorded, but tree locations are typically not determined.
263 264
The modelling of stand attributes is based on training data that consist of field attributes and 265
the ALS metrics from the same plots (Næsset 2002). The ALS metrics applied are typically 266
those related to the height distribution of the ALS echoes, but some density metrics and 267
other statistical characteristics (e.g., standard deviation, skewness, and kurtosis) are also 268
frequently used. If the interest is in separating the tree species, metrics based on the 269
intensity values of the ALS points can be applied (Ørka et al. 2007, Korpela et. al 2010). The 270
most straightforward approach is a regression analysis between ALS metrics and stand 271
attributes, such as volume, basal area, number of stems, mean diameter, and Lorey’s height 272
14
(see Næsset 2002, 2014), but the nearest neighbour (k-nn) imputation is also commonly 273
used (Lindberg et al. 2013).
274 275
Species-specific predictions 276
277
There has been a fair amount of research into species-specific estimation in boreal 278
conditions (e.g., Holmgren and Persson 2004, Packalén and Maltamo 2007, Holmgren et al.
279
2008, Korpela et al. 2010, Ørka et al. 2013, Yu et al. 2017, Budei et al. 2018, Kukkonen et al.
280
2018, 2019). Tree species may be considered in many ways. At the plot/stand level, for 281
instance, either the main tree species, the proportions of the species, or the stand attributes 282
by tree species are predicted, depending on the requirements of the end-user or the 283
established practise in the country.
284 285
In Finland, stand attributes by tree species have traditionally been produced with the field 286
inventory system (Koivuniemi and Korhonen 2006); and corresponding information also 287
needs to be provided by the current remote-sensing inventories. The stand attributes by 288
tree species are needed because mixed stands are very common in Finland. The tree species 289
(classes) separated for commercial reasons in a remote-sensing inventory are pine, spruce, 290
and the deciduous species (mostly birch). Due to the compatibility requirement of species- 291
specific estimates, the modelling is typically carried out simultaneously with the k-nn 292
imputation approach (Packalén and Maltamo 2007). This also allows the estimation of 293
species-specific tree lists (Packalén and Maltamo 2008). The process of the species-specific 294
ALS inventory is described in Fig. 2, and the remote-sensing materials included are presented 295
in Fig. 3.
296
15 297
Although ALS data include intensity values, which have some capacity to discriminate among 298
tree species (Ørka et al. 2007), and leaf-off data can distinguish between coniferous and 299
deciduous species (Kim et al. 2009, Villikka et al. 2011), tree species discrimination in 300
operational inventories is improved by also using the metrics of aerial images as predictor 301
variables (Fig. 3) (Packalén and Maltamo 2007).
302 303
Another approach is to predict the main tree species (Astrup et al. 2019) or assume them to 304
be known from other data sources or from pre-stratification (Næsset 2002). Other data 305
sources, such as older inventory data, may contain information on either the proportions of 306
the species or the main tree species. In many plantations it is sufficient to only know the 307
main tree species of a stand, but for mixed stands, more detailed information is likely to be 308
needed.
309 310 311
Wall-to-wall implementation 312
313
The ABA models constructed are applied to a systematic grid of cells overlaid on the 314
inventory area with no consideration of stand borders (Næsset 2004). In general, the size of 315
one cell is approximately the same as the size of a training plot (Packalen et al. 2019). The 316
smallest area of interest is usually an individual stand, for which the information is obtained 317
by averaging the predictions for individual cells (Fig. 3). This also means ignoring the within- 318
stand variation produced by the model predictions. Most of the stands are not represented 319
16
in the training data. In this sense, the models are characterized as synthetic (Särndal et al.
320
1992, Magnussen et al. 2016, Astrup et al. 2019).
321 322
While stand delineation in the past has usually been based on visual interpretation or 323
segmentation of aerial images, ALS data offer possibilities of developing the delineation 324
approaches by using, e.g., the canopy height model (Fig. 3) (e.g., Mustonen et al. 2008, 325
Pascual et al. 2008). By segmentation one may obtain homogeneous microstands, which are 326
then combined into operational stands in connection with forest planning or silvicultural 327
operations. But predictions of stand attributes for microstands are also made by means of a 328
systematic grid, thus using the same prediction unit as is used for stands.
329 330
During the development of an ALS-based management inventory in Norway, a large number 331
of aspects with respect to field and ALS data have been examined: plot size, number of plots, 332
geo-referencing accuracy, ALS point density, leaf-on and leaf-off data, flying altitude, and the 333
effects of different sensors (see Næsset 2014). This has led to the optimization of inventory 334
parameters. Leaf-off ALS data is suitable for the inventory (Næsset 2014, White 2015), and it 335
improves the separation of deciduous and coniferous species (Villikka et al. 2011). However, 336
the acquisition window of leaf-off data between snowmelt and leaf burst is narrow in the 337
Nordic countries, being just a few weeks in some cases. Leaf-off data may also be used in 338
land surveying for terrain modelling, which increases the cost-efficiency, provided the data 339
acquisition costs are shared among the partners.
340 341
Another possibility to reduce inventory costs is the use of existing field plots. In recent years, 342
some research has been carried out into using ground truth data from geographically 343
17
neighbouring areas inventoried earlier on (Gopalakrishnan et al. 2015, Fekety et al. 2018, 344
Kotivuori et al. 2018, Tompalski et al. 2019). The results are promising in general, but the 345
transferability of species-specific forest attributes remains a challenge. Another alternative 346
for management inventories is a combination of NFI and management inventory field data 347
(Maltamo et al. 2009, Tuominen et al. 2014). This is a realistic alternative if the plots are geo- 348
referenced and the plot properties (size and shape) and growing stock measurements are 349
similar. While NFI plots could be used as the sole source of field plots to predict stand-level 350
data (Nilsson et al 2017, Astrup et al. 2019), the sampling intensity in NFI is usually 351
insufficient for management inventories. For that purpose, the NFI sampling design must be 352
intensified locally by measuring more field plots. Currently, many NFIs are based on the 353
principle of continuous forest inventory, which means that the whole country is inventoried 354
every year by applying a rather sparse sample of field plots. This is beneficial for 355
management inventories, as it makes some up-to-date data continuously available in 356
different parts of the country. Moreover, NFI data from previous years can be utilized, 357
provided the silvicultural operations carried out within the dataset are taken into 358
consideration.
359 360
A further possibility for an ALS inventory is to use the field data obtained from harvester 361
machines. This requires accurate geo-referencing of the harvester machine and is restricted 362
to clearcut stands. Works by Hauglin et al. (2018) and Maltamo (2019) have shown that this 363
is a realistic alternative for stand volume and tree-list predictions. However, there are error 364
sources in this procedure, such as bark slippage in the stems, technical errors in the GPS and the
365
harvester head due to the field conditions, and the effect of the retention trees, which are
366
counted by ALS but not by the harvester.
367
18 368
369
Reliability 370
371
Typically, the accuracy of an ALS-based inventory has been assessed by means of training 372
plots or somewhat larger validation plots (e.g., Næsset 2002, Holmgren 2004, Packalén and 373
Maltamo 2007). The use of harvester data allows for empirical stand-level validation, as was 374
shown by Maltamo et al. (2019), but only for clearcut stands.
375 376
As the data are used at the stand level, the accuracy assessments also need to be made at 377
the stand level. In FMI, accuracy assessments have traditionally been carried out by 378
conducting a more accurate stand inventory and comparing that with the inventory results.
379
A similar approach is also possible for an ALS-based inventory. Wallenius et al. (2012), for 380
example, conducted a field inventory to assess the accuracy of an ALS inventory. The 381
drawbacks are that such an inventory is costly and that the results are only available a long 382
time after the original inventory.
383 384
Stand-level accuracy can be assessed in real time by using wall-to-wall remote-sensing data 385
and a model-based sampling theory. In principle, the errors of the model can be used 386
directly to estimate the accuracy attained at the stand level even if there are no field plots 387
available from the stand in question (synthetic estimation; see Magnussen and Breidenbach 388
2017). The within-stand correlations of the predictions, however, cause an underestimation 389
of the stand-level RMSE values. Due to these correlations, the model has a stand-level bias 390
(the so-called stand effect). It is an error component that is fixed within the stand, whereas 391
19
the residual variance decreases with an increasing number of grid cells within the stand (e.g., 392
Breidenbach et al. 2016). If the strength of the correlations can be assessed, it is possible to 393
obtain a satisfactory estimate of the accuracy attained (Magnussen et al 2016). To that end, 394
Astrup et al. (2019), for example, used information from a previous study by Rahlf et al.
395
(2014), where the stand effect accounted for about 25 % and the residual error for about 75 396
% of the total error variance in the stand-level analysis.
397 398
Instead of a stand effect, it is possible to estimate the autocorrelation function and to use 399
that for a model-based accuracy assessment (Puliti et al. 2019). Using an autocorrelation 400
function means that instead of assuming the within-stand correlation to be constant, the 401
inter-cell correlation is assumed to decrease with increasing inter-cell distance (McRoberts 402
et al. 2018).
403 404
Accuracy assessments of updated ALS inventory data have typically also been carried out by 405
means of a separate validation study (e.g., Luoma et al. 2017). Contrary to expectations, ALS 406
data updated over a short period can even have a smaller relative RMSE than a new 407
inventory. On the other hand, the levels of bias can be much higher in the updated data than 408
in new data, and that can lead to the new inventory data still carrying more value in the 409
decision-making (Kangas et al. 2019a).
410 411
Instead of a validation study, it is also possible to use a model-based approach to assessing 412
the accuracy of the updated data by employing the concept of hybrid or model-based 413
inference (e.g., Fortin et al. 2019, Melo et al. 2019). Use of the Kalman filter approach will 414
enable an assessment of the accuracy of the updated data and also an assimilation of the old 415
20
data with new data in order to obtain more accurate information (e.g., Ehlers et al. 2013, 416
Nyström et al. 2013). It is also possible to use composites of different data sources whose 417
model-based accuracy estimates are known (e.g., Ehlers et al. 2018, Lindström et al. 2017).
418 419 420
NOVEL REMOTE-SENSING DATA SOURCES 421
422
Traditionally, the inventory cycle of a management inventory has been 10 years. Nowadays, 423
more frequent updating of forest resource information is considered optimal. Rather than 424
making the inventory cycle shorter (by repeating ALS data acquisition), alternatives with 425
lower costs have been developed over the past decade. The most promising alternative is 426
photogrammetric matching of airborne images (Gobakken et al. 2015, White et al. 2015, 427
Goodbody et al. 2019).
428 429
This approach provides a point cloud comparable to ALS point data, but only from the upper 430
canopy. To utilize these data, an accurate Digital Terrain Model (DTM) is usually required to 431
scale the heights to the above-ground level. Typically, an ALS DTM is used for this purpose 432
(St-Onge et al. 2008, Bohlin et al. 2012). Moreover, old ALS DTMs can be used, for the terrain 433
level in a forest does not change rapidly over the years or even over the decades. Currently, 434
ALS DTMs are available in many countries, and thus the lack of accurate DTMs often does 435
not limit the usability of photogrammetrically created point clouds in forest inventories.
436
Several studies have shown that the accuracy of the stand attributes provided by 437
photogrammetrically created point clouds is only marginally poorer than the corresponding 438
values obtained in an ALS inventory (Bohlin et al. 2012, Gobakken et al. 2015, Tuominen et 439
21
al. 2017, Puliti et al. 2017). If this is sufficient for FMI requirements, then future inventory 440
cycles can be based on photogrammetrically created point clouds as well. In species-specific 441
FMIs, this approach provides a single-sensor solution, for the tone and texture metrics can 442
be computed from the same images (Kukkonen et al. 2019).
443 444
Very high-resolution spaceborne stereo images may be used similarly to airborne images to 445
create point clouds photogrammetrically (Straub et al. 2013). This requires a special satellite 446
constellation and data acquisition setup, such as two satellites operating in the same orbit 447
(e.g., Pléiades 1A/1B), or repeat pass of the same satellite over the AOIs (e.g., WorldView-2).
448
Research that uses point clouds based on satellite images for forest inventories is still at the 449
beginning stage (see, e.g., Persson and Perko 2016, Hosseini et al. 2020), and we are not 450
aware of any operational FMI projects on that basis. The motivation for using spaceborne 451
stereo images instead of airborne data is typically an aspiration to cover large areas at low 452
cost (e.g., Fassnacht et al. 2017), but these advantages are not apparent.
453 454
In many countries, a second round of ALS inventories has already started. This provides a 455
possibility for the use of bitemporal ALS data. Data of this type bring forth new opportunities 456
to incorporate site productivity and local growth trends into ALS inventories (Tompalski et al.
457
2018). For example, Noordermeer et al. (2018) have presented a range of approaches to 458
predicting site productivity in Norway by applying bitemporal ALS metrics and existing site 459
index curves. For the practical implementation of this approach (Noordermeer et al. 2020), 460
any disturbances between the two consecutive ALS data acquisitions need to be detected 461
(Noordermeer et al. 2019). The introduction of site productivity into applications of 462
bitemporal ALS data can be regarded as a major improvement in ALS-based management 463
22
inventories; previously, the site index has mainly been dealt with in some research (e.g., 464
Gatziolis 2007).
465 466
New laser scanning technologies are developed continually. Single-photon and Geiger-mode 467
Lidar have recently emerged on the commercial market (Leica SPL100, Harris). Both systems 468
detect backscattering photons more sensitively than conventional ALS systems do (Degnan 469
et al. 2016). Typically, just a few photons are sufficient to record a range of measurements.
470
The high sensitivity of the single-photon Lidar enables the collection of high-density point 471
clouds from higher altitudes than current ALS systems do. The most attractive property of 472
these systems is their cost-efficiency in large areas. Wästlund et al. (2018) and Yu et al.
473
(2020) obtained a level of accuracy in predicting forest attributes with single-photon Lidar 474
data that was similar to those obtained with conventional ALS data, although the flight 475
altitude was considerably higher in the single-photon acquisition. However, single-photon 476
and Geiger-mode Lidar data are not widely used at present, and with their current cost 477
structure, they are not cost-efficient options for a standard FMI project.
478 479
For discerning different species in an ALS inventory, there are now alternatives to aerial 480
images. Hyperspectral airborne Lidar data have been used successfully with ALS data 481
(Kandare et al. 2017), and modern moderate spatial-resolution satellite sensors, such as 482
Sentinel-2, also provide useful data for species discrimination (Kukkonen et al. 2018, Persson 483
et al. 2018). Correspondingly, multispectral ALS systems provide improved species 484
information (Axelsson et al. 2018, Kukkonen et al. 2019), and the use of bitemporal ALS 485
metrics may also improve the species discrimination (Räty et al.2019).
486 487
23
A new platform that has been examined extensively in the forest inventory context over the 488
recent years is unmanned aerial vehicles (UAVs) (see, e.g., the reviews by Goodbody et al.
489
2017 and Torresan et al. 2017). In general, UAVs are used to capture images, from which 490
point clouds are then generated by means of stereo matching; UAV-based laser scanning 491
systems have also gained popularity. UAVs are cost-efficient in capturing data from relatively 492
small areas (e.g., <100 ha), while obtaining a representative field sample from the same area 493
at a reasonable cost is a huge challenge. Therefore, the approach used with airborne data, 494
i.e., collecting a new set of sample plots from the inventory area, cannot be used in UAV 495
inventories. Some recent studies have focused on developing alternatives that do not need 496
local field data at all (e.g., Kotivuori et al. 2020, Puliti et al. 2020). UAVs may also be used for 497
local calibrations and detection of changes, such as windthrown trees. Although there is a 498
considerable research effort ongoing on UAV applications in forestry, the operational use of 499
UAV data in FMI has remained rare so far.
500 501
In recent years, a vast amount of research has focused on portable devices, e.g., mobile 502
(MLS) and terrestrial laser scanners (TLS). Though mobile scanners may provide accurate 503
information in situations where up-to-date information is urgently needed for a small area 504
such as single stand or estate (e.g., Gollob et al. 2020, Hyyppä et al. 2020), their usability in 505
operational FMIs is rather limited. This is because they do not provide wall-to-wall 506
information, which would require the forestry professional conducting an operational FMI to 507
visit all stands in the same way as was done in the era of field studies. Moreover, TLS or MLS 508
data do not allow for utilizing model-based inference in the same way as ALS does, which 509
would also require the accuracy assessments to be carried out with separate validation 510
studies, as was done in the field study era.
511
24 512
While portable devices are not viable for FMIs, they can provide a solution for timber sale 513
inventories, where data acquisition costs have restricted the emergence of other remote- 514
sensing applications. The information for a timber sale inventory should be up-to-date, but 515
typically it needs to cover just a very small area. ALS data typically covers considerably larger 516
areas, and ALS-based FMI predictions are up-to-date only during the growing season of the 517
data acquisition year. Different ground-based and mobile-based scanners can also take some 518
tree quality aspects into consideration (e.g., Kretschmer et al. 2013, Pyörälä et al. 2019).
519
Tree quality aspects can also be included in field measurements, field calibrations of existing 520
quality models may be conducted, or harvester information may be utilized (Karjalainen et 521
al. 2019, 2020). In addition, there have been experimental UAV data acquisitions under the 522
canopy, thereby providing critical information to improve tree stem prediction and 523
understory characterization (Hyyppä et al. 2020). The biggest remaining problem, however, 524
is the poor predictability of the quality characteristics: though these characteristics can be 525
measured, they cannot be reliably predicted from ALS data.
526 527
CONCLUDING REMARKS AND RESEARCH NEEDS 528
529
The change from field assessments and visual image interpretation to large-scale automated 530
ALS-based FMIs has already been a huge game changer in many countries. It has also 531
changed the way of thinking from stand-level subjective assessments to large-area objective 532
predictions. The differences between the FMI and the NFI have also diminished. The 533
benefits of an ALS-based management inventory are obvious: accurate stand attribute 534
predictions and improved cost-efficiencies. The wall-to-wall remote-sensing data also 535
25
enables real-time stand-level accuracy assessments by means of a model-based sampling 536
theory. Recent changes in the management inventory have fine-tuned the ALS-based 537
approach, but there is a need for further improvements.
538 539
The main drawback of ALS data has been the inability to discriminate among tree species.
540
But now, multi-spectral ALS data used in combination with novel optical images, for 541
example, has brought improvements, so that the dominant tree species can be predicted 542
correctly in most cases. Still, minor tree species and mixed stands are often predicted 543
inaccurately. This is a problem not only for tree species but also for species-specific stand 544
attributes. Predicted size distributions of mixed stands, for example, may include misleading 545
information for forest management. Thus, further technical and methodological 546
developments are still needed for the practical applications to better distinguish among tree 547
species.
548 549
Recent developments in remote-sensing technologies provide more alternatives for 550
operational FMI applications based on remote sensing than just the use of ALS and aerial 551
image data. Now the inventory can be tailored for timber sales by means of mobile Lidar 552
scanners or field measurements, even at the level of one marked stand, so that it also 553
includes tree quality aspects. On a smaller scale, even biodiversity aspects and natural 554
disturbances can be mapped and monitored by means of the newest technologies. Note, 555
however, that the methods used with the newest technologies are different from the ones 556
traditionally used in ALS-based FMIs, and the costs are substantially higher.
557 558
ACKNOWLEDGEMENTS 559
26 560
This work was supported by the Finnish Flagship Programme of the Academy of Finland for 561
the Forest-Human-Machine Interplay - Building Resilience, Redefining Value Networks and 562
Enabling Meaningful Experiences (UNITE) -project (decision number 337127), led by Prof. Jyrki 563
Kangas at the School of Forest Sciences, University of Eastern Finland.
564 565
REFERENCES 566
Anttila, P. and Lehikoinen, M. 2002. Kuvioittaisten puustotunnusten estimointi ilmakuvilta 567
puoliautomaattisella latvusten segmentoinnilla. Metsätieteen aikakauskirja 3/2002: 381–
568
389. (In Finnish) 569
Astrup, R., Rahlf, J., Bjørkelo, K., Debella-Gilo, M., Gjertsen, A.K. and Breidenbach, J. 2019.
570
Forest information at multiple scales: development, evaluation and application of the 571
Norwegian forest resources map SR16. Scandinavian Journal of Forest Research 34(6):
572
484-496.
573
Axelsson, A., Lindberg E. and Olsson H. 2018. Exploring Multispectral ALS Data for Tree 574
Species Classification. Remote Sensing 10(2): 183. https://doi.org/10.3390/rs10020183 575
Bitterlich, W. 1984. The relascope idea: relative measurements in forestry. Commonwealth 576
Agricultural Bureaux, Slough.
577
Bohlin, J., Wallerman, J. and Fransson, J.E.S. 2012. Forest variable estimation using 578
photogrammetric matching of digital aerial images in combination with a high-resolution 579
DEM, Scandinavian Journal of Forest Research 27(7): 692-699.
580
Breidenbach, J.; McRoberts, R. E., Astrup, R. 2016. Empirical coverage of model-based 581
variance estimators for remote sensing assisted estimation of stand-level timber volume.
582
Remote Sensing of Environment 173: 274-281.
583
Budei, B.C., St-Onge, B., Hopkinson, C. and Audet, F. 2018. Identifying the genus or species of 584
individual trees using a three-wavelength airborne lidar system Remote Sensing of 585
Environment 204(1): 632-647.https://doi.org/10.1016/j.rse.2017.09.037 586
Burkhart, H.E., Stuck, R.D., Leuschner, W.A. and Reynolds, M.A. 1978. Allocating inventory 587
resources for multiple-use planning. Canadian Journal of Forest Research 8: 100-110.
588
Degnan, J.J. 2016. Scanning, Multibeam, Single Photon Lidars for Rapid, Large Scale, High 589
Resolution, Topographic and Bathymetric Mapping. Remote Sensing 8: 958.
590
doi.org/10.3390/rs8110958 591
Ehlers, S., Saarela, S., Lindgren, N., Lindberg, E., Nyström, M., Persson, H.J., Olsson, H. and 592
Ståhl, G. 2018. Assessing Error Correlations in Remote Sensing-Based Estimates of Forest 593
Attributes for Improved Composite Estimation. Remote Sensing 10: 667.
594
Ehlers, S., Grafström, A., Nyström, K., Olsson, H. and Ståhl, G. 2013. Data assimilation in 595
stand-level forest inventories. Canadian Journal of Forest Research: 43: 1104–1113.
596
Fassnacht, F.E., Mangold, D., Schäfer, J., Immitzer, M., Kattenborn, T., Koch, B. and Latifi, H.
597
2017. Estimating stand density, biomass and tree species from very high resolution 598
27
stereo-imagery – towards an all-in-one sensor for forestry applications? Forestry 90(5):
599
613–631, https://doi.org/10.1093/forestry/cpx014 600
Fekety, P.A., Falkowski,M.A., Hudak, A.T., Jain, T.B. and Evans, J.S. 2018. Transferability of 601
Lidar-derived basal area and stem density models within a Northern Idaho Ecoregion.
602
Canadian Journal of Remote sensing 44 (2): 131-143, 603
Fortin, M., Manso, R. and Calama, R. 2016. Hybrid estimation based on mixed-effects models 604
in forest inventories. Canadian Journal of Forest Research 46:1310-1319.
605
Gatziolis, D. 2007. LiDAR-derived site index in the U.S. Pacific Northwest--challenges and 606
opportunities. In: Proceedings of ISPRS Workshop on Laser Scanning 2007 and SilviLaser 607
2007. IAPRS. 36(Part 3/W52): 136-143 608
Gobakken, T. and Næsset, E. 2009. Assessing effects of positioning errors and sample plot 609
size in biophysical stand properties derived from airborne laser scanner data. Canadian 610
Journal of Forest Research 39: 1036-1052 611
Gobakken, T., Bollandsås, O. M., Næsset, E. 2015. Comparing biophysical forest 612
characteristics estimated from photogrammetric matching of aerial images and airborne 613
laser scanning data. Scandinavian Journal of Forest Research 30: 73-86.
614
Gollob, C., Ritter, T. and Nothdurft, A. 2020. Forest Inventory with Long Range and High- 615
Speed Personal Laser Scanning (PLS) and Simultaneous Localization and Mapping (SLAM) 616
Technology. Remote Sensing 12, 1509; doi:10.3390/rs12091509 617
Goodbody, T.R.H., Coops, N.C., Marshall, P.L., Tompalski, P. and Crawford, P. 2017.
618
Unmanned aerial systems for precision forest inventory purposes: A review and case stud.
619
Forestry Chronicle 93(1).
620
Goodbody, T.R.H., Coops, N.C, and White, J.C. 2019. Digital Aerial Photogrammetry for 621
Updating Area-Based Forest Inventories: A Review of Opportunities, Challenges, and 622
Future Directions. Current Forestry Reports 5: 55–75.
623
Gopalakrishnan R., Thomas V., Coulston J.W. and Wynne R. 2015. Prediction of canopy 624
heights over a large region using heterogeneous lidar datasets: efficacy and challenges.
625
Remote Sensing 7: 11036–11060. doi:10.3390/rs70911036 626
Grafström, A., Saarela, S. and Ene, L. 2014. Efficient sampling strategies for forest inventories 627
by spreading the sample in auxiliary space. Canadian Journal of Forest Research 44: 1156–
628
1164.
629
Haara, A. and Leskinen, P. 2009. The assessment of the uncertainty of updated stand-level 630
inventory data. Silva Fennica 43(1): 87−112.
631
Haara, A. and Korhonen, K.T. 2004. Kuvioittaisen arvioinnin luotettavuus. Metsätieteen 632
Aikakauskirja 4/2004: 489-508. (in Finnish).
633
Harris. L3Harris Geiger-mode LiDAR. Available online:
634
https://www.harris.com/solution/geiger-mode-lidar (accessed on 26 June 2020).
635
Hauglin, M., Hansen, E., Sørngård, E., Næsset, E. and Gobakken, T. 2018. Utilizing accurately 636
positioned harvester data: Modelling forest volume with airborne laser scanning.
637
Canadian Journal of Forest Research 48(8): 913–922. https://doi.org/10.1139/cjfr-2017- 638
0467.
639
Holmgren, J. 2004. Prediction of tree height, basal area and stem volume in forest stands 640
using airborne laser scanning. Scandinavian Journal of Forest Research 19(6): 543–553.
641
https://doi.org/10.1080/02827580410019472.
642
Holmgren J., Persson Å. and Söderman, U. 2008. Species identification of individual trees by 643
combining high resolution LiDAR data with multi-spectral images. International Journal of 644
Remote Sensing 29(5): 1537–1552. https://doi.org/10.1080/0143116070173647 645
28
Holmgren, J, Joyce, S, Nilsson, M and Olsson, H. 2000. Estimating stem volume and basal 646
area in forest compartments by combining satellite image data with field data.
647
Scandinavian Journal of Forest Research 15: 103–111.
648
Holmgren, J. and Persson, Å. 2004. Identifying species of individual trees using airborne laser 649
scanner. Remote Sensing of Environment 90: 415–423.
650
Hosseini, Z., Latifi, H., Naghavi, H., Bakhtiari, S.B. and Fassnacht, F. 2020. Influence of plot 651
and sample sizes on aboveground biomass estimations in plantation forests using very 652
high resolution stereo satellite imagery. Forestry.
653
https://doi.org/10.1093/forestry/cpaa028.
654
Hyink D,.M. and Moser, J.W. 1983. A generalized framework for projecting forest yield and 655
stand structure using diameter distributions. Forest Science 29: 85–95.
656
Hyvönen, P. and Korhonen, K.T. 2003. Metsävaratiedon jatkuva ajantasaistus 657
yksityismetsissä. Metsätieteen aikakauskirja 2: 83-96. (In Finnish) 658
Hyyppä, E., Hyyppä, J., Hakala, T., Kukko, A., Wulder, M. A., White, J. C., Pyörälä, J., Yu, X., 659
Wang, Y., Virtanen, J.-P., Pohjavirta, O., Liang, X., Holopainen, M. and Kaartinen H. 2020a.
660
Under-canopy UAV laser scanning for accurate forest field measurements. ISPRS Journal 661
of Photogrammerty and. Remote Sensing 164: 41–60.
662
Hyyppä, E., Kukko, A., Kaijaluoto, R., White, J.C., Wulder, M.A., Pyörälä, J., Liang, X., Yu, X., 663
Wang, Y., Kaartinen, H., Virtanen, JP. and Hyyppä, J. 2020b. Accurate derivation of stem 664
curve and volume using backpack mobile laser scanning, ISPRS Journal of 665
Photogrammetry and Remote Sensing 161:246-262.
666
Ilvessalo, I. 1965. Metsänarvioiminen. WSOY. 400 p. (In Finnish).
667
Jonsson, B and Lindgren, O. 1978. En metod för uppskattning av ett skogsinnehav och för 668
kalibrering avokulärskattade värden. Summary: A method for estimating properties of a 669
forest and for calibrating of ocular estimates. Sveriges Skogsårdförbunds Tidskrift 76:493- 670
505.
671
Kandare, K., Dalponte, M., Ørka, H., Frizzera, L., and Næsset, E. 2017. Prediction of species- 672
specific volume using different inventory approaches by fusing airborne laser scanning 673
and hyperspectral data. Remote Sensing 9(5). http://doi.org/10.3390/rs9050400 674
Kangas, A., Astrup, R., Breidenbach, J., Fridman, J., Gobakken, T., Korhonen, K.T. Maltamo, 675
M., Nilsson, M., Nord-Larsen, T., Næsset, E and Olsson, H. 2018a. Remote sensing and 676
forest inventories in Nordic countries – roadmap for the future. Scandinavian Journal of 677
Forest Research 33:397-412.
678
Kangas, A., Gobakken, T., Puliti, S., Hauglin, M. and Næsset, E. 2018b.Value of airborne laser 679
scanning and digital aerial photogrammetry data in forest decision making. Silva Fennica 680
52(1) article id 9923.
681
Kangas, A., Haara, A., Holopainen, M., Luoma, V., Packalen, P., Packalen, T., Ruotsalainen, R., 682
and Saarinen, N. 2019a. Kaukokartoitukseen perustuvan metsävaratiedon hyötyanalyysi.
683
Luonnonvara- ja biotalouden tutkimus 6/2019. 33 p.
684
Kangas, A., Räty, M., Korhonen, K.T., Vauhkonen, J. and Packalen, T. 2019b. Catering 685
information needs from global to local scales - potential and challenges with national 686
forest inventories. Forests 10, 800.
687
Kankare V., Vauhkonen J., Tanhuanpää T., Holopainen M., Vastaranta M., Joensuu M., 688
Krooks A., Hyyppä J., Hyyppä H., Alho P. and Viitala R. 2014. Accuracy in estimation of 689
timber assortments and stem distribution – A comparison of airborne and terrestrial laser 690
scanning techniques. ISPRS Journal of Photogrammetry and Remote Sensing 97: 89–97.
691
Karjalainen, T., Packalen, P. Räty, J. and Maltamo, M. 2019. Predicting factual sawlog 692
29
volumes in Scots pine dominated forests using airborne laser scanning data. Silva Fennica 693
53 article id 10183. https://doi.org/10.14214/sf.10183 694
Karjalainen, T, Mehtätalo, L., Packalen, P,. Gobakken, T., Næsset, E. and Maltamo, M. 2020.
695
Field calibration of merchantable and sawlog volumes in airborne laser scanning -based 696
forest inventories. Canadian Journal of Forest Research.
697
Kinnunen, J., Maltamo, M and Päivinen, R. 2007. Standing volume estimates of forests in 698
Russia: how accurate is the published data? Forestry 80: 53-64.
699
Kilkki, P., Maltamo, M., Mykkänen, R. and Päivinen, R. 1989. Use of the Weibull function in 700
estimating the basal-area diameter distribution. Silva Fennica 23: 311-318 701
Kim, S., McGaughey, R. J., Andersen, H.-E. and Schreuder, G. 2009. Tree species 702
differentiation using intensity data derived from leaf-on and leaf-off airborne laser 703
scanner data. Remote Sensing of Environment 113(8): 1575–1586.
704
Kivinen, S., Koivisto, E., Keski-Saari, S., Poikolainen, L., Tanhuanpää, T., Kuzmin, A., Viinikka, 705
A., Heikkinen, R.K., Pykälä, J., Virkkala, R., Vihervaara, P. and Kumpula, T.. 2020. A 706
keystone species, European aspen (Populus tremula L.), in boreal forests: Ecological role, 707
knowledge needs and mapping using remote sensing. Forest Ecology and Management 708
462, 118008 709
Koivuniemi, J. and Korhonen, K.T. 2006. Inventory by compartments. In: Kangas, A. and 710
Maltamo, M. (Eds.) Forest Inventory. Methodology and Applications. Managing Forest 711
Ecosystems, vol 10. Springer.
712
Korpela, I., Ørka, H.O., Maltamo, M., Tokola, T. and Hyyppä, J. 2010. Tree species 713
classification using airborne LiDAR –effects of stand and tree parameters, downsizing of 714
training set, intensity normalization, and sensor type. Silva Fennica 44(2): 319–339.
715
https://doi.org/10.14214/sf.156 716
Kotivuori, E. Maltamo, M., Korhonen, L. and Packalen, P. 2018. Calibration of nationwide 717
airborne laser scanning based stem volume models. Remote Sensing of Environment 210:
718
179-192. DOI:10.1016/j.rse.2018.02.069 719
Kotivuori, E. Kukkonen, M. Mehtätalo, L., Maltamo, M., Korhonen, L. and Packalen, P. 2020.
720
Forest inventories for small areas using drone imagery without in-situ field 721
measurements. Remote Sensing of Environment 237, 111404.
722
Kretschmer, U., Kirchner, N., Morhart, C. and Spiecker, H. 2013. A new approach to assessing 723
tree stem quality characteristics using terrestrial laser scans. Silva Fennica 47(5), article-id 724
1071. http://dx.doi.org/10.14214/sf.1071 725
Kukkonen, M., Korhonen, L., Maltamo, M., Suvanto, A. and Packalen, P. 2018. How much can 726
airborne laser scanning based forest inventory by tree species benefit from auxiliary 727
optical data? International Journal of Applied Earth Observations and Geoinformation 72:
728
91-98.
729
Kukkonen, M., Maltamo, M., Korhonen, L. and Packalen, P. 2019. Comparison of 730
multispectral airborne laser scanning and stereo matching of aerial images as a single 731
sensor solution to forest inventories by tree species. Remote Sensing of Environment 231, 732
111208. https://doi.org/10.1016/j.rse.2019.05.027 733
Leica SPL100. Leica SPL100 Single Photon LiDAR Sensor. Available online: https://leica- 734
geosystems.com/en-us/products/airborne-systems/topographic-lidar-sensors/leica- 735
spl100 (accessed on 26 June 2020).
736
Lindberg, E. and Holmgren, J. 2017. Individual Tree Crown Methods for 3D Data from 737
Remote Sensing. Current Forestry Reports 3: 19–3.1 https://doi.org/10.1007/s40725-017- 738
0051-6 739
30
Lindberg, E., Holmgren, J., Olofsson, K., Wallerman, J. and Olsson, H. 2010. Estimation of tree 740
lists from airborne laser scanning by combining single-tree and area-based methods, 741
International Journal of Remote Sensing 31(5): 1175-1192, DOI:
742
10.1080/01431160903380649 743
Lindberg, E., Holmgren, J., Olofsson, K., Wallerman, J. and Olsson, H. 2013. Estimation of 744
Tree Lists from Airborne Laser Scanning Using Tree Model Clustering and k-MSN 745
Imputation. Remote Sensing. 5: 1932-1955.
746
Lindgren, N., Persson, H.J., Nyström, M., Nyström, K., Grafström, A., Muszta, A., Willén, E., 747
Fransson, J.E.S., Ståhl, G. and Olsson, H. 2017. Improved Prediction of Forest Variables 748
Using Data Assimilation of Interferometric Synthetic Aperture Radar Data, Canadian 749
Journal of Remote Sensing 43:4, 374-383, DOI: 10.1080/07038992.2017.1356220 750
Loetsch, F. and Haller and K.E. 1973. Forest Inventory. Volime 1. BLV Verlagsgesellschaft. 436 751
p.
752
Luoma, V., Vastaranta, M., Eyvindson, K., Kankare, V., Saarinen, N., Holopainen M. and 753
Hyyppä, J. 2017. Errors in the Short-Term Forest Resource Information Update. In The 754
Rise of Big Spatial Data pp. 155-166. Springer International Publishing.
755
Magnussen, S.; Frazer, G. and Penner, M. 2016. Alternative mean-squared error estimators 756
for synthetic estimators of domain means. Journal of Applied Statistics 43 (14): 2550- 757
2573.
758
Magnussen, S. and Breidenbach, J. 2017. Model-dependent forest stand-level inference with 759
and without estimates of stand-effects Forestry: An International Journal of Forest 760
Research 90: 675–685.
761
Maltamo, M., Packalén, P., Suvanto, A., Korhonen, K.T., Mehtätalo, L. and Hyvönen, P. 2009.
762
Combining ALS and NFI training data for forest management planning -a Case Study in 763
Kuortane, Western Finland. European Journal of Forest Research 128: 305-317 764
Maltamo, M., Bollandsås, O.M., Næsset, E., Gobakken, T. and Packalén, P. 2011. Different 765
plot selection strategies for field training data in ALS-assisted forest inventory. Forestry 766
84: 23-31 767
Maltamo, M. and Packalen, P. 2014. Species specific management inventory in Finland. In 768
Maltamo, M., Naesset, E. and Vauhkonen, J. (Eds.). Forestry Applications of Airborne 769
Laser Scanning –concepts and case studies. Managing Forest Ecosystems vol 27. Springer, 770
241-252.
771
Maltamo, M., Hauglin, K.M., Næsset, E. and Gobakken, T. 2019. Estimating stand level stem 772
diameter distribution utilizing accurately positioned tree-level harvester data and 773
airborne laser scanning. Silva Fennica 53 article id 10075.
774
https://doi.org/10.14214/sf.10075 775
McRoberts, R.E., Andersen, H.-E. and Næsset, E. 2014. Using Airborne Laser Scanning Data to 776
Support Forest Sample Surveys. In Maltamo, M., Naesset, E. and Vauhkonen, J. (Eds.).
777
Forestry Applications of Airborne Laser Scanning –concepts and case studies. Managing 778
Forest Ecosystems vol 27. Springer, 269-292.
779
McRoberts, R.,E., Næsset, E., Gobakken, T., Chirici, G., Condés, S., Hou, Z. Saarela, S., Chen, 780
Q., Ståhl, G. and Walters, B. 2018. Assessing components of the model-based mean 781
square error estimator for remote sensing assisted forest applications. Canadian Journal 782
of Forest Research 48(6): 642-649.
783
Melo, L., Schneider, R. and Fortin, M. 2018. Estimating model-and sampling-related 784
uncertainty in large-area growth predictions. Ecological Modelling 390: 62-69.
785
Mikhail, E.M., Bethel, J.S. and McGlone, J.C. 2001. Introduction to Modern Photogrammetry, 786
31 John Wiley & Sons. 479 p.
787
Mustonen, J. Packalén, P. and Kangas, A. 2008. Automatic segmentation of forest stands 788
using canopy height model and aerial photograph. Scandinavian Journal of Forest 789
Research 23: 534-545.
790
Næsset, E. 1997. Estimating timber volume of forest stands using airborne laser scanner 791
data. Remote Sensing of Environment 51:246–253.
792
Næsset, E. 2002. Predicting forest stand characteristics with airborne scanning laser using a 793
practical two-stage procedure and field data. Remote Sensing of Environment 80: 88–99 794
Næsset, E. 2004 Practical large-scale forest stand inventory using a small airborne scanning 795
laser. Scandinavian Journal of Forest Research 19: 164−179.
796
Næsset, E. 2005. Assessing sensor effects and effects of leaf-off and leaf-on canopy 797
conditions on biophysical stand properties derived from small-footprint airborne laser 798
data. Remote Sensing of Environment 98: 356–370. doi:10.1016/j.rse.2005.07.012.
799
Næsset, E. 2014. Area-based inventory in Norway – From Innovation to an Operational 800
reality. In Maltamo, M., Naesset, E. and Vauhkonen, J. (Eds.). Forestry Applications of 801
Airborne Laser Scanning –concepts and case studies. Managing Forest Ecosystems vol 27.
802
Springer, 215-240 803
Nelson, R. 2013. How did we get here? An early history of forestry Lidar. Canadian Journal 804
of Remote Sensing 39(S1): 6-17.
805
Nilsson, M., Nordkvist, K., Jonzén, J., Lindgren, N., Axensten, P., Wallerman, J., Egberth, M., 806
Larsson, S., Nilsson, L., Eriksson, J. and Olsson, H. 2017. A nationwide forest attribute map 807
of Sweden predicted using airborne laser scanning data and field data from the National 808
Forest Inventory. Remote Sensing of Environment 194: 447-454.
809
Noordermeer, L., Bollandsås, O. M., Gobakken, T. and Næsset, E. 2018.Direct and indirect 810
site index determination for Norway spruce and Scots pine using bitemporal airborne 811
laser scanner data. Forest Ecology and Management 428: 104-114 812
Noordermeer, L., Økseter,R., Ørka, H. O., Gobakken, T., Næsset, E., and Bollandsås, O. M.
813
2019. Classifications of forest change by using bitemporal airborne laser scanner data.
814
Remote Sensing 11(18), 2145.
815
Noordermeer, L., Gobakken, T., Næsset, E., and Bollandsås, O. M. 2020.Predicting and 816
mapping site index in operational forest inventories using bitemporal airborne laser 817
scanner data. Forest Ecology and Management 457, 117768.
818
Nord-Larsen, T., Riis-Nielsen, T. and Ottosen, M.B. 2017, Forest resource map of Denmark:
819
Mapping of Danish forest resource using ALS from 2014-2015. Department of 820
Geosciences and Natural Resource Management, University of Copenhagen. IGN Report 821
Nyström, M., Lindgren, N., Wallerman, J., Grafström, A., Muszta, A., Nyström, K., Bohlin, J., 822
Willén, E., Fransson, J.E.S., Ehlers, S., Olsson, H. and Ståhl, G. 2013. Data Assimilation in 823
Forest Inventory: First Empirical Results. Forests 6: 4540-4557.
824
Nyyssönen, A. 1955. On the estimation of of the growing stock from aerial photographs.
825
Comminicationes Instituti Forestalis Fenniae 46: 1-57 826
Ørka, H.O., Næsset, E. and Bollandsås, O.M. 2007. Utilizing airborne laser intensity for tree 827
species classification. International Archives of the Photogrammetry, Remote Sensing and 828
Spatial Information Sciences 36: 300–304.
829
Ørka, H. O., Dalponte, M.; Gobakken, T., Næsset, E. and Ene L.T. 2013.
830
Characterizing forest species composition using multiple remote sensing data sources and 831
inventory approaches. Scandinavian Journal of Forest Research 28(7): 677-688 832
Packalén, P. and Maltamo, M. 2007. The k-MSN method in the prediction of species specific 833