sensors Article
Mapping Vineyard Leaf Area Using Mobile Terrestrial Laser Scanners: Should Rows be Scanned On-the-Go or Discontinuously Sampled? Ignacio del-Moral-Martínez 1, *, Joan R. Rosell-Polo 1,3,† , Joaquim Company 1,† , Ricardo Sanz 1,3,† , Alexandre Escolà 1,3,† , Joan Masip 1,† , José A. Martínez-Casasnovas 2,3,† and Jaume Arnó 1,3,† Received: 7 December 2015; Accepted: 15 January 2016; Published: 19 January 2016 Academic Editor: Simon X. Yang 1
2 3
* †
Research Group on AgroICT & Precision Agriculture, Department of Agricultural and Forest Engineering, University of Lleida, Rovira Roure 191, Lleida 25198, Spain;
[email protected] (J.R.R.-P.);
[email protected] (J.C.);
[email protected] (R.S.);
[email protected] (A.E.);
[email protected] (J.M.);
[email protected] (J.A.) Research Group on AgroICT & Precision Agriculture, Department of Environmental and Soil Sciences, University of Lleida, Rovira Roure 191, Lleida 25198, Spain;
[email protected] Research Group on AgroICT & Precision Agriculture, Agrotecnio Center, Rovira Roure 191, Lleida 25198, Spain Correspondence:
[email protected]; Tel.: +34-973-702-861 These authors contributed equally to this work.
Abstract: The leaf area index (LAI) is defined as the one-side leaf area per unit ground area, and is probably the most widely used index to characterize grapevine vigor. However, LAI varies spatially within vineyard plots. Mapping and quantifying this variability is very important for improving management decisions and agricultural practices. In this study, a mobile terrestrial laser scanner (MTLS) was used to map the LAI of a vineyard, and then to examine how different scanning methods (on-the-go or discontinuous systematic sampling) may affect the reliability of the resulting raster maps. The use of the MTLS allows calculating the enveloping vegetative area of the canopy, which is the sum of the leaf wall areas for both sides of the row (excluding gaps) and the projected upper area. Obtaining the enveloping areas requires scanning from both sides one meter length section along the row at each systematic sampling point. By converting the enveloping areas into LAI values, a raster map of the latter can be obtained by spatial interpolation (kriging). However, the user can opt for scanning on-the-go in a continuous way and compute 1-m LAI values along the rows, or instead, perform the scanning at discontinuous systematic sampling within the plot. An analysis of correlation between maps indicated that MTLS can be used discontinuously in specific sampling sections separated by up to 15 m along the rows. This capability significantly reduces the amount of data to be acquired at field level, the data storage capacity and the processing power of computers. Keywords: LAI; LiDAR; mobile terrestrial laser scanner; precision viticulture; vegetation maps
1. Introduction Light detection and ranging (LiDAR) is a technology that is becoming widely used by researchers to characterize vineyards and other crops. LiDAR sensors can be airborne or terrestrial and measurements can be performed from stationary or mobile platforms. Different geometric parameters can be measured such as vegetation height, cross-sectional area, canopy volume [1–6], and even trunk volume [7]. Other parameters measured with terrestrial laser sensors are the tree area index [5] and the leaf wall area [8]. The accuracy and high density of the generated point clouds make LiDAR
Sensors 2016, 16, 119; doi:10.3390/s16010119
www.mdpi.com/journal/sensors
Sensors 2016, 16, 119
2 of 13
make LiDAR information a reliable way to capture detailed geometric characteristics of the canopy structure and analyze plant response to different inputs and conditions [9]. Leaf area and as Sensors 2016, 16, 119 leaf density have been related to multiple agricultural production factors 2 ofsuch 13 solar radiation [10,11], irrigation or water deficit [12,13], fertilization [14], application of plant protection products (PPP) [15,16], and yield and quality of fruits [17,18]. Studies linking LiDAR information a reliable way to capture detailed geometric characteristics of the canopy structure and point clouds or other information provided by remote sensing with the leaf area index (LAI) can be analyze plant response to different inputs and conditions [9]. found [5,6,19–21], although many of these initial applications were developed for forestry and with Leaf area and leaf density have been related to multiple agricultural production factors such airborne LiDAR sensors [22,23]. as solar radiation [10,11], irrigation or water deficit [12,13], fertilization [14], application of plant Among research viticulture, the spatiotemporal protection productsmore (PPP)specifically [15,16], andabout yield precision and quality of fruits [17,18]. Studies linkingvariability LiDAR of vineyards has been widely studied [24], often usingsensing remote sensing [25,26] point clouds or other information provided by remote with the leaftechniques area index (LAI) canto be map foundleaf [5,6,19–21], although many of very these little initialinformation applications were developed for forestry and with the vineyard area [27,28]. However, has been published [29] regarding airborne LiDAR sensors [22,23]. mapping of vineyard canopies using ground-based LiDAR sensors, and more importantly, the difficulty Among research more specifically about maps precision viticulture, the spatiotemporal variabilityinofthese of the use and interpretation of the proposed makes managing the spatial variability vineyards has been widely studied [24], often using remote sensing techniques [25,26] to map vineyard cases a remaining challenge. leaf area [27,28]. However, very little information has been published [29] regarding the mapping of The main objective of this study was to develop a methodology that allows the use of mobile vineyard canopies using ground-based LiDAR sensors, and more importantly, the difficulty of the terrestrial laser scanners (MTLS) in viticulture to estimate and map the LAI. The laser scanner used use and interpretation of the proposed maps makes managing the spatial variability in these cases a in this work has been previously designed and validated for use in tree crops [8]. There are two remaining challenge. issues covered in objective this study. Thestudy first is thetomethod to process the data. Theofsecond The main of this was develop adopted a methodology that allows the use mobile is to optimize the use MTLS in field in conditions replacing the on-the-go along the rows terrestrial laserofscanners (MTLS) viticultureby to estimate and map the LAI. scanning The laser scanner used in by discontinuous, and more affordable, scanning at only certain points. this work has been previously designed and validated for use sampling in tree crops [8]. There are two issues covered in this study. The first is the method adopted to process the data. The second is to optimize the 2. Material and Methods use of MTLS in field conditions by replacing the on-the-go scanning along the rows by discontinuous, and more affordable, scanning at only certain sampling points.
2.1. Field Trial
2. Material and Methods
A vineyard plot with contrasting vigor was chosen for the field test (Figure 1). The vineyard Field in Trial was 2.1. located Raimat (Catalonia, Spain), and was planted in 2002 with the cultivar Vitis vinifera L.
cv. Syrah. The six plot rowswith of contrasting the plot where the chosen trial was conducted north-south, A vineyard vigor was for the field test were (Figureoriented 1). The vineyard was located Raimatof (Catalonia, Spain), and was 0.70 planted 2002vines with the cultivar Vitisin vinifera L. cv. covering a totalinlength 360 m (approximately ha).inThe were trained vertical shoot Syrah. The six rows of the plot where the trial was conducted were oriented north-south, covering position, and were watered by partial rootzone drying. The test was performed at growth stage 79 a totalof length of 360 m (approximately ha). Biologische The vines were trained in vertical shoot position, und (majority berries touching) according0.70 to the Bundesanstalt, Bundessortenamt and were watered by partial rootzone drying. The test was performed at growth stage 79 (majority CHemische Industrie (BBCH)-Scale [30]. The vines were less vigorous at the northern, highest part of berries touching) according to the Biologische Bundesanstalt, Bundessortenamt und CHemische of the plot, and more vigorous at the southern area at lower elevation. Given the highly structured Industrie (BBCH)-Scale [30]. The vines were less vigorous at the northern, highest part of the plot, and spatial distribution of vigor, the plot was very suitable to assess MTLS for detecting the variability of more vigorous at the southern area at lower elevation. Given the highly structured spatial distribution the vineyard. Soils these were classified asfor Typic Haploxerepts and of Fluventic Haploxerepts of vigor, the plotin was veryfields suitable to assess MTLS detecting the variability the vineyard. Soils [31]. in The Typic Haploxerepts, located in Haploxerepts the highest part the plot, presented a[31]. paralithic contact these fields were classified as Typic and of Fluventic Haploxerepts The Typic within the first 50located cm, which bepart theof main reason for theadifferences in vigor detected in the Haploxerepts, in thecould highest the plot, presented paralithic contact within the first plot.50 cm, which could be the main reason for the differences in vigor detected in the plot.
Figure 1. Plot ofofVitis Syrah and location thescanned six rows with the Figure 1. Plot Vitis vinifera vinifera L.L.cv.cv. Syrah (left),(left), and location of the sixofrows withscanned the terrestrial terrestrial laser scanner (right) [32]. The length of the vine rows is 360 m. laser scanner (right) [32]. The length of the vine rows is 360 m.
Sensors 2016, 16, 119
2.2. Mobile Terrestrial Laser Scanner Sensors 2016, 16, 119
3 of 13
3 of 13
The sensor used in this research was described in detail in [8]. In fact, this work only involves 2.2. Mobile Terrestrial Laser Scanner the field application of the abovementioned system, comprising the LMS-200 (SICK AG, Waldkirch, sensor used this research was described kinematic in detail in global [8]. In fact, this worksystem only involves Germany)The to acquire datainon vegetation, a real-time positioning (RTK-GPS) the field application of the abovementioned system, comprising the LMS-200 (SICK AG, Waldkirch, to georeference these data, and an inertial measurement unit (IMU) to correct deviations due to Germany) to acquire data on vegetation, a real-time kinematic global positioning system (RTK-GPS) movements of the sensor. The latter was an MTi sensor (XSens, Ans Enschede, The Netherlands) to georeference these data, and an inertial measurement unit (IMU) to correct deviations due to and provided current roll and pitch rotation based on the plumb line and the earth’s magnetic field. movements of the sensor. The latter was an MTi sensor (XSens, Ans Enschede, The Netherlands) and Position and yaw of the MTLS were estimated using RTK-GPS information provided by a GNSS 1200 provided current roll and pitch rotation based on the plumb line and the earth’s magnetic field. (LeicaPosition Geosystems AG, Switzerland). All three devices supplied provided information different and yaw of St. theGallen, MTLS were estimated using RTK-GPS information by aat GNSS output rates. In our case, RTK-GPS at 5 Hz, MTLS at 15 Hz, and the IMU worked 1200 (Leica Geosystems AG, St. worked Gallen, Switzerland). Allworked three devices supplied information at at 100 Hz. All information timestamped so data could synchronized, then wasthe transferred different output rates.was In our case, RTK-GPS worked at 5beHz, MTLS workedand at 15 Hz,itand IMU at 100 Hz. All information so dataAcould be synchronized, and then it was to anworked on-board central computer viawas thetimestamped RS-232 protocol. MATLAB-based program was used to transferred to an on-board central computer via the RS-232 protocol. A MATLAB-based program control the sensors and acquire data. was the usedMTLS to control the sensorstogether and acquire data. All components with a power supply unit were mounted in a structure All the MTLS components together with a power supply unit were mounted in a structure connected to the three-point hitch of a tractor (Figure 2). The choice of using a tractor allows to better connected to the three-point hitch of a tractor (Figure 2). The choice of using a tractor allows to better drive around the vineyard and to easily choose constant travel speed. In the present work, the travel drive around the vineyard and´to easily choose constant travel speed. In the present work, the travel speedspeed was was set in about 3.6 km¨ h 1 . set in about 3.6 km·h−1.
Figure 2. Image of the different sensors (A); interaction of the MTLS components (B); and connection
Figure 2. Image of the different sensors (A); interaction of the MTLS components (B); and connection to the three-point hitch of the tractor (C). to the three-point hitch of the tractor (C).
A basic feature of the MTLS used was a LiDAR sensor emitting a laser beam used to measure thebasic distance between theMTLS objectsused and the sensor, basedsensor on the emitting time-of-flight principle This A feature of the was a LiDAR a laser beam[33]. used to time measure of flight is the travel time of the laser beam to go from and return to the LiDAR sensor after being the distance between the objects and the sensor, based on the time-of-flight principle [33]. This time reflected the object of the interest. two-dimensional scansensor was obtained of flight is theby travel time of laserSpecifically, beam to goa from and returnfan-shaped to the LiDAR after being (sweep angle of 180°) since the laser beam was pulsed with an angular resolution of 1°. In this way, reflected by the object of interest. Specifically, a two-dimensional fan-shaped scan was obtained (sweep the scanner provided up to 181 distances (points of interception) in each scan. When scanning a row, angle of 180˝ ) since the laser beam was pulsed with an angular resolution of 1˝ . In this way, the the scanner field of view was oriented towards the right hand side according to the forward scanner provided 181side distances (points in each scan.2).When a row, the direction, and up onlytoone of the row couldofbeinterception) scanned at a time (Figure Thus, scanning each vegetation scanner of view oriented the hand side according todata the was forward direction, rowfield needed to be was scanned from towards both sides. In right a subsequent step, the sensor corrected with and only one side of theand rowUTM couldcoordinates be scannedwere at a time (Figure 2). Thus, each vegetation row needed the IMU data assigned to each interception point according to the to be scanned from bothused sides. subsequent step, the sensor data was withwere the IMU data and methodology in In [8].a Once the point cloud was created, 1-m corrected long sections delimited including points obtained from both sides of the row (Figure 3). Then, a LAI estimation algorithm UTM coordinates were assigned to each interception point according to the methodology used in [8]. as described in the next Oncewas the applied point cloud was created, 1-msection. long sections were delimited including points obtained from both sides of the row (Figure 3). Then, a LAI estimation algorithm was applied as described in the 2.3. LAI Estimation next section. A pixelated area (Sj) was assigned to each of the interception points (Figure 4), calculated as the product of the vertical distance to the previous point and the horizontal distance to the previous scan [8]. 2.3. LAI Estimation The sum of the pixelated areas in each scan was finally assigned to a representative point whose
A pixelated area (Sj ) was assigned to each of the interception points (Figure 4), calculated as the product of the vertical distance to the previous point and the horizontal distance to the previous scan [8]. The sum of the pixelated areas in each scan was finally assigned to a representative point
Sensors 2016, 16, 119 Sensors 2016, 16, 119 2016, 16, 119were whoseSensors coordinates
4 of 13 4 of 13 4 of 13 was the average of the coordinates of the interception points. This procedure
coordinates were the average of the coordinates of the interception points. This procedure was repeated for each of the scans included ininaa1-m section(Figure (Figure 3). repeated for were each of the scans included 1-mlength length section coordinates the average of the coordinates of the interception3). points. This procedure was repeated for each of the scans included in a 1-m length section (Figure 3).
Figure 3. 1-m long section of the row including scans obtained from both sides of the row. (A) and (B)
Figure 3. 1-m long section of the row including scans obtained from both sides of the row. (A) and are views of the same scene different perspective. The LAI estimation is assigned to the Figure 3. 1-m long section offrom the row obtained from of the row. (A)sampling and (B) (B) are views of the same scene from including different scans perspective. The both LAIsides estimation is assigned to the point. are views of the same scene from different perspective. The LAI estimation is assigned to the sampling sampling point. point.
Figure 4. Intercepted points generated by consecutive scans and specific pixelated area (Sj) assigned to one of points generated (Pj). Figure 4. interception Intercepted points by consecutive scans and specific pixelated area (Sj) assigned
Figure 4. Intercepted points generated by consecutive scans and specific pixelated area (Sj) assigned to to one of interception points (Pj). one ofAs interception points a result, the total (Pj). leaf wall area from one side of the row was computed as the sum of the
pixelated each scan and georeferenced at the coordinates of the scans in sum 1-m length As a areas result,ofthe total leaf wall area from one sideaverage of the row was computed as the of the section. The same was the opposite of the row. Thewas ultimate goal wasas tothe obtain the pixelated areas eachrepeated scan wall andfor georeferenced atside the average coordinates of the scans in 1-m length As a result, theoftotal leaf area from one side of the row computed sum of the so-called enveloping vegetative areathe of opposite the canopy, consisting ofThe the ultimate sum of the two leaf wall areas section. The same was repeated for side of the row. goal was to obtain the pixelated areas of each scan and georeferenced at the average coordinates of the scans in 1-m length (excluding gaps) andvegetative the area area that encloses the top of the ofrow. The of topthearea computed so-called enveloping the canopy, the sum twowas leaf wall areas section. The same was repeated for theof opposite sideconsisting of the row. The ultimate goal was to obtain the considering the distance pointsthe where thethe right were (excluding gaps) and the between area thatthe encloses top of row.and Theleft topleaf areawall wasareas computed so-called enveloping vegetative area ofofthe canopy, consisting of the sumcanopy of the development two leaf wallinareas georeferenced. trainingbetween system plot where allows for important considering theThe distance thethe points the an right and left leaf wall areas were (excluding gaps) and the area that encloses the top of the row. The top area was computed considering width. This parameter should be taken representscanopy a largedevelopment leaf area. The georeferenced. The training system of theinto plotaccount allows since for anit important in the distance between the points where the right and left leaf wall areas were georeferenced. inclusion of row width data is the main difference to previous work in this [8].area. Figure 5 The width. This parameter should be taken into account since it represents a respect large leaf The training system of thewidth plot allows for main an important canopy development in respect width. This parameter shows theof different stepsdata of the LAI was then using following expression: inclusion row is process. the difference tocalculated previous work the in this [8]. Figure 5 should be taken into account since it represents a large area. The inclusion of row width data is the shows the different steps of the process. LAI was thenleaf calculated using the following expression:
Sensors 2016, 16, 119
5 of 13
main difference to previous work in this respect [8]. Figure 5 shows the different steps of the process. LAI was then calculated using the following expression: Sensors 2016, 16, 119
5 of 13
pS ` SR ` ST q ¨ C LAI “ L ( S L dSrR ˆ SLT ) C LAI
dr L
(1) (1)
where S L (m2 ) and SR (m2 ) were the left and right leaf wall areas, respectively, ST (m2 ) was the top SL (m2) and Swas R (m2the ) were the right leaf wall areas, respectively, (m2) was the top area, where C (dimensionless) ratio ofleft theand foliar surface to the enveloping area,STwhich is equal to 1.70 area, C (dimensionless) was the ratio of the foliar surface to the enveloping area, which is equal to in the vineyard (“unpublished report”), dr was the row spacing (m), and L (m) was the section length 1.70 in the vineyard (“unpublished report”), dr was the row spacing (m), and L (m) was the section (1 m in the present case). Since the areas were calculated for 1-m length sections, the final result was the length (1 m in the2present case). Since the areas were calculated for 1-m length sections, the final corresponding LAI (m ¨ m´2 ). The georeferenced LAI of this 1-m length section is called sampling point result was the corresponding LAI (m2·m−2). The georeferenced LAI of this 1-m length section is called in thesampling current point research work (Figure 3). This process allowed georeferenced LAI georeferenced values to be generated in the current research work (Figure 3). This process allowed LAI according to 1 m equidistant points placed along the line of grapevine trunks. values to be generated according to 1 m equidistant points placed along the line of grapevine trunks.
Figure 5. Diagram of data processing to compute the envelope vegetative area of the 1-m length section.
Figure 5. Diagram of data processing to compute the envelope vegetative area of the 1-m length section.
Sensors 2016, 16, 119
6 of 13
Sensors 2016, 16, 119
6 of 13
Sensors 2016, 16, 119 Clearly, point clouds
6 of 13 contained points on leaves, trunks and branches. However, no classification
Clearly, point clouds contained points on leaves, trunks and branches. However, no classification is performed because the point clouds only had position information (no intensity or color is provided Clearly, point clouds pointsonly on leaves, trunks and branches. However, no classification is performed because thecontained point clouds had position information (no intensity or color is by the system). Nevertheless, the C parameter used ininformation Equation (1)Equation has been(1) obtained taking is MTLS performed because point clouds only had intensity orhas color is provided by the MTLSthe system). Nevertheless, theposition C parameter used in(no been into account the effect of trunk and branches. providedtaking by theinto MTLS system). Nevertheless, thebranches. C parameter used in Equation (1) has been obtained account the effect of trunk and To validate theinto method forthe estimating the LAI, threevineyard vineyardblocks blocks different contrasting obtained taking account effect of trunk To validate the method for estimating theand LAI,branches. three of of different contrasting vigorvigor were selected within thethe field. Each row length 2of m (distance between To validate thewithin method for estimating the covered LAI, three vineyard blocks different contrasting were selected field. Eachblock block covered aarow length of of 2m (distance between two two vigor were selected within thescanned field. Each block covered row length of 2scanning, mscanning, (distance two consecutive trunks), and was onboth both sides with MTLS. After thebetween threethree blocks consecutive trunks), and was scanned on sides withathe the MTLS. After the blocks consecutive trunks), and was scanned on both sides with the MTLS. After scanning, the three blocks were manually defoliated to measure the actual values of LAI to compare with the values provided were manually defoliated to measure the actual values of LAI to compare with the values provided by were defoliated to measure of6.LAI by themanually electronic measurement. Results areactual shown in Figure 6. to compare with the values provided the electronic measurement. Results arethe shown invalues Figure by the electronic measurement. Results are shown in Figure 6.
Figure 6. Measured and estimated differentcontrasting contrasting vineyard blocks. Figure 6. Measured and estimatedLAI LAI in in three three different vineyard blocks. Figure 6. Measured and estimated LAI in three different contrasting vineyard blocks.
The MTLS performed good estimates of the LAI for intermediate and high vigor vines, with an The MTLS performed good estimates of the LAI for intermediate and high vigor vines, with an MTLS4% performed estimates of the LAI for intermediate withdue an errorThe between and 16%.good However, in vines with reduced vigor, the and errorhigh wasvigor high,vines, probably error between 4% and 16%. However, in vines with reduced vigor, the error was high, probably due to error between 4% and 16%. However, in vines with reduced vigor, the error was high, probably due to the horizontal scanning resolution used [33] (a scan every 7 cm along the row). This distance the horizontal scanning resolution used [33] (a[33] scan every 7leaf cmdevelopment, along the row). This distance between to the horizontal scanning resolution (a scan 7 cm along the row). This distance between scans may have been too longused in vines with poorevery computing as effective scansleaf may have been in too vines with poorwith leaf poor development, computing as effective leaf wall between scans maytoo have long vines leaf development, computing effective wall area zones in long thebeen canopy with ainconsiderable percentage of gaps. This should beasimproved area zones in the canopy with a considerable percentage of gaps. This should be improved in future leaf wall area zones canopy with a considerable percentage of gaps. This should be improved in future versions of in thethe MTLS. versions of theversions MTLS.of the MTLS. in future FIELD DATA (X, Y,DATA LAI) FIELD (X, Y, LAI)
Reduction of sampling points Reduction of sampling points Comparison of classified mapsoffor Comparison different classified sampling maps for different sampling
VESPER VESPER
LAI RASTER MAP (Kriging in 5 m blocks by using a regular 2 m grid) LAI RASTER MAP (Kriging in 5 m blocks by using a regular 2 m grid)
Interpolated values of LAI Interpolated values of LAI MZA MZA Cluster analysis (Fuzzy c-means algorithm) Cluster analysis (Fuzzy c-means algorithm)
Optimized use of MTLSuse of Optimized MTLS
LAI CLASSIFIED MAPS (Maps two or three classes) LAIofCLASSIFIED MAPS (Maps of two or three classes) Figure 7. Flowchart for obtaining and comparing LAI raster maps. Figure 7. Flowchart for obtaining and comparing LAI raster maps.
Figure 7. Flowchart for obtaining and comparing LAI raster maps. Finally, a LAI raster map was created using the previously validated data and applying a Finally, amethod LAI raster mapinterpolation was created(Figure using the previously validated dataused andfor applying geostatistical of spatial 7). VESPER software [34] was ordinarya Finally, a LAI raster created using previously validated applying a geostatistical method of map spatialwas interpolation (Figurethe 7). VESPER software [34] wasdata used and for ordinary
geostatistical method of spatial interpolation (Figure 7). VESPER software [34] was used for ordinary kriging and ArcGIS 10.1 (ESRI, Redlands, CA, USA) to import and map the interpolated values.
Sensors 2016, 16, 119 Sensors 2016, 16, 119
7 of 13 7 of 13
kriging and ArcGISon 10.1 Redlands, CA, USA) to import and map the interpolated values. Specifically, a kriging a 5(ESRI, m block basis was used having previously adjusted the global variogram Specifically, kriging on a (spherical 5 m blockmodel). basis was used having previously adjusted the global model that bestafitted the data variogram that best classification fitted the dataalgorithm (spherical (fuzzy model).c-means) was applied to the interpolated Next, anmodel unsupervised Next, an unsupervised classification algorithm (fuzzy classified c-means) was the interpolated data in order to cluster the LAI values and then generate LAIapplied maps. to Management Zone data in(MZA) order [35] to cluster the LAI andthis then generateAccording classified LAI maps. Management Analyst software wasvalues used for purpose. to previous research onZone grape Analyst (MZA) [35] software purposes was used[36], for this According to previous research on grape quality for selective harvesting the purpose. winery decided to work with only two vigor classes quality for selective harvesting purposes [36], the winery decided to work with only two vigor to produce two different wine qualities. Additionally, they also apply plant protection products classes to produce two different wine qualities. Additionally, they also apply plant protection variably based on the use of two or three different doses as much. They still use conventional sprayers, products variably based on the use of two or three different doses as much. They still use and more than two or three doses (classes) within the same plot would be difficult to handle. Hence, conventional sprayers, and more than two or three doses (classes) within the same plot would be we opted for LAI classification in two or three classes, although the sensor allows working with higher difficult to handle. Hence, we opted for LAI classification in two or three classes, although the sensor resolution to obtain a larger number of classes. allows working with higher resolution to obtain a larger number of classes.
3. Results and Discussion 3. Results and Discussion Figure 8 shows the LAI raster map. As expected, the spatial variability was remarkable (coefficient Figure 8 shows the LAI raster map. As expected, the spatial variability was remarkable of variation of 24%), but more interesting was that the variation in LAI was highly structured, allowing (coefficient of variation of 24%), but more interesting was that the variation in LAI was highly thestructured, delimitation of well-defined and compact areas within field. Adopting thethe approach suggested allowing the delimitation of well-defined andthe compact areas within field. Adopting by the [35], two additional maps weretwo obtained based on were the classification of on LAIthe into two and three approach suggested by [35], additional maps obtained based classification of classes of vigor (Figure 8). LAI into two and three classes of vigor (Figure 8).
(a)
(b)
(c)
Figure 8. (a) Sampling points, (b) LAI raster (interpolated) map and (c) LAI classified maps (2 classes Figure 8. (a) Sampling points, (b) LAI raster (interpolated) map and (c) LAI classified maps (2 classes and 3 classes) in an area of 0.70 ha within a plot in a vineyard. and 3 classes) in an area of 0.70 ha within a plot in a vineyard.
From the perspective of site-specific vineyard management, LAI classified maps provide a clear From the perspective of site-specific vineyard and, management, LAI areas classified mapspotentially provide a clear view of areas with different canopy development consequently, that could be view of areas with different canopy development and, consequently, areas that could potentially managed differently. To mention one example, LAI maps can be used in the variable-ratebe managed differently. To mention one can be used in the variable-rate application application of pesticides to adjust theexample, dosage toLAI themaps canopy characteristics. However, it is the difficult totoextensively this technology in commercial fields. The time of pesticides to adjust dosage the canopyuse characteristics. required to measure a whole plot anduse thethis large amount of data generated areThe the time two required major However, it is difficult to extensively technology in commercial fields. to measure a whole plot and the large amount of data generated are the two major drawbacks of using
Sensors 2016, 16, 119 Sensors 2016, 16, 119
8 of 13 8 of 13
drawbacks of using MTLS in agriculture onespecially a commercial especially when continuously MTLS in agriculture on a commercial basis, whenbasis, continuously scanning the rows of the scanning the rows the plot.toTherefore, alternative to is the of on-the-goscan MTLS to plot. Therefore, an of alternative the use ofan on-the-go MTLS to use discontinuously theiscanopy discontinuously scan thesampling canopy according to a specific sampling strategy thestudy vine are rows. The in according to a specific strategy along the vine rows. The resultsalong of this shown results of this study are shown in the following section. the following section.
SamplingOptimization OptimizationofofMTLS MTLSMeasurements Measurements Sampling Whenthe theMTLS MTLSwas wasused usedon-the-go, on-the-go, that scanning continuously along rows, a total When that is,is, scanning continuously along thethe rows, a total numberofof2146 2146LAI LAI sampling sampling points points were were obtained obtained (Figure 8). This number This method method took took approximately approximately 2 h 2and h and generated 1 of GBinformation of information per hectare, considering thatscanned the scanned each generated 1 GB per hectare, considering that the lengthlength at eachatsampling sampling point point was 1 m.was 1 m. To operation of of MTLS MTLSwas wassimulated. simulated.InInthis Toreduce reducethe the acquired acquired information, information, the discontinuous operation this mode, the scanner would only be used every few meters, depending on the distance between mode, the scanner would only be used every few meters, depending on the distance between sampling sampling A graphical scheme of different operating is in shown in 9. Figure 9. points. Apoints. graphical scheme of different operating modes ismodes shown Figure
Figure 9. Different modes of operation of a MTLS along a row of vines. The rectangular area Figure 9. Different modes of operation of a MTLS along a row of vines. The rectangular area indicates indicates the scanned sampling sections. (A) On-the-go scanning mode; (B) discontinuous scanning the scanned sampling sections. (A) On-the-go scanning mode; (B) discontinuous scanning mode mode separating the sampling sections 1 m. separating the sampling sections 1 m.
Sensors 2016, 16, 119
Sensors 2016, 16, 119
9 of 13
9 of 13
When the vineyard was scanned on-the-go (Figure 9A), the sampling points were located one after the When other. the Thevineyard acquired data couldon-the-go be reduced by 50% the sampling sections was scanned (Figure 9A),by theseparating sampling points were located one by a distance equal toThe the acquired scanneddata length (Figure 9B). Further in the input information would after the other. could be reduced by 50%reductions by separating sampling sections by a be achieved by equal increasing distance between sections. However, is probably limit distance to the the scanned length (Figure sampling 9B). Further reductions in input there information wouldabe achieved by increasing distance sampling sections.can However, there is without probablyaffecting a limit to the to this procedure, and thethe question isbetween how much this distance be increased thisof procedure, and the question is how much distance can be increased withoutsections affectingand the the quality the generated maps. Starting with thethis original data (Figure 9), sampling quality of the generated maps.were Starting with the original data (Figure 9), sampling sections and the corresponding sampling points progressively removed, simulating the intermittent operation corresponding sampling points were progressively removed, depending simulating the operation of the MTLS. A total of 10 different patterns were simulated on intermittent the separation distance: of the MTLS. A total of 10 different patterns were simulated depending on the separation distance: 1, 2, 3, 4, 5, 10, 15, 20, 25 and 30 m. Some of these distances are shown in Figure 9. For each of 1, these 2, 3, 4, 5, 10, 15, 20, 25 and 30 m. Some of these distances are shown in Figure 9. For each of these sampling point sets, a LAI raster map was obtained following the methodology described in Figure 7. sampling point sets, a LAI raster map was obtained following the methodology described in Figure 7. Then, the resulting maps were classified into two or three vigor classes. The final step was to assess the Then, the resulting maps were classified into two or three vigor classes. The final step was to assess correlation between the original classified maps (on-the-go sampling) and the classified maps resulting the correlation between the original classified maps (on-the-go sampling) and the classified maps fromresulting the use of fewer dataof(discontinuous sampling sampling along thealong rows). from the use fewer data (discontinuous the rows).
Figure 10. Sampling points, LAI raster maps, and LAI classified maps for different MTLS sampling
Figure 10. Sampling points, LAI raster maps, and LAI classified maps for different MTLS sampling schemes along the rows. The sampling sections are separated 1 m (A); 10 m (B); 15 m (C); and 20 m (D). schemes along the rows. The sampling sections are separated 1 m (A); 10 m (B); 15 m (C); and 20 m (D).
Sensors 2016, 16, 119
10 of 13
Figure 10 shows the maps obtained after reducing the number of sampling sections. In particular, the sampling schemes corresponded to distances of 1, 10, 15, and 20 m between scanned sections. The LAI raster maps were quite similar to the original map (on-the-go sampling). However, differences were more evident when comparing the LAI classified maps. Visual analysis of Figure 10 shows that the discrepancy between classified maps obtained by discontinuous sampling and the equivalent maps obtained with continuous sampling (Figure 8) increased with decreasing sampling points, that is, when scanned vines were separated by a greater distance. This result is expected and to some extent justifiable, but it is more important to quantify the degree of discordance as the scanned data are omitted. Comparing the initial LAI classified maps (Figure 8) with the corresponding maps obtained with a discontinuous sampling (Figure 10) was performed using the Kappa coefficient (k): k“
Po ´ Pe 1 ´ Pe
(2)
where Po is the observed class concordance between pixels from different maps, and Pe is the expected agreement. Table 1 shows the results from the comparison between maps (analysis of the degree of concordance). The number of sampling points was drastically reduced when increasing distance between sampling sections. However, good agreement with the original classified maps was achieved when the distance between sampling sections did not exceed 15 m. Both LAI classified maps (two and three classes) had a Kappa coefficient above or close to 0.8. The best results were produced with higher sampling point densities (distances between sampling sections shorter than 5 m), but probably the best option is to keep a distance between 10 and 15 m not to saturate the data acquisition system. This implies obtaining 278 and 192 sampling points per hectare, respectively, so reliable raster maps could then be obtained by ordinary kriging. In contrast, separating sampling sections by a greater distance (20 m or more) is not recommended when using MTLS. Table 1. Analysis of the correlation between the LAI classified maps obtained by on-the-go scanning (with a total of 2146 sampling points, N) and those obtained discontinuously scanning according to various sampling schemes. Distance between Sampling Sections (m)
Number of Sampling Points
Kappa Coefficient 2-Class Vigor Maps
Kappa Coefficient 3-Class Vigor Maps
1 2 3 4 5 10 15 20 25 30
N/2 = 1073 N/3 = 715 N/4 = 537 N/5 = 429 N/6 = 358 N/11 = 195 N/16 = 135 N/21 = 103 N/26 = 83 N/31 = 70
0.95 0.95 0.75 0.93 0.82 0.73 0.90 0.75 0.75 0.54
0.91 0.88 0.89 0.70 0.85 0.80 0.75 0.56 0.50 0.59
Another case of use could be the study over the time of the vineyard vigor (multiple-scans scenario). A first complete measurement could be done at the beginning of the season. After that, and once established the distance sampling better adapted to the data, the optimal sampling could be applied in following measurements. Indeed, in some cases it may be advisable to vary the sampling strategy for different areas within the plot. 4. Conclusions MTLS can be used in viticulture to obtain LAI maps. However, on-the-go sampling generates a large amount of data that can be difficult to process and store. One solution to this problem is the
Sensors 2016, 16, 119
11 of 13
discontinuous use of MTLS based on scanning specific sampling sections. Each section should be 1 m long, and the distance between sampling sections should not exceed 15 m to obtain a reliable LAI map. A greater distance yields maps that do not fit the actual spatial variability within the plot and, consequently, are not suitable for precision viticulture practices. More reliable maps can be obtained when MTLS are used more intensively, separating the sampling sections by no more than 5 m, but at the cost of acquiring, storing and processing large amounts of data. In any case, the remaining issue is to avoid the overestimation of LAI in areas where vines have low vigor. This requires reducing the distance between scans in order to better account for canopy gaps. Acknowledgments: This work was partially funded by the Spanish Ministry of Economy and Competitiveness (research projects SAFESPRAY—AGL2010-22304-C04-03 and AGVANCE—AGL2013-48297-C2-2-R). Author Contributions: Ignacio del-Moral-Martínez developed and programmed the MTLS, and supervised the acquisition of field data. He also participated in writing the paper; Joan R. Rosell-Polo contributed to the development of the scanner and its application in the field tests. He participated in drafting and revising the paper; Joaquim Company participated in field trials and in obtaining and comparing maps from sensor data. His research project within the MSc in Agricultural Engineering is based on this work; Ricardo Sanz worked significantly in programming the MTLS and data acquisition; Alexandre Escolà participated in validating the scanner in the measurements of defoliation of the vines. His contribution was important in writing the paper; Joan Masip organized field trials. He also participated during the defoliation of vines and in preparing the initial analysis of MTLS data; José A. Martínez-Casasnovas prepared LAI raster maps from data supplied by the MTLS. As a specialist in GIS, he made the cluster analysis of interpolated data to compare LAI classified maps; Jaume Arnó was involved in data acquisition, analysis, and interpretation. He planned the research work and actively contributed in writing the paper. Conflicts of Interest: The authors declare no conflict of interest.
References 1.
2.
3.
4. 5. 6.
7. 8.
9. 10.
Rosell-Polo, J.R.; Sanz, R.; Llorens, J.; Arnó, J.; Escolà, A.; Ribes-Dasi, M.; Masip, J.; Camp, F.; Gràcia, F.; Solanelles, F.; et al. A tractor-mounted scanning LIDAR for the non-destructive measurement of vegetative volume and surface area of tree-row plantations: A comparison with conventional destructive measurements. Biosyst. Eng. 2009, 102, 128–134. [CrossRef] Rosell, J.R.; Llorens, J.; Sanz, R.; Arnó, J.; Ribes-Dasi, M.; Masip, J.; Escolà, A.; Camp, F.; Solanelles, F.; Gràcia, F.; et al. Obtaining the three-dimensional structure of tree orchards from remote 2D terrestrial LIDAR scanning. Agric. For. Meteorol. 2009, 149, 1505–1515. [CrossRef] Sanz-Cortiella, R.; Llorens-Calveras, J.; Escolà, A.; Arnó-Satorra, J.; Ribes-Dasi, M.; Masip-Vilalta, J.; Camp, F.; Gràcia-Aguilá, F.; Solanelles-Batlle, F.; Planas-DeMartí, S.; et al. Innovative LIDAR 3D dynamic measurement system to estimate fruit-tree leaf area. Sensors 2011, 11, 5769–5791. [CrossRef] [PubMed] Llorens, J.; Gil, E.; Llop, J. Ultrasonic and LIDAR sensors for electronic canopy characterization in vineyards: Advances to improve pesticide application methods. Sensors 2011, 11, 2177–2194. [CrossRef] [PubMed] Arnó, J.; Escolà, A.; Vallès, J.M.; Llorens, J.; Sanz, R.; Masip, J.; Palacín, J.; Rosell-Polo, J.R. Leaf area index estimation in vineyards using a ground-based LiDAR scanner. Precis. Agric. 2013, 14, 290–306. [CrossRef] Rinaldi, M.; Llorens, J.; Gil, E. Electronic Characterization of the Phenological Stages of Grapevine Using a LIDAR Sensor. In Precision Agriculture’13; Stafford, J.V., Ed.; Wageningen Academic Publishers: Lleida, Spain, 2013; pp. 603–609. Keightley, K.E.; Bawden, G.W. 3D volumetric modelling of grapevine biomass using Tripod LiDAR. Comput. Electron. Agric. 2010, 74, 305–312. [CrossRef] Del-Moral-Martínez, I.; Arnó, J.; Escolà, A.; Sanz, R.; Masip, J.; Company-Messa, J.; Rosell-Polo, J.R. Georeferenced scanning system to estimate the leaf wall area in tree crops. Sensors 2015, 15, 8382–8405. [CrossRef] [PubMed] Omasa, K.; Hosoi, F.; Konishi, A. 3D lidar imaging for detecting and understanding plant responses and canopy structure. J. Exp. Bot. 2007, 58, 881–898. [CrossRef] [PubMed] Dokoozlian, N.K.; Kliewer, W.M. The light environment within grapevine canopies. II. Influence of leaf area density on fruit zone light environment and some canopy assessment parameters. Am. J. Enol. Vitic. 1995, 46, 219–226.
Sensors 2016, 16, 119
11.
12. 13.
14. 15.
16. 17. 18. 19. 20.
21.
22.
23. 24. 25. 26. 27. 28. 29. 30. 31.
32.
12 of 13
Louarn, G.; Lecoeur, J.; Lebon, E. A three-dimensional statistical reconstruction model of grapevine (Vitis vinifera) simulating canopy structure variability within and between cultivar/training system pairs. Ann. Bot. 2008, 101, 1167–1184. [CrossRef] [PubMed] Lebon, E.; Pellegrino, A.; Louarn, G.; Lecoeur, J. Branch development controls leaf area dynamics in grapevine (Vitis vinifera) growing in drying soil. Ann. Bot. 2006, 98, 175–185. [CrossRef] [PubMed] Bindi, M.; Bellesi, S.; Orlandini, S.; Fibbi, L.; Moriondo, M.; Sinclair, T. Influence of water deficit stress on leaf area development and transpiration of Sangiovese grapevines grown in pots. Am. J. Enol. Vitic. 2005, 56, 68–72. Rosell, J.R.; Sanz, R. A review of methods and applications of the geometric characterization of tree crops in agricultural activities. Comput. Electron. Agric. 2012, 81, 124–141. [CrossRef] Gil, E.; Arnó, J.; Llorens, J.; Sanz, R.; Llop, J.; Rosell-Polo, J.R.; Gallart, M.; Escolà, A. Advanced technologies for the improvement of spray application techniques in Spanish viticulture: An overview. Sensors 2014, 14, 691–708. [CrossRef] [PubMed] Siegfried, W.; Viret, O.; Huber, B.; Wohlhauser, R. Dosage of plant protection products adapted to leaf area index in viticulture. Crop Prot. 2007, 26, 73–82. [CrossRef] Smart, R.E. Principles of grapevine canopy microclimate manipulation with implications for yield and quality. A review. Am. J. Enol. Vitic. 1985, 36, 230–239. Kliewer, W.M.; Dokoozlian, N.K. Leaf area/crop weight ratios of grapevines: Influence on fruit composition and wine quality. Am. J. Enol. Vitic. 2005, 56, 170–181. Colombo, R.; Bellingeri, D.; Fasolini, D.; Marino, C.M. Retrieval of leaf area index in different vegetation types using high resolution satellite data. Remote Sens. Environ. 2003, 86, 120–131. [CrossRef] Mathews, A.J.; Jensen, J.L. Visualizing and quantifying vineyard canopy LAI using an unmanned aerial vehicle (UAV) collected high density structure from motion point cloud. Remote Sens. 2013, 5, 2164–2183. [CrossRef] Fuentes, S.; Poblete-Echeverría, C.; Ortega-Farias, S.; Tyerman, S.; de Bei, R. Automated estimation of leaf area index from grapevine canopies using cover photography, video and computational analysis methods. Aust. J. Grape Wine Res. 2014, 20, 465–473. [CrossRef] Zhao, K.; Popescu, S. Lidar-based mapping of leaf area index and its use for validating GLOBCARBON satellite LAI product in a temperate forest of the southern USA. Remote Sens. Environ. 2009, 113, 1628–1645. [CrossRef] Hosoi, F.; Omasa, K. Voxel-based 3-D modelling of individual trees for estimating leaf area density using high-resolution portable scanning lidar. IEEE Trans. Geosci. Remote Sens. 2006, 44, 3610–3618. [CrossRef] Arnó, J.; Martínez-Casasnovas, J.A.; Ribes-Dasi, M.; Rosell, J.R. Review. Precision viticulture. Research topics, challenges and opportunities in site-specific vineyard management. Span. J. Agric. Res. 2009, 7, 779–790. Johnson, L.F. Temporal stability of an NDVI-LAI relationship in a Napa Valley vineyard. Aust. J. Grape Wine Res. 2003, 9, 96–101. [CrossRef] Hall, A.; Lamb, D.W.; Holzapfel, B.P.; Louis, J.P. Within-season temporal variation in correlations between vineyard canopy and winegrape composition and yield. Precis. Agric. 2011, 12, 103–117. [CrossRef] Johnson, L.F.; Roczen, D.E.; Youkhana, S.K.; Nemani, R.R.; Bosch, D.F. Mapping vineyard leaf area with multispectral satellite imagery. Comput. Electron. Agric. 2003, 38, 33–44. [CrossRef] Hall, A.; Louis, J.; Lamb, D. Characterising and mapping vineyard canopy using high-spatial-resolution aerial multispectral images. Comput. Geosci. 2003, 29, 813–822. [CrossRef] Llorens, J.; Gil, E.; Llop, J.; Queraltó, M. Georeferenced LiDAR 3D vine plantation map generation. Sensors 2011, 11, 6237–6256. [CrossRef] [PubMed] Meier, U. Growth Stages of Mono-and Dicotyledonous plants. In BBCH Monograph, 2nd ed.; Federal Biological Research Centre for Agriculture and Forestry: Berlin, Germany, 2001; p. 158. Martinez-Casasnovas, J.A.; Agelet-Fernandez, J.; Arno, J.; Ramos, M.C. Analysis of vineyard differential management zones and relation to vine development, grape maturity and quality. Span. J. Agric. Res. 2012, 10, 326–337. [CrossRef] Arnó, J.; Del-Moral, I.; Escolà, A.; Company, J.; Masip, J.; Sanz, R.; Rosell, J.R. Mapping the leaf area index in vineyard using a ground-based lidar scanner. In Proceedings of the 11th International Conference on Precision Agriculture, Indianapolis, IN, USA, 15–18 July 2012.
Sensors 2016, 16, 119
33. 34. 35. 36.
13 of 13
Lee, K.H.; Ehsani, R. A laser scanner based measurement system for quantification of citrus tree geometric characteristics. Appl. Eng. Agric. 2009, 25, 777–788. [CrossRef] Minasny, B.; McBratney, A.B.; Whelan, B.M. VESPER Version 1.62. Australian Centre for Precision Agriculture, McMillan Building A05; The University of Sydney: Sydney, NSW, Australia, 2006. Fridgen, J.J.; Kitchen, N.R.; Sudduth, K.A.; Drummond, S.T.; Wiebold, W.J.; Fraisse, C.W. Management zone analyst (MZA). Agron. J. 2004, 96, 100–108. [CrossRef] Arnó, J.; Martinez-Casasnovas, J.A.; Ribes-Dasi, M.; Rosell, J.R. Clustering of grape yield maps to delineate site-specific management zones. Span. J. Agric. Res. 2011, 9, 721–729. [CrossRef] © 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).
