Close Home
iForest - Biogeosciences and Forestry
vol. 10, pp. 215-219
Copyright © 2017 by the Italian Society of Silviculture and Forest Ecology
doi: 10.3832/ifor1690-009

Technical Reports

Detecting tree water deficit by very low altitude remote sensing

Hilaire Martin (1)Corresponding author, Sylvain Labbé (2), Patrick Baldet (1), Frédéric Archaux (1), Gwénaël Philippe (1)


Important innovations in satellite sensors over the last decades have made available a great deal of information about landscape monitoring ([12]). Using plant reflectance of both high and low wavelengths in infrared and red bandwidths, many vegetation indices have been created. Among these, the NDVI (Normalized Difference Vegetation Index) developed by ([15]) is being widely used to study agricultural activities, biomass, plant productivity, chlorophyll concentrations and phenology. In forests, the NDVI has been used for about thirty years to assess fire risks ([6]). In terms of stand productivity, NDVI values are strongly linked with leaf area index, which is a good indicator of plant biomass ([18]). Moreover, other studies have used the NDVI for tree species identification on monoculture plots ([5]). In addition, Camarero et al. ([4]) used variations in NDVI to detect an increase in canopy cover preceding mast years for Quercus ilex. For drought monitoring, this vegetation index is of particular interest ([11], [17]). Drought conditions result in a reduction in soil water availability, and reduced levels of water in the soil layers alter both soil-root and leaf-atmosphere interfaces ([3]). With global climate change, droughts are expected to become more frequent, and it is important to plant tree varieties selected for their resistance to dry conditions. Douglas fir (Pseudotsuga menziesii [Mirb.] Franco) was introduced in France for its exceptional stand productivity and the quality of its wood. However, this species comes from regions where annual rainfall exceeds 800 mm year-1, a level that is not always reached in the areas where the tree was introduced ([16]). Particularly for saplings, water stress can cause mortality because the young tree’s shallow roots are subjected to high temperatures in the topsoil ([2]). In tree nurseries, we can experiment with drought effects on young trees by controlling the water supply. Coupled with the recently developed NDVI technology for crop monitoring ([8], [10]), we used modified commercial digital cameras and a mobile darkroom in order to extract NDVI at low cost. We studied the spectral response of two Douglas Fir varieties exposed to different stress levels.

Material and methods 

Plant material and experimental design

The experiment took place on two-year-old Douglas fir saplings in a greenhouse during the 2013 growing season. Twenty-four boxes were planted with saplings following a two-way randomized block design crossing two genetic varieties (A and B) and six water stress levels, with one replication. Each box had been planted in 2012 on bare soil with 30 one-year-old trees. The rectangular boxes were 1.40 m by 1.10 m and 0.78 m deep (Fig. 1). The volume of soil available for roots in each box was 0.70 m3. Stress levels were measured by probes (TRIME-IPH) and 6 categories from 1 (wettest) to 6 (driest) were distinguished by Time-Domain Reflectometry (TDR) monitoring. Tab. 1 presents the percentage of soil water content for each stress level. Note that the intermediate level (Stress 3) was characterized by normal watering (15% water content) during the first phase of the experiment, then watering was completely suspended on August 6th, 2013 in order to simulate summer drought conditions. The soil water content for this level was nearly the same as stress level 6 at the end of the experiment (November 2013).

Fig. 1 - View of a part of the experiment.
Tab. 1 - Percentage of soil water content by stress level.

Image acquisition

We used a modified commercial digital camera to access the infrared bandwidth and an identical non-modified digital camera to assess the red band. Digital camera sensors are sensitive to the full light spectrum but manufacturers affix a Color Filter Array (CFA), called “Bayer Matrix” (Fig. 2), in order to record only visible wavelengths (red, green and blue). When this filter is removed, commercial cameras become sensitive to infrared wavelengths ([14]). By photographing the same target with one modified and one non-modified Canon® EOS 350D digital camera, we recorded both near infrared (NIR) and red (from RGB) bands that could be used to calculate the NDVI. A mobile darkroom was built in order to ensure identical photography conditions: target center points (tree box), light intensity and camera height (Fig. 3a, Fig. 3b). Cameras were attached to a sliding support on the top of the dark chamber in order to move alternatively each camera on a common image acquisition vertical axis ideally located in the center of the dark chamber roof. For light intensity we chose tungsten lights due to the major part of radiation emitted in the red and infrared wavelengths. The roof of the dark room was equipped with 6 tungsten lights (model Philips Aluline 111®, 50W, G53, 12V 8D, 1CT) provided by batteries with accurate tension monitoring in order to guaranty color temperature of 3000 K. Camera settings were kept unchanged throughout the experiment and images were recorded in RAW format with a constant focal length of 34 mm and the best resolution (3456 × 2304 pixels). We photographed the 24 boxes at three dates during the 2013 growing season: August 30th, October 4-5th, November 4-5th.

Fig. 2 - Color Filter Array.
Fig. 3 - (a): Mobile darkroom; (b): Example of tree/box photography; left: NIR and right: RGB

Image treatment

We processed the images with the ArcGIS® software version 10.2. In lack of spectral calibration of cameras, we divided red and near infrared bands with a gray target (more visible in infrared spectrum) of constant reflectance in order to overcome different lighting effects related to bandwidth (Fig. 4). A panel was painted with three layers of gray matt acrylic paint (AFNOR NF T 36-005) to obtain a homogeneous reference surface with low reflectance. An image of the gray target was acquired with the modified and non-modified cameras in order to quantify lighting heterogeneity, reflectance from lateral walls and other uncontrolled effects. Due to different wavelengths in infrared and visible bandwidths, settings differ from cameras (Tab. 2). Thereafter we kept same settings of cameras for the plant material image acquisition. Only one picture of red and infrared spectrum was used as reference for correcting all pictures acquired on plant material.

Fig. 4 - Example of image processing for box no. 36 photographed on August 30th, 2013.
Tab. 2 - Canon® EOS 350D Camera settings.

We performed our analysis on a square mask centered on the middle of the image to avoid vignette effects (500 × 500 pixels). The NDVI of each box was calculated by automatic image processing for the three dates as follows (eqn. 1):

\begin{equation} NDVI = \frac{ \rho \text{NIR} - \rho \text{RED}} { \rho \text{NIR} + \rho \text{RED}} \end{equation}

where ρ is the reflectance value in spectral bands.

Statistical analyses

We used linear mixed effects models with gaussian error distribution to analyse the effects of stress (6 levels) and varieties (2 levels) and their interaction on the mean NDVI (calculated on 25 × 104 pixels per box and per date). A random “box” effect was added on the intercept to take into account a potential effect of replication (there were two boxes for a given variety and a level of stress, 12 replications in total). We used an analysis of variance (ANOVA, α = 0.05) to assess the effects of treatments (stress × variety) on the response variable. All analyses were performed in R version 3.1.0 ([13]), using the library “nlme”.


Stress effect

In general, NDVImean decreased with stress intensity during the growing season (Fig. 5a, Fig. 5b, Fig. 5c). Specifically, starting in August, NDVImean decreased almost linearly from stress level 1 (well watered) to 5 (almost never watered). Fig. 5c shows the lowest NDVImean values for stress level 6 at around 8 percent soil water content. By the end of the experiment, all the plants in this category had died. For the second period (Fig. 5b), when watering was suddenly stopped in August, NDVImean decreased for stress 3. This trend was more pronounced at the end of the growing season (Fig. 5c). Throughout the experiment, stress effect levels were significantly different (Tab. 3).

Fig. 5 - Mean NDVI at different dates. Different letters indicate significant differences in stress intensity (Tukey’s HSD test, with P < 0.05).
Tab. 3 - ANOVA results of the effects of stress and origin on mean tree NDVI per box for three dates (calculated on 25 × 104 pixels). (df): degrees of freedom; (*): p ≤ 0.05; (**): p ≤ 0.01; (***): p ≤ 0.001.

Variety effect

During the growing season, differences between varieties were slightly statistically significant starting since October. There was only one significant interaction between stress level and genetic variety in the second period (Tab. 3).


In this experiment, we studied the spectral response of two Douglas Fir varieties exposed to different water stress levels during the 2013 growing season. To be cost effective, we adapted a method used in crop studies and used a modified digital camera and a non-modified digital camera ([14], [8], [7]) and a mobile darkroom. This technology has also been used for forest monitoring with Unmanned Aerial Vehicles ([9]).The results in Tab. 3, Fig. 5a, Fig. 5b and Fig. 5c are promising for very low altitude NDVI in water deficit monitoring. Because water deficit events strongly impact young tree roots ([2]), it is important to test the most suitable tree species in a global climate change context. We were not able to provide results on genetic variety effects in our experiment due to methodological limitations (not enough replications) and similar spectral responses by the two genetic varieties. From a forestry point of view, these results are insufficient. Indeed, an experiment on only one growing season cannot predict stress behavior over the life of a stand. We should measure dendrometric variables for at least two successive years, as suggested in Becker’s work ([1]), in order to test correlations between tree sizes and spectral responses.


Very low altitude remote sensing could have a positive impact in tree nurseries. Further research may focus on testing photography with direct solar light in order to avoid the use a mobile darkroom. Other vegetation indices, including simpler ones based only on RGB bands, could also be tested, following the suggestion for crop monitoring by Meyer & Camargo Neto ([10]).


We thank Anne Villemey, Yoan Paillet and Emmanuel Cornieux for their constructive comments and Stéphane Matz, Vincent Bourlon, Pascal Croizet, Franck Stocchero, Aurélien Brochet and Cécile Joyeau for constructing the mobile darkroom and for image acquisition.

Author Contribution 

S.L., P.B., G.P., conceived and designed the experiment, H.M., F.A., analyzed data, and all authors equally contributed in writing the paper.


Becker M (1974). Étude expérimentale de la transpiration et du développement de jeunes Douglas en fonction de l’alimentation en eau [Experimental study of sweating and development of young Douglas fir controled by water supply]. Annals of Forest Science 31: 997-109. [in French]
::Online::Google Scholar::
Bréda N, Granier A, Aussenac G (2000). Evolutions possibles des contraintes climatiques et conséquences pour la croissance des arbres [Possible changes in climatic constraints and consequences for tree growth]. Revue forestière Française (numéro spécial) 52: 73-90. [in French]
::CrossRef::Google Scholar::
Bréda N, Huc R, Granier A, Dreyer E (2006). Temperate forest trees and stands under severe drought: a review of ecophysiological responses, adaptation processes and long-term consequences. Annals of Forest Science 63: 625-644.
::CrossRef::Google Scholar::
Camarero JJ, Albuixech J, López-Lozano R, Casterad MC, Montserrat-Martí G (2010). An increase in canopy cover leads to masting in Quercus ilex. Trees 24: 909-918.
::CrossRef::Google Scholar::
Carleer A, Wolff E (2004). Exploitation of very hign resolution satellite data for tree species identification. American Society for Photogrammetry and Remote Sensing 70 (1): 135-140.
::CrossRef::Google Scholar::
Gitas IZ, San-Miguel-Ayanz J, Chuvieco E, Camia A (2014). Advances in remote sensing and GIS applications in support of forest fire management. International Journal of Wildland Fire 23: 603-605.
::CrossRef::Google Scholar::
Jensen T, Apan A, Young F, Zeller L (2007). Detecting the attributes of a wheat crop using digital imagery acquired from a low-altitude platform. Computers and Electronics in Agriculture 59: 66-77.
::CrossRef::Google Scholar::
Lebourgeois V, Bégué A, Labbé S, Mallavan B, Prévot L, Roux B (2008). Can commercial digital cameras be used as multispectral sensors? A crop monitoring test. Sensors 8: 7300-7322.
::CrossRef::Google Scholar::
Lisein J, Bonnet S, Lejeune P, Pierrot-Deseiligny M (2014). Modélisation de la canopée forestière par photogrammétrie depuis les images acquises par drone [Modeling forest canopy by photogrammetry from images acquired by drone]. Revue Française de Photogrammétrie et de Télédétection 206: 45-54. [in French]
::Online::Google Scholar::
Meyer GE, Camargo Neto J (2008). Verification of color vegetation indices for automated crop imaging applications. Computers and Electronics in Agriculture 63: 282-293.
::CrossRef::Google Scholar::
Peng J, Dong W, Yuan W, Zhang Y (2012). Responses of grassland and forest to temperature and precipitation changes in northeast China. Advances in Atmospheric sciences 29 (5): 1063-1077.
::CrossRef::Google Scholar::
Pettorelli N, Laurance WF, O’Brien TG, Wegmann M, Nagendra H, Turner W (2014). Satellite remote sensing for applied ecologist: opportunities and challenges. Journal of Applied Ecology 51: 839-848.
::CrossRef::Google Scholar::
R Development Core Team (2014). R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna.
::Online::Google Scholar::
Rabatel G, Gorreta N, Labbé S (2014). Getting simultaneous red and near infrared bands from a single digital camera for plant monitoring applications: Theoretical and practical study. Biosystems Engineering 117 (1): 2-14.
::CrossRef::Google Scholar::
Rouse J, Haas RH, Schell JA, Deering DW (1973). Monitoring vegetation systems in the great plains with ERTS. In: Proceeding of the “3rd ETRS Symposium”, NASA SP-351 1, US Government Printing Office, Washington, DC, USA, pp. 309-317.
::Google Scholar::
Rozenberg P, Sergent A-S, Dalla-Salda G, Martinez-Meier A, Marin S, Ruiz-Diaz M, Bastien J-C, Sanchez L, Bréda N (2012). Analyse rétrospective de l’adaptation à la sécheresse chez le douglas [Retrospective analysis of drought adaptation in Douglas fir]. Schweiz Z Forstwes 163 (3): 88-95. [in French]
::CrossRef::Google Scholar::
Volcani A, Karnieli A, Svoray T (2005). The use of remote sensing and GIS for spatio-temporal analysis of the physiological state of semi-arid forest with respect to drought years. Forest Ecology and Management 215: 239-250.
::CrossRef::Google Scholar::
Wang Q, Tenhunen J, Granier A (2005). On the relationship of NDVI with leaf area index in a deciduous forest site. Remote sensing of Environment 94 (2): 244-255.
::CrossRef::Google Scholar::


Martin H, Labbé S, Baldet P, Archaux F, Philippe G (2017).
Detecting tree water deficit by very low altitude remote sensing
iForest - Biogeosciences and Forestry 10: 215-219. - doi: 10.3832/ifor1690-009
First Previous Next Last
© iForest

Download Reference

Paper ID# ifor1690-009
Title Detecting tree water deficit by very low altitude remote sensing
Authors Martin H, Labbé S, Baldet P, Archaux F, Philippe G
Close Download