Approach

With precise and repeatable positioning of the microscope a critical component, much thought was put into both the choice of motors and the type of drive system to be utilized.  Based on cost, size and power constraints our initial approach has focused on using small stepper motors which have so far proven effective.  We chose the ProScope microscopic video camera because its optics were sufficient for our application, the lighting system was already built-in and it is preconfigured with a USB interface and an application program interface.  (Detailed specifications are available at: AMR_Spec.doc )

Beginning in March of 2005, minirhizotron images have been recorded manually at varying temporal intervals—from monthly to weekly to intensive daily recording. By combining estimates of belowground biomass (plants and fungi) from the minirhizotrons with (1) morphological and molecular identification of the plant/fungal community and (2) data on soil temperature, moisture, water potential, nitrate concentration and CO2 flux, we will develop models to describe processes such as carbon sequestration and mycorrhizal response to environmental change.

Systems/Experiments

Below are pictures of the various system components associated with the A-MR as well as the first images taken utilizing the proof-of-concept model.  These were a test of the anti-reflective coating on the tube and confirmed that our lighting choice would be compatible with the surrounding tube.  We are currently testing positioning repeatability and mechanical reliability with results likely available by early April 2006. (Videos are available at Rotational motion and Linear motion )

Figure set 18. Pictured above is a CAD drawing of the proof-of-concept A-MR. Below is the actual device as well as one of the first images taken utilizing the ProScope.  A shortened version of the tube is being used for initial testing and, in this instance, we determined that the anti-reflective coating would prevent our light source from being reflected back into the camera.

The AMARSS project installed 10 weather stations (nodes) in an 80 x 10mts transect under the NIMS 2 project. Each node consists of an array of below and above ground sensors. The below ground sensors are: 1 minirhizotron tube (to study root and mycorrhizae growth and turnover), 3 CO2 sensors (Vaisala Carbocap GMP220) installed at 2, 8, and 16 cm depth, 3 soil temperature sensors and 3 soil moisture sensors (ECHO) installed also at 2, 8, 16 cm depth. Each Vaisala CO2 sensor required 100 mA and 12V to work. We installed 30 of these sensors at the site (3 x node x 10 nodes) that are continuously working since November 2005. The above ground sensors consist of air temperature, relative humidity and photosynthetic active radiation (PAR).

Images from the minirhizotron tubes are manually scored for counts and color variation of (a) roots, (b) mycorrhizal root tips, (c) fungal rhizomorphs, and (d) fungal hyphae. When calibrated against conventionally analyzed soil samples taken near the minirhizotron tube locations, counts of roots and fungal structures can be used to estimate total belowground biomass. Appearance and disappearance of roots and fungal structures also provide information on turnover rates. By comparing turnover at various temporal scales (daily, weekly, monthly, seasonally; see figure below), we are now building a much finer-grained dataset to describe rates of change in belowground plants and fungi.

Figure 19. Image sequence from same location within one minirhizotron tube. Images were recorded at 7-day intervals. The rapid growth of this fan of fungal hyphae occurred during the hottest and driest week of 2005. Based on the August 25 and September 1 images, we can estimate a linear hyphal growth rate of approximately 1.4 mm × d-1. Smaller time intervals between images will provide a better understanding of these dynamics.

With more frequent minirhizotron sampling, the time required to manually quantify the contents of the images becomes extremely time-consuming. When the automated cameras come on-line, there will eventually be 15 or more cameras capable of imaging the entire surface of each minirhizotron tube simultaneously and multiple times per day, increasing the information processing task beyond possible human-hours of work. To address this problem, we are collaborating with software engineers in the Vision Lab at UCLA/CENS.

The Vaisala sensors measure concentration of CO2 in the soil profile. Using published models of gas diffusivity we are able to model soil respiration as umol/CO2/m2/s. In February 2006 we acquired a Li-cor 8100 to measure soil respiration. This equipment has traditionally been used to measure soil respiration.

However, only a single measurement point can be made at a single time and, given the high cost it is notoriously difficult to continuously measure CO2 flux at multiple locations. We are using the Li-cor 8100 to better understand respiration in soil systems. These sensors have proven to be an alternative technique to continuously measure CO2 and because of its relative low cost the system can be implemented at multiple locations (10 nodes in our site).

At this moment each one of the nodes has a datalogger that records outputs from the sensors and we have the capability to communicate to each one via Ethernet connection. We are working on wireless communication and an interface that can control the sensors off site.

Belowground processes are still a black box for ecosystem models and we are interested in short time responses of root and mycorrhizae turnover. To address this question we have done two field campaigns to record daily images from the minirhizotron tubes. The first campaign was undertaken from December 2 to December 16, 2005 and the second one from February 23 to March 5, 2006. For each campaign we generated over 7000 images and continuous sensor data associated with them. Each day of work during the campaign consist of taking images in 15 minirhizotron tubes (about 60 images per tube), psychrometer readings (for soil water potential), and CO2 measurements using the Li-cor 8100 at each node. We are planning to repeat this effort monthly until the automated version of the minirhizotron is installed in the field.  

With the goal of supplementing physical data with an understanding of the organisms involved, we are employing morphological and molecular techniques to identify fungal sporocarps (mushrooms) and belowground fungal and plant tissues. Sporocarp and soil samples are collected near and within the AMARSS transect. DNA is extracted, amplified through PCR using primers for specific ITS (intervening transcribed spacers) regions. The product is purified, and either sent directly to the UCR genomics facility for sequencing or further amplified through cloning before sequencing. Sequences are checked for consistency and then identified through BLAST searches. We compare morphological and molecular identifications, and we are building a database that correlates all the information derived from each field sample.

Figure 20. Sample images demonstrating pattern-recognition software under development for automated interpretation and quantification of minirhizotron imagery. (A) 9 x 12 mm minirhizotron image. (B) Same image with "non-max" ridge detection filter. (C) Same image with "ScaleRidge" filter. (C) Digitally enlarged portion of a second image (~3 mm2). (D) Same as D but with ScaleRidge processing. (E) Same image with ScaleRidge + SeedTrace filters (circles).

To address the challenges inherent in the need to manage and interpret thousands of gigabytes of visual information, our vision software collaborators at UCLA are developing pattern-recognition and image compression software to automate the tasks of image interpretation and data storage. They have identified and developed filtering algorithms that locate structures of interest (see figure above) and are now constructing classifier algorithms to distinguish categories of structures (e.g., plant roots v. fungal rhizomorphs).

Preliminary results show spatial and temporal differences of CO2 fluxes in the study site. Figure 21 shows CO2 fluxes from three different nodes located about 20 m apart each other. Note the differences between the mean of the fluxes of each node (Node 1 –0.33, Node 3 –0.72, and Node 9 –0.54 umol/CO2/m2/s) and the amplitude of the daily variation in the flux, where Node 1 shows less daily variation than Node 3 and 9. We are interested to couple this information gathered from fix nodes with information from mobile nodes such as NIMS 2

Figure 21. Soil CO2 flux in node 1, 3, and 9. Node that flux is higher (more negative) in Node 9 and more constant in Node 1, which has a thick organic layer.

The rate of respiration is a calculated parameter using a variation of the model developed by Turchu et al. 2005, as modified by Vargas. This parameter uses outputs from 9 sensors: CO2 concentrations, soil moisture, and soil temperature at 3 depths (2, 8 and 16cm), coupled with barometric pressure and atmospheric conditions. Some of the sensor output is shown in the following figures (22-25). Shown are the sensor data taken from 12/5/2005 to 12/15/2005 in three nodes (1,3, and 9). The nodes are located in an 80 meters transect in a slope. The approximate differences in altitude from Node 1 (at the top of the slope) and Node 9 (at the bottom of the slope) is 5 meters; node 3 is located between Node 1 and 9 in the transect. Node 1 is located under trees and there is a thick organic layer on top of the soil. Node 3 has less organic matter than node 1 and node 9 is the node with least organic matter and with higher sand content in the soil.

Figure 22. Soil CO2 concentration at 2 cm depth.

Figure 23. Atmospheric pressure (Kpascals) in the three nodes. Note that atmospheric pressure is lower at Node 1 (top of the transect) and lower at node 9 (bottom of the transect).

Figure 24. Soil temperature at 8 cm depth.

Figure 25. Soil volumetric water content at 8cm depth. Note: a value of 0.4 m3/ m3 represents saturation of water in the soil and a value of 0.01 m3/m3 is oven dry soil.

Figure 26. Respiration through two precipitation events. Columns are the surface soil moisture (2cm) and line is the calculated respiration (modified by Vargas and Allen, from Turchu 2005). Respiration is net loss (negative values) from soil such that more negative is higher respiration. Not surprisingly, the variation in soil respiration is much higher under dry than wet conditions. What is surprising is that the respiration rates decline under higher moisture (less negative). This could be due to either absorption of CO2 by water, or a reduction in the allocation of C to roots and mycorrhizal fungi.

ACCOMPLISHMENTS

Prototype development of the Automated MiniRhizotron (AMR) has progressed to the lab bench testing phase. Underground testing will commence this coming summer.

Imagery collected using a manual operated mini-rhizotron (non-robotic) are providing training images for purposes of developing pattern-recognition and image compression software to automate the tasks of image interpretation and data storage for the AMR. We have identified and developed filtering algorithms that locate structures of interest and are now constructing classifier algorithms to distinguish categories of structures (e.g., plant roots v. fungal rhizomorphs).

First year data from our soil sensor arrays and coordinated manual AMR observations have discovered that mycorrhizal fungal hyphae are quite capable of growing from root tips at very low water content, as low as 2 or 3% soil moisture. This is well below the value presumed in ecosystem models. Secondly, using rapid sensing CO2 models, we have been able to see that soil respiration is highly variable through space and time. Also important, we observed a spike in soil respiration at some sites during the night. We believe that this spike results from fixed C from the plant (during the day) that pulses through the root and mycorrhizal system exiting at night. Most CO2 models assume that night and day respiration rates are uniform. By using this assumption, daytime C fixation is calculated as a difference between respiration (from nighttime measurements) and total ecosystem exchange. This assumption appears to be wrong with large impacts to existing C models.

Future Directions

When testing is completed on the proof-of-concept model, a prototype automated minirhizotron will be built and tested in a lab environment.  During this period, 15 tubes will be placed in areas of interest at the James Reserve so that the necessary soil compacting can occur around them making them useful for research after about nine months.  While the prototype is being tested, construction will begin on the production units with deployment anticipated in late fall of 2006.

Construction of a detailed taxonomic database of fungi, plants, and other soil organisms in the neighborhood of the AMARSS array will enable us to correlate relationships between biodiversity and function (based on process models from sensor data) in the ecosystem under study. Ultimately, it should be possible to use the information regarding community composition that we are acquiring to design microarrays for rapid molecular assessment of changes in community composition through time. Finally, sensor development and integration will continue over the next year.