Intelligence in Ecology : How Internet of Things Expands Insights into the Missing CO 2

1State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi 830011, China 2High Performance Computing Center, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China 3State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100101, China 4Center for Geospatial Information, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China

Recent studies of the arid and semiarid ecosystems suggest that the missing CO 2 sink can be partly attributed to unneglectable soil abiotic CO 2 uptake in arid region [21][22][23][24].Such uptake has been long-term overlooked in estimating the net ecosystem exchange of CO 2 [NEE] around the world.The global "CO 2 flux towers" employed in current micrometeorological measurements interpret NEE as biological fluxes, exactly defined as the direct sum of photosynthetic and respiratory components [20].Arid region characterizes more than 30% of the Earth's total land surface area and the area is still increasing due to the trends of global desertification, yet the extent to which it modulates the global C balance has been inadequately studied [25][26][27][28][29][30][31][32][33].

Scientific Programming
Estimates of the overall contribution of such abiotic CO 2 uptake are essentially emergent for expanding insights into the missing CO 2 sink, which further requires common huge efforts of the world scientific communities [23].The current estimates based on very limited data collected from a few sites within several typical desert ecosystems were thought to be not convincing and even problematic [24].Ecologists were cautioned to keep discreet minds in both data collection and the determination of the whole story of soil abiotic CO 2 uptake in arid region.Such abiotic uptake can be varying with predominant processes, site location, and climatic conditions.These are important factors affecting experimental designs because spatial-temporal heterogeneity must be taken into account.To treat these disturbances and simplify experimental designs, it is hence imperative to implement intelligent methods for ecologists to collect both convincing data and further evidences.
In previous publications for insights into the missing CO 2 sink and especially for insights into soil abiotic CO 2 uptake in arid regions, the utilized technologies are rather old.The emerging information technologies were hardly employed.These unemployed technologies include the wireless sensing networks [34][35][36], Internet of Things (IoT) [37][38][39][40], and cloud computing [41][42][43][44][45][46].Particularly, IoT has been further integrated with the surveillance systems and the integrated system was termed as IoT monitoring [47][48][49][50].Since IoT monitoring can generate images, videos, and other archival data, it is necessary to investigate whether IoT monitoring can serve for a better understanding of soil abiotic CO 2 uptake in arid region.The currently published studies are very limited and were thought to be not convincing.Geospatial images and videos from IoT monitoring help us to explain at which sites soil abiotic CO 2 uptake was observed and present more details of the whole experimental process.
Our objectives in this study were to examine the potentials of IoT monitoring as an emerging technology for insights into soil abiotic CO 2 uptake and in turn for expanding insights into the missing CO 2 sink in the unneglectable arid region.Utilizing geospatial, archival data, intelligent algorithms on videos and images were performed to theoretically expand insights into soil abiotic CO 2 uptake in unneglectable arid region, which has been overlooked for a long period.Additionally, the existing uncertainties and unresolved issues to develop such a thematic IoT monitoring system are also discussed.

Collection of Geospatial Data.
Analyses of the potentials of IoT monitoring for insights into the missing CO 2 sink in the present study are based on the collected geospatial images and videos from the field sites at the south edge of the Gurbantunggut Desert in the north of Xinjiang Uygur Autonomous Region, China (Figure 1).These field sites were chosen because it has been confirmed that soil abiotic CO 2 uptake can temporally dominate and cause the apparent negative soil respiration fluxes at these sites [51][52][53].Collecting the geospatial images of these sites from IoT monitoring helps us to explain at which sites soil abiotic CO 2 uptake was observed and present more details of the experimental sites.Overall, geospatial images and videos were collected from 19 field sites, 18 of which are distributed within the Manas River Basin.These sites are close to each other.Another field site is located in the Sangong River Basin [51,52].
A mobile communication tool (Redmi Note 4, with MATLAB software installed to operate the algorithms) was employed for the collection of geospatial images and videos.In total 70 geospatial images of these field sites were collected and 36 images were chosen to build the first database of geospatial images for the sites where soil abiotic CO 2 uptake can temporally dominate (Figure 2).As a first example of the utilization of geospatial videos in analyzing soil abiotic CO 2 uptake, a special experiment was designed to expand insights into soil texture at those sites where abiotic CO 2 uptake can temporally dominate in soil respiration fluxes.
The details of this experimental design are as follows.We aim to collect a video to record the process when one inserts the WET sensors of HH2 Moisture meter (Delta-T Devices Ltd., Cambridge, UK) into the soil and then utilize video tracking algorithms to analyze the movements of the sensors beneath the soil surface.This is really a challenge because we realized that the time of the collected video may be too short.However, the soil texture cannot be objectively displayed if we deliberately slowly insert the sensors.Therefore, the daughter of the first author (Wenfeng Wang), who is 6 years old and named Yanbo Wang, was invited to join the "scientific game."She saw this as an interesting game and naturally tried her best.A short video was collected when she was inserting the WET sensors into the soil.

Optimization and Control.
A histogram-based image similarity algorithm [54][55][56] was further optimized and employed to analyze the match degree between the test image and each image from the first database of geospatial images of the sites where soil abiotic CO 2 uptake can temporally dominate.This helps in finding the best match of the test image in the database.In a previous publication [54], to optimize the performance of the algorithm, the histograms of the Red-band H(R), the Green-band H(G), and the Blueband H(B), respectively, were used.In the present study, the algorithm was further optimized by taking into account the weights of H(R), H(G), and H(B) to each image, where the weights/contributions were determined by calculating the information entropy [57][58][59].R-G-B-weighted average correlation-efficient parameters that were employed to evaluate the histogram-based image similarity between the test image and each image from the database were calculated.
In order to objectively evaluate the potentials of IoT monitoring for insights into the missing CO 2 sink, a real challenge was carried out.The video object tracking algorithm was performed on the collected short video for the real-time video tracking of the WET sensors.Traditional algorithms, such as the mean-shift algorithm [60][61][62], are unsuitable for this video object tracking.Therefore, we previously specialized the video target for tracking by morphological segmentation [63], which helps to improve the performance of mean-shift algorithm.

Images' Similarity Analyses Based on IoT Monitoring.
The further optimized histogram-based images similarity algorithm was applied to search field sites where soil abiotic CO 2 uptake can temporally dominate.The performance of the similarity detection algorithm worked out the best match of the test image among images in the first images database of the sites where the abiotic CO 2 uptake can temporally dominate.Results show that the match degree between test image and best match is approximated to 90%.Therefore, the histogram-based image similarity analyses based on IoT monitoring confirmed that the test image represents a site where soil abiotic CO 2 uptake can temporally dominate in soil respiration and cause negative soil respiration fluxes.
Through further reviews of the details of the test image, it is easy to find an obvious salt accumulation on the soil surface at the test site (Figure 3).The test image can be joined to the database and the extended information can be utilized.Exactly, some previous reports of negative soil respiration fluxes in arid region do not emphasize the role of salt accumulation [21].
This helps in convincing the ecologists who were not convinced by the previous reports since they may realize that the soil and groundwater are alkaline, which is advantageous to the subterranean fixation of CO 2 .Taking into account the abiotic flux components, the soil CO 2 flux can be further reconciled as where  is the CO 2 release from roots and soil microbial respiration and  DIC and  SIC are the net CO 2 fixation in the groundwater and the soil [in inorganic forms], respectively.  and   are the net soil CO 2 influx and the net soil CO 2 efflux, respectively [52].
A sketch of soil CO 2 flux formation in arid region can be hence expanded by further mathematical analyses.First, review the mechanism of how the soil CO 2 analyzer (e.g., LI-8100; see [53]) works.Assume that, after per unit time , the CO 2 analyzer abstracts air of volume  1 from a gas room of volume  and then supplies air of the same volume sampled from atmosphere for the CO 2 pressure balance in the gas room.Go round and begin again.To compute CO 2 flux, the following is used: where () is the CO 2 concentration in the gas room at time  [64].
Let  be CO 2 concentration in the atmosphere.For the th measured value, the input and output of CO 2 are  input =   and  output /  =   , respectively, taking average within th time interval [, ( + 1)].The dynamic of CO 2 concentration in the gas room should be as follows.
Input-output balance equation: where  0 is the CO 2 concentration at starting time point.
Thus the th measured value of soil CO 2 flux is where (  ) is the CO 2 concentration from mean value theorem of integrals.
Negative soil respiration CO 2 fluxes are observed if Finally, it must be cautioned that infimum of the negative values of soil CO 2 flux may exist.Let  → 0; we obtain Hence, Stable negative fluxes may happen within a small measurement interval  when Let  → ∞; we get This is the infimum of the CO 2 concentration.

IoT Monitoring with Intelligent Video Recognition Algorithm.
The trajectory analysis of soil sensors is realized in performance of IoT monitoring with intelligent video recognition algorithm.Such video object tracking algorithm not only enables ecologists to revisit these sites and the experiments details by geospatial videos, but also helps ecologists to further understand the compact soil texture so that the whole process costs 21 seconds.The footprint of the WET sensors revealed that the process is difficult for this little girl (Figure 4).Consequently, a part of soil respiration () temporally gathers in soil (  ) or is ventilated in subterranean cavity ( V ) and contributes to the abiotic release later.This also is advantageous for a chemical fixation of CO 2 in the soilgroundwater system (Figure 5).This expands a perspective frame of IoT for insights into the missing CO 2 sink (Figure 6).
Therefore, the potentials of IoT monitoring for insights into soil abiotic CO 2 uptake and hence for the insights into the missing CO 2 sink are highlighted.A part of soil inorganic CO 2 (SIC) remained in soil layers [SR] and a part of DIC is carried away and might go out at the terminal of the The carbon assignment equation can be further hypothetically expanded.We can classify soil pores as three types: dry pore [DP], small water pore [SWSP], and big water pore [BWSP] according to their size and water content.DP is distributed in shallow soil layers and can absorb CO 2 if coupled with the condensing of vapor or the infiltration of precipitation; SWSP is distributed in moist layers around the roots system, dissolving CO 2 in it; BWSP is distributed in deep layers, dissolving CO 2 and then migrating it into groundwater.Note that these three types of soil pores may convert to each other with the changes or movements of soil water.
The balance equations can be represented as where the groundwater recharge/discharge is the major regulator of the balance.

Conclusions and Outstanding Remarks
As an emerging technology, IoT monitoring combines researchers, instruments, and field sites and generates archival data for a better understanding of soil abiotic CO 2 uptake in arid region and in turn has great potentials for insights into the missing CO 2 sink.By histogram-based image similarity analyses of image data collected from IoT monitoring, ecologists can easily find field sites where soil abiotic uptake of CO 2 can temporally dominate and further improve their understanding of the negative soil respiration flux values.Video object tracking algorithms based on IoT monitoring not only enable ecologists to revisit these sites and the experiments details by geospatial videos, but also help the ecologists to further understand other details, such as the footprint of soil sensors, which in turn can help ecologists to understand the integrated story of soil abiotic CO 2 uptake.
In subsequent studies, the employed algorithms can be more and more complex and the uncertainties of the presented algorithms must be explicitly discussed.Nevertheless, it must be pointed out that there are still considerable uncertainties and difficulties in developing such a thematic IoT monitoring system.One major challenge is how to conceptualize ecosystem as a volume with explicitly defined top, bottom, and sides.The other major challenge is how to estimate the SIC/DIC assignment proportion of the carbon fluxes in soil layers and groundwater, which should be also analyzed in complicated cases due to great difference in soil types and the groundwater levels.The possible scheme is relating the field sites, instruments, and researchers together by a stable IoT monitoring system and conceptualizing each block of terrestrial ecosystem.In this case, net ecosystem carbon balance equals the total C input minus the total C output from the ecosystem over a specified time interval.
To reduce the increased complexity, one can analyze the situation in the different layers of the local groundwater-soil system.A research priority is the explicit characterization of the situation in different layers of the local groundwatersoil system, which deserves subsequent studies on the field collection of geospatial data for soil abiotic CO 2 uptake [65][66][67][68][69][70][71][72][73], the visualization of CO 2 footprints [74], and 2D-3D video treatments technology to enhance the visualization effect [75].

Figure 1 :
Figure 1: Distribution of the field sites where geospatial images and videos are collected in this study.

Figure 2 :
Figure 2: The first images database of the sites where soil abiotic CO 2 uptake can temporally dominate.

Figure 3 :
Figure3: To find the best match for test image in the first images database of the sites where soil abiotic CO 2 uptake can temporally dominate, utilizing the image analysis algorithm referred to in this study.

Figure 4 :Figure 5 :
Figure 4: Trajectory analysis of soil sensors by the video object tracking algorithm referred to in this study.