Soil Moisture Monitoring Through UAS-Assisted Internet of Things LoRaWAN Wireless Underground Sensors

With growing usage and demand for the global freshwater supply, there is an increasing need for technologies that facilitate water conservation and environmental stewardship in irrigated agriculture. Toward this end, recent demonstrations of Internet of Things (IoT) sensors have revealed the value of wireless soil moisture content sensors. However, existing wireless solutions often employ above-ground wireless communication modules that physically interfere with routine farming operations. Underground wireless network solutions are severely challenged by the large radio-frequency (RF) propagation loss through soil. This paper presents a method that overcomes both problems by employing completely buried underground IoT sensors and communication modules with UAS (uncrewed aircraft system) mounted LoRaWAN gateways. The UAS mounted LoRaWAN gateway eliminates the need for any in-field base stations and also allows the LoRa enabled sensors to transmit data over short distances with very low energy. Field tests were carried out using this approach to serve as a proof of concept. The RSSI (received signal strength indicator) demonstrates that the proposed solution has good communication link margin and a significantly larger communication range than is necessary for reliable operation. Moreover, this solution is easy to build, scalable, cost-effective, and can be implemented in a highly power efficient fashion.

taining some minimum required soil moisture is important 23 in various biophysical processes like plant growth, germina- 24 tion of seeds, nutrient cycling and sustenance of the natural 25 The associate editor coordinating the review of this manuscript and approving it for publication was Ghufran Ahmed . biodiversity in soil [1]. On the other hand, over-irrigation can 26 be equally destructive, causing salinization [2] of the land and 27 pollution [3] of freshwater sources when chemigated water 28 is used for irrigation. The objective then becomes to always 29 provide the optimal level of irrigation, just enough to maintain 30 required soil moisture level. This is referred to as scientific 31 irrigation scheduling [4] which falls under the umbrella of 32 precision agriculture, and soil moisture monitoring is an 33 important element in meeting this objective [5], [6]. 34 There are different ways of sensing and monitoring soil 35 moisture levels in agricultural fields. Generally, crop fields 36 have large areas, therefore implementing a wireless solution 37 IoT-based wireless sensor network to develop a Greenhouse LoRa enabled sensors [16]. After analyzing many different 92 technologies, some researchers have concluded that for smart 93 agricultural applications, LoRa is the best option [17]. LoRa 94 uses chirped spread-spectrum modulation, with 433 MHz, 95 868 MHz, or 915 MHz as the carrier frequency -depending 96 on region -with a maximum data rate of 50 kbps [18], 97 which is more than adequate for soil moisture monitoring 98 and occasional reporting. The relatively low carrier frequen-99 cies translate to manageable free-space path loss and some 100 immunity to line-of-sight issues. This generally gives LoRa 101 a coverage radius of between 1 to 10 km above ground. In a 102 work published by Renzone et al. [19], LoRaWAN sensors 103 were buried underground with different soil types and depths 104 between 10 and 50 cm. In their experiment, they measured 105 the corresponding path and packet loss using a receiver with 106 its antenna placed on the ground 15 m horizontally away 107 from the buried sensor. Additionally, in [19], the RSSI was 108 measured over a 20 day time period to measure the impacts 109 of soil compaction. 110 Another research avenue of environmental and agricultural 111 monitoring involves uncrewed aircraft systems (UASs) [20]. 112 The applications of UAS in the field of irrigation management 113 have been comprehensively reviewed in [21]. UAS-based, 114 low-altitude remote sensing technologies and wireless sen-115 sor networks have been studied for precision weed manage-116 ment [22] and for preventing frost in fragmented vineyards 117 by monitoring temperature and humidity [23]. Soil moisture 118 content monitoring using UAS-based hyperspectral imagery 119 has also been demonstrated [24]. Ground sensor communi-120 cation with UAS-based small cell networks was developed 121 in [25]. The ground sensors have low energy and low cov-122 erage area while the UAS-SC (UAS-Small Cell) acts as a 123 mobile transceiver that collects the data from the ground sen-124 sors by flying over them [25]. In this way a UAS can be used 125 as a mobile gateway for low power IoT sensors. But in [25], 126 above-ground communication modules were employed along 127 with solar cells for powering the modules and sensors. These 128 pose a significant practical hindrance for farm machinery 129 operations.

130
In this paper, we present a soil moisture monitoring system 131 which employs completely buried IoT sensors with integrated 132 communication modules, no in-field base stations or gate-133 ways, and a UAS-based mobile gateway for retrieving the soil 134 moisture data from the buried sensors. The buried sensors 135 and the gateway aboard the UAS use LoRa-based wireless 136 communication. A similar work [26] has been published by 137 Cariou et al.; however, there are some important differences 138 between our work and [26]. The operating frequency in our 139 work is 916 MHz whereas in [26], the operating frequency 140 is 868 MHz. Our sensor node was buried at a depth of 141 0.3 meters whereas the sensor node in [26] was buried at a 142 depth of 0.15 meters. Generally, sensor nodes buried deeper 143 into the ground can prove more beneficial as they are more 144 likely to be cleared from farming machinery operations. 145 In our work, from experimental data and path loss analysis, 146 we have demonstrated that the relative antenna orientation 147 of the buried sensor and the above ground gateway has little 148 significance in communicating data packets -which in [26] 149  has not been presented. We have also presented our analysis 150 of percentage packet loss with varying distances which in [26] 151 has been mentioned to be done in future work. In [26], it has 152 been shown that the RSSI (received signal strength indicator) 153 values after a particular horizontal distance is higher at a 154 higher altitude -the impact of the soil on the propagation of 155 the signal has been attributed as the reason for this -whereas 156 we have elaborated on the mechanisms of signal propaga-157 tion from soil to air from an electromagnetic perspective 158 and also using an electromagnetic simulation. In addition, 159 unlike in [26], we have presented a plausible estimation of  The LoRaWAN gateway used to collect data from the LoRa 178 enabled sensor was primarily based on a Raspberry Pi 4. 179 A LoRa module was added to function as the communica-180 tion module along with a dipole antenna. When mounted 181 on a UAS, the Raspberry Pi was also accompanied by a 182 GPS/GLONASS (Global Positioning System/ Global Nav-183 igation Satellite System) module. The reason for using a 184 GPS/GLONASS module will be apparent in a later section. 185 The Raspberry Pi contained the programs necessary to con-186 trol the LoRa module to collect data packets sent from the 187 LoRa enabled sensor and it also acted as storage for the 188 collected data. The Raspbian operating system was used to 189 govern operations. A block diagram of the LoRaWAN gate-190 way is shown in Fig. 1.

VOLUME 10, 2022
The LoRa radio was a bonnet-style RFM95W radio from   unperturbed sidewall of the hole at a depth of approximately 246 30.5 cm. The hole was filled with soil so that the LoRa 247 enabled sensor and broadcasting antenna were under 30.5 cm 248 (1 foot) of soil. The LoRa enabled sensor was powered from 249 a car battery, whose 12 V DC supply was converted into 5 V DC 250 supply by a common USB charging pod. The DC power 251 wires were run through the soil from the battery and volt-252 age converter above the ground to the LoRa enabled sensor 253 underground.

254
The portable LoRaWAN gateway was fixed at the end of a 255 wooden pole with the antenna having the same orientation 256 (co-polarized) as the LoRa enabled sensor's antenna. The 257 LoRaWAN gateway was powered by a small, portable bat-258 tery. Data sent from the buried sensor were received by the 259 LoRaWAN gateway and the received signal strength indica-260 tors (RSSI) were logged as a function of several LoRaWAN 261 gateway positions relative to the end-node, both at different 262 distances and heights. Numerous distances were selected in 263 both the south and west directions, all at several heights. 264 In addition, to get an estimate of the maximum coverage 265 area of the buried LoRa enabled sensor, single-point mea-266 surements were collected when the LoRaWAN gateway was 267 moved as far as possible to the north, south, east, and west 268 directions until the data packets were no longer received. 269 All the same measurements were repeated after changing 270 the orientation of the LoRaWAN gateway antenna to east-271 west, which put the transmit and receive antennas in a 272 cross-polarized relative orientation. Fig. 3 shows an image 273 of the research team collecting data during the preliminary 274 experiment.   enabled sensor antenna orientation while measuring the RSSI 317 data. The maximum usable distance -beyond which packet 318 loss became 100% -was found to be greater than 36.5 m 319 (120 feet).

320
Four heatmaps are shown in Fig. 6. Fig. 6(a) shows the 321 RSSI values at different heights from ground and at different 322 distances toward the west from the LoRa enabled sensor 323 burial position. Fig. 6(b) shows the RSSI values at differ-324 ent heights from ground and at different distances towards 325 the south from the LoRa enabled sensor burial position. 326 Fig. 6(a) and Fig. 6(b) show that, at observation points not 327 directly above the buried sensor, the RSSI value increases 328 with increasing height. It is noted that the observation window 329 is within 2 m height from ground surface and 5 m horizontal 330 distance from the burial position of the sensor. However, 331 when the LoRaWAN gateway antenna is directly above the 332 LoRa enabled sensor antenna, the RSSI values decrease with 333 increase in height. At a particular height, the RSSI values 334 mostly decrease with increasing horizontal distance from the 335 burial position of the LoRa enabled sensor. The mechanisms 336 behind this behavior are discussed in detail in the Comparison 337 VOLUME 10, 2022 with Numerical Simulation section. In short, however, the 338 radiation pattern of the buried dipole near the air-soil interface 339 appears to be the principal factor in this observed behavior.  Therefore, very careful measurements with tightly controlled 354 soil parameters and system design would be required to iso-355 late this mechanism, which is beyond the scope of this work.  The LoRa enabled sensor's longitude and latitude were 366 normalized to zero to present the measured data in Fig. 7(a) 367 and Fig. 7(b). Here, it is observed that for a particular height, 368 the RSSI values decrease with increasing distance from the 369 LoRa enabled sensor's position. But, at a particular distance 370 from the LoRa enabled sensor's position, the RSSI values 371 generally increase with increase in height. However, directly 372 above the LoRa enabled sensor, the increase in height accom-373 panied reducing RSSI values. These results are congruent 374 with the results of our preliminary measurement campaign. 375

376
The measurements were then analyzed in terms of packet 377 loss. The LoRaWAN gateway encountered a best case packet 378 loss of 12% when the antennas were co-polarized and 10% 379 when the antennas were cross-polarized. Packet loss can also 380 be quantified for different intervals of distances. Fig. 8(a) 381 and Fig. 8(b) show the percentage packet loss computed 382 for various intervals of distances between the UAS-mounted 383 LoRaWAN gateway and the buried LoRa enabled sensor. 384 Fig. 8(a) indicates that when the antennas were co-polarized, 385 percentage packet loss was no more than 15% for distances 386 smaller than 61 m and after this distance the percentage 387 packet loss increases dramatically. From Fig. 8(b), it can be 388 seen that when the antennas were cross-polarized, percentage 389 packet loss was no more than 17% for distances smaller than 390 82 m. The horizontal distance is contributing the most to 391 the packet loss, compared to the height component, since 392 the UAS never reached a height greater than 12 ft (3.66 m) 393 from the ground. Summarily, the LoRaWAN gateway and the 394 buried device could communicate without experiencing more 395 than 17% packet loss at distances ≤61 m from each other. 396 This is more than adequate for the intended purpose.   The soil sensor itself can also be powered for only one second, transmission is set to 4.7 s (typical values), then over the 418 course of the five minute beacon mode, the total transmis-419 sion time will be approximately 8.5 s. The current draw for 420 each component device was obtained from their respective 421 data sheets. Using these currents, the radio's delivered power 422 (+20 dBm) to the antenna, and supply voltage (3.3 V), the 423 average power draw by the LoRa enabled sensor can be 424 estimated for both its active and dormant period. Under these 425 conditions and assuming a battery total capacity of approxi-426 mately 9,000 J and a 1.5 V to 3.3 V converter circuit efficiency 427 of 80%, the LoRa enabled sensor should be able to operate 428 without battery maintenance for about 3.5 years. It is noted 429 that the duration the sensor node stays awake every time 430 it wakes up and the number of times the sensor wakes up 431 in a day can vary depending on the application, which will 432 impact the battery lifetime. Moreover, the sensor nodes may 433 be required to wake up at different times on different days to 434 send their stored data to the UAS-mounted gateway. As an 435 example, inclement weather may prohibit UAS flight on a 436 particular day. Data storage with on-board sensor memory 437 would usually permit this data to be collected later, but such 438 considerations remain important in a system-wide design.

439
In another application, the gateway could actively wake the 440 sensor nodes when they reach close proximity. Reference [28]  The K dB value signifies the minimum loss at reference 460 distance. The path loss, P LdB may then be quantified as

462
where P L dB = 10 log 10 (P T /P R ) is the path loss in decibels.

463
In our case, d 0 = 1.2192 m, which is the minimum measured In the ideal case, γ = 2, which indicates a path loss exponent 476 of free space. The fact that our values of γ are greater than 477 two in both cases is a consequence of the transmitted signal 478 partially traveling through soil and also experiencing the 479 effects of the soil-air boundary, as described further below. 480 The γ value in the case of cross-polarized antennas is larger 481 than the co-polarized case which is expected. This means 482 that for a real-world application, keeping the transmitter and 483 receiver antenna in parallel will be statistically advantageous, 484 although this benefit is surprisingly slight. In practice, the 485 polarization state of the antennas would have little significant 486 effect.

488
A numerical simulation was performed in COMSOL Mul-489 tiphysics to obtain the radiated power pattern of the buried 490 antenna and thereby better understand the mechanisms 491 limiting performance. The soil properties assumed in these 492 simulations were comparable to those of the site of our pre-493 liminary measurements. To get these properties, we entered 494 the coordinates of the site in the USDA Web Soil Survey 495 website [30] and obtained an estimate of the soil composition. 496 We also measured the volumetric water content of the soil on 497 the site, which was found to be 41.4% at 12 inches depth into 498 soil, obtained using a calibrated meter, Campbell Scientific 499 Hydrosense II. Incidentally, this is quite wet and lossy soil, 500 having just experienced rain in previous days. From these we 501 obtained values of the complex relative electric permittivity 502 as˜ r = 24.5 − j2.23 (where j = √ −1) by exploiting 503 the empirical equations in [31]. The complex relative mag-504 netic permeability was assumed to be unity. The complex 505 permittivity and permeability were assumed to be frequency 506 independent as our operating bandwidth was only 125 kHz. 507 Consequently, the simulation was performed for only one 508 frequency, i.e. 915 MHz. From the simulation, radiated power 509 was examined in two vertical planes (both perpendicular to 510 ground) -one plane is parallel to the antenna axis and contains 511 the antenna axis center line while the other plane bisects the 512 antenna and is perpendicular to its axis. In both cases, the 513  due to the strongly refracted waves. We point out that a 535 strict plane wave interpretation of these phenomena is not 536 possible due to the close proximity of the antenna to the 537 interface. It is worth mentioning that the simulated power 538 pattern inside the soil just next to the antenna is dominated 539 by near field components, which would normally die off 540 and become insignificant in the far-field (air). The near field 541 components can act as energy storage (resonant) fields which 542 leads to artificially inflated values of RSSI inside the soil next 543 to the antenna using our simple power calculation method 544 (P ∝ |E| 2 ).

545
In addition, Figs. 10 and 11 show that the power pattern 546 within the soil looks similar to that of a λ/2 dipole antenna. 547 On the other hand, it is reshaped in air by the discontinuity 548 at the soil-air interface. Of particular note is that the critical 549 angle interface produces strongly refracted waves that prop-550 agate parallel to the interface on the air side (see Fig. 10(a)). 551 This represents power flow in directions that would be 552 forbidden from an unbounded dipole antenna (parallel to 553 the antenna axis), which is possibly another benefit to this 554 application.

555
Simulation further permits a one-to-one comparison of 556 power flow results with the experimental results obtained 557 in preliminary measurements. These corresponding power 558 patterns are compared in Fig. 12.  Since the cross-polarized simulation results would imply 576 undetectable signal levels, in contrast to measurements, there 577 is no significant value to presenting this comparison. It is 578 notable, however, that the experimental results proved the 579 system more robust than expectations derived from simula-580 tion. In many cases, the measured RSSI was actually stronger 581 than expected.

582
The similarity of the general trend in simulated power 583 patterns and experimental power patterns in Fig. 12 sug-584 gests that the experimental power patterns were the result of 585 various mechanisms including wave reflections from soil-air 586 interface, complete reflection beyond critical angle, and large 587 soil losses. A perfect match between simulated and measured 588 power is certainly not expected. Some of the possible sources 589 of error include the following. First, in the experiment the 590 buried antenna was not immediately surrounded by soil; 591 rather it was inside an air-filled (albeit small) plastic enclo-592 sure. Second, the buried antenna was placed next to other 593 electronics including a Raspberry Pi and ADC which may 594 have affected the radiation behavior. Finally, there may have 595 been air pockets present in the soil above the buried enclosure 596 containing the LoRa enabled sensor, an unavoidable result of 597 perturbing the soil during digging and burying.

599
The concept and experimental verification of using a UAS 600 mounted LoRaWAN gateway with completely buried LoRa 601 enabled sensors for soil monitoring is presented. Both 602 manual and automated field measurements using a UAS 603 revealed highly reliable communications over large distances 604 (>36.5 meters), even with wet, lossy soil. Communication 605 was found to be largely insensitive to antenna polarizations 606 as evidenced by total packet losses of no more than 12%, 607 regardless of antenna orientations. Having measured multiple 608 heights, these tests also prove that this concept is viable 609 even when the UAS needs to clear the heights of almost 610 all kinds of crops. The simulations revealed that the signal 611 crosses the soil-air boundary in a relatively narrow region 612 FAHIM FERDOUS HOSSAIN received the B.Sc.
Internet of Things applications.