^{1}

^{*}

^{1}

^{1}

^{1}

^{1}

^{2}

^{1}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

The aim of this study was to investigate the suitability of active infrared thermography and thermometry in combination with multivariate statistical partial least squares analysis as rapid soil water content detection techniques both in the laboratory and the field. Such techniques allow fast soil water content measurements helpful in both agricultural and environmental fields. These techniques, based on the theory of heat dissipation, were tested by directly measuring temperature dynamic variation of samples after heating. For the assessment of temperature dynamic variations data were collected during three intervals (3, 6 and 10 s). To account for the presence of specific heats differences between water and soil, the analyses were regulated using slopes to linearly describe their trends. For all analyses, the best model was achieved for a 10 s slope. Three different approaches were considered, two in the laboratory and one in the field. The first laboratory-based one was centred on active infrared thermography, considered measurement of temperature variation as independent variable and reported

Recently, the need to measure in-field the variability of soil characteristics has increased following both sensor engineering developments, as well as the necessity to apply innovative crop management systems [

An important soil property is the spatial variation of water content measured at a proper depth and time [

Another widely used method for small spatial scale estimates of SWC is the measurement of soil thermal properties such as the heat dissipation technique and the heat pulse technique [

In order to overcome the limits of heat dissipation sensors, in this study we propose the use of a new technique based on the same underlying theory of the heat dissipation methods. Unlike heat dissipation sensors, we propose to directly measure temperature changes of soil samples, after heating, by using active infrared thermography and thermometry. The assumption is that these techniques could lead to the development of a faster SWC measurement system and could represent informative and non-destructive tools to remotely assess the dynamic variation of soil temperature [_{n} is the net radiation at soil surface, M represents the supply of energy to the surface by metabolism or absorption of energy by photosynthesis, H is the sensible heat flux, λE is the latent heat flux by evapotranspiration and G is the soil heat flux.

Adapting the energy balance _{s} is the heat variation of soil surface and G_{l} the heat flux in the soil by contact. The surface thermal variation will be related to G_{s}, H, λE and G_{l}. In this case, these parameters will be dependent on agro-pedological and meteorological parameters such as air temperatures and humidity, SWC, irradiance, wind regimes, soil water potential and soil roughness. The deterministic modelling of the environmental variables influencing the physical process which is developing in such a short time of analysis (few seconds) would have been very complex.

For the above mentioned reasons the system could be approached in a statistical way and the estimation of SWC innovatively implemented by using a multivariate analysis [

In this study, a multivariate statistic approach (Partial Least Squares regression, PLS, and Discriminant Analysis, PLSDA) is used to estimate the SWC with active infrared thermal methods by warming up and measuring, at different time steps, several non-factorial soil samples with different water contents. Three different hypotheses were considered, two in the laboratory and one in the field. The laboratory experiments were carried out to determine the best performing one. The latter was then chosen in order to be applied in-field. The first one tested in the laboratory is based on active infrared thermography, which considers only the measurement of temperature variation as independent (observed) variable. The second one examined in the laboratory added the irradiation of soil samples as independent variable and it was based on active infrared thermometry. Finally the in-field experiment was based on active infrared thermometry and also considered some meteorological parameters as independent variables (

In order to develop models for the statistical interpretation of the phenomenon, according to the previously indicated thermo-physical context, a series of progressive laboratory tests were performed. These laboratory tests were developed to highlight the limits and possibilities of the techniques and chose among them the most suitable one for an in-field application.

The experimental laboratory protocol consists in warming up soil samples with different initial temperatures and water contents and in measuring for a few seconds the dynamic temperature variations. This investigation was carried out in two different steps: the first with an infrared thermocamera considering as dependent variable the percentage of water content and as independent ones the initial soil temperature and the exposition time at constant irradiance.

In a second step, an infrared thermometer was used to simplify the measuring system by introducing among the independent variables also the irradiance produced by photographic bulbs (200 W and 2,800 K) to approach in-field applications. In both cases, air temperature and air relative humidity were considered as constant.

Soil samples, number of soil samples (N) = 250, were collected from the CRA-ING experimental field (Lat. 42°06′11.00″N, Long. 12°37′40.81″E) at a depth of 30 cm. The operative soil status being normally unknown, different initial levels of water content and temperature were achieved by hydrating, dehydrating, warming up (stove, 40–60 °C; controlled environment, 20–25 °C) and cooling down (fridge, 2–4 °C) different plastic trays (20 × 30 cm) previously filled with soil samples.

The dynamic variation of sampled soil temperature was identified by an operator analyzing a specific thermal image area called the Region Of Interest (ROI). The temperature values were collected at four different intervals: 0, 3, 6 and 10 s. The water content (%) was expressed gravimetrically as percentage of grams (g) of water on g of dry soil (θ g, water g/dry soil g). The water content reference measurements of samples were obtained through the official oven-drying gravimetric technique [

The soil surface temperature dynamic variation was acquired using a FLIR (S40) thermocamera [

As in the previous case, soil samples (N = 50) were collected in the CRA-ING experimental field at a depth of 30 cm. In addition to different initial levels of water content and temperature, different irradiance values were considered for all the samples. Irradiance was measured by a radiation sensor whose sensible element is a photodiode that converts incident radiation into a voltage (LP-9021 RAD, Delta Ohm, Padova, Italy). The signal is then acquired by a portable microprocessor-controlled multifunction quantum-photo-radiometric indicator with LCD indication (HD-9021, Delta Ohm). The sensor measures the flux of incident radiation in the spectral region spanning from 450 nm to 950 nm, ranging from 0 to 2,000 W/m^{2} and having a precision of ±3.5%. The portable indicator has a resolution of 0.1 W/m^{2} for values minor than 200 W/m^{2} and 1 W/m^{2} for values greater than 200 W/m^{2}.

Soil surface temperature dynamic variation was measured at four intervals (0, 3, 6 and 10 s) by an infrared thermometer measuring the amount of radiant energy emitted by the samples (IRtec P500, Eurotron, Milano, Italy). The instrument has a measurement range from −30 °C to 930 °C, a resolution of 0.1 °C and an accuracy of ±1% + 1 °C. Also in this case the water content reference measurements of samples were obtained through the oven-drying gravimetric technique [

Soil samples (N = 40) were collected in the CRA-ING experimental field facilities during three different days and times in order to obtain a high SWC and solar irradiance variability. The samples were collected on the surface of bare soil after deep ploughing (60 cm) [

For the datasets creation, the temperature dynamic variations were collected at four intervals (0, 3, 6 and 10 s), named hereafter t_{0}, t_{3}, t_{6} and t_{10}, respectively. For the presence of different specific heats between water and soil, the analyses were regulated with the addition of slopes obtained by interpolation values for each interval (t_{3} slope, t_{6} slope and t_{10} slope). The t-slopes were calculated from the initial temperature (t_{0}) to the final one, including all the internal steps (_{10} slope was calculated from the t_{0}, t_{3}, t_{6} and t_{10} values).

Three different datasets were hence created. The first in the laboratory method based on the active infrared thermography considered only the measurement of temperature variation as independent (observed) variable. The second laboratory one added the irradiation of soil samples as independent variable and it was based on active infrared thermometry. Finally, the in-field experiment was based on active infrared thermometry and it also considered as independent variables some meteorological parameters (

The SWC estimation in all the analysis, both in laboratory (_{3}, t_{3} slope, t_{6}, t_{6} slope, t_{10} and t_{10} slope for the thermographic analysis and only at the interval t_{10} slope for both laboratory and in-field thermometric analysis taking a temperature reading every second. For these two last analyses only the t_{10} slope was considered because it performed better than the t_{3} and t_{6} ones. This is due to the presence in-field of the environmental variables producing noises that can be lowered with a longer acquisition.

The procedure of PLS [

The predictive ability of the model is partially dependent on the number of Latent Vectors (LV) used and was assessed by the prediction efficiency parameters: Root Mean Square Error (RMSE), Standard Error of Prevision (SEP) and correlation coefficient (

In order to obtain more general and wide (

_{3}, t_{6} and t_{10}) and for the slopes obtained by value interpolation for each interval (t_{3} slope, t_{6} slope and t_{10} slope) with maximum

_{10} slope interval, assessed with maximum _{10} slope.

_{10} slope. _{10} slope considering three different classes of SWC (low < 11%; 11% < medium < 14% and high > 14%).

As reported by Bittelli [

Generally among the results shown by both laboratory and in-field applications, the best performing models were the t_{10} slope ones. The laboratory analysis showed that active infrared thermometry performed better than thermography, probably due to the variable represented by the irradiance present in the statistical model. This latter increased the correlation coefficient (

Since the best performing laboratory model was thermometry, we have chosen to use only this methodology for the in-field applications in order to measure SWC from the dynamic variation of surface temperature after deep ploughing and primary tillage. Both ploughing and tillage were used only for practical reasons. This in-field measurement is influenced by soil-atmosphere interactions as reported by Campbell and Norman [

In particular, the proposed in-field method performed worse than the laboratory ones (

Therefore, relationships between soil water and environmental factors need to be studied over wider time- and spatial-scales [

Finally, both thermography and thermometry are of fast execution and could produce highly informative results if paired with a Geographic Information System (GIS). Also, as reported by Schmidhalter

This study is part of the Ph.D. thesis of Francesca Antonucci on “Environmental sciences” (XXIV cycle) at the University of Tuscia (Viterbo), Italy.

Regression between observed and predicted values of soil water content for the intervals t_{10} slope in the independent test for the thermometric analysis (

List of the different X and Y pre-processing techniques applied in the analysis.

| |
---|---|

None | No pre-processing |

Baseline | Baseline (Weighted Least Squares) |

Abs | Takes the absolute value of the data |

Autoscale | Centres columns to zero mean and scales to unit variance |

Detrend | Remove a linear trend |

Groupscale | Group/block scaling |

mean center | Center columns to have zero mean |

median centre | Centre columns to have zero median |

Normalize | Normalization of the rows |

SNV | Standard Normal Deviate |

Centering | Multiway Center |

Partial Least Squares (PLS) results for the prediction of soil water content (SWC) obtained with laboratory thermographic analysis for the three time intervals (t_{3}, t_{6} and t_{10}) and for the slopes obtained by values interpolation for each interval (t_{3} slope, t_{6} slope and t_{10} slope). The table reports n° of Latent Vectors (LV); first and second pre-processing for the X-block and one for the Y-block; the correlation coefficient (

_{3} |
_{3} slope |
_{6} |
_{6} slope |
_{10} |
_{10} slope | |
---|---|---|---|---|---|---|

| ||||||

| ||||||

3 | 3 | 3 | 2 | 10 | 9 | |

autoscale | none | autoscale | autoscale | autoscale | median center | |

normalize | none | none | none | median center | none | |

median center | autoscale | none | autoscale | none | median center | |

0.3051 | 0.5765 | 0.6016 | 0.6113 | 0.7524 | 0.7756 | |

1.0476 | 1.2209 | 1.209 | 1.2606 | 1.5133 | 1.5804 | |

15.093 | 12.929 | 13.045 | 12.37 | 10.323 | 9.9083 | |

16.125 | 12.898 | 21.675 | 12.341 | 10.301 | 9.8858 | |

| ||||||

| ||||||

0.2993 | 0.5198 | 0.5784 | 0.542 | 0.7227 | 0.7417 | |

1.0322 | 1.1013 | 1.209 | 1.129 | 1.2163 | 1.4524 | |

16.682 | 15.438 | 15.407 | 15.335 | 14.729 | 12.314 | |

19.444 | 15.286 | 27.478 | 16.258 | 26.281 | 15.512 |

Results of Partial Least Squares (PLS) for the prediction of soil water content (SWC) obtained with laboratory thermometric analysis for the interval t_{10} slope. In the table are reported: n° of Latent Vectors (LV); first and second pre-processing for the X-block and one for the Y-block; the correlation coefficient (

| |
---|---|

| |

4 | |

autoscale | |

none | |

autoscale | |

0.7095 | |

1.4024 | |

2.125 | |

2.1001 | |

| |

| |

0.7634 | |

1.2868 | |

4.5316 | |

4.2138 |

Results of Partial Least Squares (PLS) for the prediction of SWC obtained with thermometric analysis performed in field for the interval t_{10} slope. In the table are reported: n° of Latent Vectors (LV); first and second pre-processing for the X-block and one for the Y-block; the correlation coefficient (

| |
---|---|

| |

5 | |

mean center | |

baseline | |

autoscale | |

0.6383 | |

1.2803 | |

1.6924 | |

1.6726 | |

| |

| |

0.6063 | |

0.9742 | |

3.6123 | |

3.3194 |

Results of Partial Least Squares Discriminant Analysis (PLSDA) for the in-field prediction of SWC obtained with thermometric analysis for the interval t_{10} slope considering three different classes of soil water content (SWC) (low < 11%; 11% < medium < 14% and high > 14%). N is the number of samples; n° units (Y-block) is the number of units to be discriminated by the PLSDA; n° LV is the number of latent vectors. Random Probability (%) is the probability of random assignment of an individual into a unit.

13 | |

19 | |

9 | |

3 | |

6 | |

100 | |

89.033 | |

86.667 | |

33.333 | |

12.143 | |

86.349 | |

88.889 |