Soil Colour Capture

Inspiration

Current land management techniques are reducing the levels of nutrients in the soil at alarming rates. Monoculture, the practice of repeatedly planting the same crop on the same farmland, removes nutrients from the soil that must be replenished with additional fertilizers. In Australia, "80% of farmed soils' carbon has been lost. The use of Nitrogen fertilizer has decreased in efficiency by 66% since 1970 and has caused lower nutrient uptake in crops and pastures." (YLAD, 2014, para 3). Considering farmers and graziers own 135,997 farms, covering 61% of Australia's landmass, this causes a significant problem for agriculture in finding accessible and viable agricultural land. (Europarl, 2015). Soil colour is a crucial factor in agriculture, often determining land for crop yield and efficiency. (Bot & Benites, 2005). Soil colour values can be valuable for predictively estimating total carbon (C) and total nitrogen (N). The ability to quickly and accurately measure carbon and nitrogen levels in soils would greatly aid agriculture by rapidly categorizing soil types. (Johns, 2017). Soil colours are most conveniently measured by comparison with a Munsell soil colour chart (MSCC). The one generally used is a modification of the Munsell colour chart that includes only the portion needed for soil colours, about one-fifth of the entire range of colours. The arrangement is by hue, value, and chroma (HVC) – three simple variables combined to give all colours. The method of manually comparing the MSCC to soil samples is subjective, and the results can widely vary between users (Milotta, 2020). Researchers have established three significant difficulties in obtaining an unbiased classification of colour using Munsell soil colour charts (Post et al., 2006; Rabenhorst et al., 2014; Kirillova et al., 2018)

Colour discernment varies between users due to experience and colour perceptions.
The colour chips in the Munsell charts do not fully cover the range of natural soil colours.
A printed colour cannot be accurately reproduced, and colour chips may differ between copies of the charts. Therefore an automated, reliable solution was required that will produce repeatable results independent of the users or devices implemented

What it does

This aim was to develop an Android application that can process images to read the Munsell colour values for soil colour analysis and the calculation of soil properties. The completed application can utilise captured images to calculate the dominant RGB colour value and then apply an algorithm to find the closest matching Munsell soil colour value from a samples database. It will then be able to calculate a range of soil properties including pH, Carbon, Nitrogen and Moisture. This type of application could potentially be used in a range of fields including, but not limited to, farming, wastewater management, forestry services, environmental protection services, colour matching services, colour space theory research and soil analysis.

How we built it

We built the app using Android Studio and Java.

Challenges we ran into

The feature of utmost importance required from an application of this type should be its precision and the repeatability of results. The colour values returned from the soil samples must be as close to a match as possible to the MSCC. The need for an automated system for comparing soil samples to the MSCC becomes apparent when manually comparing the MSCC to soil samples. This manual comparison is very subjective, and results can widely vary between users (Milotta, 2020; Bloch et al., 2021). Several other factors affect the accuracy of readings that have been identified when researching the possibility of utilising smartphones as an alternative to the MSCC human comparisons. The main two identified issues that affected accuracy included lighting and moisture variables. We have learned that further research and development are needed for the prototype app to enhance the accuracy and capability of the app.

Accomplishments that we're proud of

We are most proud of the potential of our app to be used world wide as an easy and affordable method of soil property analysis. Currently, there are a range of devices that provide soil property analysis and whilst these devices proved accurate in terms of matching soils samples to the MSCC. They are also expensive and too cumbersome for use in the field. The number of smartphone users will reach 5.22 billion by the end of 2021, which represents 66% of the world's population (DataReportal, 2021). Due to the availability and affordability of these devices, they are the obvious choice for soil colour research. More meaningfully, soil colour measured with smartphone cameras might also be used to predict other essential soil properties (soil organic matter content and soil fertility) (Wills et al., 2007). This validated the method of creating a mobile application that analyses soil colour and different soil properties.

What we learned

Due to the issues affecting accuracy, further research is needed in this area to determine how to ensure the reliability and accuracy of samples. Several areas of interest have arisen from this app development, including illumination effects, moisture effects, the relationship between soil colour and other soil properties and the use of different colour spaces. There is contesting evidence that the lighting values of the soil samples affect the determination of the Munsell soil colour classification. For this reason, it is advised that further research be undertaken in this area. An illumination box could be created to block out all-natural light similar to the studies of (4, 5). Illumination could then be increased/decreased via the use of an inbuilt light source on the same soil sample to quantify the effects of illumination on the chosen colour space values. Furthermore, by utilising a reference image that is a white card to display the illumination levels, one could then develop an algorithm to compensate the "L" lightness value of the CIELAB colour space. Milotta (2020) has proposed a similar idea in mentioning their intentions to create a program that will automatically recognise a reference marker (a White Calibration Plate CM-A145, or a Macbeth Color Checker) for use in an automatic illumination calibration phase. Pegalajar (2020) also support this idea hypothesising that "a reference gamut might improve the HVC approximations." (pg. 52). Han (2016) also used a colour reference card to calibrate samples specifically designed to deduct "colour drift" and external interference effectively. Further research on the effects of moisture should also be undertaken. These could include taking soil samples and introducing water to gauge the impact of moisture percentage on the colour changes of models. Schmidt (2021) advocates a similar approach hypothesising that samples could be saturated with water into a solution to remove the effects of moisture by hydrating all samples to the same level to achieve a baseline reading in terms of moisture. Whilst the purpose of this project is to create an application that can be deployed into the field and used in-situ, the preparation of samples before analyses could prove a useful technique to remove several of the inaccuracy issues. Schmidt (2021) has proposed the process of baking soil samples to remove moisture before analysis. In the study conducted by Fu (2020), "SPC images were pre-processed using illumination normalisation to avoid illumination inconsistencies and segmentation techniques employed to remove non-soil parts of the images including black cracks, leaf residues and specular reflection before modelling." (pg. 3). Schmidt (2021) also remarks that in the field a "higher priority should be placed on creating a smooth surface for each soil sample using a knife while minimising the mixing of colours present on the soil surface." (pg. 10). The relationship between soil colour and other soil properties also needs to be explored, including soil organic matter (SOM). Fan (2017) supports this, stating that "further studies are needed to explore and establish the relationships between soil colour and other soil properties (e.g., soil organic matter content)." (pg. 1145). Stiglitz (2017) lists this as a priority for the expansion of their GPS mapping soil colour analysis solution, stating that the "soil colour application may be extended in the future to include data entry options for other soil, and land cover attributes to further augment spatial databases." (pg. 114). These" land cover attributes" include SOM. Kirillova suggests a novel reason for the lack of accuracy in soil colour readings. Attributing the lack of MSCC colours to the misreading of values. “ A major disadvantage of the MSCC is the lack of a sufficient number of chips for precise colour determination.” (Kirillova, 2018, pg. 381). Indeed, only 437 colours are represented in the MSCC. This is only a portion of the available colours represented in the Munsell colour space. The full-colour space could easily be implemented by including the interpolation works of Centore (2013) and Hemink (2021). While these values fall outside the MSCC, they could prove helpful in colour calibration and testing procedures.