Remote Sensing Based Soil Moisture Estimation for Agricultural Productivity: A note from Lake Tana Sub Basin, NW Ethiopia

Abstract

Soil moisture availability affects agricultural productivity and in turn food security. Estimating the moisture content of soil is imperative for proper water resource management and agricultural productivity. However, field based method is expensive and covers limited spatial variation. The advancement of remote sensing technology eases the soil moisture estimation over large geographic area. Hence, this study intended to apply the optical and thermal remote sensing data for estimating SM in the Lake Tana sub basin. Temperature vegetation dryness index (TVDI) model which is used in this study to estimate soil moisture is derived from the wet and dry edge of the LST-NDVI triangular scatterplot. The finding revealed that NDVI and LST have inverse relationship where LST decrease with increasing NDVI. Spatially, northern and north western part has experienced high LST. The estimated soil moisture result ranging from 0 to 1 where the soil moisture is higher in areas with TVDI value is near 1. Thus, soil moisture is higher in the east, and northeast part of the sub basin whereas the central, western and northwest part experienced low soil moisture. Therefore, applying remote sensing enables estimation of soil moisture across large geographical area with scarcity of field data (in-situ observations).

1. Introduction

Soil moisture (SM) is defined as retained water within the unsaturated soil zone [1] which serves as an essential variable for hydrological management and understanding the energy balance of terrestrial water [2]. It makes up 0.15% of the global fresh water [3] and is affected by factors such as precipitation, temperature, and other soil characteristics [4].

Accurate estimation of SM is of great relevance to a wide range of disciplines and practical applications. It has high relevance to different bio-physical processes related to the exchanges of energy and mass between the hydrosphere, atmosphere, and biosphere [5]. Soil moisture has also been recognized as a key state variable within the global energy cycle due to its control on the partitioning of available energy at the Earth’s surface into latent (LE) and sensible (H) heat exchange [6].

SM can be estimated using field-based point measurement or spatial distribution using remote sensing. A field based SM measurement includes gravimetric, neutronic, time-domain refractometry, capacitance, tensiometer, and hygrometric techniques [7]. These methods are the most reliable approach for determining soil moisture content in terms of its accuracy [3]. Despite the many advantages of point measurement techniques, they are destructive, time taking, have limited coverage and not able to estimate temporal soil moisture [8, 9]. Whereas, remote sensing based soil moisture estimations which is based on the basic principle of energy absorbed or reflected from the surface to the sensor defines the amount of moisture content present in the soil. The advent of spaceborne remote sensing and the development of sensors and algorithms since the late 1970s [10] become a credible approach for securing SM data with spatial and temporal consistency [11, 12]. The remote sensing based approach provides estimated moisture content of the soil from local to global scales efficiently with little cost of time and effort [13].

A variety of space borne remote sensing based techniques from optical and both active and passive microwave remote sensing satellites have been utilized to analyze the spatio-temporal dynamics and distribution of soil moisture properties across a broad range of scales [6]. These can be classified as microwave and optical-thermal techniques. Microwave RS techniques have shown greater potential for monitoring global scale soil moisture dynamics because microwaves can penetrate through vegetation canopy and underlying soil, especially at lower frequencies [14]. The advanced microwave sensors like microwave scanning radiometer (AMSRE and AMSR2), the advanced scatterometer (ASCAT), soil moisture and ocean salinity (SMOS), and soil moisture active passive (SMAP) have been successfully used by Kolassa et al. [15] and Rodriguez-Fernandez et al. [16] to retrieve soil moisture. However, the soil moisture products from these sensors have a coarse spatial resolution (more than 25 km) which is not suitable for applying in small scale and heterogeneous environments [17].

On the other hand, thermal-optical remote sensing-based soil moisture estimation applies the visible, near infrared, shortwave infrared, and thermal infrared spectral region. The most common models used to estimate soil moisture using optical and thermal wavebands are shortwave infrared water indices (SPSI) [18], shortwave infrared water stress index (SIWSI) [19], vegetation condition index (VCI) [20], thermal inertia method (TIM) [21], and temperature vegetation dryness index (TVDI) [22].

These methods have achieved good results in estimating SM, however, some of them have certain limitations based on land surface conditions. The use of satellite-based vegetation indexes (VIs) extracted from the optical bands, such as the vegetation condition index (VCI) [20], normalized difference vegetation index (NDVI), and enhanced vegetation index (EVI) [23], have proven useful for monitoring soil moisture [10]. The method uses the land surface temperature (LST), extracted from the thermal infrared bands, and vegetation indices from optical bands to monitor soil moisture which was originally proposed to monitor canopy water stress. Studies [24, 25, 26] have confirmed that the combination of LST and VI provides better information on the surface soil moisture conditions. Thus, to improve the deficiency in estimating profile soil moisture distribution over heterogeneous conditions, this study proposed an approach to estimate profile soil moisture by integrating high resolution remote sensing LST and NDVI products.

2. Methods and materials

2.1 Description of the study area

This study was conducted in the Lake Tana sub-basin which is found in the Upper Blue Nile Basin (UBNB) covering a total catchment area of 15,140 km² (Figure 1). It is the second largest sub-basin in the Upper Blue Nile which has immense potential for irrigation and has high value crops. Beside the majority of the sub-basin has been designated as a key economic growth corridor in the Ethiopian Growth and Transformation Plan, and it is also part of UNESCO’s world biodiversity reserve [27]. Lake Tana sub-basin has high potential for agriculture, livestock, water resources, forest and wildlife, tourism, and fishery development. The climate of the study area ranges from temperate to cool semi humid zone (75%) and the remaining 25% is found in cool to cold humid zone [28]. The climate structure of the sub basin is divided into rainy and dry seasons. Annual rainfall distribution of the basin ranges from 964 mm up to 2000 mm. The mean annual maximum and minimum temperature is 25.5°C and of 15°C, respectively. Topographically, it lies over the area altitudinally ranging from nearly 4000 masl (at Guna mountain) to 1700 masl (at the cost of the lake and Fogera-Dembia plain) which consist gro-ecologically it consist from the cool zone (Dega) and sub-tropical zone warm and wet (Woyina Dega) agro-ecological regions in the basin. Vertisols dominate the lower plain of the basin whereas lower plateau, foot of sloppy areas and mountain systems are dominated by Cambisol and Luvisols nitosols, lithosols, and regosols [29].

Figure 1.

Map of Lake Tana sub basin.

2.2 Data sets

To estimate the spatio-temporal variation of soil moisture content of Lake Tana sub basin, both optical and thermal remote sensing data sets were used from USGS. A cloud free Landsat 8 OLI/TIRS from October up to December 2021 were used for the study. Since the temporal resolution of the Landsat satellite is 16 days, two images per month totally six images were obtained from the USGS archive (https://earthexplorer.usgs.gov/). The visible (Band 4 Red (0.64–0.67 μm)) and near-infrared (Band 5 (0.85–0.88 μm)) optical categories were used for calculating NDVI and the thermal Band 10 and Band 11 were used for estimating land surface temperature.

2.3 Estimating vegetation indices

Vegetation indices are dimensionless, radiometric measures that function as indicators of relative abundance and activity in green vegetation. Normalized difference vegetation index (NDVI) is the parameter in thermal optical trapezoidal models (TOTRAM) that can be computed using the different bands of Landsat 8. NDVI was calculated from satellite images by converting the spectral radiance planetary reflectance (DN) to top of canopy reflectance (TOC) by using the reflectance rescaling coefficients. Computational equation of NDVI is described in Eq. (1) below:

$$\mathit{\text{NDVI}}=\frac{\mathit{NIR}-\mathit{Red}}{\mathit{NIR}+\mathit{Red}}$$E1

Where: NDVI is normalized difference vegetation index, NIR is near infrared.

2.4 Land surface temperature estimation

LST is a key factor in the TOTRAM model for determining the soil moisture as it has relation with vegetation indices and which gives the measure of root zone soil moisture in vegetation. Computation of LST could be performed using the thermal bands of Landsat 8, that is, Band 10 and Band 11. Hence the equation used is as follows (Eq. (3)):

$$\mathit{\text{LST}}=\frac{\mathit{BT}}{\left\{1,+,\left[\left(\mathit{\lambda },\frac{\mathit{BT}}{\mathit{P}}\right),\mathit{*},\mathit{LSE}\right]\right\}}$$E2

Where: λ = wavelength of emitted radiance. BT = Brightness temperature. LSE = Land surface emissivity. p = h*c/s = 14380 mK. H = Plank’s constant (6.626*10–34 Js). S = Boltzmann constant (1.38*10–23 J/K). C = Velocity of light (2.998*108 m/s).

2.5 The thermal-optical trapezoid model (TOTRAM)

The thermal-optical trapezoid model is a widely applied approach to estimate soil moisture based on synergy of thermal (i.e., land surface temperature) and optical remote sensing satellite observations [30]. It is based on the interpretation of the pixel distribution within the LST-VI space, where LST is the land surface temperature and VI is a RS-based vegetation index [24, 25, 26] were among the first to apply the LST-VI space for estimating surface soil moisture. The basic principle of the model is if a sufficiently large number of pixels exist and cloud and standing surface water pixels are removed from the pixel distribution, the shape of the pixel envelope resembles a triangle or a trapezoid [24]. We employed the Thermal-Optical Trapezoid Model described in Figure 2 to estimate the soil moisture in heterogeneous topographic regions. Using this method relative soil moisture was estimated using the following equation derived by Sandholt et al. [22].

$$\mathit{SMI}=\left(\frac{\mathit{\text{LSTmax}}-\mathit{LST}}{\mathit{\text{LSTmax}}-\mathit{\text{LSTmin}}}\right)$$E3

Figure 2.

Schematic illustration of the TVDI method (adapted from Sandholt et al. [22]).

Where LST is the observed LST at a specific pixel and LSTmax and LSTmin are maximum and minimum surface temperatures for a given NDVI. The surface temperature of a pixel for a given NDVI derived using remote sensing data. LSTmax and LSTmin were calculated using the following formulas (Eqs. (4) and (5)).

$$\mathit{\text{LSTmax}}=\mathrm{sd}\ast \mathit{\text{NDVI}}+\mathrm{id}$$E4

$$\mathit{\text{LSTmin}}=\mathrm{Sw}\ast \mathit{\text{NDVI}}+\mathrm{iw}$$E5

Then relative soil moisture can be estimated by combining LSTmax and LSTmin (Eq. (4) and (5)).

$$\mathit{SMI}=\left(\frac{\mathit{Id}+\mathit{Sd}\ast \mathit{\text{NDVI}}-\mathit{LST}}{\mathit{Id}-\mathit{Iw}+\left(\mathit{Sd}-\mathit{Sw}\right)\mathit{\text{NDVI}}}\right)$$E6

Where Id and Iw are intercepts in linear regression equation at dry and wet edges, and Sd and Sw are slope values in the regression equation at dry and wet edges. NDVI is normalized difference vegetation index and LST is land surface temperature.

Hence, using the SMI equation (Eq. (6)) the soil moisture extracted based on the TOTRAM model (Figure 3).

Figure 3.

Analytical framework of soil moisture estimations.

3. Results

3.1 NDVI and LST time series

In the present study, the spatio-temporal variation of NDVI and LST of Lake Tana Sub basin from October to December was derived from the Landsat 8 multispectral images (Figures 4 and 5). The vegetation cover had a significant impact on surface temperature (LST) as high values of LST were associated with low NDVI values. The low values of NDVI and high LST is associated with bare land and sparse vegetation cover. The NDVI values reached a maximum value (0.89) during the 1st weeks of November and declined for the forthcoming months. This this probably because the moisture content of the soil declined as the temperature increased which increase the rate of evapotranspiration. Thus, the higher values of LST correlate with regions with a lower vegetation cover. Spatially, the maximum values of LST were observed in the northern and north western part of the sub basin which has relatively low elevation and inversely the LST minimum values were observed in the southern and eastern regions of the sub basin where dominated by higher topography (about 4000 masl). As Mekonnen [31] reported elevation and topography highly affects land surface temperature and soil moisture estimations. Similarly, Liu et al. [32] has confirmed that surface temperature is considerably affected by altitude and usually assumed to decrease in a linear manner.

Figure 4.

NDVI 16 days map estimated from Landsat 8 OLI data from September 25, 2021 to December 21, 2021.

Figure 5.

LST time series maps estimated from Landsat 8 TIRS data from September 25, 2021 to December 21, 2021.

3.2 Estimating wet and dry edge for TVDI

So as to set the parameters describing dry and wet edges, maximum and minimum temperatures computed for different intervals of NDVI were extracted from LST-NDVI triangular spaces. The dry and wet edges for each month under study (15 September 2021 to 30 December 2021) were determined from LST-NDVI scatterplots to obtain TVDI (Figure 6) and the slope and intercept (Table 1) from the linear regression model. The cluster of pixels with uniform NDVI values was considered to eliminate extreme values for building a true relationship between LST and NDVI. From the scatter plot NDVI and LST have negative relationships for most of the pixels found in triangle created, as NDVI increases the LST decreases and vice versa. The findings were in line with the concept that LST decreases with an increase of NDVI [33]. Similar findings have been reported by Sadeghi et al. [30] and Liu et al. [32] which reveal a strong negative relationship between land surface temperature and the normalized difference vegetation index (NDVI) for all biome types. As vegetation greenness increases the NDVI value increases and surface temperature will decreases.

Figure 6.

Scatter plot between NDVI vs LST in oct 2021, Nov 2021 and December 2021.

Date	Wet edge	R^2	Dry edge	R^2
Oct 02 2021	y = −1.21x + 18.92	0.89	y = 0.34x + 8.87	0.75
Oct 18 2021	y = −11.08x + 28.84	0.85	y = −24.45x + 43.68	0.91
Nov 03 2021	y = −16.42x + 34.85	0.90	y = −12.62x + 41.56	0.83
Nov 19 2021	y = −20.84x + 36.33	0.87	y = −23.62x + 45.28	0.86
Dec 05 2021	y = −0.94x + 22.76	0.94	y = −11.31x + 37.86	0.91
Dec 21 2021	y = 0.85x +20.62	0.78	y = −24.13x + 44.23	0.85

Table 1.

Wet and dry edges in NDVI vs. LST space estimated by linear regression.

3.3 Soil moisture estimation

Soil moisture (SM) is estimated from remote sensing data by relating NDVI and LST derived from Landsat 8 using the trapezoid model. Studies by Liu et al. [32] and Gu et al. [34] have proven the potential of using the trapezoid model for SM estimation at large scales. The trapezoid models well fit the image data with clear boundaries defined by the wet and dry edges as indicated by the high R² values of the linear regression model for the dry and wet edges (Table 1). The wet edge for each date was calculated from LSTmin values and dry edge was calculated from LSTmax values.

The estimated TVDI result represents the sub surface soil moisture which the value ranges from 0 to 1 (Figure 7). Though the values sometimes recorded less than 1, which is in agreement with a previous finding of Ryu et al. [35] and Sandholt et al. [22] that TVDI values often exceed 1 in regions where LST is rapidly increasing because LST highly affects TVDI values. However, values near to 1 indicate regions with a higher amount of soil moisture and those close to 0 represent areas with a high surface temperature and low level of soil moisture. In the resent study, soil moisture shows both spatial and temporal variability across the sub basin. SM was higher in the east, and in the northeast part of Lake Tana sub basin. The central, western and northwest part of the sub basin has experiencing low SM [34] confirms SM varied significantly across the sub basin, which could be attributed to environmental factors, including topography, vegetation and ground cover, and soil properties, especially in soil texture ranging from 0 to 15 cm depth.

Figure 7.

Soil moisture maps generated with TOTRAM (a) September 25, 2021, (b) October 11, 2021, (c) November 3, 2021, (d) November 19, 2021, (e) December 05, 2021, and (f) December 21, 2021.

4. Conclusion

The recent advancement of remote sensing technology enabled to estimate the spatial and temporal distribution of soil moisture for proper farm scale management. Satellite data, which were acquired from optical/thermal (Landsat 8) satellite sensors was used to retrieve soil moisture. The LST/VIs method which is a remote sensing-based approach was employed for estimating SM in Lake Tana sub basin. The estimated SM value ranges from 0 to 1 although the values sometimes exceeds 1, in in regions where LST is rapidly increasing because of LST highly affects TVDI values. The value significantly varies across the sub basin, which could be attributed to environmental factors, including topography, vegetation and ground cover, and soil properties. The study proof the use of optical-thermal remote sensing data for mapping soil moisture in a large geographical area in a high spatial resolution (30 m), however the high cloud cover during the late September and October affects the SM estimations. The study recommends further studies on the basin using high resolution radar images for SM estimation during the cloudy season.

Conflict of interest

The authors declare that they have no competing interests.