Spatial Distribution of Soil Moisture Content and Tree Volume Estimation in International Institute of Tropical Agriculture Forest, Ibadan, Nigeria

The role of soil moisture in the survival and growth of trees cannot be over-emphasized and it contributes to the net productivity of the forest. However, information on the spatial distribution of the soil moisture content regarding the tree volume in forest ecosystems especially in Nigeria is limited. Therefore, this study combined spatial and ground data to determine soil moisture distribution and tree volume in the International Institute of Tropical Agriculture (IITA) forest, Ibadan. Satellite images of 1989, 1999, 2009 and 2019 were obtained and processed using topographic and vegetation-based models to examine the soil moisture status of the forest. Satellite-based soil moisture obtained was validated with ground soil moisture data collected in 2019. Tree growth variables were obtained for tree volume computation using Newton’s formular. Forest soil moisture models employed in this study include Topographic Wetness Index (TWI), Temperature Dryness Vegetation Index (TDVI) and Modified Normalized Difference Wetness Index (MNDWI). Relationships between index-based and ground base Soil Moisture Content (SMC), as well as the correlation between soil moisture and tree volume, were examined. The study revealed strong relationships between tree volume and TDVI, SMC, TWI with R 2 values of 0.91, 0.85, and 0.75, respectively. The regression values of 0.89 between in-situ tionship (R 2 = 0.03) with ground data. The strong relationships between soil moisture and tree volume suggest tree volume can be predicted based on available soil moisture content. Any slight undesirable change in soil moisture could lead to severe forest conditions.


Introduction
Forest Soil moisture content (SMC) is one of the important environmental factors that affect natural ecosystems (Briggs, 2016). Soil is the most treasured nonrenewable natural resource on earth and the most diverse part of the biosphere. Knowledge of soil moisture plays a crucial role in the field of hydrology (Bai et al., 2020;Gruhier et al., 2008), meteorology, climatology, ecology, land surface modeling, and studies in environmental changes (Gruhier et al., 2008;Verstraeten et al., 2006). Soil moisture content in forest ecology is influenced by the existing forest species, altitude, climatic conditions, density, age, and soil conditions (Briggs, 2016). According to Zwartendijk et al. (2017), the physico-chemical and biological soil properties such as temperature, ventilation, microbiological soil activity, nutrient uptake capacity, and accumulation of toxic substances are influenced indirectly by SMC. Xu et al. (2006), determined the contributions of soil water to forest biomass and the results of their study revealed a strong relationship between these ecological components.
Soil moisture is often expressed spatially as indices and such indices used include the Topographic Wetness Index (TWI), The Temperature-Vegetation Dryness Index (TVDI) and modified normalized soil water index (MNDWI) (Haas, 2010;Maselli & Chiesi, 2007;Seelig et al., 2008). The TWI was first developed in 1979 by Beven and Kirkby. It is based on surrounding topography and describes the proclivity of an area to become water-saturated. Remote Sensing based soil moisture retrieval has been a promising research field since the early 1970s. Sandholt et al. (2002), developed the TVDI that describes the relations between measured land surface temperature, vegetation and soil moisture, which could be expressed by the "triangle" method (Haas, 2010). These indices have shown to be reasonable estimators of surface soil moisture, although some questions of validity remained. However, there seems to be limited knowledge about soil moisture index-tree volume correlations and if one particular index can be considered to give superior results in the region of interest.
Accurate estimation of soil moisture content is crucial for planning and management of water resources, particularly in forest ecosystems where water is a major resource. However, information on the spatial distribution of the soil  (Egbinola & Amobichukwu, 2013). It is characterized by surface elevation between 160 m and 240 m above sea level, with a rainforest vegetation type.

Data Collection
Since there are no permanent sample plots in the IITA forest, temporary sample plots were adopted for this study as was done by Akindele (1991) Landsat images of 1989, 1999, 1999, and 2019 were downloaded from the official website of US Geological Survey (USGS). The study area is within Landsat path 191 and row 55. Table 1 shows the specifications of AsterDem, Landsat TM, ETM+ and OLI images used.
Soil samples were collected with the use of a soil core sampler at every angle and middle of each sample plot, so to achieve a good mixture of the overall soil sample of each plot. The data sets in Table 1 were used as inputs to the ArcGIS and Idrisi software. Model setup includes delineation of the study area, topographical modeling from Asterdem, vegetation indexes from Landsat images, and index-ground data comparison (Haas, 2010). The primary and secondary data were acquired and used for this study. The primary data collected were soil and biomass data from samples plots in the field, point coordinates to identify different sample locations for effective data modeling and comparison. The Soil samples were taken from five locations of each plot at 30 cm depth by using a soil core sampler. The five soil sub-samples were bulked together to make a composite soil sample for each plot. The moist soil samples were first sieved

Soil Moisture Retrieval in Remote Sensing
With readily available long-term remote sensing records and progress in digital image processing techniques as well as the tendency towards macro-scale modeling, remote sensing can provide large-scale distributed data sets where groundbased measurements are unavailable (Makinde & Agbor, 2019). Attempts have been made to combine remote sensing and hydrologic modeling (Bauer et al., 2006;Houser et al., 1998;Milly & Kabala, 1985). Studies have been carried to compare remotely sensed and simulated soil moisture over a heterogeneous watershed and found both techniques to be consistent with ground measurements (Haas, 2010). Soil moisture retrieval with remote sensing techniques can be achieved in all regions of the electromagnetic spectrum. Comprehensive comparisons between different retrieval techniques can be found in Bryant et al. (2003) and Moran et al. (2014). In the same vein, Haas (2010) reported soil moisture retrieval techniques and analysed their capabilities, advantages and disadvantages.
The following methods mentioned have been successfully used to model soil moisture and are briefly described:

Dryness Index (TVDI)
Studies have revealed that remotely sensed surface temperature as measured by thermal infrared (TIR) emissions and vegetation index has a strong relation with surface soil moisture Naira et al., 2007;Sandholt et al., 2002;Wang et al., 2007;Zeng et al., 2004). The relationship has proved to be important in obtaining further information. For instance, Nemani and Running (1989) and Price (1990) were the initial studies that made use of the temperature/vegetation relationship to estimate evapotranspiration. Numerous investigations regarding the validity of the relationship have been made and modifications to improve soil moisture estimates have also been tested (Carlson, 2007;Hassan et al., 2007;Kimura, 2007). The approach of a simplified land surface dryness index, often referred to as the "triangle" method was chosen in this study (Haas, 2010). It interprets the Ts/FCD space in terms of surface soil moisture status and is based on the assumption that remotely sensed surface temperatures are related to vegetation canopy cover. The TVDI was calculated with Equation (2) where T s = surface temperature; T smin = minimum surface temperature; b fcd = coefficient of vegetation index; a = the intercept.
The dry edge was developed from Equation (3) s fcd T s = surface temperature a, b define the parameters defining the dry edge of the triangle The forest canopy density fcd was derived using Equation (4) where SVD represents scaled vegetation index and SSI is the scaled shadow index calculated using a linear transformation function from normalized Advance vegetation index (AVI), Shadow index (SI) and Bare soil index (BI) (Agbor & Makinde, 2018).
The surface dryness index known as the "triangle" method as proposed by Sandholt et al. (2002) was used in this study to interpret the Ts/Fcd space in terms of forest canopy status. Sandholt et al. (2002) adopted NDVI as the vegetation index, but this study used FCD instead because NDVI is unable to highlight subtle differences in canopy density (Agbor et al., 2017).

Topographic Wetness Index (TWI)
The DEM was resampled to match the dimensions of Landsat images The DEM is the basis for TWI calculation and therefore requires the removal of spurious sinks and pits. TWI surface generation was performed using Equation (5) where Ln = natural log; FLOWACC = flow accumulationl

Modified Normalized Difference Water Index
The assessment of water status of vegetation canopies from spectral remote sensing data is a major goal in ecology and agriculture. Over the past decades, various studies have assessed whether soil water status, defined by leaf water content or canopy water content, can be measured using light reflected from leaves (Equation (6)) (Rokni et al., 2014).

Surface Classification
In order to explore the distribution of soil moisture content in the area, the TWI,

Data Validation
The sample points were compared to the corresponding pixel values from the generated surfaces using the regression analysis tool in SPSS. The SPSS was used to perform linear regression analysis by using the "least squares" method to fit a line through a set of observations. It explains how a single dependent variable is affected by the values of one or more independent variables. For example, it analyzes how tree volume is affected by soil moisture level. It also explains in this

Image Processing
In raw remote sensing data, each pixel has digital number value that corresponds to a raw measurement required by the sensor (Giannini et al., 2015). To obtain quantitative information from images, there is a need to convert images from their raw state to reflectance measures, using Equation (7) (Chander et al., 2009).

SUN sz
TOAr cos where: θ SZ = the solar zenith angle (degree). The cosine of this angle is equal to the sine of the sun elevation θ SE . therefore, θ SZ = 90 − θ SE . These are rescaling factors given in the metadata.

Temperature Retrieval
All the image bands are quantized as 8-bit data except Landsat 8 which is 16 bit, thus; all information is stored in DN which will then be converted to radiance with a linear Equation (8). The linear Equation (8) (Giannini et al., 2015;Makinde & Agbor, 2019) is given as: where Y = TOAr (Top of Atmosphere) radiance-the radiance measured by the sensor; m = Radiance multiplicative value; x = Raw band; b = Radiance additive value.
By applying the inverse of the Planck function, thermal bands' radiance values will be converted to brightness temperature values using Equation (8) (Agbor & Makinde, 2018 (1321.08 for OLI band 10).

Tree Volume Estimation
The volume of the trees was calculated using a common formula for volume computation, which includes; Newton's, Huber's, and Smalian's formulae of volume computation. The volume obtained with formulae were compared with volume obtained by summing up all the bolt of each felled tree for significant differences using SPSS 10.0 for window • Newton's formula (Avery & Burkhart, 1983) ( ) A. A. Alo et al.
where V = tree volume (m 3 ); h = total height of the tree (m); A b = Area at the base (m 2 ); A m = Area at the middle (m 2 ); A t = Area at the top (m 2 ).

Soil Moisture Distribution and Impacts on Tree Volume
Results of the soil moisture calculated with varying settings as previously described are shown in Figures 2-11. The presentation of soil moisture distribution in sub-pixels form is to establish the relationships between soil moisture from different models and calculated forest volume. Figures 12-17 Table 2. The prediction of tree volume using TVDI was poor while that of MNDWI was not considered since it exhibits a poor correlation value. Figure 13 and Figure   The Ts/Fcd values plotted against each other, as shown in Figure 18, result in a triangle whose edges represent either dry (low canopy and low evapotranspiration) or moist (high canopy and high evapotranspiration) conditions.
The dry edge was developed from Equation (10).
s fcd T a b = + (10) Figure 11 indicates a strong relationship between SMC, which is field-based soil moisture content and tree volume. Interestingly the satellite-based soil moisture level content relates strongly with field-based soil moisture content ( Figure   13 and Figure 14), and this made the prediction of tree volume possible.

Data Validation
The sample points were compared to the corresponding pixel values in the generated surfaces. Regressions values are shown in figures are 0.90 for TWI and 0.83 for the TVDI surface, respectively. Both correlation coefficients are rather high, although there seems to be a slightly higher correlation with the TWI surface. P-values of 0.02 for the TVDI correlation and 0.04 for the TWI correlation show that the sample size was large to achieve a statistically significant result. It is important to note that weather conditions (dry) at the time the samples were collected are similar to the ones of the satellite image acquisition date in 2019.

Conclusion and Recommendations
This study assessed soil moisture distribution using three different methods: Topographic Wetness Index (TWI), Temperature-Vegetation Dryness Index (TVDI) and Modified Normalized Difference Water Indexes (MNDWI) for IITA forest reserve. In-situ soil moisture was carried out and co-located to the derived moisture indices from satellite images. Index dependencies on in-situ soil moisture and tree volume were investigated and significant correlations were detected.
Only MNDWI showed weak correlations with in-situ measured soil moisture and tree volume. However, since both TWI and TVDI indexes correlate strongly with in-situ measured soil moisture and that of tree volume, this suggests that both methods can be used to model soil moisture and tree volume for this area.
The establishment of combined effective models for soil moisture determination over large areas requires more extensive in situ measurements and methods to fully assess the models' capabilities, limitations and value for hydrological and tree volume predictions.