Proxy quantification and mapping of seasonal rangeland herbage using grass cover in the cattle corridor of Uganda

Abstract

This study aimed at estimating herbage mass functional relationship with rainfall and selected soil morphological properties using Geographic Information Systems (GIS). Herbage mass as dry matter (DM) measurements from quadrat harvesting method and estimations based on grass cover were done. Herbage on 1 × 1 m quadrats was harvested and oven dried at 60°C until constant weight. Grass cover data collection was based on spectral strata of vegetation cover types obtained from Landsat image. Rainfall data were collected from 14 rain gauge stations. Soil profiles were described and their drainage was determined. A multi-linear regression analysis was performed to determine the relationship among DM, grass cover, rainfall effectiveness and soil drainage. The relationships among the variables were integrated using ModelBuilder in ArcGIS 9.3 to come up with spatial and seasonal herbage distribution. Results showed a strong relationship among the above variables and DM (R ² = 0.76; p < 0.001). Grass cover was found to be the important variable for herbage DM determination (R ² = 0.73; p < 0.001). A correlation analysis of estimated and harvested DM resulted in a positive relationship (R ² = 0.85). Estimated DM ranged between 895 and 1923 kg/ha with the highest DM during March–May and the lowest during September–November. This study demonstrated that grass is an important proxy measure of spatial and seasonal patterns of rangeland herbage mass. Therefore, rangeland herbage can in reality be monitored based on grass cover measurements to avoid or at least minimize the cost, destruction and information timeliness implications that are known to be associated with harvesting methods.

Keywords:

• rangeland herbage
• grass cover
• rainfall
• Landsat image
• GIS

1. Introduction

Herbage production in rain-fed rangelands is characterized by seasonal and annual disparities which are generally attributed to the erratic nature of rainfall (Tueller 1993; Fujita et al. 2009). Such changeability calls for continuous quantification and monitoring of herbage availability so as to provide reliable information for livestock and wildlife management. The recent common and unpredictable occurrences of droughts and the associated deaths of livestock and wild animals in Ugandan rangelands require quick and accurate management responses based on reliable spatial and temporal information. Knowledge of patterns of herbage production in space and time is essential for monitoring and predicting its availability and quality. With such understanding, the likely responses to herbage variations can easily be explored for sustainable utilization of rangeland ecosystems (de Ridder and Breman 1993; Teague and Foy 2002). Rangeland resource inventory and monitoring information is needed for decision-making on land use and promoting practices and strategies that minimize the impact of climate variability and change (MoLHUD 2007).

Rangeland herbage quantification and monitoring processes require methodological approaches that provide reliable information for decision-making and management processes (Reeves, Winslow, and Running 2001). Martin, Astatkie, and Fredeen (2005) highlight the challenges of identifying a reliable and consistent method for monitoring and predicting herbage mass in a variety of field conditions. These challenges include differences in sampling dates that affect accuracy, differences in instrument calibrations for different harvest occasions, inconsistencies in results (Roy and Ravan 1996) and rigorous and expensive data collection procedures (Li Jianlong and Quangong 1998; Vermeire and Gillen 2001; Martin, Astatkie, and Fredeen 2005).

Several studies have attempted to use remote-sensing (RS) imagery and Geographic Information Systems (GIS) to map rangeland vegetation and herbage mass (Liang and Chen 1999; Price, Guo, and Stiles 2002; Moreau et al. 2003; Reeves, Zhao, and Running, 2006; Zhang et al. 2006). It is apparent that RS and GIS technologies have proved to be efficient tools that enable decision-makers to address grazing land environmental and developmental issues. RS along with GIS can contribute information for a variety of rangeland resources management and estimation applications. Annual or inter-annual maps of herbage can guide decision-makers to understand how rangelands vary spatially and temporally (Woodward, Wake, and McCall 1995; Beeri et al. 2007).

It has been demonstrated that there is a relationship between satellites-measured spectral responses and plant biomass (Roy and Ravan 1996; Price, Guo, Stiles 2002). However, most studies have extensively been centred on low-resolution satellite sensor data, especially National Oceanic and Atmospheric Administration-Advanced Very High Resolution Radiometer (NOAA-AVHRR) and the Moderate Resolution Imaging Spectroradiometer (MODIS). Satellite sensors like Landsat, ASTER, SPOT, IKONOS and Indian Remote Sensing Satellite (IRS), which provide relatively larger-scale spatial information, are not yet fully exploited for the purposes of understanding rangeland herbage dynamics (Reeves, Zhao, and Running, 2006). Forage mapping is usually derived from global public domain datasets that do not reflect landscape differences in radiation use efficiency, which can vary significantly across landscapes and between rangeland sites. As such, productivity estimates over areas with variable weather conditions and landscape characteristics may be subject to error.

Beeri et al. (2007) highlight the need for research that integrates RS and ground-based data including herbage quantity-related measurements to establish forage availability for large herbivores. Grass cover measurement is one of the key indicators of herbage quantity at a given rangeland site (Mueller-Dombos and Ellenberg 1974). Understanding spatial patterns of grass cover is thus very important in estimating herbage available for grazers (Castillo et al. 2009). Grass cover observations based on satellite images can facilitate large-scale hypothesis testing necessary for scaling up spatial and temporal information on rangeland condition from field plots to eco-regions (Beeri et al. 2007). Such information is useful to livestock owners, agricultural extension workers and wildlife managers in determining how many, and when, animals can be grazed on specific sites.

In this study, an attempt is made to quantify spatial and seasonal patterns of herbage mass using grass cover proportions of vegetation spectral strata derived from Landsat Enhanced Thematic Mapper. It is assumed that herbage production (expressed as grass cover) is a direct result of interactions among rainfall, soil and topographic characteristics (van Wijngaarden 1985; Canton et al. 2003; Santos et al. 2003). Here, the differences in the site interactions are referred to as rainfall effectiveness. The specific objectives of this study were (1) to estimate spatial and seasonal herbage mass from grass cover and its functional relationship with rainfall and soil physical properties and (2) to test whether grass mass measurements from the quadrat harvesting method are significantly different from proxy herbage estimations based on grass cover.

2. Materials and methods

2.1. Study area description

This study was conducted in two selected areas of the Ugandan cattle corridor, namely Kiruhura and Rakai districts (30° 53′ E and 0° 13′ S to 31° 14′ E and 0° 41′ S), South Western Uganda. This area is part of the ‘cattle corridor’ where the majority of the population depends on livestock for most of their livelihood. The area is important for the grazing of both wild and domesticated ungulates. Rainfall varies considerably in space and time and is rather unpredictable. It generally occurs in a bimodal pattern with mean annual rainfall (1975–2009) of 948 mm with peaks during March–May and September–November, which are usually the growing seasons. The months of June–August are usually very dry. In January and February, little rainfall is usually received. Mean annual temperature is 22.9°C. Daily temperatures fluctuate between 12°C and 34°C with negligible seasonal fluctuations. Vegetation cover is mainly composed of alternations of woodland and grassland patches with varying proportions of herbaceous and woody species. The dominant woody vegetation species included Acacia sp., Rhus natalensis and Carissa edulis (personal observations). Dominant grass species are Themeda triandra, Cynodon dactylon, Panicum maximum, Brachiaria decumbens, Sporobolus pyramidalis, Laudetia kagerensis and Hyparrhenia filipendula (Okello, Sabiiti, and Schwartz 2005). The soils are mainly classified as Leptosols, Acric ferrasols, Lixic ferrasols, Gleyic arenosols, Gleysols and Histosols (FAO 1990).

2.2. Data collection, preparation and analysis

2.2.1. Grass cover sampling

Grass cover data collection was based on Landsat TM image spectral strata. The number of strata was obtained based on experience from field reconnaissance and visual inspection of different combinations of 5, 4, 3 and 2 bands. The image was stratified into 10 spectral pattern cover classes using unsupervised classification in ERDAS IMAGINE 9.1. The classes (strata) were verified and modified using vegetation survey data from the field to ensure that sampling strata belonged to similar and realistic cover types in the field. Two of the cover classes were water and wetland. Eight other classes were used as the sampling strata for grass cover data collection. The grass cover data from the strata were collected using clustered random sampling (Figure 1). Sampling sites were selected (clustered) in such way that a maximum number of strata were represented at a given data collection area (cluster) as a way of minimizing the time spent on movement between strata (Mueller-Dombos and Ellenberg 1974). All strata of minimum mapping size of 60 m × 60 m (16 pixels) or greater (Townshend 1983) were considered for grass cover sampling irrespective of their size. Sampling point locations for the grass cover were randomly selected and saved in a Garmin 12 Global Positioning System (GPS) for navigation during field data collection. For each stratum, at least 12 quadrats of 2 m × 2 m were placed every month from December 2008 to November 2009. In each quadrat, percent grass cover was estimated and recorded. Percentage cover estimates were done using calculated percentage cover charts (Kercher, Frieswyk, and Zedler 2003). Grass cover data for the different strata and months were averaged according to seasons (Table 1). The seasons used were December–February, March–May, June–August and September–November. For each season, the averaged grass cover percentages were used to assign grass cover classes to all the eight vegetation strata to come up with seasonal herbage cover maps (Figure 2).

Table 1. Average seasonal grass cover

Season	Grass cover (%)
December–February	53.4
March–May	58.1
June–August	56.3
September–November	51.4

Figure 1. Image stratification and location of sampling sites in the different vegetation strata.

Figure 2. Spatial and seasonal patterns of herbage cover.

2.2.2. Rainfall data

Fourteen rain gauge stations were set up in the study area. Recording of rainfall (in mm) was done every morning at 09:00 for the whole period of grass cover data collection. The rainfall for each station was computed for monthly (Figure 3) and seasonal totals (Table 2). The station seasonal rainfall data were interpolated using the spline tool in ArcGIS 9.3 to estimate rainfall values for every 30 m × 30 m grid on a map (ESRI 2010).

Table 2. Seasonal rainfall totals

Season	Rainfall (mm)
December–February	194.6
March–May	202.3
June–August	58.3
September–November	302.1

Figure 3. Monthly rainfall distribution during the study period.

2.2.3. Soil and rainfall effectiveness data

Rainfall effectiveness was based on the premise that soil water availability for herbage growth at a given rangeland site is directly related to rainfall and water holding capacity properties of soil. During rainfall, water may infiltrate or percolate through the soil profile or run off the soil surface to be lost from a specific site depending on soil characteristics. Such differences are associated with differences in rainfall effectiveness (McGinty, Thurow, and Taylor 1991) and herbage production at a given location (Santos et al. 2003). In order to classify soils according to rainfall effectiveness groups, soil profiles were described following the standard methodology (FAO 1990). Sampling was based on strata resulting from GIS spatial overlay of the six FAO soil and two slope (%) classes (≤8.65 and >8.65). The source of elevation data for slope classes was a 90-m resolution digital elevation model derived from Shuttle Radar Topographic Mission . For each of the resulting 12 soil-slope combination strata, at least three sites were randomly sampled for soil profile description and soil sample collection for texture, hydraulic conductivity and bulk density determination in laboratory. Soil depth was described in the field as very shallow (0–15 cm), shallow (15–30 cm) or very deep (>30 cm) using mini-pits of 1 m × 1 m dug in each of the sampled strata. The laboratory samples were analysed using methods described by Okalebo, Gathua, and Woomer (2002). The results were used to classify the FAO soil map units into soil drainage (rocky, poorly drained or well drained) categories. Combinations of soil profile and drainage classes were grouped into three soil physical groups (Table 3) according to FAO (1990) and assigned rainfall effectiveness factors as described by van Wijngaarden (1985). The effectiveness factor for a soil map unit was an outcome of a combination of its drainage, slope and soil depth properties as determined above. Each soil map unit was subsequently assigned respective rainfall effectiveness value to come up with a rainfall effectiveness map. The higher the value of the effectiveness factor, the higher the level of expected herbage production.

Table 3. FAO soil physical groups and respective rainfall effectiveness factors

Soil group	Rainfall effectiveness factor
Deep well-drained soils; in general no run-off; occasionally some run-on	0.0120
Deep poorly drained soils; run-off is common and during heavy storms may be considerable	0.0137
Shallow well-drained soils; in general only slight run-off occurs	0.0059

2.2.4. Grass mass from clipping

Quadrats of 1 m × 1 m were randomly laid at three locations of each of the eight sampling strata that were selected for grass cover measurements during the same month. Grass in the quadrats was harvested for above-ground DM determination. All residual litter and other non-grass plant materials were removed by hand picking from the harvest. The harvested fresh grass samples were oven dried at 60°C until constant weight. The weight of dry samples (in g-DM m⁻²) was recorded for the resulting 24 samples.

2.3. Herbage mass estimation

Seasonal grass mass was estimated from seasonal grass cover, rainfall and rainfall effectiveness factor. The estimations were done using the following equation by Toxopeus (1999):

(1)

where PSC is the peak standing herbage mass (in gm⁻²), RF is the seasonal rainfall (in mm), RE is the rainfall effectiveness factor and PG is the grass cover (%).

The relationships among the harvested grass above-ground DM and seasonal grass cover, rainfall and rainfall effectiveness were assessed using regression analysis with half of the quadrat clipping dataset. Before the analysis, all data were transformed using natural log function. Coefficients of variation from the regression were used to modify Equation (1)(1) resulting in a new equation (Equation (2)(2)), which was also used for the estimations of herbage mass spatial and seasonal patterns.

(2)

where e ^k is the exponential of regression intercept and α, β and γ are regression coefficients for RF, RE and PG, respectively.

The relationship between harvested and the estimated grass mass (DM) values at locations of quadrat clipping was evaluated using correlation analysis with the half of clipped DM dataset that was not used for regression analysis. The same dataset was used to test whether grass mass DM measurements from the quadrat harvesting method were significantly different from those estimated using a t-test.

The estimation equation variables were integrated using ModelBuilder in ArcGIS 9.3 (ESRI 2008) to come up with spatial and seasonal grass mass distributions. To establish the relationship between the estimated and harvested grass mass, a correlation analysis was done using Genstat discovery version software.

2.3.1. Sensitivity and uncertainty analysis

A stepwise approach to multi-linear regression analysis was used for model sensitivity analysis for the individual explanatory variables. Uncertainty was analysed based on root-mean-square errors (RMSEs) for the different variables using Equation (3)(3).

(3)

where ŷ_i is the observed value for the ith observation, y_i is the predicted value and n is the number of observations.

3. Results

The highest amount of rainfall was during September–November season while the lowest was during June–August (Figure 3, Table 2). In contrast, grass cover was highest during March–May with the lowest cover during September–November (Table 1). The correlation analysis between the estimated and harvested grass mass (DM) using Equation (1)(1) resulted in a very low correlation with an R ² of 0.23.

When transformed data were used, regression analysis of harvested DM with rainfall, grass cover and rainfall effectiveness resulted in an R ² of 0.76 (p < 0.001). The coefficients used in grass mass estimation are shown in Table 4. Results of correlation analysis between estimated and harvested grass mass showed that there is a significant positive relationship (R ² = 0.85) (Figure 4). A t-test showed that there is no significant difference between estimated and measured DM (p = 0.5). The highest grass mass average estimate (1923 kg-DM ha⁻¹) was during March–May season and the lowest was during September–November (895 kg-DM ha⁻¹) (Figure 5, Table 5).

Table 4. Regression coefficients and statistics of rainfall, grass cover and rainfall effectiveness

	Coefficient	Standard error	t Statistic	p-Value
Intercept	5.2608	4.0372	1.3031	0.2081
Rainfall	−0.7884	0.5724	−1.3774	0.1844
Grass cover	1.4354	0.2048	7.0102	<0.001
Effectiveness	0.3938	0.2720	1.4478	0.1640

Table 5. Seasonal grass mass yield

Season	Average grass mass (kg-DM ha^−1)
December–February	1591
March–May	1923
June–August	1194
September–November	895

Figure 4. Correlation between estimated and measured herbage mass.

Figure 5. Seasonal and spatial distribution of herbage mass.

Results from stepwise multi-linear regression analysis showed that grass cover was the most important and significant variable for herbage mass estimation. It registered an R ² value of 0.73 (p < 0.001) with harvested herbage mass. The regression results with rainfall and rainfall effectiveness were insignificant (Table 6). RMSEs from uncertainty analysis indicated that grass cover, rainfall and rainfall effectiveness were reliable predictors of herbage mass. The highest level of certainty was registered by grass cover (Table 6).

Table 6. Model sensitivity (R ² and slope p-value) and uncertainty RMSEs

Variable	R ^2	Slope p-value	RMSE (g-DM m^−2)
Grass cover	0.73	<0.001	0.57
Rainfall	0.09	0.16	0.68
Rainfall affectivity	0.13	0.09	0.64

4. Discussion

Results of this study show that there is a consistent positive relationship between rainfall and grass cover. However, this relationship is associated with a consistent time lag in grass cover response to dry and wet seasons. Consequently, this relationship is translated into grass mass seasonal patterns. The lag can be explained by the time taken before rain effect on growth and the dry season effect on grass cover decline due to desiccation and grazing pressure. Whereas the September–November season received higher amounts of rainfall, it recorded the lowest herbage yields compared with June–August season, which was the driest season. The herbage mass during June–August was as a result of accumulated growth during the preceding March–May rainy season. Similarly, the June–August dry season effect was realized during the September–November season. The recovery effect of the September–November rains was realized during December–February season. Such seasonal patterns exhibited by the results are not only a reflection of the importance of rainfall seasons in determining herbage production (Schönbach et al. 2012) but also the influence of productivity conditions of soil, topographic conditions among other factors (Canton et al. 2003; Diaz-Solis et al. 2003).

The observed inter-seasonal herbage mass transition lag requires management measures that will optimize herbage utilization while minimizing the negative effects from grazing. Different rangeland sites have different rates of response to transition from dry or wet season and hence require different grazing management measures. Areas with longer time lag from dry to wet season are more susceptible to degradation and therefore need to be exposed to less grazing pressure, especially during the dry season, to minimize soil exposure to erosion and loss of soil moisture which may lead to rangeland retrogression (Ao et al. 2008). On the other hand, rangeland sites that take a shorter time to recover from dry seasons are less likely to be affected by grazing pressure. Such sites are usually areas of grazing concentration at the start of wet seasons. When not well managed, such grazing concentration may eliminate some herbage species that are sensitive to grazing pressure (Facelli and Springbett 2009) such as Themeda triandra (McIntyre, Heard, and Martin 2003). However, well managed and moderate grazing distribution at such stage may enhance herbage quality while maintaining high production and herbage cover as was noted by Osborne (2000).

The 76% explanation of herbage mass patterns by the variables considered in this study indicates that the relationship among rainfall and grass cover offers a good estimator of herbage DM. Interestingly, this was comparable to other results that have been reported based on small cartographic scales for East Africa (Jama et al. 2003; Mulindwa et al. 2009). The determination coefficients were higher than what was reported in related studies for rangelands in Asia by Suzuki et al. (2012) and Li Jianlong and Quangong (1998). In Australia, an estimation model by Trnka et al. (2006) explained 78% variability in harvested herbage DM production. Results from stepwise regression analysis with harvested DM yield indicated that grass cover measurements alone could explain 73% of herbage DM yield. Such a relationship shows that grass cover can reasonably be used as a proxy measurement of herbage mass.

The inclusion of rainfall and rainfall effectiveness in the equation improved the estimation of herbage mass, but the improvement was not significant. However, from uncertainty analysis, results indicated that all the variables considered were reliable predictors of herbage mass with grass cover having the lowest RMSE. The significant influence and reliability of grass cover underscores the importance of cover measurements in estimating herbage quantity and its distribution (Mueller-Dombos and Ellenberg 1974; Tueller 1993). With grass cover estimates based on satellite image spectral strata as used in this study, spatial and temporal herbage patterns can easily and accurately be monitored (Herlocker 1999).

The herbage mass seasonal trends exhibited by results of this study are related to the findings by Mulindwa et al. (2009) in which they reported the highest carrying capacity during March–May season in the same area. The low herbage during September–November season due to the effect of June–August dry season poses productivity management challenges to livestock managers and users. While there was sufficient herbage during other seasons, many animals died at the end of the dry season in September 2009. Currently, the dry season response mechanisms to herbage scarcity by livestock owners involve selling of some of their animals to avoid death-related losses. Usually during such dry spells, there is overgrazing with most of rangeland sites left bare which has caused soil erosion, especially at the beginning of rainy seasons. If not well addressed, this cycle of events and processes is likely to continue reducing the productivity of the rangelands, compromise water quality and quantity due to sediment and nutrient loading in water bodies (NEMA 2007). One of the potential remedies is irrigation and increased watering points, but this will require financial investment in water distribution and storage infrastructure. Herbage conservation especially during March–May growing season to be used at the end of the dry season is another possible way of availing feed for animals. Moreover, most of the herbage is lost through trumpling during the growing season (Osborne 2000).

Whereas the estimations were based on the regression coefficients for one season, it is possible that differences might occur with clipping data from other seasons and more so in different years. For instance, December to February, which received relatively higher rainfall and considered a growing season in this study, has been reported to be drier in some years (Bloesch 2002). Therefore, the estimation procedures outlined in the study will have to change accordingly based on seasonal rainfall patterns.

5. Conclusions

The results of this study clearly indicated that rainfall and rainfall effectiveness were insignificant in the estimation of herbage yield. This study demonstrated that grass cover was the most important proxy measurement of spatial and seasonal patterns of rangeland herbage. DM predictions obtained using grass cover were not significantly different from those measured using the classical quadrat clipping method. Hence, herbage mass can in reality be monitored based on grass cover measurements to avoid or at least minimize the cost, destruction and information timeliness implications that are known to be associated with harvesting methods. Since the estimation method is reasonably accurate and eliminates the lengthy weighing and drying procedures, the intensity of data collection for monitoring can be increased tremendously to ensure regular and more precise updates on rangeland productivity conditions. It was, therefore, recommended that the approach described in this study can be adopted and up-scaled for herbage mass estimations in Uganda. Moreover, the method can potentially be easily adopted by ordinary farmers with fewer and cheaper materials through training. However, it should be noted that the estimation procedures could be improved with seasonal-specific prediction equations which should be regularly updated whenever there are drastic shifts in rangeland conditions.