The feasibility of using a low-cost near-infrared, sensitive, consumer-grade digital camera mounted on a commercial UAV to assess Bambara groundnut yield

Figures & data

Figure 1. Study area and experimental design. Field site at the Field Research Centre of Crops for the Future in Semenyih, Malaysia. Experimental layout of plots as shown in aerial photo mosaic RGB in bird’s-eye view taken at a height of 10 meters on 22 June 2018 at flowering stage using the integrated DJI Phantom 4 Pro camera. B1G1R1 means; plot is in block 1, genotype is genotype 1 and replicate is the first replicate.

Table 1. Remote sensing and proximal sensing data acquisition during Bambara groundnut growing season in 2018

Acquisition Date (d/m/y)	UAV flight missions				ASD handheld spectroradiometer
	Growth Stages	DAS	Flight altitude (m)	Total no. of images captured	GSD* (mm)	Height above canopy (m)	Number of spectra	Footprint diameter (m)
5 June 2018	Vegetative	41	10	400	4.0	1.1	36	0.5
22 June 2018	Flowering	58	10	380	4.0	1.1	36	0.5
18 July 2018	Podding	84	10	395	4.1	1.1	36	0.5
31 July 2018	Podfilling	97	10	400	4.0	1.1	36	0.5
8 August 2018	Maturity	105	10	385	4.1	1.1	36	0.5
17 August 2018	Senescence	114	10	390	4.1	1.1	36	0.5

DAS means Days After Sowing; *Ground sampling distance (GSD) was averaged for all images in a flight. Variations in GSD are due to slight fluctuations of UAV altitude in the air.

Figure 2. Image processing workflow and proposed method for estimating Bambara groundnut yield using the low cost UAV imaging system.

Figure 3. Spatial variability of Bambara groundnut yield and the productivity of different plots within one field. The value ranges of yield are presented according to the four quartiles.

Table 2. Descriptive statistics of the UAV based spectral VIs at different crop stages for 2018 Bambara groundnut growing season

Growth Stages	NDVI			EVI2			SR			GNDVI
	Mean ± SD	Range	CV (%)	Mean ± SD	Range	CV (%)	Mean ± SD	Range	CV (%)	Mean ± SD	Range	CV (%)
Vegetative	0.56 ± 0.03	0.50–0.62	5.2	0.24 ± 0.05	0.13–0.35	22.0	2.25 ± 0.10	2.09–2.50	4.2	0.47 ± 0.03	0.43–0.51	5.3
Flowering	0.75 ± 0.03	0.70–0.80	4.2	0.65 ± 0.08	0.53–0.77	11.6	3.86 ± 0.39	3.12–4.44	10.2	0.58 ± 0.03	0.54–0.63	5.3
Podding	0.80 ± 0.04	0.73–0.85	4.8	0.74 ± 0.09	0.59–0.88	12.5	4.07 ± 0.47	3.25–4.71	11.6	0.60 ± 0.03	0.55–0.64	5.4
Podfilling	0.78 ± 0.04	0.72–0.83	5.1	0.73 ± 0.09	0.59–0.85	12.0	4.25 ± 0.56	3.41–5.06	13.2	0.60 ± 0.04	0.54–0.65	6.5
Maturity	0.80 ± 0.04	0.73–0.85	5.0	0.74 ± 0.09	0.60–0.87	12.7	4.25 ± 0.45	3.51–4.80	10.7	0.59 ± 0.03	0.54–0.63	5.0
Senescence	0.76 ± 0.05	0.68–0.82	6.3	0.67 ± 0.11	0.50–0.81	16.9	3.70 ± 0.57	2.92–4.56	15.3	0.56 ± 0.04	0.50–0.62	7.5

SD: standard deviation; CV: coefficient of variation; Kolmogorov–Smirnov test is non-significant for all indices at each growth stages.

Table 3. Correlation coefficients between VIs extracted from remote sensing UAV imagery data (UAV) and proximal sensing BER values derived from data acquired by the spectroradiometer (gt)

	NDVIUAV	EVI2UAV	SRUAV	GNDVIUAV
NDVIgt	0.88**	0.87**	0.78**	0.73**
EVI2gt	0.85**	0.85**	0.77**	0.71**
SRgt	0.77**	0.79**	0.84**	0.75**
GNDVIgt	0.82**	0.83**	0.78**	0.78**

**Correlation is significant at the P < 0.01 level (2-tailed).

Figure 4. Scatterplots indicate the linear relationship between the indices derived from data acquired by Canon S100 and BER indices derived from data acquired by spectroradiometer; (a) NDVI, (b) EVI2, (c) SR, and (d) GNDVI.

Figure 5. Bland–Altman plots comparing VIs calculated from data acquired by UAV based Canon S100 multispectral images and BER indices derived from data acquired by Spectroradiometer; (a) NDVI, (b) EVI2, (c) SR, and (d) GNDVI. Blue dots are individual random samples; black solid lines indicate mean bias and red solid lines indicate the upper and lower limits of agreement.

Table 4. Pearson correlation coefficients between Bambara groundnut dry seed yield and spectral VIs at different growth stages

Growth stage	Spectral VIs
	NDVI	EVI2	SR	GNDVI
Vegetative	0.37*	0.36*	0.38*	0.32
Flowering	0.46**	0.45**	0.48**	0.40*
Podding	0.71**	0.67**	0.72**	0.63**
Podfilling	0.80**	0.78**	0.81**	0.77**
Maturity	0.75**	0.71**	0.75**	0.71**
Senescence	0.49**	0.45**	0.51**	0.43**

**Correlation is significant at the P < 0.01 level (2-tailed); *Correlation is significant at the P < 0.05 level (2-tailed); NIR, R and G represents the NIR, red and green bands, respectively.

Figure 6. Correlation of Bambara groundnut yield with the cumulative VIs from vegetative stage to senescence stage.

Table 5. Quantitative relationships between yield (y) and spectral VIs (x) in Bambara groundnut for 2018 growing season

Spectral vegetation index	Growth stage	Model calibration		Model validation
		Fitting model	R^2	RMSE	MAPE (%)
SR	Podfilling	y = exp(−1.37 + 0.064 × (SR)^2)	0.68	0.24	15.1
NDVI	Podfilling	y = exp(−5.97 + 7.40 × NDVI)	0.66	0.21	14.7
∑SR	Podfilling to maturity	y = exp(−1.56 + 0.0012 × (SR)^2)	0.71	0.20	14.2
∑NDVI	Podfilling to maturity	y = 1/(16.3–8.15 × ln(NDVI))	0.70	0.22	14.5

Figure 7. (a) Relationship between yield and ∑SR from podfilling to maturity stage; (b) Correlation between predicted field-scale yield estimated by regression model developed and observed field yield using independent validation (test) data set.

Table 6. Predicted yield versus observed yield for the growing season

	Predicted dry seed yield (tonnes/ha)	Observed dry seed yield (tonnes/ha)
Min yield	0.53	0.38
Max yield	1.36	1.62
Average yield	0.90	0.87
Standard error	0.05	0.05
Standard deviation	0.27	0.32
Prediction error (%)	13.9

Figure 8. Map of predicted Bambara groundnut yield expressed as dry seed weight in tonnes/ha.

Data availability statement

The data that support the findings of this study are available from the corresponding author, [A.S.], upon reasonable request by e-mail: [email protected].