**5. Discussion**

Five methods for the estimation of chlorophyll content were applied to the collection of over 1100 leaf samples from the Ecuadorian Amazon rainforest, which represents a wide range of vegetation species growing in a disturbed and a pristine lowland rainforest. The first method is an optical method based on transmittance from the SPAD-502 chlorophyll meter index, the second method, also optical, is based on reflectance measurements collected by a spectroradiometer, and the third method is based on radiative transfer approach using the inversion process of the PROSPECT model. The other two methods are based on vegetation indices derived from satellite images.

For the first method, seven models that account for a wide range of vegetation species, phenological stage, and leaf structure showed close estimations between them until 80 SPAD-502 index (**Table 3** and **Figure 5**). At higher indices the differences increase. This can be explained by the fact that the calibration models considered a maximum SPAD-502 range of 80 units, meanwhile our database register readings beyond this range until 95 units. The best accuracy claimed by the instrument reaches its maxima until 50 units; therefore, higher values may be less accurate.

Based on the results of the seven SPAD-502 published calibration models, we compute their average in order to obtain a general model for chlorophyll content estimation which accomplish for a wide range of vegetation species and physiological stage. The resulting general model is a second-order polynomial in a range of 15–95 SPAD index readings. This general model is proposed as ground truth chlorophyll which is assessed by comparing it to a reference published generalized model based on SPAD-502 readings and traditional methods in a laboratory. The first reference model is a homographic model proposed by Cerovic et al. [36] and computed from seven (polynomial, exponential, and homographic) models applied to a variety of plant species. The second model is the generalized homographic model for tropical trees proposed by Coste et al. [15], which was discussed before as Model 2. **Figure 11** illustrates the comparison of the three models.

The proposed second-order polynomial model has the same concave shape and very close chlorophyll estimations along the range 15-95 SPAD-502 readings than the two homographic models. Homographic models have the generalized equation proposed by Cerovic et al. [36] and claims to be probably more accurate and certainly more rapid and portable than wet methods when used in crop plants. The model proposed by Coste et al. [15] was developed for the tropical forest from the Amazon region and has been a reference model for estimating chlorophyll content based on SPAD-502 readings.

model is a second order polynomial in a range of 15 to 95 SPAD index readings. This general model is proposed as ground truth chlorophyll which is assessed by comparing it to a reference published generalized model based on SPAD-502 readings and traditional methods in a laboratory. The first reference model is a homographic model proposed by Cerovic et al. (2012) and computed from seven (polynomial, exponential and homographic) models applied to a variety of plant species. The second model is the generalised homographic model for tropical trees proposed by Coste et al. (2010) which was discussed before as Model 2 in **Table 1**. **Figure**

68 Tropical Forests - The Challenges of Maintaining Ecosystem Services while Managing the Landscape

Five methods for the estimation of chlorophyll content were applied to the collection of over 1100 leaf samples from the Ecuadorian Amazon rainforest, which represents a wide range of vegetation species growing in a disturbed and a pristine lowland rainforest. The first method is an optical method based on transmittance from the SPAD-502 chlorophyll meter index, the second method, also optical, is based on reflectance measurements collected by a spectroradiometer, and the third method is based on radiative transfer approach using the inversion process of the PROSPECT model. The other two methods are based on vegetation indices

For the first method, seven models that account for a wide range of vegetation species, phenological stage, and leaf structure showed close estimations between them until 80 SPAD-502 index (**Table 3** and **Figure 5**). At higher indices the differences increase. This can be explained by the fact that the calibration models considered a maximum SPAD-502 range of 80 units, meanwhile our database register readings beyond this range until 95 units. The best accuracy claimed by the instrument reaches its maxima until 50 units; therefore, higher values

Based on the results of the seven SPAD-502 published calibration models, we compute their average in order to obtain a general model for chlorophyll content estimation which accomplish for a wide range of vegetation species and physiological stage. The resulting general model is a second-order polynomial in a range of 15–95 SPAD index readings. This general model is proposed as ground truth chlorophyll which is assessed by comparing it to a reference published generalized model based on SPAD-502 readings and traditional methods in a laboratory. The first reference model is a homographic model proposed by Cerovic et al. [36] and computed from seven (polynomial, exponential, and homographic) models applied to a variety of plant species. The second model is the generalized homographic model for tropical trees proposed by Coste et al. [15], which was discussed before as Model 2. **Figure 11** illustrates

The proposed second-order polynomial model has the same concave shape and very close chlorophyll estimations along the range 15-95 SPAD-502 readings than the two homographic models. Homographic models have the generalized equation proposed by Cerovic et al. [36] and claims to be probably more accurate and certainly more rapid and portable than wet

**13** illustrates the comparison of the three models.

**5. Discussion**

derived from satellite images.

may be less accurate.

the comparison of the three models.

Indeed, published SPAD-502 models applied to tropical rainforest vegetation are rare. A literature search by the authors only found two models (Model 1 and Model 2) developed for several species of the Amazon forest. Both experiments with tropical trees of the Amazon exhibited higher SPAD-502 readings which are comparable with our dataset. Those models account for a wide range of species, leaf structure, and phenology, and claim good accuracy for chlorophyll content estimation in multispecies forest stands. The homographic model proposed by Coste et al. [15] (Model 2) has been used to estimate chlorophyll content in a study that considered 1084 trees from 758 species across a broad environment gradient of 13 sites (seasonal flooded, clay terra firma, and white-sand forest) at opposite ends of Amazonia in Guiana and Peru [63]. The study relies on chlorophyll estimations based on the SPAD-502 model without considering traditional methods in a laboratory which prove the ability of a rapid and portable method of chlorophyll content in remote areas where analysis in a laboratory is not available.

Based on the comparison to published homographic models for multispecies, it is derived that the second-order polynomial calibration model offers a good approximation of chlorophyll content in tropical forest species. This is because of its close performance compared to the models proposed by Cerovic et al. [36] and Coste et al. [15] (**Figure 10**), and its homographic nature takes into consideration the reduced performance of chlorophyll meters at high chlorophyll contents. Indeed a homographic nature of SPAD-502 model has been applied to a wide range of tropical species from the Amazonia [63].

Estimations from the second method based on five reflectance models illustrate good agreements along all range of SPAD-bins (15–95 units). **Table 4** and **Figure 6** illustrate the results of these methods showing a saturation curve at the higher SPAD bind (80–95).

The observed maximum values of chlorophyll estimation from SPAD-502 (**Table 3**) are considerably higher than maximum values from reflectance indices (**Table 4**), which reflect the exponential increase of SPAD-502 models after 80 SPAD-502 units and the asymptotic nature of reflectance indices after this range. Differences between average estimations are less distinctive.

The first two methods are compared with the third method which is based on the inversion process of the PROSPECT model. **Figure 7** illustrates that the mean values are close to each other until 50–60, and after that the estimations based on SPAD-502 models increase faster than the other two methods. The method based on reflectance models and the PROSPECT model show close mean values until bin 70–80. Analysis of variance (ANOVA) and pairwise comparison between the three methods shown in **Table 5** indicate significant difference between the methods. Results from the lower SPAD-502 bin reported no differences between the methods.


ns, nonsignificant.

\*\*\*Strongly significant (0.1%) or lowest significant (10%).

\*\*Highly significant (1%).

\*Significant (5%).

**Table 5.** ANOVA and pairwise comparison between the three chlorophyll methods for chlorophyll estimation based on the binned SPAD-502 index.

**Table 5** shows ANOVA and pairwise comparison between the three chlorophyll methods for chlorophyll estimation based on the binned SPAD-502 index.

**Figure 10** presented that the chlorophyll estimations at leave level (SPAD-502, reflectance indices, and PROSPECT model) and estimations at regional level (satellite images) applied in this study show strong correlations between them. This finding demonstrates that a combination of field-based methods at leaf level with remote sensing methods at regional level may provide a good opportunity to evaluate forest health caused by land use changes. As it was stated in the introduction, forest degradation and its related changes in ecosystem services have not been fully assessed using remote sensing techniques, especially in high diverse tropical forest. The estimations of MTCI index in Site 1 and Site 2 shown in **Figure 12** have demonstrated lower levels of chlorophyll content caused by land use changes, specifically due the influence of petroleum facilities cause forest degradation. Therefore, in those areas accurate estimations of photosynthetic activity of forested areas are needed to quantify forest degradation and evaluate environmental services provided by flora in the tropical forest.

#### **6. Conclusion**

Three optical methods for estimation of chlorophyll content at leaf level were applied to the collection of over 1100 leaf samples collected in the Ecuadorian Amazon rainforest, which represents a wide range of vegetation species growing in a disturbed and a pristine lowland rainforest. The first method is based on transmittance from the SPAD-502 chlorophyll meter index, the second method is based on reflectance measurements collected by a spectroradiometer, and the third method estimates chlorophyll content from the radiative transfer PROSPECT model. For the first method, seven models that account for a wide range of vegetation species showed similar average leaf chlorophyll contents until 80 units of SPAD-502. An average of the results of these models was computed and used as ground truth from where a generalized second-order polynomial model was created. For the second method, five chlorophyll indices based on reflectance measurements provided similar chlorophyll content estimations for all SPAD range (15–95 units). The third method estimates chlorophyll content based on the inversion process of the PROSPECT model.

Comparison between the three methods shows that estimations until bin 50–60 are relatively similar, and estimations from SPAD increased exponentially. Estimations from reflectance and the PROSPECT model are close to each other until bin 70–80, after that differences increased since the asymptotic behavior of reflectance models estimations. A strong coefficient of correlations between the proposed generalized model and reflectance and PROSPECT approaches result in 0.76 and 0.71, respectively. Comparisons with MTCI and REP indicate correlations of 0.74 and 0.66, respectively.

The results of this study show that the relatively lightweight handheld field spectroradiometer can be used at field level to estimate leaf chlorophyll content in remote tropical rainforest ecosystems that are difficult to access. They provide a rapid and portable method for such remote areas where traditional chemical extraction methods for chlorophyll estimation are not viable. A general second-order polynomial calibration model for chlorophyll content estimation which accounts for a wide range of plant species, phenological stage, and leaf structure based on spectral measures offers an alternative approach for chlorophyll estimation. At a regional level, vegetation indices derived from satellite images are an efficient approach to detect chlorophyll content differences in vegetation exposed to main impacts of land use changes in the Amazon forest. These methods can be applied to regional scale to monitor the effects environmental services provided by the tropical forest and to detect forest degradation caused by land use changes.
